Academic Article Archives

A biomimetic sperm selection device for routine sperm selection

January 16, 2026August 27, 2024 by CadWorks3DAdmin

Academic Article

A biomimetic sperm selection device for routine sperm selection

by Steven A. Vasilescu, Dale M. Goss, Kathryn H. Gurner, Rebecca L. Kelley, Maria Mazi, Fabrice K. De Bond, Jennifer Lorimer, Fabrizzio Horta, Farin Y. Parast, David K. Gardner, Reza Nosrati and Majid E. Warkiani

Abstract

Research question: Can a biomimetic microfluidic sperm sorter isolate motile sperm while minimizing DNA damage in comparison with density gradient centrifugation (DGC)?

Design: This was a two-phase study of 61 men, consisting of a proof-of-concept study with 21 donated semen samples in a university research laboratory, followed by a diagnostic andrology study with 40 consenting patients who presented at a fertility clinic for semen diagnostics. Each sample was split to perform DGC and microfluidic sperm selection (one-step sperm selection with 15 min of incubation) side-by-side. Outcomes evaluated included concentration, progressive motility, and DNA fragmentation index (DFI) of raw semen, and sperm isolated using DGC and the microfluidic device. Results were analysed using Friedman’s test for non-parametric data (significant when P < 0.05). DFI values were assessed by sperm chromatin dispersion assay.

Results: Sperm isolated using DGC and the microfluidic device showed improved DFI values and motility compared with the raw semen sample in both cohorts. However, the microfluidic device was significantly better than DGC at reducing DFI values in both the proof-of-concept study (P = 0.012) and the diagnostic andrology study (P < 0.001). Progressive motility was significantly higher for sperm isolated using the microfluidic device in the proof-of-concept study (P = 0.0061) but not the diagnostic andrology study. Sperm concentration was significantly lower for samples isolated using the microfluidic device compared with DGC for both cohorts (P < 0.001).

Conclusion: Channel-based biomimetic sperm selection can passively select motile sperm with low DNA fragmentation. When compared with DGC, this method isolates fewer sperm but with a higher proportion of progressively motile cells and greater DNA integrity.

Keywords: sperm selection; microfluidics; density gradient centrifugation; DNA fragmentation

We kindly thank the researchers at University of Technology Sydney for this collaboration, and for sharing the results obtained with their system.

Introduction

Infertility affects approximately 15% of couples worldwide, with approximately 55% of those having a male contributing factor (WHO, 2023). The use of medical intervention in the form of assisted reproductive technology (ART) is growing annually, yet the success rate of assisted reproduction cycles per embryo transfer has stagnated at approximately 33% per cycle, and the proportion of live births has plateaued at approximately 26% per cycle over the last two decades (Adamson et al., 2018; Nosrati et al., 2017). Many factors play a role in the success of a cycle, and one crucial aspect of all methods is sperm selection, where sperm quality can have a direct effect on outcomes (Donnelly et al., 1998; Tandara et al., 2014).

An increased DNA fragmentation index (DFI) is prevalent in infertile men and in men aged >40 years, and is even higher in those with abnormalities in conventional semen parameters such as motility, morphology and concentration (Moskovtsev et al., 2009). Furthermore, couples with idiopathic infertility and recurrent pregnancy loss, where conventional sperm parameters lie within healthy norms, show a higher incidence of DNA fragmentation (Esteves et al., 2021; Tan et al., 2019). Studies have shown, using the sperm chromatin structure assay (SCSA), that approximately 28% of men from infertile couples have moderate (>20%) or high (>30%) DFI values (Erenpreiss et al., 2008), while in couples with ‘unexplained infertility’, the percentage of men with moderate-to-high DFI values is 26.1% (Oleszczuk et al., 2013); this incidence increases with age. High levels of DNA fragmentation (>30%) in sperm have been shown to increase the risk factors involved in IVF by reducing embryo quality, lowering implantation rates, and increasing the chance of miscarriage up to 3.9 times that of patients using sperm with low DFI values (Borges Jr et al., 2019; Duran et al., 2002; Robinson et al., 2012; Sedó et al., 2017).

Density gradient centrifugation (DGC) and swim-up methods are the most common techniques used for processing semen samples and selecting sperm for use in assisted reproduction. However, in recent studies, these methods have been implicated in the increase in sperm DNA fragmentation, purportedly due to the generation of reactive oxygen species (Aitken et al., 2014; Muratori et al., 2016; Stevanato et al., 2008). As it is not yet possible to assess sperm DNA fragmentation non-invasively prior to use in IVF or intracytoplasmic sperm injection (ICSI), innovative methods are required to select and isolate sperm with low DFI values for use in IVF or ICSI at appropriate concentrations. There have been several attempts to create an alternative sperm selection method, including the MACS ART Annexin V System (Miltenyi Biotec, Australia), Physiological Intracytoplasmic Sperm Injection dish (CooperSurgical, USA) and the DGC-zeta potential method (Hasanen et al., 2020; Karimi et al., 2020), and many innovative approaches have originated from the application of microfluidics (Keskin et al., 2022; Quinn et al., 2018; Suh et al., 2006; Villeneuve et al., 2023).

Microfluidic devices have been developed to select high-quality sperm using flow or sperm migration behaviour, without the need for active input such as centrifugation or electrophoretic fields. These devices purport to reduce exposure to oxidative stress and subsequent DNA fragmentation (Dadkhah et al., 2023; Nosrati et al., 2014; Simchi et al., 2021). However, the clinical translation and adoption of many of these technologies have been limited (Nosrati, 2022), largely due to their complexity of workflow, operational instability and/or inconsistent results. Without an intuitive user interface, many devices have not seen further side-by-side clinical testing to evaluate their performance in the hands of clinical scientists. An effective sperm selection platform must not only provide high-quality sperm in a timely manner, but must also be simple to use and consistent in its performance over a range of sperm motility levels (Nosrati et al., 2017; Rappa et al., 2016).

This study aimed to test a novel microfluidic channel-based biomimetic sperm selection device against the gold standard, DGC, to isolate motile sperm from a range of semen samples. This radial microfluidic device selects sperm based on their preference to follow boundaries in a static fluid environment mimicking the geometries of the female reproductive tract (Figure 1). This device consists of a radial array of hundreds of media-filled channels, whereby semen interfaces with the entrances to these channels, selecting sperm by their ability to traverse corners, boundaries and troughs, mimicking the endometrium and epithelium of the oviduct, towards a central outlet port. It was postulated, based on the authors’ previous insights into motile sperm behaviour in confined geometries and channels (Eamer et al., 2016; Nosrati et al., 2014; Vasilescu et al., 2023), that the boundary-following tendencies of sperm will lead to the isolation of a sample with high motility and low DFI values.

Figure 1. Overview of the microfluidic device selection process. (A) Schematic representation of the stages of operation of the microfluidic device. (i) Loading the device with media from the centre outlet, (ii) raw semen is then loaded into the outer inlet to create the semen–media interface, (iii) the centre outlet is sealed with tape and the device is left untouched for 15 min, and (iv) the tape is removed, and isolated and washed sperm are aspirated from the centre well ready for assessment and use. (B) A representative image of the microfluidic sperm selection device pre-loading with gamete buffer labelled with the key stages of device operation. Red sperm and round cells indicate non-motile cells and debris, blue sperm represent low motility sperm, and green sperm represent highly motile sperm.

Materials and Methods

Device fabrication

The microfluidic device contained a semen reservoir designed to hold 1 ml of raw semen. The semen reservoir was connected to a sperm trapping and collection area via a radial array of microchannels (Figure 1 and Supplementary Video 1). Microfluidic devices were fabricated via a moulding process using three-dimensional-printed moulds adapted from Shrestha et al. (2019). The moulds were fabricated using a digital light processing three-dimensional printer (MiiCraft Ultra 50; MiiCraft, Taiwan) and a photopolymer resin (BV-007; Creative CADWorks, Canada). The microfluidic device design was created in Solidworks 2019 (SolidWorks Corp., USA) and sliced at 25 µm. The device was made from two separate moulds comprising the top and bottom layers of the device. After printing, each mould was washed with isopropanol, and dried with an air gun to remove any excess liquid resin from the microchannels. Washing with isopropanol was repeated three times, and the moulds were cured in an ultraviolet curing box for 2 min prior to being placed in a 70% ethanol bath for 2 h. The moulds were subsequently dried with an air gun and treated with oxygen plasma (Basic Plasma Cleaner PDC-002; Harrick Plasma, USA) for 2 min, followed by salinization using trichloro (1H, 1H, 2H-perfluoro-octyl) silane (Sigma-Aldrich, USA) in a desiccator under vacuum for 1.5 h. The moulds were cast with polydimethylsiloxane (PDMS) (Sylgard 184; Dow Corning, USA), a non-toxic polymer (Colucci et al., 2018), prepared using a mixture of base and curing agents in a ratio of 1:10. The mixture was degassed to remove all bubbles before casting on to the moulds and cured in a hot air oven (75°C) for 2 h. After curing, the PDMS layers were gently peeled off the moulds; holes were punched for the semen inlet, overflow reservoir and sperm collection outlet; and then oxygen plasma treatment was used for bonding.

Apparatus Used

Clear Microfluidic Resin

Learn More

Ultra 50
3D Printer

Legacy

Semen collection and patient demographics

Human semen samples were obtained through ejaculation after 2–4 days of sexual abstinence as recommended by the World Health Organization (WHO, 2021). Raw semen samples were kept on a shaker at room temperature for 15–20 min to allow for full liquefaction. The proof-of-concept study was approved by Monash University Human Research Ethics Committee (ID 26713, approval date 9 December 2020), and the diagnostic andrology study was approved by Melbourne IVF Human Research Ethics Committee (88-22-MIVF, approved 10 May 2022). Samples were obtained from donors for the proof-of-concept study, and from patients presenting for semen analysis for the diagnostic andrology study. Specific age information is not available for the proof-of-concept study, but these donors were students between the ages of 18 and 24 years. Patients in the diagnostic andrology study were between 26 and 54 years of age (mean 36.3 years) (Table 1). Differences were found between the two populations; specifically, the average semen volume was greater in the diagnostic andrology group (P = 0.012). All participants provided informed consent for their inclusion as completely de-identified participants in this study.

The exclusion criteria for both groups were azoospermia, severe oligozoospermia (<1 million/ml sperm concentration) or absolute asthenozoospermia (0% motile sperm). Additionally, participants providing samples <1 ml volume, and those designated for cryopreservation in the diagnostic andrology study were excluded. Finally, patients who reported active infection or had samples with signs of active infection or inflammation (presence of many round cells) were also excluded.

Experimental procedure

In total, 61 patient and donor samples were processed across two separate studies. Although the two studies were performed at different sites and by different personnel (research embryologists in the proof-of-concept study, and clinical embryologists and andrologists in the diagnostic andrology study), both studies followed the same approach whereby all samples were split into three groups – raw semen, and sperm isolated using DGC and the microfluidic device – and processed in parallel. In cases where the semen volume was <2.1 ml, the unprocessed sample was diluted to 2.1 ml in G-MOPS Plus medium (Vitrolife, Sweden) to enable parallelized processing (nine of 61 samples), as 1 ml was required for both DGC and the microfluidic device, and 100 μl of raw semen was required as the control.

To perform sperm selection with the microfluidic device (see Figure 1Ai–iv), the device was pre-filled with Sydney IVF Gamete Buffer (Cook Medical, USA) at room temperature for the proof-of-concept study, or G-MOPS Plus medium (Vitrolife) for the diagnostic andrology study by injecting 1.5 ml through the central outlet using a filled 3-ml plastic syringe (Becton, Dickinson and Company, USA). These two media are the sperm processing buffers used routinely at each site. A strip of AS-110 acrylic medical grade adhesive tape (AR Care Ltd, UK) was placed over the central outlet to create an airtight seal for channel stability, preventing undesired flow of fluids. The device was then left for 5–10 min on a warm plate (37°C) to equilibrate. Next, 1 ml of liquified semen was injected into the device using a 1-ml plastic syringe, and the device was left undisturbed at 37°C (on a heated stage) for 15 min. After incubation, the tape was removed, and 200 µl of sperm suspended in media was collected from the central outlet using a 200-µl pipette and transferred to a final tube for analysis. The migration of sperm through various stages of the device can be seen in Supplementary Video 1.

In the proof-of-concept study, DGC was performed using Sydney IVF 80/40 gradients (Cook Medical, USA). First, 1 ml of 40% gradient was placed in a conical 15-ml tube and underlaid with 80% gradient. Next, 1 ml of semen was layered carefully on top of the 40% density gradient layer using a 1-ml pipette. The solution was centrifuged at 500 x g for 10 min, and the pellet was aspirated directly from the bottom of the tube by collecting 200 µl of fluid. The pellet was resuspended in 3 ml of Sydney IVF Gamete Buffer, centrifuged for 5 min at 500 x g, the subsequent pellet was aspirated from the bottom by collecting 200 µl of sperm in Sydney IVF Gamete Buffer, followed by transfer to a final tube for analysis.

In the diagnostic andrology study, DGC was performed using PureSperm 80/40 gradient (Nidacon, Sweden). With aseptic techniques, 1 ml of 40% PureSperm density gradient was pipetted into a 15-ml conical tube. Using a new pipette, 1 ml of PureSperm 80% density gradient was underlaid carefully to avoid mixing the two layers. After layering 1 ml of semen sample carefully on to the gradient, without disrupting the density gradient, the gradient tube was centrifuged at 500 x g for 10 min. After centrifugation, the gradient tube was removed carefully from the centrifuge to avoid mixing the layers. The pellet was removed using a clean Pasteur pipette, transferred to a clean 15-ml blue-capped Falcon conical tube containing 8 ml of G-MOPS Plus media, and centrifuged at 500 x g for 5 min. Finally, the pellet was transferred to the final tube with 200 µl of G-MOPS Plus media. The same media were used at both sites for both the microfluidic device and the wash centrifugation step of DGC processing.

Sperm chromatin dispersion assay and motility analysis

DFI was assessed with a modified sperm chromatin dispersion (SCD) test, using the HT-HSG2 kit (Halotech DNA, Spain) as reported previously (Vasilescu et al., 2021). Sperm DNA fragmentation was obtained for the raw semen sample and for sperm isolated using the microfluidic device and DGC. Briefly, 80 μl of sperm suspension was added to 80 μl of pre-aliquoted warmed agarose in a 2-ml Eppendorf tube (Eppendorf, Germany). Thereafter, 10 μl of the semen–agarose mixture was pipetted on to pre-coated slides and covered with a coverslip. The slides were placed on a cold plate at 4°C for 5 min to allow the agarose to set. The coverslips were then slid gently off the slides, and the slides were immediately immersed horizontally in an acid solution and incubated for 7 min with a new coverslip placed on top. Next, slides were gently tilted vertically to allow the acid solution to run off. Slides were then immersed horizontally in the lysing solution for 20 min, washed with distilled water for 5 min, and then dehydrated in increasing concentrations of ethanol (70% and 100%) for 2 min each, air-dried, and stored at room temperature in the dark. Sperm were categorized into one of five groups during counting following SCD: (i) sperm with a halo width equal to or larger than the minor diameter of the core; (ii) sperm with a small halo, similar to or smaller than one-third of the minor diameter of the core; (iii) sperm with a medium halo, between the size of small and large halos from Groups i or ii; (iv) sperm with no halo; and (v) sperm with a degraded halo. Sperm with a small, degraded or absent halo contained fragmented DNA (Groups ii, iv and v). DFI values were recorded as the percentage of sperm cells with fragmented DNA. Three hundred sperm were counted per sample, and each sample was counted twice; counts were considered to be accurate if the difference in DFI value was within 5%. Seven samples were removed from the DFI results due to inadequate staining for reliable counting.

Motility and concentration for each group were assessed using the WHO guidelines (WHO, 2021), Sperm were classified as either progressively motile, non-progressive or immotile, with a minimum of 200 sperm assessed for motility. This method for assessing sperm concentration and motility was performed consistently at both sites to ensure accurate comparisons and minimize confounding variables. For assessing sperm concentration, a 1:10 dilution of raw semen with gamete buffer (Sydney IVF Gamete Buffer for the proof-of-concept study; G-MOPS Plus for the diagnostic andrology study) was required for 10 of the 50 samples. Concentration was assessed under a phase contrast microscope (Olympus CKX53; Evident, Japan) using a haemocytometer at 200 X, and a duplicate count was performed between two scientists, with the counts repeated if the difference between the two counts exceeded 5%. All sperm were counted in the four corner squares and the centre square, and the total number of sperm was multiplied by the dilution factor for each sample where applicable. For motility assessment, samples were assessed under similar conditions after liquefaction of semen within 1 h of collection, and immediately after processing with DGC and the microfluidic device. Sperm were also assessed by two scientists, and the assessment was repeated if the difference between counts exceeded 5%.

Statistical analysis

All statistical analyses were performed using GraphPad Prism 6.0 (GraphPad Software, USA). The normality of distribution was assessed using the Shapiro–Wilk test. The significance of differences between values for demographic data (Table 1) was assessed using the Mann–Whitney U-test. Experimental analysis (Table 2) was assessed using Friedman’s test for non-parametric data to account for repeated measures from the same patient after Dunn’s multiple comparisons test to correct for multiple comparisons. Pearson’s correlation test was performed to assess the linear relationship between DFI values of sperm isolated using the microfluidic device and DGC. Pearson’s correlation coefficient (r-value) was calculated to quantify the strength and direction of the linear association between these two variables. Data are presented as median and interquartile range (IQR) and mean ± SEM. P < 0.05 was considered to indicate significance.

Table 2. Sperm assessments for raw semen, and sperm isolated using density gradient centrifugation and the biomimetic microfluidic device. Data presented as median (interquartile range) and mean ± SEM. Data were assessed using Friedman's test for non-parametric data after Dunn's multiple comparisons test. Bold type indicates a significant result. a (superscript): Compared with DGC. b (superscript): Compared with sperm isolated using the microfluidic device. DGC, density gradient centrifugation.

Results

The radial microfluidic device selects sperm based on their preference to follow boundaries in a confined space within a static fluid environment, as shown in Figure 1Bii and iii. Table 2 shows the sperm quality metric values for DFI, motility and concentration for the three groups processed in parallel (raw semen, DGC and microfluidic device) for both the proof-of-concept study and the diagnostic andrology study.

Proof-of-concept study

The proof-of-concept study showed that sperm isolated using the microfluidic device had significantly lower median DFI values compared with sperm isolated using DGC (0.7% versus 4.1% respectively, P = 0.0012) (Figure 2A). Sperm isolated using the microfluidic device also showed consistently lower DFI values than sperm isolated using DGC in all 21 samples, as the distribution of DFI values compared between split samples shows in Figure 2B. There is a positive correlation with a slope <1 and r = 0.81, showing that increases in DFI values for sperm isolated using DGC were accompanied by increases in DFI values for sperm isolated using the microfluidic device, although the latter increased at a lower rate. Sperm isolated using the microfluidic device yielded a 92.2% decrease in the average DFI value (calculated as the percentage reduction in DFI value compared with the raw semen sample), outperforming DGC which reduced the average DFI value by 57.4% (P < 0.0001). A significantly higher percentage of sperm isolated using the microfluidic device were progressively motile compared with sperm isolated using DGC (92.8% versus 77.7%, respectively; P = 0.0061), with the former showing improvement in 20 of 21 samples (Supplementary Figure 1 and Figure 2C). This represents a two-fold average increase in progressive motility in sperm isolated using the microfluidic device compared with DGC (61.1% for microfluidic device, 30.5% for DGC). Microfluidic selection reduced sperm concentration significantly compared with DGC selection (2.9 × 10⁶ sperm/ml versus 61.0 × 10⁶ sperm/ml, respectively; P < 0.0001) (Figure 2D). The concentration difference between raw semen and sperm isolated using DGC was not significant. Each selection method resulted in a 200-µl suspension of sperm in gamete buffer. Results for before and after microfluidic device processing can be seen in Figure 2E and F, respectively.

Diagnostic andrology study

The diagnostic andrology study repeated trends observed in the proof-of-concept study in a population of patients presenting for infertility. Of the 33 samples processed for DFI values, the sperm isolated using the microfluidic device had significantly lower median DFI values compared with sperm isolated using DGC (1.0% versus 3.9%, respectively; P < 0.001), and a reduction was seen in 30 of 33 samples assessed (Supplementary Figure 2 and Figure 3A,B). Although the DFI values of sperm isolated using the microfluidic device and DGC were significantly reduced compared with the raw semen sample (P < 0.0001 and P = 0.0022, respectively), use of the microfluidic device yielded an 82.9% average improvement (calculated as the percentage reduction in DFI value compared with the raw semen sample), significantly outperforming DGC (P < 0.001) which reduced the DFI value by an average of 44.4%. Furthermore, irrespective of the DFI value of raw semen, DFI values of sperm isolated using the microfluidic device were consistently reduced to <10%; in comparison, the use of DGC resulted in 10 of 33 samples with DFI values >10% (Figure 3B). Although a weak positive correlation was found, with a slope < 1 and r = 0.44, a less pronounced increase in the DFI values of sperm isolated using the microfluidic device was observed compared with sperm isolated using DGC. Unlike the proof-of-concept study, there was no significant difference in progressive motility of sperm isolated using the microfluidic device compared with DGC, and both the microfluidic device and DGC improved progressive motility significantly compared with the raw semen sample (Figure 3C, both P < 0.0001). Furthermore, the sperm concentration reflected similar results as the proof-of-concept study, with samples isolated using the microfluidic device yielding a lower median sperm concentration compared with samples isolated using DGC (2.0 × 10⁶ sperm/ml versus 20.0 × 10⁶ sperm/ml, respectively; P < 0.0001) (Table 2 and Figure 3D).

Site comparison

Results from each site were compared (Supplementary Figures 3A–C), and both progressive motility (P < 0.0001) and concentration (P = 0.0015) of sperm isolated using DGC were significantly lower in the proof-of-concept study compared with the diagnostic andrology study. No other significant differences were observed between the sites.

Discussion

This study demonstrated that passive biomimetic microfluidic channel-based processing of semen, from healthy donors as well as patients attending an IVF clinic for diagnostic andrology, enables the selection of a high proportion of progressively motile sperm with significantly lower DFI values compared with conventional DGC. A biomimetic mode of sperm selection offers consistent results and can be performed with minimal training, with the operation of the microfluidic device requiring only a syringe, a pipette, and a heated stage or incubator to operate the device (Figure 1A,B). During the incubation time, only motile sperm make their way from the semen reservoir down the microchannels via boundary-following behaviour towards the collection chamber, and in doing so, are resuspended in the gamete buffer (Figure 1Aiii). A sharp decrease in height and a gradual reduction in the channel height towards the centre of the device effectively limits the chance for sperm to exit the 200-µL collection zone. Passive sperm selection may also minimize iatrogenic DNA damage by avoiding any centrifugal forces, and selects sperm based on previously reported boundary-following behaviour which correlates with reduced DNA fragmentation (Denissenko et al., 2012; Eamer et al., 2016).

The present research group has tested a similar device against swim-up sperm selection on a smaller cohort of donors, which harnessed MACS ART Annexin V beads (Miltenyi Biotec) and opposing neodymium magnetic plates (AMF Magnetics, Australia) to negatively select apoptotic sperm (Vasilescu et al., 2023). This previous device was more complex in operation, required multiple reagents, was fabricated from a different material (three-dimensional-printed photopolymer resin), and was designed with different internal geometry. The current device uses a simpler, more accessible approach aimed at routine use. To test this microfluidic device, donor samples were used to compare DGC against the microfluidic method of sperm selection, and the latter allowed selection of sperm with higher DNA integrity and progressively motile sperm from this cohort. While conventional semen processing with DGC did result in an average improvement in motility and DFI values compared with unprocessed semen, it did so with a higher level of variability in sperm quality. Specifically, three of 21 samples processed via DGC showed an increase in DFI values, and many samples only showed an incremental reduction (<10%) (Supplementary Figure 1). Conversely, all samples processed with the microfluidic device showed a significant improvement, irrespective of the starting DFI value and motility. The average DFI value of sperm isolated using the microfluidic device was <1%, demonstrating that this method of sperm selection, when applied to motile sperm populations, is effective regardless of the starting DFI value. Similarly, the motility of sperm isolated using the microfluidic device was consistently higher compared with the motility of sperm isolated using DGC (Figure 2C). These results were consistent with previous studies indicating a high level of variance in recovered sperm motility when using DGC (Malvezzi et al., 2014; Muratori et al., 2016).

When performing a study on 40 consenting patients undergoing diagnostic andrology, similar trends were observed. Despite being a more clinically diverse cohort, notable improvements in DFI values were observed in 30 of 33 samples of sperm isolated using the microfluidic device compared with DGC (two samples had identical DFI values for both groups). This consistency highlights the usability and standardization achievable with a biomimetic device. Additionally, although overall improvements in DFI values were observed in sperm isolated using DGC (44.4% average), three samples had increased DFI values, possibly due to the iatrogenic damage caused by centrifugation on particularly susceptible samples, but this would require further investigation (Supplementary Figure 2). This was not observed when using the biomimetic device, which showed an average improvement in DFI values of 82.9%, with only one sample showing a reduction in DFI value <60%. Another noteworthy observation made in both studies was that, although DFI values were reduced in most samples isolated using DGC, the average reductions in DFI values of 57.4% and 44.4% for the proof-of-concept study and diagnostic andrology study, respectively, were inefficient compared with the average reductions observed in the samples isolated using the biomimetic device (92.2% and 82.9%, respectively). Compared with DGC, use of the microfluidic device increases the chance of selecting sperm with high DFI values, and this creates a population of sperm for fertilization which has lower DNA damage and may provide clinical benefit within IVF workflows by reducing the incidence of miscarriage and failed implantation, as suggested in the literature (Borges Jr et al., 2019; Duran et al., 2002; Robinson et al., 2012; Sedó et al., 2017).

The microfluidic device performed consistently between sites for all three key parameters measured (Supplementary Figures 3A–C). Although differences were observed in progressive motility and concentration between the DGC groups, these differences can be attributed to multiple factors, including operator experience between research scientists at the university research laboratory for the proof-of-concept study versus clinical embryologists in the diagnostic andrology study. Although there were differences in the density gradients used at the two sites, both gradients were a 40% and 80% gradient solution combination, and were silane-coated silica-based.

The average DFI values in sperm isolated using DGC vary in the literature, and depend largely on sample populations. Some studies have indicated an increase in total DNA fragmentation (Muratori et al., 2019), and others have suggested an average improvement in DFI values, with a subpopulation of samples experiencing an increase in DFI values or no improvement in DFI values, which is consistent with the results of this study (Rappa et al., 2016; Stevanato et al., 2008). DNA fragmentation in sperm is commonly attributed to oxidative stress, plausibly induced by repetitive centrifugation used in conventional sperm selection methods (Henkel et al., 2005; Hernández-Silva et al., 2021; Nabi et al., 2014). High DNA fragmentation is associated with pregnancy loss in conventional IVF and ICSI, as well as lower implantation rates and a reduction in average embryo quality (Benagiano et al., 2017; Coughlan et al., 2015; Larson-Cook et al., 2003; Robinson et al., 2012). What is perhaps more concerning is that sperm DNA fragmentation has no obvious effect on fertilization, but becomes apparent during blastocyst development by reducing the generation of good-quality blastocysts and ability to achieve successful implantation (Newman et al., 2022; Seli et al., 2004). As a result, the risk of using compromised sperm remains present in clinical practice, and highlights the need for the selection of sperm with high DNA integrity. Importantly, this is of relevance when considering that advanced reproductive age has an increased negative effect on sperm DNA damage (Horta et al., 2020). Furthermore, male ageing has been linked to a significant increase in miscarriage rate, and a decrease in live birth rate, with a larger impact in women of advanced reproductive age (Horta et al., 2019).

A clear limitation of the output of the current microfluidic device, and arguably of microfluidic motility-based sperm selection in general, is the smaller number of sperm isolated when using the microfluidic device when compared with DGC. In conventional IVF, 50,000 sperm are typically required per oocyte, and an average of 10–12 oocytes are harvested per stimulated cycle (Oseguera-López et al., 2019). However, it has been shown that high fertilization and cleavage rates are possible with as few as 2000–4000 sperm per oocyte (Fiorentino et al., 1994). The average number of sperm isolated using the microfluidic device was approximately 720,000, which may be too few for many conventional IVF cases if clinics were to adhere to the requirement of approximately 50,000 motile sperm per oocyte (Li et al., 2018). Logically speaking, sperm selected using passive biomimetic selection, such as the device in this study, may have higher fertilization efficiency than those selected using active measures such as centrifugation; therefore, fewer sperm may be required for conventional IVF, similar to that of in-vivo fertilization whereby only approximately 200 sperm fertilize the oocyte (Chaffey, 2003). Nevertheless, in future prototypes of this biomimetic device, improvement in the yield of motile sperm for high-responder women for whom many oocytes are collected will improve the potential for clinical adoption, as the yield and high selectivity of this device is better suited for ICSI cases which require fewer sperm for insemination, and ICSI is often prescribed for cases where the male partner has a high DFI value. The form factor of the device also enables the adherance of an 18 mm x 36 mm automatic witnessing tag and patient label for seamless clinical integration.

Semen processing using DGC requires several manual interventions during sample handling, each with the potential for human error. Passive devices such as that used in this study, as well as ZyMōt (ZyMōt Fertility, USA) and Lenshooke CA0 (Hamilton Thorne, USA), limit human interaction in semen to sample injection and sperm collection, usually without centrifugation (Shirota et al., 2016). The microfluidic device used in this study, with a simple three-step operation, will reduce the clinical workload while offering a greater reduction in DFI value after processing. While many studies have investigated the impacts of commercialized microfluidic devices, and these have been reviewed systematically (Ferreira Aderaldo et al., 2023), the present biomimetic device takes a different approach to sperm selection by leveraging the boundary-following behaviour of sperm to perform selection, and requiring sperm to travel several millimetres to a collection zone. Reductions in DNA fragmentation and the simplicity of this device are comparable to existing commercial devices such as ZyMōt and LensHooke (Parrella et al., 2019; Hsu et al., 2023), both of which exploit sperm motility via membrane filtration. In a recent study comparing DGC, ZyMōt and LensHooke CA0, Hsu et al. (2023) demonstrated progressively motile sperm counts of 80.6%, 85.6% and 90.8%, respectively, and DFI values of 11.8%, 3.7% and 2.4%, respectively, in normospermic samples (Hsu et al., 2023). Further comparative studies are now required to determine whether the use of a biomimetic device that leverages boundary-following behaviour in sperm will lead to improved outcomes.

This study has several limitations which can be addressed in larger follow-up studies. Firstly, limited access to samples with high DFI values (>25%) prevented a robust testing approach on extreme cases, which are perhaps those which would benefit the most from a reduction in DFI value. Secondly, to prove the clinical usefulness of this approach, clinical outcomes are required when assessing the device on a range of patients, whereby the effect of each sperm selection method on fertilization and embryo development is evaluated thoroughly. Ideally, a randomized controlled trial or sister oocyte study (whereby half the oocytes are inseminated with sperm isolated using conventional methods, and half the oocytes are inseminated with sperm isolated using the microfluidic device) will better display the clinical utility of this microfluidic device in IVF workflows. The prototype in its current format does show utility for ICSI cases, for which lower numbers of high-quality sperm are sufficient. This format suits a side-by-side study for an ICSI cohort, but may not suit an IVF insemination side-by-side comparison with DGC. Thirdly, the method of DNA fragmentation assessment, SCD, only identifies single-stranded DNA breaks, and has limitations in the subjective nature of the assessment. Future studies and validation are needed using SCSA for a more robust assessment of DFI values by detecting double-stranded DNA breaks. SCD is also susceptible to human error and sperm concentration restraints during the preparation and staining of samples, as shown by seven of 40 patients with inadequate staining for SCD assessment in this study. Finally, the current biomimetic prototype does limit the output of sperm by only processing 1 ml of semen, whereas conventional methods process the entire ejaculate. The purpose of this was to enable side-by-side testing against DGC; however, further studies are currently evaluating a larger volume platform capable of processing an entire semen sample to maximize the sperm yield for use in IVF and intrauterine insemination.

This study reports a novel, highly selective biomimetic method for sperm selection in a simple-to-use, single-use chip format for isolating highly motile sperm with minimal DNA fragmentation, without the need for centrifugation or other active mechanisms. Considering the limitations of this study, this proof-of-concept test shows that highly selective, lower output sperm isolation, such as channel-based microfluidic selection in its current form, may prove to be a practical alternative for ICSI cycles if higher motile concentrations for larger oocyte numbers are preferred for conventional IVF. As many patients with high DFI values tend to be prescribed ICSI as a method of fertilization, this device does have the potential to address these cases, considering its ability to reduce DFI values consistently compared with DGC. The novel selectiveness of mimicking the female reproductive system provides a high-quality population of sperm for use in treatments. Clinical studies have now been initiated to validate the proposed benefits of this selection mechanism.

Apparatus Used

Clear Microfluidic Resin

Learn More

Ultra 50
3D Printer

Legacy

Supplementary Material

Download:

References

Adamson, G.D., Zegers-Hochschild, F., Dyer, S., Chambers, G.M., De Mouzon, J., Ishihara, O., Kupka, M., Banker, M., Jwa, S.C., Elgindy, E., Baker, V., 2018. International committee for monitoring assisted reproductive technology: World report on assisted reproductive technology. ICMART. [Google Scholar]
Aitken, R., Finnie, J., Muscio, L., Whiting, S., Connaughton, H., Kuczera, L., Rothkirch, T., De Iuliis, G., 2014. Potential importance of transition metals in the induction of DNA damage by sperm preparation media. Hum. Reprod. 29, 2136–2147. [Google Scholar]
Benagiano, G., Paoli, D., Lombardo, F., Brosens, J.J., Brosens, I.A., 2017. DNA fragmentation and the ultimate success of a pregnancy. Translational andrology and urology (6), S539. [Google Scholar]
Borges, Jr, E., Zanetti, B.F., Setti, A.S., Braga, D.P.D.a.F., Provenza, R.R., Iaconelli, Jr, A., 2019. Sperm DNA fragmentation is correlated
with poor embryo development, lower implantation rate, and higher miscarriage rate in reproductive cycles of non-male factor infertility. Fertil. Steril. 112, 483–490. [Google Scholar]
Chaffey, N., Alberts, B., Johnson, A., Lewis, J., Raff, M., Roberts, K., Walter, P., 2003. 4th edn. Molecular biology of the cell, 91. Annals of Botany. 401-401. [Google Scholar]
Colucci, F., Mckeegan, P., Picton, H., Pensabene, V., 2018. Mouse embryo assay to evaluate polydimethylsiloxane (PDMS) embryo-toxicity. Annu Int Conf IEEE Eng Med Biol Soc 2018, 4484–4487. [Google Scholar]
Coughlan, C., Clarke, H., Cutting, R., Saxton, J., Waite, S., Ledger, W., Li, T., Pacey, A.A., 2015. Sperm DNA fragmentation, recurrent implantation failure and recurrent miscarriage. Asian J. Androl. 17, 681–685. [Google Scholar]
Dadkhah, E., Hajari, M.A., Abdorahimzadeh, S., Shahverdi, A., Esfandiari, F., Ziarati, N., Taghipoor, M., Montazeri, L., 2023. Development of a novel cervix-inspired tortuous microfluidic system for efficient, high-quality sperm selection. Lab. Chip 23, 3080–3091. [Google Scholar]
Denissenko, P., Kantsler, V., Smith, D.J., Kirkman-Brown, J., 2012. Human spermatozoa migration in microchannels reveals boundaryfollowing navigation. Proceedings of the National Academy of Sciences 109, 8007–8010. [Google Scholar]
Donnelly, E.T., Lewis, S.E., Mcnally, J.A., Thompson, W., 1998. In vitro fertilization and pregnancy rates: The influence of sperm motility and morphology on IVF outcome. Fertil. Steril. 70, 305–314. [Google Scholar]
Duran, E., Morshedi, M., Taylor, S., Oehninger, S., 2002. Sperm DNA quality predicts intrauterine insemination outcome: A prospective cohort study. Hum. Reprod. 17, 3122–3128. [Google Scholar]
Eamer, L., Vollmer, M., Nosrati, R., San Gabriel, M.C., Zeidan, K., Zini, A., Sinton, D., 2016. Turning the corner in fertility: High DNA integrity of boundary-following sperm. Lab. Chip 16, 2418–2422. [Google Scholar]
Erenpreiss, J., Elzanaty, S., Giwercman, A., 2008. Sperm DNA damage in men from infertile couples. Asian J. Androl. 10, 786–790. [Google Scholar]
Esteves, S.C., Zini, A., Coward, R.M., Evenson, D.P., Gosalvez, J., Lewis, S.E., Sharma, R., Humaidan, P., 2021. Sperm DNA fragmentation testing: Summary evidence and clinical practice recommendations. Andrologia 53, e13874. [Google Scholar]
Ferreira Aderaldo, J., Da Silva Maranh~ao, K., Ferreira Lanza, D.C., 2023. Does microfluidic sperm selection improve clinical pregnancy and miscarriage outcomes in assisted reproductive treatments? A systematic review and metaanalysis. PLoS One 18, e0292891. [Google Scholar]
Fiorentino, A., Magli, M., Fortini, D., Feliciani, E., Ferraretti, A., Dale, B., Gianaroli, L., 1994. Sperm: Oocyte ratios in an in vitro fertilization (IVF) program. J. Assist. Reprod. Genet. 11, 97–103. [Google Scholar]
Hasanen, E., Elqusi, K., Eltanbouly, S., Hussin, A.E., Alkhadr, H., Zaki, H., Henkel, R., Agarwal, A., 2020. PICSI versus MACS for abnormal sperm DNA fragmentation ICSI cases: A prospective randomized trial. J. Assist. Reprod. Genet. 37, 2605–2613. [Google Scholar]
Henkel, R., Kierspel, E., Stalf, T., Mehnert, C., Menkveld, R., Tinneberg, H.-R., Schill, W.-B., Kruger, T.F., 2005. Effect of reactive oxygen species produced by spermatozoa and leukocytes on sperm functions in nonleukocytospermic patients. Fertil. Steril. 83, 635–642. [Google Scholar]
Hernandez-Silva, G., Lopez-Torres, A.S., Maldonado-Rosas, I., Mata-Martínez, E., Larrea, F., Torres-Flores, V., Trevino, C.L., Chirinos, M., 2021. Effects of semen processing on sperm function: Differences between swim-up and density gradient centrifugation. The World Journal of Men’s Health 39, 740. [Google Scholar]
Horta, F., Catt, S., Ramachandran, P., Vollenhoven, B., Temple-Smith, P., 2020. Female ageing affects the DNA repair capacity of oocytes in IVF using a controlled model of sperm DNA damage in mice. Hum. Reprod. 35, 529–544. [Google Scholar]
Horta, F., Vollenhoven, B., Healey, M., Busija, L., Catt, S., Temple-Smith, P., 2019. Male ageing is negatively associated with the chance of live birth in IVF/ICSI cycles for idiopathic infertility. Hum. Reprod. 34, 2523–2532. [Google Scholar]
Hsu, C.-T., Lee, C.-I., Lin, F.-S., Wang, F.-Z., Chang, H.-C., Wang, T.-E., Huang, C.-C., Tsao, H.-M., Lee, M.-S., Agarwal, A., 2023. Live motile sperm sorting device for enhanced sperm fertilization competency: Comparative analysis with density-gradient centrifugation and microfluidic sperm sorting. J. Assist. Reprod. Genet. 40, 1855–1864. [Google Scholar]
Karimi, N., Kouchesfahani, H.M., Nasr-Esfahani, M.H., Tavalaee, M., Shahverdi, A., Choobineh, H., 2020. DGC/zeta as a new strategy to improve clinical outcome in male factor infertility patients following intracytoplasmic sperm injection: A randomized, single-blind, clinical trial. Cell Journal (Yakhteh) 22, 55. [Google Scholar]
Keskin, M., Pabuccu, E.G., Arslanca, T., ¸ Demirkıran, O.D., Pabuccu, R., 2022. Does microfluidic sperm sorting affect embryo euploidy rates in couples with high sperm DNA fragmentation? Reprod. Sci. 29, 1801–1808. [Google Scholar]
Larson-Cook, K.L., Brannian, J.D., Hansen, K.A., Kasperson, K.M., Aamold, E.T., Evenson, D.P., 2003. Relationship between the outcomes of assisted reproductive techniques and sperm DNA fragmentation as measured by the sperm chromatin structure assay. Fertil. Steril. 80, 895–902. [Google Scholar]
Li, D., Wang, T., Wang, X., 2018. Characterization of sperm proteome and reproductive outcomes after reduced male abstinence in ivf treatment. Fertil. Steril. 110, e296. [Google Scholar]
Malvezzi, H., Sharma, R., Agarwal, A., Abuzenadah, A.M., Abu-Elmagd, M., 2014. Sperm quality after density gradient centrifugation with three commercially available media: A controlled trial. Reprod. Biol. Endocrinol. 12, 1–7. [Google Scholar]
Moskovtsev, S.I., Willis, J., White, J., Mullen, J.B.M., 2009. Sperm DNA damage: Correlation to severity of semen abnormalities. Urology 74, 789–793. [Google Scholar]
Muratori, M., Marchiani, S., Tamburrino, L., Baldi, E., 2019. Sperm DNA fragmentation: Mechanisms of origin. Adv Exp Med Biol 1166, 75–85. [Google Scholar]
Muratori, M., Tarozzi, N., Cambi, M., Boni, L., Iorio, A.L., Passaro, C., Luppino, B., Nadalini, M., Marchiani, S., Tamburrino, L., 2016. Variation of DNA fragmentation levels during density gradient sperm selection for assisted reproduction techniques: A possible new male predictive parameter of pregnancy? Medicine (Baltimore) 95, e3624. [Google Scholar]
Nabi, A., Khalili, M., Halvaei, I., Roodbari, F., 2014. Prolonged incubation of processed human spermatozoa will increase DNA fragmentation. Andrologia 46, 374–379. [Google Scholar]
Newman, H., Catt, S., Vining, B., Vollenhoven, B., Horta, F., 2022. DNA repair and response to sperm DNA damage in oocytes and embryos, and the potential consequences in art: A systematic review. Molecular Human Reproduction 28, gaab071. [Google Scholar]
Nosrati, R., 2022. Lab on a chip devices for fertility: From proof-of-concept to clinical impact. Lab. Chip 22, 1680–1689. [Google Scholar]
Nosrati, R., Graham, P.J., Zhang, B., Riordon, J., Lagunov, A., Hannam, T.G., Escobedo, C., Jarvi, K., Sinton, D., 2017. Microfluidics for sperm analysis and selection. Nat Rev Urol 14, 707–730. [Google Scholar]
Nosrati, R., Vollmer, M., Eamer, L., San Gabriel, M.C., Zeidan, K., Zini, A., Sinton, D., 2014. Rapid selection of sperm with high DNA integrity. Lab. Chip 14, 1142–1150. [Google Scholar]
Oleszczuk, K., Augustinsson, L., Bayat, N., Giwercman, A., Bungum, M., 2013. Prevalence of high DNA fragmentation index in male partners of unexplained infertile couples. Andrology 1, 357–360. [Google Scholar]
Oseguera-Lopez, I., Ruiz-Díaz, S., Ramos-Ibeas, P., Perez-Cerezales, S., 2019. Novel techniques of sperm selection for improving IVF and ICSI outcomes. Frontiers in cell and developmental biology 7, 298. [Google Scholar]
Parrella, A., Keating, D., Cheung, S., Xie, P., Stewart, J.D., Rosenwaks, Z., Palermo, G.D., 2019. A treatment approach for couples with disrupted sperm DNA integrity and recurrent ART failure. J. Assist. Reprod. Genet. 36, 2057–2066. [Google Scholar]
Quinn, M.M., Jalalian, L., Ribeiro, S., Ona, K., Demirci, U., Cedars, M.I., Rosen, M.P., 2018. Microfluidic sorting selects sperm for clinical use with reduced DNA damage compared to density gradient centrifugation with swim-up in split semen samples. Hum. Reprod. 33, 1388–1393. [Google Scholar]
Rappa, K.L., Rodriguez, H.F., Hakkarainen, G.C., Anchan, R.M., Mutter, G.L., Asghar, W., 2016. Sperm processing for advanced reproductive technologies: Where are we today? Biotechnol. Adv. 34, 578–587. [Google Scholar]
Robinson, L., Gallos, I.D., Conner, S.J., Rajkhowa, M., Miller, D., Lewis, S., Kirkman-Brown, J., Coomarasamy, A., 2012. The effect of sperm DNA fragmentation on miscarriage rates: A systematic review and metaanalysis. Hum. Reprod. 27, 2908–2917. [Google Scholar]
Sedo, C.A., Bilinski, M., Lorenzi, D., Uriondo, H., Noblía, F., Longobucco, V., Lagar, E.V., Nodar, F., 2017. Effect of sperm DNA fragmentation on embryo development: Clinical and biological aspects. JBRA assisted reproduction 21, 343. [Google Scholar]
Seli, E., Gardner, D.K., Schoolcraft, W.B., Moffatt, O., Sakkas, D., 2004. Extent of nuclear DNA damage in ejaculated spermatozoa impacts on blastocyst development after in vitro fertilization. Fertil. Steril. 82, 378–383. [Google Scholar]
Shirota, K., Yotsumoto, F., Itoh, H., Obama, H., Hidaka, N., Nakajima, K., Miyamoto, S., 2016. Separation efficiency of a microfluidic sperm sorter to minimize sperm DNA damage. Fertil. Steril. 105, 315–321.e1. [Google Scholar]
Shrestha, J., Ghadiri, M., Shanmugavel, M., Bazaz, S.R., Vasilescu, S., Ding, L., Warkiani, M.E., 2019. A rapidly prototyped lung-on-a-chip model using 3d-printed molds. Organs-on-a-Chip 1, 100001. [Google Scholar]
Simchi, M., Riordon, J., You, J.B., Wang, Y., Xiao, S., Lagunov, A., Hannam, T., Jarvi, K., Nosrati, R., Sinton, D., 2021. Selection of high-quality sperm with thousands of parallel channels. Lab. Chip 21, 2464–2475. [Google Scholar]
Stevanato, J., Bertolla, R.P., Barradas, V., Spaine, D.M., Cedenho, A.P., Ortiz, V., 2008. Semen processing by density gradient centrifugation does not improve sperm apoptotic deoxyribonucleic acid fragmentation rates. Fertil. Steril. 90, 889–890. [Google Scholar]
Suh, R.S., Zhu, X., Phadke, N., Ohl, D.A., Takayama, S., Smith, G.D., 2006. IVF within microfluidic channels requires lower total numbers and lower concentrations of sperm. Hum. Reprod. 21, 477–483. [Google Scholar]
Tan, J., Taskin, O., Albert, A., Bedaiwy, M.A., 2019. Association between sperm DNA fragmentation and idiopathic recurrent pregnancy loss: A systematic review and meta-analysis. Reprod. Biomed. Online 38, 951–960. [Google Scholar]
Tandara, M., Bajic, A., Tandara, L., Bili c-Zulle, L., Sunj, M., Kozina, V., Goluza, T., Jukic, M., 2014. Sperm DNA integrity testing: Big halo is a good predictor of embryo quality and pregnancy after conventional IVF. Andrology 2, 678–686. [Google Scholar]
Vasilescu, S.A., Ding, L., Parast, F.Y., Nosrati, R., Warkiani, M.E., 2023. Sperm quality metrics were improved by a biomimetic microfluidic selection platform compared to swim-up methods. Microsystems & Nanoengineering 9, 37. [Google Scholar]
Vasilescu, S.A., Khorsandi, S., Ding, L., Bazaz, S.R., Nosrati, R., Gook, D., Warkiani, M.E., 2021. A microfluidic approach to rapid sperm recovery from heterogeneous cell suspensions. Sci Rep 11, 7917. [Google Scholar]
Villeneuve, P., Saez, F., Hug, E., Chorfa, A., Guiton, R., Schubert, B., Force, A., Drevet, J.R., 2023. Spermatozoa isolation with FelixTM outperforms conventional density gradient centrifugation preparation in selecting cells with low DNA damage. Andrology 11, 1593–1604. [Google Scholar]
WHO, 2021. Who laboratory manual for the examination and processing of human semen, 6th ed. World Health Organization, Geneva. [Google Scholar]
WHO, 2023. Infertility prevalence estimates: 1990-2021. World Health Organization, Geneva. [Google Scholar]

Offsetting Dense Particle Sedimentation in Microfluidic Systems

January 16, 2026August 21, 2024 by CadWorks3DAdmin

Academic Article

Offsetting Dense Particle Sedimentation in Microfluidic Systems

by Tochukwu Dubem Anyaduba and Jesus Rodriguez-Manzano

Abstract: Sedimentation is an undesirable phenomenon that complicates the design of microsystems that exploit dense microparticles as delivery tools, especially in biotechnological applications. It often informs the integration of continuous mixing modules, consequently impacting the system footprint, cost, and complexity. The impact of sedimentation is significantly worse in systems designed with the intent of particle metering or binary encapsulation in droplets. Circumventing this problem involves the unsatisfactory adoption of gel microparticles as an alternative. This paper presents two solutions—a hydrodynamic solution that changes the particle sedimentation trajectory relative to a flow-rate dependent resultant force, and induced hindered settling (i-HS), which exploits Richardson–Zaki (RZ) corrections of Stokes’ law. The hydrodynamic solution was validated using a multi-well fluidic multiplexing and particle metering manifold. Computational image analysis of multiplex metering efficiency using this method showed an average reduction in well-to-well variation in particle concentration from 45% (Q = 1 mL/min, n = 32 total wells) to 17% (Q = 10 mL/min, n = 48 total wells). By exploiting a physical property (cloud point) of surfactants in the bead suspension in vials, the i-HS achieved a 58% reduction in the sedimentation rate. This effect results from the surfactant phase change, which increases the turbidity (transient increase in particle concentration), thereby exploiting the RZ theories. Both methods can be used independently or synergistically to eliminate bead settling in microsystems or to minimize particle sedimentation.

Keywords: microfluidics; beads, sedimentation; cloud point; droplet microfluidics; phase change; surfactant; hindered settling; Richardson-Zaki; Stokes law; fluid dynamics; fluid splitting; fluid metering; dense particles

We kindly thank the researchers at the Imperial College London for this collaboration, and for sharing the results obtained with their system.

1. Introduction

Interest in using microparticles as delivery systems in various technologies has been widely researched, especially in combination with microdroplets for biological applications [1,2,3,4,5,6]. This is partly due to the high surface-to-volume ratio and the ease of immobilizing bio-recognition molecules on these materials, as well as the potential for compartmentalized single-molecule assays [7,8]. Unfortunately, challenges with bead settling confound these applications [3,6,9]. Offsetting particle density poses a challenge when loading microparticles into encapsulation devices because the higher-density particles sediment in the fluidic channels, causing a non-homogeneous distribution of microparticles in droplets. One method of resolving this challenge involves suspending the particles in equally dense fluids or introducing humectants such as glycerol [3,10]. However, an adequate amount of the humectants for increased bead buoyancy may be required at concentrations that may be inhibitory to the intended bio-applications, such as nucleic acid amplification technologies [11]. Researchers have also circumvented sedimentation problems by using gel beads [12,13,14,15]. While these have been used successfully in ensuring the binary distribution of beads in droplets without sedimentation issues, their non-Newtonian rheological properties make them difficult to handle. The use of channels with aspect ratios close to the particle diameter is another method for maintaining a single streamline, ensuring that only one particle is queried by the continuous phase at the point of encapsulation. However, considering that these beads are hard-shelled, their packing density may prohibit the possibility of closed packing in narrow channels. Additionally, as sedimentation velocity depends on the mass and size of the particles, the use of smaller particles is also an option; however, this may impact the capacity to carry an adequate amount of biomolecules of interest. Price and Paegel [3] presented a potentially simple solution by exploiting the sedimentation potential of the beads using a hopper system. However, they found that it took 0.8 h (17 µm TetanGel resin beads) and 3.8 h (2.8 µm magnetic beads) to introduce the beads before single-bead encapsulation. Kim et al. [2] successfully developed a pneumatic system that was capable of trapping and releasing beads, thus creating a deterministic encapsulation of a defined number of beads per droplet. This system, however, involves complex implementations of pumps and valves, thus making it unfit for low-cost and low-complexity applications. Mechanical agitation has also been successfully adopted; however, this complicates the system and could make integration into a unified product difficult. Applications requiring equal spatial distribution of particles are also impacted by sedimentation, which is compounded by non-slip conditions in laminar flow between parallel plates. For particles in such systems, wall lift and drag forces have been shown to depend on shear rate, especially at very low Reynold’s numbers [16]. In this paper, simplistic solutions to sedimentation, which can be applied to most particle-based systems, are exemplified in two different forms. A flow-rate-dependent method that alters the sedimentation trajectory of suspended particles was applied to a microfluidic particle metering system while induced hindered settling was applied to particles in suspension.

Apparatus Used

ProFluidics 285D

Learn More

2. Materials and Methods

2.1. Design and Fabrication of Fluid-Metering Chip

The chip was designed as a 16-well manifold for fluid and particle metering devices in which the metering chambers are separated from a lower storage chamber by a capillary valve (Figure 1A). Having both chambers was necessary to prevent one of the consequences of manifold systems, which is sequential filling. This would entail that each well will be filled to the brim before the next, thereby leaving no headspace to allow for further fluid manipulations such as mixing. The lower storage chambers were perforated at positions modeled to provide a convex meniscus at the approximate intended fill volume. To achieve this on a 3D plane, a 60° hemisphere mimicking the hydrophilic contact angle between the chip surface and the buffer was used to cut an extrusion of the 3D-model infill until the desired fill volume was achieved on the model (Figure 1B). During assembly, the perforations were plugged with polytetrafluoroethylene (PTFE) membranes such that wicking of the metered volume into the membranes triggered an increase in the chamber pressure, thereby preventing further emptying of the top metering chamber (Figure 1C). All 3D models were designed using SolidWorks (Dassault Systèmes) and printed using ProFluidics 285D digital light processing (DLP) 3D printer (CADworks3D). The design files are included as Supplementary File (SF1).

Figure 1. Minimal illustrations of the chip sub-units showing critical design elements. (A) Metering and storage chambers separated by a capillary valve (B) Illustration showing method used to determine the position of the venting holes in the bottom (storage) chamber (C) Once fluid in the storage chamber wicks through the venting holes to a PTFE filter, the pressure within the chamber increases above the Laplace pressure. this prevents further filling of the storage chamber.

As illustrated in Figure 2, a consequence of the increased number of metering wells, n, is an increase in length of the flow path, l (Figure 2i), which consequently leads to a particle concentration gradient, in which the first well (Figure 2ii) contains a significantly greater concentration of suspended particles than the nth well due to bead settling. This challenge necessitates the need for a mechanism of counteracting or reducing the sedimentation of the beads to improve metering efficiency (Figure 2ii,iii).

2.2. Hydrodynamic Interruption of Sedimentation

The interaction of particles of different sizes and shapes have been the subject of plenty of research, especially in environmental studies. However, there is a dearth of empirical demonstration of hydrodynamic interruption of particle sedimentation, especially in microfluidic systems. Here, the fabricated fluidic and particle-metering chips were used to demonstrate this phenomenon. The fabricated chips were connected to a syringe pump (Legato, KD Scientific, USA) via a modified Eppendorf tube, which held the bead suspension. For each experimental run, the syringe pump was programmed to run at a defined, arbitrarily chosen volumetric rate, Q = 1, 3.5, 5, and 10 mL/min until the feed channel was completely emptied. Each experimental condition was replicated (for Q = 1 mL/min, n = 2; for Q = 3.5, 5 and 10 mL/min, n = 3) to determine the reproducibility of the results. Before each run, the bead vial was vortexed to ensure uniform bead distribution, then connected to the metering device and pumped within 10 s, thereby preventing pre-settling of the beads. Visual data were collected using M50 Mark II mirrorless camera (Canon, USA) at 60 frames/s for computational image analysis via the algorithm described in the data acquisition and analysis section.

2.3. Induced Hindered Settling (i-HS)

The Richardson–Zaki modification of Stokes’ law [17,18] (Equation (1)) shows that at higher particle concentrations, the particle settling velocity decreases due to particle–particle interactions. This principle suggests that if we could increase the particle concentration in the buffer, we could delay or overcome particle settling. For biomedical applications, however, the particle concentrations are predetermined empirically for optimal performance and cost reasons and, as such, cannot be indefinitely increased to exploit the Richardson–Zaki principle. This principle could be exploited only if there was a way to transiently increase the particle concentration (or increase the turbidity of the particle-suspending solution) without introducing additional particles. This is the basis of the i-HS alternative.

where $V_{t}$ = settling velocity in turbid fluid,
$V_{t}$ = settling velocity in clear fluid,
n = exponent of reduction in settling velocity,
⌀ = volume fraction of suspended particles in the fluid (turbidity)

To ascertain the feasibility of the i-HS, the cloud point of the surfactant in the bead suspension buffer was exploited. Cloud point refers to a physical property (temperature) of non-ionic surfactants at which their dissolution in liquid reverses, leading to liquid–liquid phase separation. At this temperature, as illustrated in Figure 3, the surfactants form micelles, thereby increasing the turbidity (cloudiness) of the suspending solution. These micelles transiently act as particles, thereby allowing the empirical exploitation of the Richard–Zaki theory. As a preliminary analysis of this concept, the buffer was modified with and without ECOSURF EH-9 and designated with the prefixes -WS and -NS, respectively, and analysed via a spectrophotometric thermal gradient (STG) using BioTek Epoch 2 microplate spectrophotometer (Agilent, USA) in a 96-well plate. The choice of the non-ionic surfactant, ECOSURF EH-9, was informed by its biodegradability, among other functions such as wetting. Hypothetically, an increase in the turbidity of the -WS buffer relative to the turbidity of the -NS buffer would signal the feasibility of the concept.

Subsequently, bead suspensions of the same concentration as in previous experiments were added to clear glass vials and grouped according to the experimental protocol This setup enabled the investigation of the effect of the surfactant cloud point on bead settling. Test temperatures that can be tolerated by downstream processes (50 ± 10 °C), and at which cloud point was induced were chosen from different temperature points on the STG data for further sedimentation analyses. Raw STG data are included as Supplemetary Data, SD1.

2.4. Data Acquisition and Analysis

Replicate (n = 12) turbidimetric measurements at different temperatures and data collection were achieved using a spectrophotometer at a near-infrared (NIR) wavelength of 850 nm. For bead settling experiments, heating of the glass vials for bead settling was performed using a Benchmark Multitherm shaker and cooling device. Visual data in the form of video recordings were collected for the assessment of bead settling and suspension homogeneity in the vials using M50 Mark II Mirrorless camera (Canon, USA) at 60 frames per second). Each test was run in duplicates.

The collected videos were analysed using Python 3.6 computer vision library (OpenCV) following the algorithm shown in Figure 4. To determine the bead settling velocity in the vials, the intensity of the pixels within the region of interest (region covered by the bead suspension) was monitored throughout the video timeline. By monitoring the pixel intensity over time, the presence and later absence of beads within a pixel signals the settling of the beads. The sedimentation rate was determined as the rate of change of gray value (Equation (2)).

where ∂𝐺/∂𝑡 = rate of change of gray values over time, $V_{t}$ = sedimentation rate

The pixel intensity (mean gray value) refers to the brightness of a pixel in an image. From the grayscale images, the intensity was measured on a scale of 0–255, where 0 represents black and 255 represents white. The mean gray values correspond to the concentration of the beads or the homogeneity of the bead suspension.

3. Results

3.1. Fabrication of Particle Metering Chip

To prevent premature emptying of the top metering wells, the dimensions of the feed channel were modeled using 2023 SolidWorks Flow Simulations software (Dassault Systèmes). The capillary stop valves were designed to have a Laplace pressure of 403 Pa, calculated using Equation (3).

where $θ_{c}$ = fluid contact angle = 2, w = valve width = 0.5 mm, h = valve height = 1.25 mm, $γ = γ_{water}$ = surface tension = 0.072 N/m.

After printing, the geometric conformance of the 3D-printed capillary valves to the CAD model was determined using optical metrology (Keyence). This was determined to be 0.49 ± 0.02 mm (n = 10), thus conforming with the design parameters (Equation (3)).

A major consequence of manifold splitting results from the interaction between the buffer, the shared walls between wells, the material surface properties, and the flow regime (Figure 5iii). This interaction led to an undesired siphon effect, whereby asynchronously emptied wells drew fluid from neighboring wells, leading to metering inefficiencies. This was resolved by reprogramming the flow regime to ensure synchronous emptying of the wells via an instantaneous increase in Laplace pressure (Figure 5iv).

Figure 5. (i–iii) shows the effect of asynchronous emptying of the metering wells. (iv) shows the pump ramp program: (2) Ramp from 0.1 mL/min to 1 mL/min in 5 s. (3) Hold at 1 mL/min for 72 s (72 s is the time required to completely empty the surrogate elution chamber per my setup + time to empty the feed well) (4) Ramp from 1 mL/min to 10 mL/min in 5 s (5 s–arbitrarily chosen—being careful to ensure the syringe pump could handle that ramp) (5) Hold at 10 mL/min for 6 s (10 mL/min for 6 s ensures 690 µL is pushed from the all metering wells while keeping the unit pressurized). The positive pressure is maintained momentarily to ensure all wells are emptied simultaneously.

3.2. Hydrodynamic Interruption of Sedimentation

As shown in Figure 6A,B, at a volumetric flow rate of Q = 1 mL/min, much of the beads eluted early in the experimental run, forcing the mean gray value (MGV) of wells 1 through 7 to be significantly lower (higher bead concentration) than that of wells 8–16 (Figure 6D). Pairwise comparisons (Student’s t test, alpha = 0.05) of the MGVs of the 16 wells from Q = 1 mL/min to the other tested Qs revealed a significant difference between them, with p values < 0.0001 (Figure 6C). While increasing Q from 3.5 mL/min to 10 mL/min did not significantly affect the average concentration of the beads (p values: Q3.5 − Q10 = 0.107; Q5 − Q10 = 0.231; Q3.5 − Q5 = 0.675), analyses of the data from each flow rate showed improved bead distribution (Figure 6D). For Q = 3.5 mL, wells 14–16 had significantly lower bead concentrations; for Q = 5 mL/min, wells 15 and 16 had significantly lower bead concentrations; and for Q = 10 mL/min, only well 16 had a significantly lower bead concentration. Video recording of the analyses can be found in the Supplementary Folder, SD2.

3.3. Sedimentation Offset via Induced Hindered Settling

Phase separation in the particle suspension buffer was induced and confirmed via spectrophotometric analyses, which showed a temperature-driven increase in turbidity (Figure 7A). As this change is reversible [19], it offers a perfect solution for increasing the probability of particle–particle interactions and, consequently, hindering settling. Incubation of the WS buffer at 55 °C resulted in a ~45% increase in turbidity (Figure 7B).

Figure 7. (A). Spectrophotometric thermal gradient (STG) analysis of buffers with (-WS) and without (-NS) surfactants. (B). Pairwise comparison of STG results. (C). Timelapse images showing bead sedimentation in glass vials. (D). XY plots of different experimental conditions showing the temporal rate of change in pixel intensity. (E). Analysis of means (ANOM) of the sedimentation gradients (𝛼 = 0.05). Video data of the experimental setup is included in Supplementary Folder, SD3.

As shown in Figure 7C,D, this phenomenon translated to a reduction in bead settling velocity. In accordance with the theory of hindered settling, an increase in the bead suspension buffer temperature and, consequently, the cloud point of the surfactant reduced the sedimentation rate of the beads by 58% (Figure 7D,E). Continuous or controlled heating of the beads in a vial showed a similar result of improved particle metering. Notably, the presence of surfactants, while necessary for fluid flow, increases the bead settling velocity, as shown in Figure 7C–E—WS_RT (slope = 0.4341) vs. NS_RT (slope = 0.2744). As shown in Figure 7E, there was no significant difference in sedimentation rate between -NS buffer at room temperature and one heated to 46 °C (NS_46 °C vs. NS_RT); however, increasing the temperature of the -NS group to 55 °C led to an increase in the rate of sedimentation (NS_46—slope: 0.267, NS_55—slope: 0.296).

Apparatus Used

ProFluidics 285D

Learn More

4. Discussion

4.1. Hydrodynamic Interruption of Particle Sedimentation

Particle sedimentation occurs because of the action of gravitational pull on the particles. The rate at which particles sediment, $V_{s},$ is influenced by factors such as particle volume and the density of the suspending fluid.

$V_{s},$ = is Sedimentation velocity at infinite dilution, r = particle radius, $η_{0}$ = viscosity of suspending fluid, $m_{p}$ = mass of particle, $v_{p}$ = volume of particle, $ρ_{f}$ = density of suspending fluid, g = |g| = 9.8 ${m / s}^{2}$ .

These, in turn, also influence the drag force, ( $F_{d}$ ), which in simple terms, the drag force refers to the resistance force exerted by a fluid to the downward motion of the particles.

where $F_{d}$ = drag force, V = particle velocity, A = particle area, $C_{d}$ = drag coefficient = $\frac{24}{Re}$ for spheres [20], therefore:

Substituting $g \frac{2 r^{2}}{9 η_{0}} (\frac{m_{p}}{v_{p}} - ρ_{f})$

As shown in Equation (7), several factors contribute to the drag force exerted on the particles. Consider particles at a steady state, as illustrated in Figure 8, the particles sediment in response to the resultant of opposing forces, drag force, and gravity.

However, in a fluidic state, the rate and direction of sedimentation is largely determined by the magnitude of the force, $F_{p}$ applied to the particle due to pumping.

As the deviation of the particle from a straight downward sedimentation is given by:

Theoretically, at constant |g’|, increasing $F_{p}$ would result in a concomitant increase in $θ$ towards 90°.

Considering this theory, as $F_{p}$ = ma, where m=mass; and a = acceleration = $\frac{Fluid Velocity}{time}$ ,
$F_{p}$ can be increased via the volumetric flow rate.

Empirical findings from the data above validate the working theory as stated in Equations (9)–(11); thus, as the acceleration of the beads in solution increase as a result of the increase in Q of the suspending fluid, gravitational pull on the particles is counteracted. The particles therefore remain homogenized in solution long enough to be evenly distributed. This method, although simplistic and easy to implement in most systems, may be impractical for shear-averse processes where high flow rates (deviation from laminar flow) could impact the structural integrity of the suspended particles. Additionally, as shown in Figure 6C, increasing Q beyond a certain threshold yields very minimal improvement, which may not justify the high volumetric flow rate requirement. These limitations, therefore, necessitate the need for a more shear-tolerant alternative that would not only alter the sedimentation trajectory, but also significantly reduce it.

4.2. Induced Hindered Settling

Although ECOSURF EH-9 has a cloud point of 64 °C at 10 wt% actives aqueous solution, the spectrophotometric thermal gradient data (Figure 7A) show that at 0.1 wt%, turbidimetric changes could be initiated at temperatures lower than the cloud point. This finding improves on the existing knowledge that heating solutions containing non-ionic surfactants to temperatures above the surfactant cloud point (T_c) could induce a phase change [21]. While the exact role of the salts in the proprietary buffer on the lower cloud point was not explored, data from [22] suggest that the cloud point of non-ionic surfactants could be lowered via the addition of salts. Figure 7A also shows a biphasic change in buffer turbidity relative to temperature increase. The primary reason for this biphasic relationship could not be determined from secondary data; however, a thermally-induced phase change due to surfactant cloud point could explain an increase in turbidity. This phase change results in the formation of surfactant micelles that interact with the particles, thereby conforming to the Richardson–Zaki theory. Although this phase change can also be induced chemically via the addition of ionic surfactant and electrolytes, for this application, micellar aggregation occurred because of thermally induced reduction in the hydration of oxyethylene oxygen in hydrophilic groups, which leads to micellar aggregation. This phenomenon has been widely adopted in environmental studies and in separation science as a form of cloud point extraction (CPE) [21,23,24,25,26]. While encouraging, most applications requiring microparticle manipulation may not be tolerant to heating. For such applications, the choice of surfactants with cloud points closer to tolerable temperatures or chemical tuning [22] of the surfactant cloud point may be ideal.

5. Conclusions

Various biomedical applications, especially in microfluidic systems, require microparticle handling [27]. The development of specialized devices for continuous mixing [28,29,30,31] or a complete switch to gel beads has been mostly adopted due to the challenge of bead settling. However, these factors increase the cost and complexity of the system. This paper presents two simplistic solutions, hydrodynamic and i-HS solutions. Both solutions exploit rudimentary components in biomedical platforms, thereby not adding to the cost. While they can be applied independently, they can also be combined to achieve a uniform distribution of particles of various sizes or to lower their settling velocity for fluidic applications. This was successfully applied in particle metering and distribution manifold. A limitation of both systems is the requirement for an initial homogenization step and continued heating for the i-HS (optional). Moreso, the thermal induction of hindered settling as demonstrated in this research may not be ideal for most biomedical systems. However, thermal i-HS may not be necessary, considering the possibility of non-thermal cloud point tuning [22]. These principles can be adopted in particle-laden flows involving the need for mechanistic encapsulation of microparticles. More so, to the best of our knowledge, there are no published studies that explore the integration of mechanical agitation and multi-well particle metering to set a comparative baseline. This study establishes such a baseline. Further, continuous mechanical agitation of particle suspensions upstream does not influence particle behaviours downstream where convective flows are negligible; as such, regardless of mechanical mixing in bulk solution, additional measures are required to forestall undesirable sedimentation of particles.

Supplementary Materials

Download

References

Iwai, K.; Sochol, D.R.; Lin, L. A bead-in-droplet solution exchange system via continuous flow microfluidic railing. In Proceedings of the 2013 IEEE 26th International Conference on Micro Electro Mechanical Systems (MEMS), Taipei, Taiwan, 20–24 January 2013; pp. 1203–1206. [Google Scholar]
Kim, H.; Choi, I.H.; Lee, S.; Won, D.J.; Oh, Y.S.; Kwon, D.; Sung, H.J.; Jeon, S.; Kim, J. Deterministic bead-in-droplet ejection utilizing an integrated plug-in bead dispenser for single bead-based applications. Sci. Rep. 2017, 7, 1–9. [Google Scholar] [CrossRef]
Price, A.K.; Macconnell, A.B.; Paegel, B.M. Microfluidic bead suspension hopper. Anal. Chem. 2014, 86, 5039–5044. [Google Scholar] [CrossRef] [PubMed]
Myers, K.K.; Herich, J.P.; Chavez, J.E.; Berkey, K.G.; Loi, A.J.; Cleveland, P.H. A Novel Method to Gently Mix and Uniformly Suspend Particulates for Automated Assays. SLAS Technol. 2021, 26, 498–509. [Google Scholar] [CrossRef]
Wang, C.H.; Lien, K.Y.; Wu, J.J.; Lee, G.B. A magnetic bead-based assay for the rapid detection of methicillin-resistant Staphylococcus aureus by using a microfluidic system with integrated loop-mediated isothermal amplification. Lab A Chip 2011, 11, 1521–1531. [Google Scholar] [CrossRef]
Anyaduba, T.D.; Otoo, J.A.; Schlappi, T.S. Picoliter Droplet Generation and Dense Bead-in-Droplet Encapsulation via Microfluidic Devices Fabricated via 3D Printed Molds. Micromachines 2022, 13, 1946. [Google Scholar] [CrossRef] [PubMed]
Sassolas, A.; Hayat, A.; Marty, J.L. Immobilization of enzymes on magnetic beads through affinity interactions. Methods Mol. Biol. 2013, 1051, 139–148. [Google Scholar] [CrossRef]
Vashist, S.K.; Luong, J.H. Antibody Immobilization and Surface Functionalization Chemistries for Immunodiagnostics; Elsevier: Amsterdam, The Netherlands, 2018; pp. 19–46. [Google Scholar] [CrossRef]
Poles, M.; Meggiolaro, A.; Cremaschini, S.; Marinello, F.; Filippi, D.; Pierno, M.; Mistura, G.; Ferraro, D. Shaking Device for Homogeneous Dispersion of Magnetic Beads in Droplet Microfluidics. Sensors 2023, 23, 5399. [Google Scholar] [CrossRef] [PubMed]
Banerjee, U.; Jain, S.K.; Sen, A.K. Particle encapsulation in aqueous ferrofluid drops and sorting of particle-encapsulating drops from empty drops using a magnetic field. Soft Matter 2021, 17, 6020–6028. [Google Scholar] [CrossRef]
Anyaduba, T. Primer Payload System for Higher-Order Multiplex LAMP: Design and Development of Unit Processes. Ph.D. Dissertation, Keck Graduate Institute, Claremont, CA, USA, 2021. [Google Scholar]
Shintaku, H.; Kuwabara, T.; Kawano, S.; Suzuki, T.; Kanno, I.; Kotera, H. Micro cell encapsulation and its hydrogel-beads production using microfluidic device. Microsyst. Technol. 2007, 13, 951–958. [Google Scholar] [CrossRef]
Klein, A.M.; Mazutis, L.; Akartuna, I.; Tallapragada, N.; Veres, A.; Li, V.; Peshkin, L.; Weitz, D.A.; Kirschner, M.W. Droplet barcoding for single-cell transcriptomics applied to embryonic stem cells. Cell 2015, 161, 1187–1201. [Google Scholar] [CrossRef]
Zilionis, R.; Nainys, J.; Veres, A.; Savova, V.; Zemmour, D.; Klein, A.M.; Mazutis, L. Single-cell barcoding and sequencing using droplet microfluidics. Nat. Protoc. 2017, 12, 44–73. [Google Scholar] [CrossRef]
Clark, I.C.; Abate, A.R. Microfluidic bead encapsulation above 20 kHz with triggered drop formation. Lab A Chip 2018, 18, 3598–3605. [Google Scholar] [CrossRef] [PubMed]
Ekanayake, N.I.; Berry, J.D.; Stickland, A.D.; Dunstan, D.E.; Muir, I.L.; Dower, S.K.; Harvie, D.J. Lift and drag forces acting on a particle moving with zero slip in a linear shear flow near a wall. J. Fluid Mech. 2020, 904, A6. [Google Scholar] [CrossRef]
Riuxardson, J.F.; Zaki, W.N. The sedimentation of a suspension of uniform spheres under conditions of viscous flow. Chem. Eng. Sci. 1954, 3, 65–73. [Google Scholar] [CrossRef]
Zhu, Z.; Wang, H.; Peng, D.; Dou, J. Modelling the hindered settling velocity of a falling particle in a particle-fluid mixture by the Tsallis entropy theory. Entropy 2019, 21, 55. [Google Scholar] [CrossRef] [PubMed]
Ghaedi, M.; Shokrollahi, A.; Mehrnoosh, R.; Hossaini, O.; Soylak, M. Combination of cloud point extraction and flame atomic absorption spectrometry for preconcentration and determination of trace iron in environmental and biological samples. Cent. Eur. J. Chem. 2008, 6, 488–496. [Google Scholar] [CrossRef]
Mariusz, R.; Ewelina, L. Modeling of the sedimentation process of monodisperse suspension. Int. J. Comput. Methods Exp. Meas. 2022, 10, 50–61. [Google Scholar] [CrossRef]
Miyake, M.; Yamashita, Y. Molecular Structure and Phase Behavior of Surfactants. In Cosmetic Science and Technology: Theoretical Principles and Applications; Elsevier: Amsterdam, The Netherlands, 2017; pp. 389–414. [Google Scholar] [CrossRef]
Cloud Point of Non-Ionic Surfactants—METTLER TOLEDO. Available online: https://www.mt.com/gb/en/home/library/applications/lab-analytical-instruments/cloud-point-non-ionic-surfactant.html#:~:text=Importance%20of%20the%20Cloud%20Point,non%2Dionic%20surfactants%20in%20water. (accessed on 19 August 2024).
Nazar, M.F.; Shah, S.S.; Eastoe, J.; Khan, A.M.; Shah, A. Separation and recycling of nanoparticles using cloud point extraction with non-ionic surfactant mixtures. J. Colloid Interface Sci. 2011, 363, 490–496. [Google Scholar] [CrossRef]
Duester, L.; Fabricius, A.L.; Jakobtorweihen, S.; Philippe, A.; Weigl, F.; Wimmer, A.; Schuster, M.; Nazar, M.F. Can cloud point-based enrichment, preservation, and detection methods help to bridge gaps in aquatic nanometrology? Anal. Bioanal. Chem. 2016, 408, 7551–7557. [Google Scholar] [CrossRef]
Na, G.C.; Yuan, B.O.; Stevens, H.J.; Weekley, B.S.; Rajagopalan, N. Cloud point of nonionic surfactants: Modulation with pharmaceutical excipients. Pharm. Res. 1999, 16, 562–568. [Google Scholar] [CrossRef]
Al-Saadi, M.R.; Al-Garawi, Z.S.; Thani, M.Z. Promising technique, cloud point extraction: Technology & applications. J. Phys. Conf. Ser. 2021, 1853, 012064. [Google Scholar]
Cremaschini, S.; Torriero, N.; Maceri, C.; Poles, M.; Cleve, S.; Crestani, B.; Meggiolaro, A.; Pierno, M.; Mistura, G.; Brun, P.; et al. Magnetic Stirring Device for Limiting the Sedimentation of Cells inside Microfluidic Devices. Sensors 2024, 24, 5014. [Google Scholar] [CrossRef] [PubMed]
Burgoyne, F. Preventing Suspension Settling during Injection—Chips and Tips. Available online: https://blogs.rsc.org/chipsandtips/2007/08/21/preventing-suspension-settling-during-injection/?doing_wp_cron=1724225330.9217240810394287109375 (accessed on 19 August 2024).
Chong, W.H.; Chin, L.K.; Tan, R.L.S.; Wang, H.; Liu, A.Q.; Chen, H. Stirring in Suspension: Nanometer-Sized Magnetic Stir Bars. Angew. Chem. Int. Ed. 2013, 52, 8570–8573. [Google Scholar] [CrossRef]
Suk Ryu, K.; Shaikh, K.; Goluch, E.; Fan, Z.; Liu, C. Micro magnetic stir-bar mixer integrated with parylene microfluidic channels. Lab A Chip 2004, 4, 608–613. [Google Scholar] [CrossRef]
Lane, S.I.R.; Butement, J.; Harrington, J.; Underwood, T.; Shrimpton, J.; West, J. Perpetual sedimentation for the continuous delivery of particulate suspensions. Lab A Chip 2019, 19, 3771–3775. [Google Scholar] [CrossRef] [PubMed]

Microfabricated dynamic brain organoid cocultures to assess the effects of surface geometry on assembloid formation

January 16, 2026August 21, 2024 by CadWorks3DAdmin

Academic Article

Microfabricated dynamic brain organoid cocultures to assess the effects of surface geometry on assembloid formation

by Camille Cassel de Camps, Sabra Rostami, Vanessa Xu, Chen Li, Paula Lépine, Thomas M. Durcan and Christopher Moraes

Abstract: Organoids have emerged as valuable tools for the study of development and disease. Assembloids are formed by integrating multiple organoid types to create more complex models. However, the process by which organoids integrate to form assembloids remains unclear and may play an important role in the resulting organoid structure. Here, a microfluidic platform is developed that allows separate culture of distinct organoid types and provides the capacity to partially control the geometry of the resulting organoid surfaces. Removal of a microfabricated barrier then allows the shaped and positioned organoids to interact and form an assembloid. When midbrain and unguided brain organoids were allowed to assemble with a defined spacing between them, axonal projections from midbrain organoids and cell migration out of unguided organoids were observed and quantitatively measured as the two types of organoids fused together. Axonal projection directions were statistically biased toward other midbrain organoids, and unguided organoid surface geometry was found to affect cell invasion. This platform provides a tool to observe cellular interactions between organoid surfaces that are spaced apart in a controlled manner, and may ultimately have value in exploring neuronal migration, axon targeting, and assembloid formation mechanisms.

We kindly thank the researchers at the McGill University for this collaboration, and for sharing the results obtained with their system.

1. Introduction

Organoids have gained popularity as experimental models for developmental and disease studies.^[^1-11^] Grown from stem cells, these 3D tissue-engineered cultures can differentiate toward diverse lineages that capture the complexity of in vivo tissues.^[^1-9^] Multiple organoid types can also be assembled to interact, fuse, and mature^[^12-14^] and these “assembloids” can hence capture some of the additional cellular diversity and architectural complexity of multi-component organ systems, compared to single-type organoids.^[^{12, 13, 15, 16}^] In the developing brain, for example, complex circuits are established by neuronal projection and migration to create both local and long-distance connections.^[^13, 17-19^] Regionalized organoids can hence be assembled to create in vitro models of the circuits that run throughout the brain. For example, functional synaptic connections can form between cortical and striatal organoids, specific neurons can migrate from ventral to dorsal forebrain organoids,^[^20, 21^] and muscle contraction can be stimulated by brain organoid activity.^[²²^] Assembloids can hence be powerful in vitro models for a wide variety of neurodevelopmental disease processes.^[^20, 21^]

Tissue geometry is now well-established to influence fundamental cellular processes, such as proliferation, differentiation, branching, and invasion.^[^23-30^] Driven by endogenous mechanical cellular stresses that spontaneously arise in three-dimensional tissues,^[^24, 25, 31^] these cellular phenotypes drive feedback loops that govern tissue organization, specification, and maturation.^[^23, 26, 32^] While previous studies have demonstrated that geometric confinement and associated mechanical stresses drive the organization of developing neural structures,^[^33, 34^] whether these geometric features play a role in neural organoid development and assembloid formation remains an open question. Such experiments would require the technical capacity to simultaneously impose a specific geometry on independently cultured organoids, and control their relative positions before allowing them to interact. Moreover, such experiments would require long-term culture in biologically permissive and optically addressable formats. Given the intrinsic challenges associated with precisely manipulating soft living matter, technical innovations are required to better understand assembloid formation.

Recent developments in organoid culture models suggest a path to achieve these goals. Park et al. recently developed a microfabricated culture approach in which oxygen-permeable silicone inserts are used to restrict the size and shape of intestinal organoids as they grow into a hydrogel matrix.^[³⁵^] This approach was successfully used to allow stem cell proliferation and maturation, while controlling the global geometry of mature intestinal organoids, such that diffusive transport of oxygen, nutrients, and waste was sufficient to prevent the formation of a necrotic core that commonly arises in large, dense tissues.^[³⁵^] Although very promising, this strategy has yet to be demonstrated for other types of organoids. Inspired by this approach, here we develop a strategy to separately culture distinct brain organoid types in adjacent compartments, while shaping the surface geometry of the tissues; and explore this concept using two types of brain organoids. After establishing mature organoids, an insert separating the organoid compartments can be manually removed and replaced with a thin layer of extracellular matrix, allowing the precisely positioned organoids to begin forming an assembloid (Figure 1). To prove this concept, here we use various channel geometries to shape unguided and midbrain organoids. We demonstrate simultaneous axonal projections emanating from the midbrain organoids, and surface geometry-specific cell migration from unguided organoids. We hence propose that this technical innovation allows systematic investigation of the role of interacting surface geometries in assembloid formation.

2. Methods

Unless otherwise stated, all cell culture materials and supplies were purchased from Fisher Scientific (Ottawa, ON) and chemicals were from Sigma-Aldrich (Oakville, ON). The use of induced pluripotent stem cells (iPSCs) in this research was approved by the McGill University Health Centre Research Ethics Board (DURCAN_IPSC/2019-5374).

2.1. Device fabrication process

Molds were designed in Fusion 360 (AutoDesk), and printed on a ProFluidics 285D 3D resin printer using Master Mold Resin (CADworks3D) with a layer thickness of 50 µm. After washing with isopropanol, mold pieces were cured in a 36 W ultraviolet (UV) chamber overnight. Molds were designed for assembly into chambers with patterned features on both the base and lid (Figure 2A). Polydimethylsiloxane (PDMS) prepolymer and curing agent were mixed at a ratio of 10:1 w/w, poured into the chamber, and degassed under vacuum. The molded lid was then lowered slowly from one side to avoid trapping air bubbles in the chamber. The lid was then pressed down to displace excess PDMS. Tongue-and-groove convex/concave features in the chamber base and lid contained the PDMS prepolymer after chamber assembly. PDMS was cured overnight in an oven at 40°C to minimize shrinkage^[³⁶^] and de-molded using 70% ethanol to help release the devices from the 3D printed resin.

Figure 2. Functional, separate channels and reservoirs. (A) Base and insert parts of displacement mold for casting devices for single cell culture. (B) Replica molded PDMS devices, shown with channels filled with dye, left, and reservoirs filled, middle and right. (C) T47D cells were stained with CellTracker Red or Green, and then loaded into channels with Matrigel. After 3 days of culture, cells remain separated in their respective channels. (D) T47D cells were loaded with Matrigel into both channels of a device, with cells in one channel dyed once inside the channel by adding CellTracker Green to that media reservoir. (E) iPSCs were loaded into channels with Matrigel, and cultured with midbrain organoid media. Live/dead staining shows high viability after 7 days in culture.

Apparatus Used

Master Mold for PDMS

Learn More

ProFluidics 285D

Learn More

2.2. Device preparation for organoid culture

Base devices were coated with dopamine to improve adhesion to Matrigel.^[³⁷^] Briefly, dopamine hydrochloride was dissolved in 10 mM Tris buffer (pH 8.5; 2 mg mL⁻¹), pipetted into the devices, and incubated overnight at room temperature. After treatment, devices were rinsed in reverse osmosis water, and dried with a stream of dry compressed air. To facilitate release from the devices and reduce adhesion to tissue cultures, the removable inserts were passivated with Pluronic® F-68 (5% in water) overnight at room temperature.^[^38, 39^] Treated devices were rinsed in water, and dried with compressed air. All components were sterilized for 20 min in a 36 W UV chamber before assembly. Devices were assembled on a coverslip, which formed the bottom of the media reservoirs and allowed the device to be easily manipulated as a unit. For experiments without a removable insert, glass coverslips were used as the base surface, after functionalization with dopamine. Assembled devices were sterilized by UV for 45 min.

2.3. Cell culture

The AIW002-2 iPSC line (male) was used to generate unguided (previously referred to as “cerebral” organoids), and the TD22 iPSC line (male) was used to generate midbrain organoids, following established and characterized protocols^[^40, 41^] with some previously-described modifications.^[⁴²^] Both lines were obtained from The Neuro’s C-BIG repository and had undergone multistep quality control.^[⁴³^] T47D human breast cancer cells (ATCC HTB-133) were used for preliminary validation experiments with the microfluidic device.

All cell cultures were maintained at 37°C with 5% CO₂. T47D cells were cultured in Dulbecco’s Modified Eagle Medium with 10% fetal bovine serum and 1% antibiotic–antimycotic (complete DMEM). Media was changed every 3–4 days, and cells were passaged using 0.25% trypsin–EDTA at 80% confluence in a 1:5 dilution. The iPSC lines TD22 and AIW002-02 were maintained on Matrigel-coated plates in mTeSR1 complete kit media (Basal medium with supplement; STEMCELL Technologies, Cat No. 85850) with daily media changes. iPSC plated petri dishes were checked every day for spontaneous differentiation using a brightfield microscope. In cases where cells with morphologies different from iPSC colonies were detected, locations were marked and then in the BSC, cells on the marked locations were scraped off using a P200 pipette tip. Then, the media was collected, and the plate was washed with DMEM-F12 media supplemented with 1% antibiotic-antimycotic to ensure the removal of the scraped cells from the culture. iPSC cells were passaged with Gentle Cell Dissociation Reagent (StemCell Technologies) at 70% confluence. A ratio of 1:10 was maintained throughout all passages in which the colony pellet was broken down in such a way that each fragment contained between 10 and 15 cells. The homogeneity of the colony sizes in the sub-culture was assessed the next day by imaging with a brightfield microscope. At this step, colonies that were either too large or too small were scraped off and removed from the culture using the same method mentioned for removal of spontaneous differentiation.

2.4. Organoid culture

When iPSC cultures reached 70% confluence, cells were detached with Accutase (Gibco), resuspended in the appropriate media, seeded at 10,000 cells per well in 96-well round-bottom ultralow attachment plates (Corning Costar), and centrifuged for 10 min at 1200 rpm to aggregate the cells. Organoids were seeded so that they would be ready for Matrigel embedding simultaneously (Day 7 for midbrain, and Day 12 for unguided).

Media was changed every other day according to published protocols.^[^40, 41^] For unguided organoids: human embryonic stem cell (hES) media (low basic fibroblast growth factor (bFGF), with ROCK inhibitor) was used on Day 0 (consisting of 400 mL DMEM-F12 + Glutamax, 100 mL Knockout Serum Replacement, 15 mL hESC-quality FBS (Gibco), 5 mL MEM- non-essential amino acids, 3.5 µL 2-mercaptoethanol, bFGF at 4 ng mL⁻¹ final concentration, and ROCK inhibitor at 50 µM final concentration); hES media (low bFGF, no ROCK inhibitor) on Day 2; hES media (no bFGF, no ROCK inhibitor) on Day 4; neural induction media on Day 7 and 9 (consisting of DMEM-F12 + Glutamax, 1% N2 supplement, 1% MEM-non-essential amino acids (MEM-NEAA), and heparin at 1 µg mL⁻¹ final concentration); cerebral organoid differentiation media without vitamin A on Day 12 and 14 (consisting of 125 mL DMEM-F12 + Glutamax, 125 mL Neurobasal, 1.25 mL N2 supplement, 62.5 µL insulin, 1.25 mL MEM-NEAA, 2.5 mL penicillin-streptomycin, 1.75 µL of 1/10 2-mercaptoethanol dilution in neurobasal, and 2.5 mL B27 supplement without vitamin A); cerebral organoid differentiation media with vitamin A on Day 16 onwards (made using B27 supplement with vitamin A).

For midbrain organoids: neuronal induction media was used on Day 0 (consisting of 25 mL DMEM-F12 + Glutamax + 1% antibiotic-antimycotic, 25 mL neurobasal, 500 µL N2 supplement, 1 mL B27 without vitamin A, 500 mL MEM-NEAA, 1.75 µL of 1/10 2-mercaptoethanol dilution in neurobasal, heparin at 1 µg mL⁻¹ final concentration, SB431542 at 10 µM final concentration, noggin at 200 ng mL⁻¹ final concentration, CHIR99021 at 0.8 µM final concentration, and ROCK inhibitor at 10 µM final concentration); neuronal induction media without ROCK inhibitor was used on Day 2; midbrain patterning media was used on Day 4 (consisting of neuronal induction media without ROCK inhibitor with the addition of Sonic Hedgehog (SHH) at 100 ng mL⁻¹ final concentration, and Fibroblast Growth Factor 8 (FGF8) at 100 ng mL⁻¹ final concentration); tissue induction media was used on Day 7 (consisting of 50 mL neurobasal, 500 µL N2 supplement, 1 mL B27 without vitamin A, 500 mL MEM-NEAA, 1.75 µL of 1/10 2-mercaptoethanol dilution in neurobasal, 12.5 µL insulin, laminin at 200 ng mL⁻¹ final concentration, SHH at 100 ng mL⁻¹ final concentration, FGF8 at 100 ng mL⁻¹ final concentration, and 50 µL penicillin-streptomycin); final differentiation media was used on Day 8 onwards (consisting of 50 mL neurobasal, 500 µL N2 supplement, 1 mL B27 without vitamin A, 500 mL MEM-NEAA, 1.75 µL of 1/10 2-mercaptoethanol dilution in neurobasal, brain-derived neurotrophic factor (BDNF) at 10 ng mL⁻¹ final concentration, glial cell line-derived neurotrophic factor (GDNF) at 10 ng mL⁻¹ final concentration, ascorbic acid at 100 µM final concentration, db-cAMP at 125 µM final concentration, and 50 µL penicillin-streptomycin).

2.5. Device loading

Cell cultures were either loaded as single cells or as pre-formed organoids into the devices. Single cell cultures were obtained by trypsinization (T-47D, breast cancer line) or detachment (TD22 iPSCs, using Accutase as previously described^[⁴²^]) and resuspended in undiluted Matrigel (Corning 356230) at concentrations of 8 × 10⁶ cells mL⁻¹ for T-47D, or 1 × 10⁶, 3 × 10⁶, or 1 × 10⁷ cells mL⁻¹ for iPSCs. All pipetting steps were performed with chilled pipette tips to prevent premature polymerization of the Matrigel. To load the organoids into the devices, they were pipetted directly into the loading ports in media with cut P200 pipette tips on Day 12 or 13 after seeding cerebral organoids, and on Day 7 or 8 for midbrain organoids. Once in the chamber, they were too big to pass through the channel restriction. Media was aspirated, leaving the organoids in the device, and replaced with undiluted Matrigel. All devices were incubated for 20 min at 37°C to polymerize the Matrigel. The appropriate media was added after polymerization, and replaced every 2 days.^[⁴⁰^]

2.6. Insert removal

Once the organoids had grown and adopted the shapes defined by the compartment dimensions, the inserts separating the compartments were removed. One pair of tweezers was used to hold the base device down, while another was used to slowly pull the insert away. Media was gently aspirated from between the organoids, and replaced with Matrigel to fill in the space. Devices were left at room temperature for 5 min to allow the newly added liquid Matrigel to seep into the existing Matrigel, and then incubated for 20 min at 37°C. Final differentiation medium from the midbrain protocol,^[⁴⁰^] with the addition of insulin at a concentration of 0.25 µL mL⁻¹, was added to the well for this stage of combined culture. This media formulation was selected based on consultation and comparison of existing organoid and assembloid protocols and the function of each component,^[^21, 41, 44^] and would need to be adjusted if other types of organoids were grown in the coculture device.

2.7. Tissue staining

Live CellTracker Green or Red were loaded into cells at 20 µM in media overnight at 37°C. Live/dead viability assays were performed with calcein AM and ethidium homodimer-1 (Life Technologies), diluted in media to 4 µM each and incubated with devices for 30 min at 37°C.

For immunostaining, Matrigel was first dissolved using Cell Recovery Solution (Corning; at 4°C for 20 min, twice). Devices were washed twice with phosphate buffered saline (PBS), fixed in 4% paraformaldehyde for 1 h at room temperature, and washed three times for 15 min each with PBS before storage at 4°C until staining. Whole mount staining was performed on organoids directly in the devices, using standard protocols.^[⁴²^] Briefly, organoids were incubated in blocking buffer (0.5% Triton X-100 + 5% donkey serum in PBS) for 5 h at room temperature, then with primary antibodies diluted in blocking buffer overnight at 4°C. Organoids were then washed with PBS, three times for 15 min each, and then incubated with secondary antibodies and Hoechst diluted in blocking buffer overnight at 4°C. Organoids were washed again as before. Antibodies and stains were used as follows: anti-tyrosine hydroxylase (TH) at 1:200 (rabbit polyclonal, Pel Freez P40101-150), anti-β-tubulin III (Tuj1) at 1:300 (chicken polyclonal, Millipore Sigma AB9354), anti-tau-1 clone PC1C6 at 1:200 (mouse monoclonal, Millipore Sigma MAB3420), anti-glial fibrillary acidic protein (GFAP) at 1:250 (rabbit polyclonal, Millipore Sigma AB5804), anti-microtubule-associated protein 2 (MAP2) at 1:400 (chicken polyclonal, EnCor Biotechnology CPCA-MAP2), goat anti-chicken IgY H&L (DyLight® 488) at 1:500 (Abcam ab96947), donkey anti-rabbit IgG H&L (DyLight® 488) at 1:500 (Abcam ab96891), donkey anti-mouse IgG H&L (DyLight® 594) at 1:500 (Abcam ab96877), donkey anti-Rabbit IgG (H+L) Alexa Fluor™ 594 at 1:500 (Invitrogen A-21207), and Hoescht 33342 at 1:5000 (Invitrogen H3570). Immunostains were performed with a negative control (staining without primary antibody) to confirm that under these imaging conditions, any detected signals were not the result of non-specific binding or autofluorescence.

2.8. Microscopy and image analysis

Devices were imaged using an EVOS transmitted light microscope (XL Core) or an EVOS M7000 fluorescent Imaging System. Images were processed and analyzed in Fiji (NIH).^[⁴⁵^] Pairwise stitching was performed using the Stitching plugin^[⁴⁶^] when needed. Axonal projections were measured from the organoid surface to projection tip to obtain their length and angle. Cell migration distances were measured from the organoid surface to the edge of the migrating cell front, 2–3 days after removal of the separating wall.

2.9. Statistical analysis

Analyses were performed in R statistical software.^[⁴⁷^] All data was confirmed to be normally distributed, with equal variances. The measurements of axonal projection lengths toward nearby organoids were normalized by lengths not directed towards a nearby organoid within samples, and then a two-sample, two-tailed t-test was used to compare between those axons that were directed towards midbrain organoids against those directed toward unguided organoids. The measurements of axonal projection angle for each organoid were used to calculate kurtosis, after centering distributions around the angle defined as toward the nearby organoid; one-sample t-tests were used to compare against random chance (i.e., uniform distribution, kurtosis of 1.8), and a two-sample, two-tailed t-test was used to compare between groups. One-way ANOVA was performed with Tukey post hoc comparisons for measurements of cell migration distance. All analyses for significance were carried out with α = 0.05.

3. Results

3.1. Device design for separated adjacent co-cultures

Double-sided 3D-printed molding chambers (Figure 2A) were essential for the successful operation of these devices, as complex 3D geometries and multiple overhanging and double-sided layer features were required, which could only be achieved by designing interlocking surfaces for double-sided PDMS molding. The PDMS devices themselves were designed to facilitate pipetting of Matrigel and cell/organoid solutions into the channels via inlet ports, while leaving the tops of the channels open for nutrient exchange. This was achieved using an overhanging phase guide that allows surface tension to hold injected liquid (prepolymerized Matrigel) in a confined space, while leaving a slit in the top of the channels open for media exchange (slit was 600 µm across). We adapted this design to create two adjacent channels, each fed by an integrated and independent media reservoir to support growth of organoids with separate media requirements (Figure 2B; shown with red and blue liquids to represent different media formulations).

We first verified that our devices were operational, suitable for cell culture, and that the media reservoirs were functionally isolated from each other by loading an available cell line (T47D breast cancer cells, used as a model cell line for preliminary experiments). We verified that the devices successfully separated cell compartments by dying the T47D breast cancer cells with either CellTracker Red or Green, and loading them in Matrigel into adjacent channels (each 1 mm wide, separated by ≈500 µm; Figure 2C). Devices were cultured for 3 days, and no color exchange was observed between compartments. Next, we sought to validate media reservoir function. T47D cells were suspended in Matrigel and loaded into channels. One reservoir was filled with regular media, and the other was filled with media containing CellTracker Green. Over several days in culture, only the cells in the channel fed by that reservoir were dyed green, demonstrating functional separation of the reservoirs produced by this fabrication technique (Figure 2D).

3.2. Devices support iPSC culture

Once device design was validated, we next sought to verify that the devices could support iPSC culture, which is typically much more stringent and would be required to grow developing organoids within the compartments. iPSCs were suspended in Matrigel at several different densities, loaded into the device channels as described, and cultured with midbrain organoid media.^[⁴⁰^] Initially loaded at 10 million cells mL⁻¹, the density of cells in the channels increased over days in culture, and multiple cells aggregated together to form numerous small clusters within the channels (Figure 2E). At 1 and 3 million cells mL⁻¹, cells also aggregated to form extremely small clusters that grew over time (data not shown). We also confirmed that iPSCs were viable for at least 1 week in culture (Figure 2E) before proceeding with organoid culture experiments.

3.3. Removable inserts for dynamic organoid co-culture

Devices for dynamic organoid co-culture were fabricated as two separate pieces, the lower of which acts as a base to hold the organoids, while the upper piece includes the separating wall and media reservoirs. The separating wall can be designed with a variety of geometries to shape the growing organoids. The organoid loading ports were designed to be sufficiently large to load pre-formed organoids, as required in standard brain organoid culture protocols (Figure 3A–C), while a small outlet port was designed to allow Matrigel loading, while keeping the organoids in the channels. We then confirmed that midbrain organoids remained viable for at least one week in culture after loading (Figure 3D).

Figure 3. Assembloid formation in two-piece separated devices. (A) 3D schematic of removable insert piece; shown here with a triangular wall geometry. (B) Replica molded two-piece PDMS devices, with base and insert pieces shown. (C) Assembled two-piece device, imaged from below. (D) Midbrain organoids were loaded into channels with Matrigel, and maintained viability for 7 days in culture. (E) Unguided and midbrain organoids were loaded into channels with Matrigel and cultured for 7 days before removing the separating wall. Organoids maintained the shape and spacing imposed by the separating wall before beginning to grow toward each other to form an assembloid. (F) Astrocytes identified with glial fibrillary acidic protein (GFAP; magenta) are observed on the edges of unguided organoids only. (G) Midbrain organoids uniquely express dopaminergic neuron marker tyrosine hydroxylase (TH; magenta). Both organoid types express neural marker β-tubulin III (Tuj1; green), which is observed across the separating bridge within 3 days of insert removal.

3.4. Devices support assembloid formation

As a first proof-of-concept, midbrain and unguided organoids were loaded in Matrigel in adjacent channels to form assembloids from two distinct organoid types. After 5–28 days, when the organoids had expanded to contact and mold themselves against the channel wall, the inserts were removed from the bases, leaving the organoids and surrounding Matrigel separated by the width of the separating wall (600 µm, but could be varied by mold design). This gap was back-filled with Matrigel, and the assembloids were then monitored during growth. We first confirmed that using this system, the organoids retained the shape of the insert wall after removal (Figure 3E). Immunostaining of fixed samples at this stage demonstrated that glial fibrillary acidic protein (GFAP)-positive astrocytes arise in unguided organoids only (Figure 3F), and tyrosine hydroxylase (TH)-positive dopaminergic neurons in the midbrain organoids only (Figure 3G), as expected for this type of organoid.^[⁴⁰^] Within 3 days, the organoids bridged the space between them to initiate formation of an assembloid. These results confirm (1) appropriate and expected differentiation of these neural organoids within our devices, including differentiation towards the non-neuronal lineages expected to arise in unguided organoids^[⁹^]; (2) that our separated device supports distinct media-driven differentiation patterns in adjacent compartments; and (3) that assembloids can form after removal of the separating wall.

3.5. Axonal projections arise from midbrain organoids during assembloid formation

Having confirmed that the device architecture enables us to reliably examine a reproducible interface between organoids during assembloid formation, we then characterized the early stages of assembloid integration in terms of behavior of cells from each of the two organoids. We noted long and thin cellular processes arising from midbrain organoids, that began to appear after ≈7–9 days of co-culture in our devices (≈15–16 days after organoid seeding; Figure 4A). We confirmed that these were axonal projections by immunostaining for tau-1, which localizes to axons only,^[^48, 49^] as well as for TH, which can also be found in axons^[⁵⁰^] (Figure 4B).

Figure 4. Axonal projection from midbrain organoids. (A) Axonal projections extending from midbrain organoid, and (B) staining positive for axonal marker tau-1 (green). (C) Axonal projections arise from all sides of the midbrain organoid. (D) Representative frequency distribution of angles of axonal projections from a midbrain organoid, showing majority of axons angled towards nearby organoid (distribution is centered around angle towards nearby organoids, 180°). Compared against a statistically random orientation, the distribution is biased toward other midbrain organoids (p < 0.05) and unguided organoids (p < 0.1; n = 50–159 axons from 3–4 organoids, one-sample t-test). (E) No significant differences in axon lengths directed toward an unguided or another midbrain organoid were observed (data presented as individual data points with an overlaid bar graph showing mean ± standard deviation; Ø symbol used to denote no observations in that category; n = 3–57 axons from seven organoids; p > 0.1 by two-sample t-test comparing axon lengths that were directed toward either midbrain or unguided organoids). — Figure 4. Axonal projection from midbrain organoids. (A) Axonal projections extending from midbrain organoid, and (B) staining positive for axonal marker tau-1 (green). (C) Axonal projections arise from all sides of the midbrain organoid. (D) Representative frequency distribution of angles of axonal projections from a midbrain organoid, showing majority of axons angled towards nearby organoid (distribution is centered around angle towards nearby organoids, 180°). Compared against a statistically random orientation, the distribution is biased toward other midbrain organoids (p < 0.05) and unguided organoids (p 0.1 by two-sample t-test comparing axon lengths that were directed toward either midbrain or unguided organoids).

Based on mechanisms of axon guidance^[^17-19^] it is reasonable to suppose that factors secreted during co-culture may direct axonal outgrowth. We first asked whether quantitative analysis would allow us to better understand the factors that might affect axon targeting behaviors. Axonal projections grew from all sides of the midbrain organoid (Figure 4C) toward midbrain organoids in the same channel (oriented at 180°), unguided organoids in the adjacent channel (oriented at 270°), and toward spaces without organoids in them. By comparing the frequency distributions of axonal direction (Figure 4D) against random chance, we found that axonal projections from midbrain organoids were significantly statistically biased toward other midbrain organoids (n = 50–159 axons from 3 to 4 organoids; p < 0.05 by one-sample t-test), while projections toward unguided organoids only approached significance (p < 0.1). In contrast, axon lengths were not significantly different regardless of direction toward which they grow (Figure 4E). This analysis would therefore suggest that when allowed to form in co-culture, the distribution of directed axon targeting may be biased by brain region-specific secreted factors.

3.6. Organoid surface geometry influences cell migration

We also observed invasive migration of cells from the unguided organoids into the inter-organoidal space, and given the known impact of tissue geometry on cellular invasion and migration,^[^23-25^] we asked whether organoid surface geometry might influence this invasive behavior during assembloid formation. We therefore tested flat versus triangular geometric shapes of the separating wall (Figure 5A), and the organoids grew to adopt the shapes provided by the channel wall (Figure 5B,C). Once the organoids had reached this stage, the separating inserts were removed. Interestingly, cells migrated out from the unguided organoids into the Matrigel towards the midbrain organoids regardless of organoid peripheral geometry (Figure 5C), and immunostaining indicated that some of these migrating cells were astrocytes, positive for GFAP (Figure 5D). However, migration distance was significantly different based on the global originating tissue shape (Figure 5E,F). Cells migrating from a flat organoid periphery travelled significantly farther than cells migrating from the flat midpoint of a triangle edge. Given that flat surfaces are predicted to have lower mechanical tension than surfaces of high curvature, it seems likely that differential mechanical priming arising from shape could lead to different migration distances.

Figure 5. Geometrical shaping of organoids and cell migration. (A) Schematic showing channel-separating walls with different geometries. (B) Unguided and midbrain organoids with flat peripheries, shaped by flat separating wall. (C) Unguided organoid shaped into triangles by separating wall with points, and maintaining shape after wall removal. Cells migrate out of organoid at periphery, regardless of geometry. (D) Stained unguided organoid shows expression of astrocyte marker glial fibrillary acidic protein (GFAP), with astrocytes having migrated out of organoid. The dashed line indicates edge of unguided organoid. (E) Schematic showing locations of measurements of cell migration front from unguided organoid. (F) Measurements of migration distance of cells leaving unguided organoids from locations with different geometries 2–3 days after wall removal (side refers to location at midpoint of triangle edge) (data presented as mean ± standard deviation; n = 5–8 locations; **p < 0.01 by one-way ANOVA with Tukey post hoc comparisons).

Apparatus Used

Master Mold for PDMS

Learn More

ProFluidics 285D

Learn More

4. Discussion

In this work, we build on recent advances by Park et al. in culturing shaped organoids^[³⁵^] and develop a platform and methodology to individually culture multiple organoids of distinct types and control surface shape and spacing of the organoids, before allowing them to interact and form an assembloid on demand. Since the distance between shaped organoids can be precisely defined, the process of assembly can be closely observed in situ across multiple similarly-shaped live cultures during assembloid formation. Constructing devices capable of supporting long-term growth of organoids into predefined shapes, while affording the ability to (1) allow simultaneous but separate culture protocols for each of the organoid types, (2) support distinct surface modification strategies to enhance or prevent adhesion, and (3) to gently remove a separating barrier on demand required complex device geometries. To meet these fabrication demands, we developed a 3D printer-supported double-sided molding technique, which we successfully demonstrated to create barriers as thin as ≈200 µm. This fabrication method has the advantage of being extremely rapid and versatile, allowing design-to-device turnaround times of less than 8 h, while creating novel structures that would be extremely difficult to produce using conventional single-side replica molding approaches.

As a first demonstration of this technology, we investigated assembloid formation processes between midbrain and unguided organoids. Although we focused our experiments on controlling the space between organoids in separate channels, in principle this approach could in future be adapted to control the spacing between individual organoids, by reducing the length of the channel. Despite the simplicity of the current design, we were able to position organoids sufficiently close together to observe axonal projections from the midbrain organoids, and invasion of individual cells from the unguided organoids into the surrounding matrix. Together, these processes should capture key features of the assembloid formation process as two organoids merge with each other, which would be quite challenging to observe over time and quantitatively evaluate when simply placing optically-dense organoids against each other. Using the capacity for microfluidics to position and support these interactions, and to facilitate longitudinal observation using brightfield imaging, we were then able to demonstrate that axonal projections from the midbrain organoids display differential targeted outgrowth. Speculatively, since dopaminergic neurons project to a variety of brain regions in vivo,^[⁵¹^] these findings may ultimately be relevant in understanding how and why neuronal connections form differently in different regions of the brain. We were also able to observe that cell invasion from the unguided organoids was affected by global surface geometries. Unexpectedly, cells leaving flat organoids displayed higher invasive potential than cells leaving the flat regions of triangular shapes. This was unexpected because higher endogenous mechanical stress levels at shape-driven stress points such as triangular apexes have previously been thought of as “launching sites” for invasive cells in cancer models,^[^23-25^] but the opposite was observed in our neural cultures. Taken together, these results suggest that the process of assembloid formation can be dissected using microfluidic systems, and that this general approach might be leveraged to improve our understanding of the development of neural circuitry, in healthy and diseased states, both within the brain and targeting of other organs such as muscle or gut. More generally, our observations that assembloid-formation behaviors can be directed through physical cues in the local microenvironment suggests that such approaches may ultimately be useful in establishing predictive control over complex assembloid formation processes.

Several limitations should be considered in the utility of these devices. First, although the experiments here were designed primarily to prove the concept of these devices using brain organoid components, applying this strategy to other organoids may present unexpected complications. For example, the device architecture was sufficient to support the metabolic needs for brain organoid maturation, but other more energy-intensive processes may require alternative designs or other strategies to enhance metabolic transport and availability. Second, the capacity to observe assembloid-associated processes between organoids is enhanced through our microfluidic system, but is ultimately limited by difficulties in imaging through optically-dense organoids. Strategies such as brain organoid tissue clearing for end-point analysis^[⁵²^] and genetically-encoded markers to monitor processes in real time may be useful here, but cannot be used in parallel to allow for live, high-resolution imaging of such structures. Alternative imaging modalities such as MicroCT or ultrasound imaging may also have significant value in addressing this specific issue. Finally, the true capacity for assembloid formation to accurately capture in vivo processes and final structures remains unclear. While we hope that the devices presented here will provide important tools in answering such questions, much remains to be done in establishing the fundamental utility of assembloids as in vitro models of development and disease.

References

Marton, R. M., & Pașca, S. P. (2020). Organoid and assembloid technologies for investigating cellular crosstalk in human brain development and disease. Trends in Cell Biology, 30(2), 133–143. https://doi.org/10.1016/j.tcb.2019.11.004
Qian, X., Nguyen, H. N., Jacob, F., Song, H., & Ming, G. (2017). Using brain organoids to understand Zika virus-induced microcephaly. Development (Cambridge, England), 144(6), 952–957. https://doi.org/10.1242/DEV.140707
Di Lullo, E., & Kriegstein, A. R. (2017). The use of brain organoids to investigate neural development and disease. Nature Reviews Neuroscience, 18(10), 573–584. https://doi.org/10.1038/nrn.2017.107
Velasco, S., Paulsen, B., & Arlotta, P. (2020). 3D brain organoids: Studying brain development and disease outside the embryo. Annual Review of Neuroscience, 43, 375–389. https://doi.org/10.1146/annurev-neuro-070918-050154
Hofer, M., & Lutolf, M. P. (2021). Engineering organoids. Nature Reviews Materials, 2021 65, 6(5), 402–420. https://doi.org/10.1038/s41578-021-00279-y
Rossi, G., Manfrin, A., & Lutolf, M. P. (2018). Progress and potential in organoid research. Nature Reviews Genetics, 19, 671–687. https://doi.org/10.1038/s41576-018-0051-9
Otani, T., Marchetto, M. C., Gage, F. H., Simons, B. D., & Livesey, F. J. (2016). 2D and 3D stem cell models of primate cortical development identify species-specific differences in progenitor behavior contributing to brain size. Cell Stem Cell, 18(4), 467–480. https://doi.org/10.1016/j.stem.2016.03.003
Pasca, A. M., Sloan, S. A., Clarke, L. E., Tian, Y., Makinson, C. D., Huber, N., Kim, C. H., Park, J. Y., O’Rourke, N. A., Nguyen, K. D., Smith, S. J., Huguenard, J. R., Geschwind, D. H., Barres, B. A., & Pasca, S. P. (2015). Functional cortical neurons and astrocytes from human pluripotent stem cells in 3D culture. Nature Methods, 12(7), 671–678. https://doi.org/10.1038/nmeth.3415
Sloan, S. A., Darmanis, S., Huber, N., Khan, T. A., Birey, F., Caneda, C., Reimer, R., Quake, S. R., Barres, B. A., & Paşca, S. P. (2017). Human astrocyte maturation captured in 3D cerebral cortical spheroids derived from pluripotent stem cells. Neuron, 95(4), 779–790.e6. https://doi.org/10.1016/J.NEURON.2017.07.035
Zhang, W., Ma, L., Yang, M., Shao, Q., Xu, J., Lu, Z., Zhao, Z., Chen, R., Chai, Y., & Chen, J. F. (2020). Cerebral organoid and mouse models reveal a RAB39b–PI3K–mTOR pathway-dependent dysregulation of cortical development leading to macrocephaly/autism phenotypes. Genes & Development, 34(7–8), 580–597. https://doi.org/10.1101/gad.332494.119
Chen, X., Sun, G., Tian, E., Zhang, M., Davtyan, H., Beach, T. G., Reiman, E. M., Blurton-Jones, M., Holtzman, D. M., & Shi, Y. (2021). Modeling sporadic Alzheimer’s disease in human brain organoids under serum exposure. Advancement of Science, 8(18), 2101462. https://doi.org/10.1002/ADVS.202101462
Vogt, N. (2021). Assembloids. Nature Methods. Nature Publishing Group, 18, 27. https://doi.org/10.1038/s41592-020-01026-x
Miura, Y., Li, M. Y., Revah, O., Yoon, S. J., Narazaki, G., & Pașca, S. P. (2022). Engineering brain assembloids to interrogate human neural circuits. Nature Protocols. Nature Publishing Group, 17, 15–35. https://doi.org/10.1038/s41596-021-00632-z
Pașca, S. P., Arlotta, P., Bateup, H. S., Camp, J. G., Cappello, S., Gage, F. H., Knoblich, J. A., Kriegstein, A. R., Lancaster, M. A., Ming, G. L., Muotri, A. R., Park, I. H., Reiner, O., Song, H., Studer, L., Temple, S., Testa, G., Treutlein, B., & Vaccarino, F. M. (2022). A nomenclature consensus for nervous system organoids and assembloids. Nature. 2022 6097929, 609(7929), 907–910. https://doi.org/10.1038/s41586-022-05219-6
Kim, E., Choi, S., Kang, B., Kong, J. H., Kim, Y., Yoon, W. H., Lee, H. R., Kim, S. E., Kim, H. M., Lee, H. S., Yang, C., Lee, Y. J., Kang, M., Roh, T. Y., Jung, S., Kim, S., Ku, J. H., & Shin, K. (2020). Creation of bladder assembloids mimicking tissue regeneration and cancer. Nature, 588(7839), 664–669. https://doi.org/10.1038/s41586-020-3034-x
Rawlings, T. M., Makwana, K., Taylor, D. M., Molè, M. A., Fishwick, K. J., Tryfonos, M., Odendaal, J., Hawkes, A., Zernicka-Goetz, M., Hartshorne, G. M., Brosens, J. J., & Lucas, E. S. (2021). Modelling the impact of decidual senescence on embryo implantation in human endometrial assembloids. Elife, 10, e69603. https://doi.org/10.7554/eLife.69603
Stoeckli, E. T. (2018). Understanding axon guidance: Are we nearly there yet? Dev, 145(10), dev151415. https://doi.org/10.1242/DEV.151415/48515
Huberman, A. D., Clandinin, T. R., & Baier, H. (2010). Molecular and cellular mechanisms of lamina-specific axon targeting. Cold Spring Harbor Perspectives in Biology, 2(3), a001743. https://doi.org/10.1101/CSHPERSPECT.A001743
Accogli, A., Addour-Boudrahem, N., & Srour, M. (2020). Neurogenesis, neuronal migration, and axon guidance. Handbook of Clinical Neurology, 173, 25–42. https://doi.org/10.1016/B978-0-444-64150-2.00004-6
Birey, F., Andersen, J., Makinson, C. D., Islam, S., Wei, W., Huber, N., Fan, H. C., Metzler, K. R. C., Panagiotakos, G., Thom, N., O’Rourke, N. A., Steinmetz, L. M., Bernstein, J. A., Hallmayer, J., Huguenard, J. R., & Pasca, S. P. (2017). Assembly of functionally integrated human forebrain spheroids. Nature, 545(7652), 54–59. https://doi.org/10.1038/nature22330
Miura, Y., Li, M. Y., Birey, F., Ikeda, K., Revah, O., Thete, M. V., Park, J. Y., Puno, A., Lee, S. H., Porteus, M. H., & Pașca, S. P. (2020). Generation of human striatal organoids and cortico-striatal assembloids from human pluripotent stem cells. Nature Biotechnology, 38(12), 1421–1430. https://doi.org/10.1038/s41587-020-00763-w
Andersen, J., Revah, O., Miura, Y., Thom, N., Amin, N. D., Kelley, K. W., Singh, M., Chen, X., Thete, M. V., Walczak, E. M., Vogel, H., Fan, H. C., & Paşca, S. P. (2020). Generation of functional human 3D cortico-motor assembloids. Cell, 183(7), 1913–1929.e26. https://doi.org/10.1016/j.cell.2020.11.017
Nelson, C. M., VanDuijn, M. M., Inman, J. L., Fletcher, D. A., & Bissell, M. J. (2006). Tissue geometry determines sites of mammary branching morphogenesis in organotypic cultures. Science, 314(5797), 298. https://doi.org/10.1126/SCIENCE.1131000
Boghaert, E., Gleghorn, J. P., Lee, K. A., Gjorevski, N., Radisky, D. C., & Nelson, C. M. (2012). Host epithelial geometry regulates breast cancer cell invasiveness. Proceedings of the National Academy of Sciences of the United States of America, 109(48), 19632–19637. https://doi.org/10.1073/PNAS.1118872109/-/DCSUPPLEMENTAL
Gjorevski, N., & Nelson, C. M. (2010). Endogenous patterns of mechanical stress are required for branching morphogenesis. Integrative Biology, 2(9), 424–434. https://doi.org/10.1039/C0IB00040J
Gomez, E. W., Chen, Q. K., Gjorevski, N., & Nelson, C. M. (2010). Tissue geometry patterns epithelial-mesenchymal transition via intercellular mechanotransduction. Journal of Cellular Biochemistry, 110(1), 44. https://doi.org/10.1002/JCB.22545
Raghavan, S., Nelson, C. M., Baranski, J. D., Lim, E., & Chen, C. S. (2010). Geometrically controlled endothelial tubulogenesis in micropatterned gels. Tissue Engineering Part A, 16(7), 2255–2263. https://doi.org/10.1089/ten.tea.2009.0584
Nelson, C. M., Jean, R. P., Tan, J. L., Liu, W. F., Sniadecki, N. J., Spector, A. A., & Chen, C. S. (2005). Emergent patterns of growth controlled by multicellular form and mechanics. PNAS, 102(33), 11594–11599. https://doi.org/10.1073/PNAS.0502575102
Tran, R., Moraes, C., & Hoesli, C. A. (2020). Controlled clustering enhances PDX1 and NKX6.1 expression in pancreatic endoderm cells derived from pluripotent stem cells. Scientific Reports, 10(1), 1190. https://doi.org/10.1038/s41598-020-57787-0
Ma, Z., Sagrillo-Fagundes, L., Tran, R., Parameshwar, P. K., Kalashnikov, N., Vaillancourt, C., & Moraes, C. (2019). Biomimetic micropatterned adhesive surfaces to mechanobiologically regulate placental trophoblast fusion. ACS Applied Materials & Interfaces, 11(51), 47810–47821. https://doi.org/10.1021/acsami.9b19906
Lee, W., Kalashnikov, N., Mok, S., Halaoui, R., Kuzmin, E., Putnam, A. J., Takayama, S., Park, M., McCaffrey, L., Zhao, R., Leask, R. L., & Moraes, C. (2019). Dispersible hydrogel force sensors reveal patterns of solid mechanical stress in multicellular spheroid cultures. Nature Communications, 10(1), 1–14. https://doi.org/10.1038/s41467-018-07967-4
Boghdady, C. M., Kalashnikov, N., Mok, S., McCaffrey, L., & Moraes, C. (2021). Revisiting tissue tensegrity: Biomaterial-based approaches to measure forces across length scales. APL Bioengineering, 5(4), 41501. https://doi.org/10.1063/5.0046093/1025169
Knight, G. T., Lundin, B. F., Iyer, N., Ashton, L. M. T., Sethares, W. A., Willett, R. M., & Ashton, R. S. (2018). Engineering induction of singular neural rosette emergence within HPSC-derived tissues. Elife, 7, e37549. https://doi.org/10.7554/eLife.37549
Haremaki, T., Metzger, J. J., Rito, T., Ozair, M. Z., Etoc, F., & Brivanlou, A. H. (2019). Self-organizing neuruloids model developmental aspects of Huntington’s disease in the ectodermal compartment. Nature Biotechnology, 2019 3710, 37(10), 1198–1208. https://doi.org/10.1038/s41587-019-0237-5
Park, S. E., Kang, S., Paek, J., Georgescu, A., Chang, J., Yi, A. Y., Wilkins, B. J., Karakasheva, T. A., Hamilton, K. E., & Huh, D. D. (2022). Geometric engineering of organoid culture for enhanced organogenesis in a dish. Nature Methods, 19(November), 1449–1460. https://doi.org/10.1038/s41592-022-01643-8
Moraes, C., Sun, Y., & Simmons, C. A. (2009). Solving the shrinkage-induced PDMS alignment registration issue in multilayer soft lithography. Journal of Micromechanics and Microengineering, 19(6), 065015. https://doi.org/10.1088/0960-1317/19/6/065015
Park, S. E., Georgescu, A., Oh, J. M., Kwon, K. W., & Huh, D. (2019). Polydopamine-based interfacial engineering of extracellular matrix hydrogels for the construction and long-term maintenance of living three-dimensional tissues. ACS Applied Materials & Interfaces, 11(27), 23919–23925. https://doi.org/10.1021/acsami.9b07912
Azizipour, N., Avazpour, R., Sawan, M., Rosenzweig, D. H., & Ajji, A. (2022). Uniformity of spheroids-on-a-chip by surface treatment of PDMS microfluidic platforms. Sensors & Diagnostics, 1(4), 750–764. https://doi.org/10.1039/D2SD00004K
Boxshall, K., Wu, M. H., Cui, Z., Cui, Z., Watts, J. F., & Baker, M. A. (2006). Simple surface treatments to modify protein adsorption and cell attachment properties within a poly(dimethylsiloxane) micro-bioreactor. Surface and Interface Analysis, 38, 198–201. https://doi.org/10.1002/sia.2274
Mohamed, N.-V., Mathur, M., da Silva, R. V., Thomas, R. A., Lepine, P., Beitel, L. K., Fon, E. A., & Durcan, T. M. (2021). Generation of human midbrain organoids from induced pluripotent stem cells. MNI Open Research, 3, 1. https://doi.org/10.12688/mniopenres.12816.2
Lancaster, M. A., & Knoblich, J. A. (2014). Generation of cerebral organoids from human pluripotent stem cells. Nature Protocols, 9(10), 2329–2340. https://doi.org/10.1038/nprot.2014.158.Generation
Cassel de Camps, C., Mok, S., Ashby, E., Li, C., Lépine, P., Durcan, T. M., & Moraes, C. (2023). Compressive molding of engineered tissues via thermoresponsive hydrogel devices. Lab on a Chip, 23(8), 2057–2067. https://doi.org/10.1039/D3LC00007A
Chen, C. X. Q., Abdian, N., Maussion, G., Thomas, R. A., Demirova, I., Cai, E., Tabatabaei, M., Beitel, L. K., Karamchandani, J., Fon, E. A., & Durcan, T. M. (2021). A multistep workflow to evaluate newly generated iPSCs and their ability to generate different cell types. Methods and Protocols, 4(3), 50. https://doi.org/10.3390/MPS4030050/S1
Yeap, Y. J., Teddy, T. J. W., Lee, M. J., Goh, M., & Lim, K. L. (2023). From 2D to 3D: Development of monolayer dopaminergic neuronal and midbrain organoid cultures for Parkinson’s disease modeling and regenerative therapy. International Journal of Molecular Sciences, 24(3), 2523. https://doi.org/10.3390/IJMS24032523/S1
Schindelin, J., Arganda-Carreras, I., Frise, E., Kaynig, V., Longair, M., Pietzsch, T., Preibisch, S., Rueden, C., Saalfeld, S., Schmid, B., Tinevez, J. Y., White, D. J., Hartenstein, V., Eliceiri, K., Tomancak, P., & Cardona, A. (2012). Fiji: An open-source platform for biological-image analysis. Nature Methods, 9, 676–682. https://doi.org/10.1038/nmeth.2019
Preibisch, S., Saalfeld, S., & Tomancak, P. (2009). Globally optimal stitching of tiled 3D microscopic image acquisitions. Bioinformatics, 25(11), 1463–1465. https://doi.org/10.1093/BIOINFORMATICS/BTP184
R Core Team. (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing.
Binder, L. I., Frankfurter, A., & Rebhun, L. I. (1985). The distribution of Tau in the mammalian central nervous system. Journal of Cell Biology, 101(4), 1371–1378. https://doi.org/10.1083/jcb.101.4.1371
Papasozomenos, S. C., & Binder, L. I. (1987). Phosphorylation determines two distinct species of Tau in the central nervous system. Cell Motility and the Cytoskeleton, 8(3), 210–226. https://doi.org/10.1002/cm.970080303
Gervasi, N. M., Scott, S. S., Aschrafi, A., Gale, J., Vohra, S. N., Macgibeny, M. A., Kar, A. N., Gioio, A. E., & Kaplan, B. B. (2016). The local expression and trafficking of tyrosine hydroxylase MRNA in the axons of sympathetic neurons. RNA, 22(6), 883–895. https://doi.org/10.1261/rna.053272.115
Prakash, N., & Wurst, W. (2006). Development of dopaminergic neurons in the mammalian brain. Cellular and Molecular Life Sciences, 63(2), 187–206. https://doi.org/10.1007/s00018-005-5387-6
Mohamed, N. V., Lépine, P., Lacalle-Aurioles, M., Sirois, J., Mathur, M., Reintsch, W., Beitel, L. K., Fon, E. A., & Durcan, T. M. (2022). Microfabricated disk technology: Rapid scale up in midbrain organoid generation. Methods (San Diego, Calif.), 203, 465–477. https://doi.org/10.1016/J.YMETH.2021.07.008

Evaluation of industrial and consumer 3-D resin printer fabrication of microdevices for quality management of genetic resources in aquatic species

January 16, 2026July 23, 2024 by CadWorks3DAdmin

Academic Article

Evaluation of industrial and consumer 3-D resin printer fabrication of microdevices for quality management of genetic resources in aquatic species

by Seyedmajid Hosseini, Jack C. Koch, Yue Liu, Ignatius Semmes, Isabelina Nahmens, W. Todd Monroe, Jian Xu and Terrence R. Tiersch

Abstract: Aquatic germplasm repositories can play a pivotal role in securing the genetic diversity of natural populations and agriculturally important aquatic species. However, existing technologies for repository development and operation face challenges in terms of accuracy, precision, efficiency, and cost-effectiveness, especially for microdevices used in gamete quality evaluation. Quality management is critical throughout genetic resource protection processes from sample collection to final usage. In this study, we examined the potential of using three-dimensional (3-D) stereolithography resin printing to address these challenges and evaluated the overall capabilities and limitations of a representative industrial 3-D resin printer with a price of US$18,000, a consumer-level printer with a price <US$700, and soft lithography, a conventional microfabrication method. A standardized test object, the Integrated Geometry Sampler (IGS), and a device with application in repository quality management, the Single-piece Sperm Counting Chamber (SSCC), were printed to determine capabilities and evaluate differences in targeted versus printed depths and heights. The IGS design had an array of negative and positive features with dimensions ranging from 1 mm to 0.02 mm in width and depth. The SSCC consisted of grid and wall features to facilitate cell counting. The SSCC was evaluated with polydimethylsiloxane (PDMS) devices cast from a typical photoresist and silicon mold. Fabrication quality was evaluated by optical profilometry for parameters such as dimensional accuracy, precision, and visual morphology. Fabrication time and cost were also evaluated. The precision, reliability, and surface quality of industrial-grade 3-D resin printing were satisfactory for operations requiring depths or heights larger than 0.1 mm due to a low discrepancy between targeted and measured dimensions across a range of 1 mm to 0.1 mm. Meanwhile, consumer-grade printers were suitable for microdevices with depths or heights larger than 0.2 mm. While the performance of either of these printers could be further optimized, their current capabilities, broad availability, low cost of operation, high throughput, and simplicity offer great promise for rapid development and widespread use of standardized microdevices for numerous applications, including gamete quality evaluation and “laboratory-on-a-chip” applications in support of aquatic germplasm repositories.

Keywords: germplasm repository; aquatic species; 3-D resin printing; soft lithography; photolithography; aquaculture industry; genetic resources

We kindly thank the researchers at Louisiana State University for this collaboration, and for sharing the results obtained with their system.

1. Introduction

Throughout history, ensuring the protection of economically important agricultural species has involved the storage, assessment, and distribution of genetic resources. One preservation method for these resources involves placing them in a frozen state, a technique known as cryopreservation. Cryopreserved samples are commonly stored in collections or repositories [[1], [2], [3], [4]]. However, scalable cryopreservation technologies and germplasm repositories are not in place for most aquatic species despite the urgent need to protect the genetic resources that provide the foundation for aquaculture, food security, biomedical research, conservation, and wild fisheries. The genetic resources that support billions of dollars [5] of capture fisheries and human livelihoods are not protected, and the risk and expense of maintaining live animals (rather than frozen samples) hinder the growth of numerous aquatic industries [6]. These risks and expenses can be minimized by developing proprietary or shared (open) hardware devices that are capable of accelerating repository development and aiding in management and processing operations for the protection of genetic resources [7].

The growing climate crisis has exacerbated costs, risks, and needs associated with safeguarding genetic resources of aquatic species around the world. A vast majority of aquatic species that are important for aquaculture, food security, biomedical research, conservation, and wild fisheries are native to low-to-middle income nations where genetic resource protection is not a long-term priority or where equipment and reliable resources are scarce. Policy and long-term agendas must be addressed at scales beyond the individual. With the rapid growth of open-additive manufacturing, sustainable capabilities and resources can become widely accessible, and can be developed, customized, and fabricated by anyone.

Reliable tools and devices are essential for safeguarding genetic resources because they enable critical processing and quality management (QM) steps from sample collection to final usage. A relevant example is Bangladesh which is home to >600 species of freshwater and marine fishes. These fishes provide a primary protein source to sustain a growing human population of 171 million. Land use changes, introduced species, overharvesting, and other anthropogenic effects have strained open-water fisheries, and the country now relies heavily on farmed fishes (i.e., aquaculture) [8] with a narrowing gene pool. Cryopreservation is essential for preserving quality genetics, sustaining livelihoods, and ensuring sustainable production and improvement of aquatic species in Bangladesh and abroad. There are ongoing efforts to develop germplasm repositories for aquatic species in Bangladesh [8,9], but access to reliable tools and supplies, especially for quality management, are major roadblocks to these efforts. These same urgent needs for protection of aquatic genetic resources exist throughout the world, including the United States.

Capability needs for cryopreservation are driven by processing steps such as germplasm collection (e.g., sperm, eggs, early life stages, cells), quality evaluation, cryopreservation, storage, thawing, and final usage. Sample quality is of critical importance as the samples frozen today may be stored for decades and processing of poor quality samples wastes time and resources today and in the future. Quality management is a major driving force behind the need to develop novel, customizable, and accessible microdevices (e.g., micromixers [10], microfluidic lab-on-a-chip systems [11], and micro-separators [12]) to assist in safeguarding aquatic genetic resources. Such microdevices need to be versatile and practical for activities centered around germplasm QM, including quality planning, quality assurance (QA, process oriented), quality control (QC, product oriented), quality evaluation, and quality improvement.

There are existing devices to accomplish these processes, but they are often fixed in design, not suitable for multiple species, and prohibitively expensive for global deployment. For example, the process of counting sperm to calculate concentration can be accomplished by use of commercial devices such as a hemocytometer (>US$100) or a Makler chamber (>US$750) with counting by eye (which requires experience and is prone to variation) or by use of a computer-assisted sperm analysis (CASA) system (highly repeatable but >US$25,000). Integration of open-hardware microfluidic and microdevice systems would play a pivotal role in ensuring the dependable quality of germplasm materials, facilitating the isolation and culture of gametes and embryos, and optimizing the efficiency of sperm sorting and separation [13].

The use of soft lithography to develop microdevice systems (e.g., the Microfabricated Enumeration Grid Chamber [MEGC, 14] or the Single-piece Sperm Counting Chamber [SSCC, 15]) began to address some of the issues with sperm counting devices but suffer from prohibitively expensive initial costs and a lack of efficient options for iterative customization. Soft lithography typically makes use of the material, polydimethylsiloxane (PDMS) which yields high-resolution parts with excellent surface finish and low cytotoxicity [[14], [15], [16], [17], [18]]. In traditional PDMS-based soft lithography, microdevice fabrication relies on a master mold created through intensive soft lithography processes (e.g., photolithography and etching) [19]. Microdevice creation entails pouring the PDMS onto the master mold, letting it cure, and peeling it off to replicate the mold pattern. Despite its effectiveness, this process is expensive, time-consuming, complex [20], and has a number of drawbacks that limit use, especially for rapid prototyping. Soft lithography is more than capable of fabricating high-resolution devices for germplasm samples that exist at the smallest size ranges of aquatic germplasm (e.g., sperm of zebrafish [0.002 mm head width] or swordtails [0.001 mm head width]). However, it is not reliable for creating devices with larger or varied heights and depths that are required for most other species, and is slow, costly, and restricted to specialized facilities. In this study, we did not directly compare PDMS and resin prints, although such evaluations have been conducted in the past [[21], [22], [23], [24], [25], [26]]. Instead, our focus was on the evaluation of the potential for shifting to new fabrication techniques, including overall consideration of factors such as cost reduction, improved fabrication accessibility, and development of open-hardware communities based on the sharing of digital design files.

Three-dimensional (3-D) resin printing techniques such as stereolithography (SLA) and digital light projection (DLP) offer a promising and effective alternative to soft lithography and are gaining traction in the development and prototyping of microdevices. These rapidly advancing technologies can play a crucial role in addressing the creation of hardware devices with broad applications in genetic resource protection [14,15,27,28]. Three-dimensional resin printers surpass many of the constraints of soft lithography and other traditional methods through layer-by-layer transformation of computer-aided designs into tangible hardware, crafting accurate 3-D shapes. This process eliminates the need for photo masks, alignment processes, etching, and bonding which require specialized facilities and well-trained personnel, offering a more efficient and flexible manufacturing approach [29]. In addition, resin printers have access to thousands of resin types, including those developed for application in human medicine (e.g., dental-grade resins) and for use with germplasm [e.g., [30]].

Two major levels of resin printers are industrial-grade and consumer-grade. In general, consumer-grade printers have lower prices (US$400 – US$1,000) and lower-grade components, often limiting the resolution that can be achieved. Industrial-grade printers come with greater up-front costs (>US$10,000) but have higher-grade components, access to customizable resin materials, optimized resin polymerization, and system processing features for faster and more successful prints. Even the higher-priced machines, however, are much more widely available and less expensive than traditional soft lithography. Although some groups have taken the approach of pushing the capabilities of consumer-grade printers, a slow and resource-intensive process [31], there are few studies that evaluate the accuracy and precision available and resources required for device fabrication using these different techniques (resin printing and soft lithography) and printer types (industrial and consumer). This understanding is vital for aquaculture and aquatic research communities outside of traditional engineering departments. Consumer-grade products are beneficial because of their accessibility, but for technology developers, it may be more advantageous to prototype quickly with industrial-grade printers before pushing the boundaries of consumer-grade products to make devices widely available. By finding a balance among these factors, 3-D resin printing can offer new opportunities for rapid prototyping and production of micro-scale devices as an alternative to conventional soft lithographic methods.

Thus, the goal of this study was to evaluate the capabilities of 3-D resin printers and demonstrate the fabrication quality of microdevices using industrial and consumer 3-D resin printers and conventional soft lithography (photolithography) techniques. The specific objectives were to: 1) evaluate accuracy and precision in feature fabrication with opaque and clear resins; 2) assess the accuracy and precision between fabrication techniques (resin printing and photolithography), particularly for small features (<1 mm); 3) analyze the visual morphology of features produced by different methods, and 4) evaluate the utility, time, and cost requirements for overall comparison of microfabrication among the methods.

2. Materials and methods

2.1. Device description and fabrication

There is a wide range of forms and functions of microdevices. This study evaluated two representative devices: one with real-world application and one specifically designed to test the limits of resin 3-D printers. Both devices were designed using computer-aided design (CAD) software (Fusion 360, Autodesk, San Rafael, CA, USA) for systematic evaluations. The first was the Single-piece Sperm Counting Chamber (SSCC) [15], which consisted of grid and wall features with height of 0.01 mm (photolithography) or 0.1 mm (resin printing), including gaps in the gridlines that connect the squares to allow better distribution of samples for counting or quality evaluation (e.g., sperm motility) (Fig. 1, left column images). This difference in wall height between fabrication technologies prevented direct comparison but was necessary based on the current limitations and nature of the technological processes (e.g., photolithography spinning and resin printing layer height). The SSCC was specifically designed to accurately count sperm concentration in aquatic species such as zebrafish [15]. Its functionality heavily depends on the chamber volume, ensuring precise counting. This chamber provides a simple, customizable, and cost-effective alternative to traditional counting methods, aiding research in reproductive biology and assisted reproduction technologies.

The second was the Integrated Geometry Sampler (IGS), which featured an array of negative and positive features such as a semi-spheres, channels with dimensions ranging from 1 mm to 0.02 mm in width and depth, and concentric circles ranging in diameter from 4.2 mm to 0.6 mm and with step sizes of 0.2 mm (Fig. 1, right column images). Fabricating the IGS with traditional photolithography techniques would be cumbersome due to the many different feature heights on the device, each of which necessitates a separate masking, exposure, and development process. The IGS was designed specifically to evaluate the fabrication quality of 3-D resin printers.

2.1.1. Fabrication by use of soft lithography

The photolithography-based SSCC devices evaluated in this study were fabricated using a master mold created previously [12] on a silicon wafer. The process is briefly described below. A clean silicon wafer (UniversityWafer Inc., South Boston, MA, USA) served as a mold substrate. A 0.01-mm layer of SU-8 photoresist (MicroChem Corp., Newton, MA, USA) was spin-coated (Laurell Technologies Corporation, North Wales, PA, USA) on the wafer. This photoresist was chosen for its resolution and ease of microfabrication [32]. Spin-coater settings (e.g., rotational velocity) and photoresist type [33] determine the minimum and maximum feature height that can be achieved. A precisely aligned mask was set on top of the wafer to transfer the pattern during UV light exposure (American Ultraviolet®, Lebanon, IN, USA). The unexposed SU-8 photoresist was subsequently removed by application of SU-8 developer (MicroChem Corp., Newton, MA, USA), revealing the pattern on the wafer. A 10:1 mixture of PDMS and curing agent (Sylgard-184, Sigma-Aldrich, Inc., MO, USA) was prepared according to the manufacturer’s specifications and was poured onto the mold, covering the SSCC pattern. To remove air bubbles, the PDMS was degassed in a vacuum chamber, followed by curing in an oven at 70 °C for 2 h to solidify the PDMS. The de-molded PDMS was cleaned with 70% isopropyl alcohol (IPA), deionized (DI) water, and dried with nitrogen gas.

2.1.2. Selection of 3-D printers and resin materials

This study evaluated two 3-D resin printer models representative of current (as of January 2024) industrial and consumer levels. This work was not intended as a direct comparison of manufacturers or models. To broadly evaluate the capabilities of 3-D resin printers and microfabrication, we chose an industrial 3-D resin printer (ProFluidics 285D, CADworks3D, Concord, ON) and a consumer-grade printer (Sonic Mighty 8K, Phrozen, Hsinchu City, Taiwan). The IGS produced using the industrial 3-D printer was evaluated against another made with the consumer 3-D printer. Versions of the SSCC made with industrial and consumer 3-D printers were also evaluated with a photolithography-based SSCC, although this was again not intended to be a direct comparison because of a pre-selected difference in the SSCC grid-wall heights (0.1 mm for resin printers and 0.01 mm for photolithography).

The industrial 3-D printer had a 28.5-μm dynamic pixel size and an approximate cost of US$18,000. For the fabrication of devices with the industrial 3-D printer, opaque green mastermold resin (CADworks3D, Concord, Ontario) and clear microfluidic resin (CADworks3D) were used. The consumer 3-D resin printer had a 28-μm pixel size and cost approximately US$700. For fabrication of devices with the consumer 3-D printer, opaque gray Aqua 8K resin (Phrozen, Hsinchu City, Taiwan) and Nova3D ultra-clear resin (Nova3D, Guangdong, China) were used.

Apparatus Used

Clear Microfluidic Resin

Learn More

Master Mold for PDMS

Learn More

ProFluidics 285D

Learn More

2.1.3. Three-dimensional resin printing process

Pre-processing: During the printer slicing process for the SSCC and IGS models, variations were introduced to the dimensions because the slicer needed to adjust the lateral measurements to align with the pixel sizes (∼28 μm) of the LCD array – i.e., the printer cannot use half a pixel to accommodate a specific dimension [34]. This device-pixel alignment improved grid line consistency and uniformity in square designs. Thus, the X-Y dimensions of each device were scaled by a factor of 28. For example, the SSCC width was set at 392 μm, which is divisible by 28. Designs were converted to STL format for slicing using Utility (Ver 6.4.4.t12) for the industrial 3-D printer and LycheeSlicer (Ver 5.2.201) for the consumer 3-D printer. Multiple printing configurations were evaluated for both printers and the chosen settings are listed in Supplementary Table 1. These settings were selected to balance printing of positive and negative features. Based on printer behavior with specific geometries (e.g., a channel printed 20% deeper than expected), device dimensions (e.g., decreased channel depth) and slicer settings can be optimized to target positive or negative features. This study thus evaluated print quality without focus on single fine-tuned adjustments, illustrating the mechanical precision and accuracy differences of industrial and consumer-grade resin printers across a composite range of feature types and sizes that would occur in quality-management devices for aquatic species.

Post-processing: After printing, residual resin was removed by immersing the devices in a plastic bag containing 70% IPA and placing the bag into an ultrasonic water bath for 4 min. The devices were rinsed with fresh 70% IPA and DI water to ensure the complete removal of uncured resins. After 2 min of air drying, devices were exposed to a 405 nm UV light (Elegoo, Mercury X Cure, China) for 1 min to complete curing and post-processing.

2.2. Fabrication quality assessment

Comparison of multiple device features was used to provide insight into fabrication quality, accuracy, and precision among the different fabrication technologies and served as a preliminary evaluation for the widespread transition from conventional microfabrication approaches to 3-D resin printing. In fabrication of microdevices such as microchannels, clarity in the final device is crucial for use in light microscopy. While clear resin is typically preferred for this purpose in 3-D resin printing, we encountered challenges from reflected light when conducting profilometry on devices crafted from clear resin, which hampered dimensional measurements. To address this, we conducted experiments wherein devices were fabricated in parallel using opaque and clear resins.

To directly address the profiling of clear resin, a thin coating (<0.005 mm) of titanium dioxide (TiO₂) nanoparticles was applied to devices. Titanium dioxide nanoparticles were suspended in isopropanol (0.1 mg/ml) and spray coated by use of an airbrush to enhance the surface optical properties (Semmes et al., unpublished data). The gray devices did not require the addition of a TiO₂ coating layer. All devices were scanned by use of an optical profiler (Keyence VR-6100, Osaka, Japan) that used non-destructive, non-contact analysis principles. The profiler utilized light to examine surface topography, splitting the light source into two paths: one directed at the surface and the other at a reference mirror. Upon recombination, reflections were projected onto an array detector, enabling precise (0.001 mm) measurements with minimal interference. Dimensional measurements were analyzed using 3D Optical Profilometer VR-6000 software (Keyence). The reference plane for the SSCC was at the bottom of the counting chamber, and for the IGS was at the middle surface between negative and positive features.

While simple 3-D printed parts with large feature sizes can be assessed by visual observation for suitability, microdevices with features and dimensions that differ from the target dimensions (e.g., micromixers) can show altered performance and require closer inspection upon fabrication. Thus, this study evaluated accuracy and precision of printed device features. Accuracy refers to the closeness of measured values to target values, while precision indicates the consistency and reproducibility of measurements. These metrics are crucial for understanding the reliability and performance of fabrication processes, particularly in the context of transitioning from conventional microfabrication methods to emerging 3-D printing techniques. In this study, accuracy was evaluated by the difference between the target dimension and the mean of measured dimensions (photolithography, n = 6 measurements on 1 device; resin printing, n = 6 measurements each on 4 devices). Precision was assessed by calculating the standard deviation to quantify consistency and reproducibility of the fabrication processes.

To visually assess surface morphology and examine small changes in residual resin, samples were examined by use of a scanning electron microscope (SEM) (JSM -6610 LV SEM, Jeol USA, Peabody, MA, USA). Preparing devices for SEM imaging involved several steps to achieve high-resolution images. Devices were cleaned with IPA, rinsed with DI water, and air dried. A thin layer of titanium was sputter-coated onto the sample to prevent charging and improve image quality. Prepared devices were loaded into the SEM chamber. A high vacuum was pulled on the chamber in preparation for imaging. Devices were positioned and images from several angles and magnifications were captured for later visual analyses.

2.3. Time and cost requirements of microfabrication techniques

An evaluation was conducted of the time and cost associated with fabrication, assessing photolithography-based SSCC and devices printed by use of 3-D resin printing. For the photolithography time estimate, we assumed that all necessary equipment was at one facility to perform typical tasks as follows: mask preparation, spin-coating SU-8 photoresist, aligning and UV exposing, developing and curing the photoresist, pouring the PDMS mixture onto the mold, vacuum degassing, curing in an oven, and cleaning and drying the de-molded PDMS. For evaluation, the total printing duration comprised the active printing time (provided by the 3-D printer) and the associated pre-processing (e.g., slicing) and post-processing (e.g., cleaning and curing) steps. For photolithography, costs included mask creation, silicon wafer production, SU-8, SU-8 developer, and PDMS; for the 3-D printers, the cost calculation incorporated the expenses of resin materials.

3. Results

3.1. Accuracy and precision in feature fabrication with opaque resin

Dimensional accuracy (difference between target dimensions and the mean of measured dimensions) and precision (standard deviation) were assessed in fabrication of negative features (microchannel depth) on the IGS using opaque resins. For channel depths of 0.4–1 mm, the industrial 3-D printer depth showed a difference of <2% between the target and measured dimensions (Fig. 2a). In contrast, the consumer 3-D printer displayed a 3–8% discrepancy in channel depths across this range. For smaller channels of 0.1–0.2 mm, the industrial 3-D printer depth showed a difference of about 13% (Fig. 2b). Using the settings described herein, the consumer printer failed to reliably fabricate channels <0.2 mm. The standard deviation of measurements for all channel depths ranged from 0.011 to 0.039 mm for the industrial 3-D printer configuration. In comparison, the consumer printer exhibited standard deviation values ranging from 0.018 to 0.046 mm for all channel depths.

Figure 2. Comparative analysis of IGS positive and negative microchannel features of devices fabricated with industrial-level (white) and consumer-level (black) 3-D resin printers compared to target dimensions (gray). Devices were printed using opaque resin with features ranging from 1 to 0.4 mm (panel a), and ranging from 0.2 to 0.05 mm (b); and clear resin with features ranging from 1 to 0.4 mm (c), and ranging from 0.2 to 0.05 mm (d). Sample size of four devices with six measurements per device were averaged and error bars were reported as standard deviation.

Analysis of the accuracy of positive features (raised microchannels) on the IGS using opaque resins for heights of 0.4–1 mm fabricated by the industrial 3-D printer showed a difference of <2% between the target and measured dimensions (Fig. 2a). In contrast, evaluation of the accuracy of positive (raised) features produced by the consumer 3-D printer revealed fluctuations exceeding 30%. For smaller heights of 0.1–0.2 mm, the industrial 3-D printer depth showed a difference of about 19% (Fig. 2b). The standard deviation of measurements for all channel heights ranged from 0.01 to 0.031 mm for the industrial 3-D printer. In comparison, the consumer 3-D printer exhibited standard deviation values ranging from 0.007 to 0.029 mm.

Seven positive and seven negative stepped features were designed in the IGS with heights and depths of 0.2 mm per step (1.4 mm overall). For the first five negative stepped features the depths fabricated with the industrial 3-D printer showed a difference of <6% between the target and measured dimensions (Fig. 3a). In contrast, the consumer-level printer displayed a 12–26% discrepancy between target and measured dimensions in seven-stepped feature depths. Using the settings described herein, the industrial printer failed for the last two bottom (6th and 7th) stepped features. The standard deviation of depth measurement ranged from 0.011 to 0.042 mm for the industrial printer. In comparison, the consumer printer exhibited standard deviation values ranging from 0.024 to 0.041 mm.

For the positive stepped features (diameter range from 4.2 mm to 0.6 mm), accuracy of the industrial 3-D printer had a difference of <5% between the target and measured heights. In contrast, accuracy of positive stepped features produced by the consumer printer was <11% other than the first and second round features (diameters of 4.2 and 3.6 mm) which deviated from the target values by over 40%. The standard deviation of height measurement ranged from 0.011 to 0.019 mm for the industrial printer. In comparison, the consumer printer exhibited standard deviation values ranging from 0.01 to 0.041 mm (Fig. 3a).

Analysis of the accuracy of the opaque SSCC (with a grid height of 0.1 mm) fabricated with the industrial 3-D printer had a 4% discrepancy between target and measured heights (Fig. 4). The consumer printer had a discrepancy of 37% in dimensional height when utilizing opaque resin. The standard deviation for the samples fabricated with the industrial printer was 0.003 mm. Despite the discrepancies with the consumer printer, the standard deviation was 0.004 mm.

Figure 4. Comparative analysis of the target depth versus the mean of measured depth using opaque resin and clear resin for SSCC fabricated with an industrial 3-D printer (white), consumer 3-D printer (black), and photolithography (cross-hatched), compared to target dimensions (gray). Percent difference of the mean measured depth from the target depth was indicated at the top of each bar. Standard deviation was indicated by error bars. The target height for SSCC fabricated by use of resin printers was 0.1 mm and the target height for SSCC fabricated by use of photolithography was 0.01 mm. For resin prints, four devices with six measurements per device were averaged. For photolithography, one device with six measurements were averaged.

3.2. Accuracy and precision in feature fabrication with clear resin

While clear resin proved to be an ideal choice for fabricating microfluidic channels, SSCCs, or any devices requiring transparency (e.g., for visual observation), there were several challenges related to printing and profilometry that must be taken into consideration. During printing, parts in clear resins were vulnerable to distortion from additional light exposure “bleed” from layers above and below the intended layer. Also, during profilometry, reflection and refraction can distort the measurements due to changes in the optical properties of the medium.

For channel depths of 0.1–1 mm, the industrial 3-D printed IGS showed a depth difference of <4% between target and measured dimensions (Fig. 2c and d). In contrast, for channel depths of 0.4–1 mm, the consumer-level printer had less than a 6% discrepancy. It exhibited a 12–15% discrepancy for features of 0.1–0.2 mm. The standard deviation of depth measurement ranged from 0.004 to 0.03 mm for the industrial printer. In comparison, the consumer printer had standard deviation values ranging from 0.003 and 0.027 mm.

Analysis of the accuracy of positive features (raised microchannels) of the IGS fabricated with the industrial 3-D printer using clear resins for heights of 0.1–1 mm showed a difference of 0.5–5% between target and measured dimensions (Fig. 2c and d). In contrast, for channel heights of 0.4–1 mm the consumer-level printer had a 15–17% discrepancy. However, this exceeded 33% for channels ranging from 0.1 to 0.2 mm. The standard deviation of height measurements ranged from 0.005 to 0.031 mm for the industrial printer. In comparison, the consumer printer had standard deviation values ranging from 0.017 to 0.063 mm.

For the first five negative stepped features (diameter range from 4.2 mm to 1.8 mm) the depths with the industrial 3-D printer using clear resin had a difference of <4% between target and measured dimensions (Fig. 3b). In contrast, the consumer-level printer had a 2–15% discrepancy in seven stepped-feature depths (diameter range from 4.2 mm to 0.6 mm). Using the settings described herein, the industrial printer failed for the last two bottom (6th and 7th) stepped features with diameters of 1.2 and 0.6 mm. The standard deviation of depth measurements ranged from 0.009 to 0.044 mm for the industrial printer. In comparison, the consumer printer had standard deviation values ranging from 0.07 to 0.039 mm.

For the positive stepped features (diameter range from 4.2 mm to 0.6 mm), accuracy of the industrial 3-D printer showed a difference of <6% between target and measured heights. In contrast, accuracy of positive stepped features produced with the consumer printer was <12% other than the first and second stepped features (diameters of 4.2 and 3.6 mm) which failed to print. The standard deviation of height measurements ranged from 0.01 to 0.022 mm for the industrial printer. In comparison, the consumer printer had standard deviation values ranging from 0.014 to 0.029 mm (Fig. 3b).

Analysis of the accuracy of the clear SSCC (with grid height of 0.1 mm) fabricated with the industrial 3-D printer had a 2% discrepancy between target and measured heights (Fig. 4). The consumer printer had a discrepancy of 24% in dimensional height when utilizing clear resin. The standard deviation for the samples fabricated with the industrial printer was 0.008 mm. Despite the discrepancies for samples fabricated with the consumer printer, the standard deviation was 0.012 mm.

For the photolithography-based SSCC (cast from the wafer), a 0.8% difference between target and measured heights was observed (Fig. 4). The standard deviation was 0.001 mm.

3.3. Visual morphology

Achieving optimal squareness in 3-D printed features posed a notable challenge. Images captured with the optical profilometer illustrated differences in IGS negative features fabricated with industrial and consumer 3-D resin printers using opaque and clear resins (Fig. 5a-e).

Figure 5. Two-dimensional images produced by optical profiling of IGS channels with square cross-sectional design (ranging from 1 mm to 0.05 mm) fabricated with industrial 3-D printer using clear resin (panel a), industrial 3-D printer using opaque resin (b), consumer printer using clear resin (c), consumer printer using opaque resin (d), in comparison to a representative 3-D image (industrial printer using opaque resin) of the negative features (e). Blue colors indicate values below the reference plane and green colors indicate depths closer or equal to the reference plane. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

In general, deeper features appeared to have better squareness at their bottoms. Visually, the industrial resin printer (Fig. 5a and b) was able to produce sharper angles at the bottom and top of negative features compared with the consumer resin printer (Fig. 5c and d).

Visual analysis of stepped features identified different problems than square features. To facilitate visual analysis, a Keyence tool called “CAD Compare” was used (Fig. 6). The line plots (top of each panel) showed that the industrial printer with opaque resin performed best at depth and height feature fabrication. The consumer printer did well when printing negative features with clear resin, while the industrial printer did better when printing positive features with clear resin. This CAD Compare analysis also assessed the roundness of the cylinders which is important in some applications but was not directly addressed herein. Visually, the roundness of the cylinders was good with both printer types and resins, until the smallest (deepest and tallest) layers. This was also where the printers struggled to fabricate accurate depths and heights (Fig. 3).

Figure 6. Images produced by optical profiling of IGS stepped features for target profile, sample profile, and CAD comparison (height difference between target and printed sample) for the industrial 3-D printer with clear resin (panel a), industrial 3-D printer with opaque resin (b), consumer printer with clear resin (c) and, consumer printer with opaque resin (d). Reds indicated measured depths that were shallower than the target, green indicated measured depths and heights that were near or equal to the target, and blues indicated measured heights that were shorter than the target. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

The surface quality of the SSCC imaged by use of scanning electron microscopy (SEM) revealed differences in performance among the fabrication methods (Fig. 7a-c). This showed the pre-selected difference in grid-wall height between the photolithography-based (0.01 mm) and resin-based (0.1 mm) SSCCs. The channels between grids were reliably fabricated with photolithography and the surface detail at the bottom of each cell was smooth (Fig. 7a). In resin-based devices, the channels between the grids were not reliably created, although the surface detail was relatively smooth (Fig. 7b and c). A texturing is visible resembling the pixel pattern from the printer LCD. Of note are the visible layer lines produced with the industrial 3-D resin printer (Fig. 7b). The grid walls appear to be 2-layer lines thick (0.03 mm layer height and 0.1 mm target grid height).

Figure 7. Scanning electron microscopy of SSCC devices fabricated using: photolithography (panel a), industrial 3-D resin printing (b), and consumer 3-D printing (c) (scale bars = 0.1 mm). The SSCC consisted of grid and wall features with heights of 0.01 mm (photolithography) or 0.1 mm (resin printing), including gaps in the gridlines that connected the squares to allow better distribution of sample for counting or quality evaluation (e.g., sperm motility) of biological samples.

To evaluate post-processing, SEM images of SSCC were captured before and after removal of residual resin (Fig. 8a and b). One interesting observation was that post-processing revealed the channels between the grids, which were important for even filling and cellular distribution within the device.

In the final phase of post-processing, UV exposure was applied to the samples for 1 min, enhancing mechanical properties but sometimes introducing a curvature [35]. Various printing parameters including print time, single-layer height, post-curing UV intensity, and total thickness have been reported to play substantial roles in this curvature phenomenon [36]. Images of SSCC with and without post-processing (UV exposure) were captured with the profiler (Fig. 9a and b). Curling was evident in the UV-exposed sample. To mitigate this, an approach was developed allowing the sample to remain affixed to the build plate for approximately 24 h after UV exposure (data not shown). This served as an effective method to alleviate stress in the printed samples, contributing to the reduction of curvature induced by post-processing.

Figure 9. Profiled height of a SSCC device produced by the industrial printer, before curing (panel a), and after curing demonstrating curling (b) (UV exposure). The colour indicates the depth relative to a baseline of zero. Note the difference in scales. Orange indicates measurements equal to the reference plane (a). Reds indicate measurements above, the reference plane and greens and blues indicate measurements below the reference plane (a). Blue indicates measurements equal to the reference plane (b). Reds, oranges, and greens indicate measurements above the reference plane (upward curvature) (b). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

3.4. Comparison of time and cost of microfabrication

Efficiency and cost-effectiveness are pivotal factors in selecting microfabrication techniques for creation of microdevices [31]. In terms of fabrication time, fabrication of a single SSCC with photolithography typically required around 2 d and cost US$555 (Table 1). However, utilizing industrial and consumer-grade 3-D resin printers significantly reduced time and material costs. For example, the industrial printer fabrication time was <38 min and cost as little as US$0.08 per unit (Table 1). The cost calculations (Supplementary Table 2) did not include salary, electrical, or other facilities and personnel costs because these vary by location. The time calculations did not take into consideration the new approach to mitigate post-UV exposure curvature because it may not be necessary for all device configurations.

4. Discussion

This study evaluated dimensional accuracy and precision, visual morphology, and squareness of depth and height features in devices fabricated by use of traditional photolithography and industrial and consumer-grade 3-D resin printers using opaque and clear resin. In addition, the time and cost requirements were evaluated for microfabrication among the methods. The findings of this study underscored the comprehensive capabilities of industrial-grade and consumer-grade 3-D printers, in relation to photolithography in meeting complex dimensional requirements for specific applications. For this evaluation, the SSCC provided a device with direct relevance to biological usage, and the IGS allowed evaluation of fabrication quality of 3-D resin printers across a range of features. In the selection of IGS features, square geometries were chosen due to their prevalence and significance in microfluidic applications which were the primary focus of this study. Depth and height of stepped features also hold significance in certain applications and pose challenges during fabrication and were thus also chosen for evaluation. Dimensional accuracy and precision and squareness of depth and height features are critical factors for the use of quality management devices for germplasm. Differences in fabrication quality translate to functional performance changes, affecting accuracy of metrics such as sperm concentration which requires accurate volumetric calculations or microfluidic mixing efficiency needing precise feature angles and placement.

The industrial 3-D printer consistently exhibited close accuracy and precision aligned with the photolithography-based SSCC for microchannel fabrication using clear resin, even with minimum feature sizes as small as 0.1 mm. Conversely, the consumer 3-D printer, although less accurate and precise for features smaller than 0.2 mm, demonstrated reliability for features above this threshold. The larger variations, especially for raised features, were likely attributable to the specific printer settings. Throughout preliminary preparation for this study, numerous settings were explored to establish a balance for the evaluation of positive and negative features in the same device. The settings outlined in Supplementary Table 1 were identified as being effective for the specific purposes of this study.

As with traditional fabrication approaches, 3-D printing has a trade-off between extensive optimization for accuracy and precision, and the need for rapid prototyping or production of parts that do not require high tolerances [37]. With targeted refinements it would be possible to identify specific print settings for the consumer-grade printer that would further enhance the quality of the positive or negative features, but this was not the intent of the study. Industrial printers require less adjustment of such settings as they are already optimized for specific applications (e.g., microfabrication in the present study). In fabricating opaque SSCCs using consumer-grade printers, we encountered instances where the devices exhibited high precision comparable to the industrial-grade printer but lacked in accuracy (Fig. 4). Through experimentation involving iterative prototyping and adjustments to factors such as resin types and print settings, achieving a practical balance between accuracy and precision appeared to be achievable.

Stepped features printed with an industrial printer, had positive features (using opaque and clear resins) within 6% of the target dimensions, and thus exhibited reliable accuracy. However, the 6th (1.2 mm diameter) and 7th (0.6 mm diameter) negative stepped features failed to print reliably. This could be attributed to several factors including the round shape of features with small diameters, or the print settings, which may pose challenges for printing. Surprisingly, this was less challenging with the consumer printer. Instead, the consumer printer had problems with the 1st (4.2 mm diameter) and 2nd (3.6 mm diameter) positive stepped features using opaque and clear resins. Due to the higher sensitivity of consumer machines to print settings, achieving better results may be possible through adjustments to these settings. Future evaluations of the roundness (X-Y orientation) of round features on the IGS will provide insight into the potential limitations of LCD-based resin 3-D printers to fabricate round features.

These results highlighted the capabilities of different fabrication methods in terms of accuracy and precision across different feature sizes and types. It established industrial 3-D resin printers as a strong option for applications requiring high-resolution microfabrication, while consumer printers remain a viable choice for less demanding applications where feature sizes are not as critical or where there is more time for optimization of settings and resin types. The rapid advancements being made in consumer 3-D resin printing open substantial opportunities for achieving reliable microfabrication in the future. With respect to aquatic organisms, even the consumer-level printers provided sufficient accuracy and precision for routine practical use with most species and germplasm types.

Apparatus Used

Clear Microfluidic Resin

Learn More

Master Mold for PDMS

Learn More

ProFluidics 285D

Learn More

The width resolution (not directly examined herein), is intricately linked to the X-Y resolution and is influenced by the size of the projected pixels. It operates in conjunction with depth resolution, which is closely tied to Z resolution, representing the thickness of each cured layer [38]. These interrelated factors collectively contributed to the overall accuracy, precision, and level of detail attainable in printed objects. In exploring the relationship between printing orientation and dimensional accuracy and precision, altering printing orientation by 90 degrees could potentially result in a reduction in the variance of depth dimensions compared to width [39,40]. Changing the print orientation may thus influence the dimensional accuracy and precision of features within devices like the SSCC. These considerations should be investigated further and offer another avenue for reducing differences and variations observed in the depth dimensions from the printers.

In terms of visual squareness, the industrial 3-D resin printer using clear resin produced sharper corners in channels compared to prints with opaque resin (Fig. 5). When assessing the final output, it was essential to consider the influence of the slicer settings on dimensions and printer resolution. The industrial printer produced channels of ≥0.1 mm with minimal defects. However, potential defects, particularly in the roundness of the edges, become noticeable for channels smaller than 0.1 mm, affecting positive and negative features under these conditions. The consumer 3-D resin printer using opaque resin produced sharper corners in channels compared to prints with clear resin but was able to produce smaller channels better with the clear resins (Fig. 5). These visual observations of squareness between the two printer types demonstrates the opportunities to advance their capabilities and could be helpful in deciding which printer type would be best suited for specific applications that require square-angle features in clear or opaque.

Surface morphology plays a crucial role in determining the functionality and performance of microdevices. For instance, in biomedical devices it can affect cell adhesion and proliferation [41]. The industrial 3-D printer produced visually smoother surfaces compared to prints with the consumer 3-D printer (Fig. 7b and c). This superior surface quality could contribute to enhanced functionality and reliability of microfluidic channels fabricated using industrial-grade 3-D printing, highlighting its potential for various applications requiring high-quality surface finishes. Comparatively, the photolithography-based SSCC exhibited a visually smoother surface (Fig. 7a) with fewer defects than devices created by the industrial 3-D resin printer. The effect of these differences in surface morphology on the functionality and performance of resin-printed devices requires further investigation and will vary across the range of devices and functionalities desired.

For stepped features, squareness and roundness of the edges were significant factors. These parameters appeared to be satisfactory for industrial 3D-printed samples. However, in the case of consumer 3D-printed samples, while the roundness appeared acceptable, there were noticeable issues with edge squareness. The CAD comparisons showed differences in depth or height of negative or positive features compared to the design. The industrial and consumer printers generally produced better surface morphology for positive stepped features compared to negative features.

The photolithography-based method, despite its time-intensive 2-day fabrication process and substantial cost of approximately US$555 per unit for SSCC fabrication, distinguished itself through accuracy, precision, and control. This makes it particularly well-suited for applications that prioritize exacting accuracy and precision, such as highly quantitative analysis of the smallest of aquatic germplasm types (e.g., zebrafish sperm). It should be noted that the unit cost associated with photolithography decreases once a finalized mold is generated because many devices can be cast from one mold [15]. Although, the unit cost could also increase drastically if customization or changes need to be made to a mold which would require repeating the mold creation process. Industrial resin 3-D printing was efficient, requiring <30 min for IGS and SSCC fabrication with material costs estimated at US$0.32 per unit for clear IGS and US$0.08 for clear SSCC. This method enhanced rapid prototyping and fabrication. On the other hand, consumer resin 3-D printing struck a balance between fabrication time and cost, providing relatively fast times (<150 min) for IGS and SSCC (clear and opaque), with costs estimated at US$0.03 per unit for clear IGS and US$0.01 for clear SSCC. This presents a cost-effective solution for open-hardware applications with budgetary constraints while providing satisfactory fabrication times across the size and resolution range needed for aquatic species. Overall, photolithography can provide accuracy and precision, industrial resin 3-D printing can be useful for rapid prototyping, and consumer resin 3-D printing offers a balance between efficiency and cost-effectiveness, with these differences operating along a gradient of scale.

Aquatic species are in great need for powerful and innovative solutions requiring rapid prototyping and, at a minimum, batch fabrication. Currently, the resolution for resin printing is sufficient for these applications, and eventually printers that can compete directly with soft lithography will become cheap enough to be accessible to the broader community. Moving forward, devices created by use of 3-D resin printing will require testing to ensure accuracy and functionality, and resins will need to be evaluated for cytotoxicity. Though it is important to note that resin contact with sperm is for a short duration (1–2 s) and devices can make use of a disposable sub-sample rather than the entire sample. Even with the current capabilities of resin printing, we can move seamlessly from micro- to milli- to macro- scales in the design process without changing materials or equipment. Overall, the dynamic range of resin printing demonstrated herein enables high precision and throughput that can extend into open hardware, promoting inclusive access to critical technology and fostering community-driven innovation for aquatic species.

In addition, all the architecture necessary for a single device (e.g., Luer locks and other connections) can be printed at one time and in one material enabling direct single-step production of devices. However, resin printers are not limited to direct production and, like photolithography, can also print device molds facilitating access, distribution, and community development. By adding 3-D resin printing to the roster, designers, fabricators, and users now have multiple options for developing and obtaining devices. The greatly broadened accessibility of resin 3-D printers compared to photolithography provides new opportunities for standardized designs and procedures. This approach empowers wider dissemination and adoption within research and aquaculture communities, accelerating the transition to open hardware and sustainable germplasm repository development. By cultivating an international community capable of independent fabrication and design, a collaborative system can emerge to enhance genetic resource protection and promote much-needed innovation in culture, breeding, conservation, and research of aquatic species.

5. Conclusions

This study provided an overall evaluation of the capabilities of representative industrial-grade and consumer-level 3-D resin printing compared to conventional photolithography in fabrication of microdevices with a specific focus on germplasm repository development for aquatic species. This provides a platform for improving critical QM strategies throughout genetic resource protection processes. Fabrication quality distinctions were highlighted through the assessment of various components in microdevices including positive and negative features, channels, and complex structures. Visual morphologic analysis revealed differences in squareness and roundness, emphasizing the importance of considering these factors in the design of devices and the selection of fabrication technologies.

Overall, this research contributes to the ongoing exploration of emerging technologies in microdevice development and prototyping. By understanding the strengths and limitations of 3-D resin printing technologies, researchers and industry professionals can make informed decisions when selecting fabrication methods for specific applications. As the field progresses, studies of this type can provide a solid foundation for future advancements in custom microdevice creation, particularly with essential germplasm repository technology for aquatic species. By pioneering a shift from traditional methods to 3-D resin printing, the stage can be set for more efficient and precise fabrication processes. In addition, insights into the trade-offs between consumer-grade and industrial-grade printers can guide technology selection facilitating broader accessibility and innovation. Such findings will fuel development of standardized protocols, open-hardware designs, and novel solutions, ultimately enhancing aquatic species conservation efforts and genetic resource preservation. Interdisciplinary approaches such as these integrating biology and engineering (e.g., [42]) in a real-world context offer powerful mechanisms for producing innovation to address future challenges including climate change.

Supplementary Materials

Download

References

N. Coxe, Y. Liu, L. Arregui, R. Upton, S. Bodenstein, S.R. Voss, M.T. Gutierrez-Wing, T.R. Tiersch, Establishment of a practical sperm cryopreservation pathway for the axolotl (Ambystoma mexicanum): a community-level approach to germplasm repository development, Animals (Basel) 14 (2024) 206. [Google Scholar]
I. Haagen, H. Blackburn, Efforts to cryopreserve shrimp (Penaeid) genetic resources and the potential for a shrimp germplasm bank in the United States, Aquaculture 580 (2023) 740298. [Google Scholar]
J.C. Koch, A.M. Oune, S. Bodenstein, T.R. Tiersch, Untangling the Gordian Knot of Aplysia sea hare egg masses: an integrated open-hardware system for standardized egg strand sizing and packaging for cryopreservation research and application, HardwareX 16 (2023) e00476. [Google Scholar]
Y. Liu, J.C. Koch, L. Arregui, A. Oune, S. Bodenstein, M.T. Gutierrez-Wing, T. R. Tiersch, Exploring pathways toward open-hardware ecosystems to safeguard genetic resources for biomedical research communities using aquatic model species, J. Exp. Zool. B Mol. Dev. Evol. (2024) 278–290. [Google Scholar]
R.S.V. Pullin, Genetic resources for aquaculture: Status and trends, in: Status and Trends in Aquatic Genetic Resources: A Basis for International Policy, 2006. [Google Scholar]
T.R. Tiersch, C.C. Green, Cryopreservation in Aquatic Species: A Comprehensive Overview of Current Practices, Programmatic Development and Future Directions for Cryopreservation of Gametes Embryos and Larvae of Aquatic Species, World Aquaculture Society, 2011. [Google Scholar]
Y. Liu, W.T. Monroe, J.A. Belgodere, J.-W. Choi, M.T. Gutierrez-Wing, T.R. Tiersch, The emerging role of open technologies for community-based improvement of cryopreservation and quality management for repository development in aquatic species, Anim. Reprod. Sci. 246 (2022) 106871. [Google Scholar]
M.R.I. Sarder, Potential of Fish Gamete Cryopreservation in Conservation Programs in Bangladesh, in: Cryopreservation of Fish Gametes, Springer Singapore, Singapore, 2020, pp. 337–344. [Google Scholar]
S. Bodenstein, T.R. Tiersch, M.A.R. Hossain, M.G. Hamilton, M. Yeasin, M. M. Akhter, T.Q. Trinh, M. Mahmuddin, A cryopreserved sperm repository strategy for WorldFish genetically improved carp, 2023. [Google Scholar]
W. Wang, J. Zhang, C. Xu, Oscillating feedback micromixer: a short review, Chemical Engineering and Processing-Process Intensification 109812 (2024). [Google Scholar]
R. Ghosh, A. Arnheim, M. van Zee, L. Shang, C. Soemardy, R.-C. Tang, M. Mellody, S. Baghdasarian, E. Sanchez Ochoa, S. Ye, Lab on a particle technologies, Anal. Chem. 96 (2024) 7817–7839. [Google Scholar]
A. Ebrahimi, K. Icoz, R. Didarian, C. Shih, E.A. Tarim, B. Nasseri, A. Akpek, B. Cecen, A. Bal-Ozturk, K. Güleç, Molecular separation by using active and passive microfluidic chip designs: a comprehensive review, Adv. Mater. Interfaces 11 (2024) 2300492. [Google Scholar]
M. Hagedorn, Z. Varga, R.B. Walter, T.R. Tiersch, Workshop report: cryopreservation of aquatic biomedical models, Cryobiology 86 (2019) 120–129. [Google Scholar]
Y. Liu, M. Chesnut, A. Guitreau, J. Beckham, A. Melvin, J. Eades, T.R. Tiersch, W. T. Monroe, Microfabrication of low-cost customisable counting chambers for standardised estimation of sperm concentration, Reprod. Fertil. Dev. 32 (2020) 873–878. [Google Scholar]
J.A. Belgodere, Y. Liu, E.L. Reich, J. Eades, T.R. Tiersch, W.T. Monroe, Development of a single-piece sperm counting chamber (SSCC) for aquatic species, G 7 (2022) 231. [Google Scholar]
Beckham, F. Alam, V. Omojola, T. Scherr, A. Guitreau, A. Melvin, D.S. Park, J. W. Choi, T.R. Tiersch, W. Todd Monroe, A microfluidic device for motility and osmolality analysis of zebrafish sperm, Biomed. Microdevices 20 (2018) 67. [Google Scholar]
D.S. Park, R.A. Egnatchik, H. Bordelon, T.R. Tiersch, W.T. Monroe, Microfluidic mixing for sperm activation and motility analysis of pearl Danio zebrafish, Theriogenology 78 (2012) 334–344. [Google Scholar]
D.S. Park, C. Quitadamo, T.R. Tiersch, W.T. Monroe, Microfluidic mixers for standardization of computer-assisted sperm analysis, Cryopreservation in Aquatic Species 2 (2011) 261–272. [Google Scholar]
S. Razavi Bazaz, N. Kashaninejad, S. Azadi, K. Patel, M. Asadnia, D. Jin, M. Ebrahimi Warkiani, Rapid softlithography using 3D-printed molds, Advanced Materials Technologies 4 (2019) 1900425. [Google Scholar]
R. Rahul, N. Prasad, R.R. Ajith, P. Sajeesh, R.S. Mini, R.S. Kumar, A mould-free soft-lithography approach for rapid, low-cost and bulk fabrication of microfluidic chips using photopolymer sheets, Microfluid. Nanofluid. 27 (2023) 78. [Google Scholar]
C. He, S. Li, B. Jiang, F. Chen, W. Hu, F. Deng, Surface hydrophobicity and guest permeability in polydimethylsiloxane-coated MIL-53 as studied by solid-state nuclear magnetic resonance spectroscopy, ACS Appl. Mater. Interfaces 15 (2023) 37936–37945. [Google Scholar]
J. Lee, J. Kim, H. Kim, Y.M. Bae, K.-H. Lee, H.J. Cho, Effect of thermal treatment on the chemical resistance of polydimethylsiloxane for microfluidic devices, J. Micromech. Microeng. 23 (2013) 035007. [Google Scholar]
A.E. Lenhart, R.T. Kennedy, Evaluation of surface treatments of PDMS microfluidic devices for improving small-molecule recovery with application to monitoring metabolites secreted from islets of Langerhans, ACS Measurement Science Au 3 (2023) 380–389. [Google Scholar]
A. Mata, A.J. Fleischman, S. Roy, Characterization of polydimethylsiloxane (PDMS) properties for biomedical micro/nanosystems, Biomed. Microdevices 7 (2005) 281–293. [Google Scholar]
A.-G. Niculescu, C. Chircov, A.C. Bîrca, ˘ A.M. Grumezescu, Fabrication and applications of microfluidic devices: a review, Int. J. Mol. Sci. 22 (2021), https://doi.org/10.3390/ijms22042011. [Google Scholar]
M.W. Toepke, D.J. Beebe, PDMS absorption of small molecules and consequences in microfluidic applications, Lab Chip 6 (2006) 1484–1486. [Google Scholar]
W.M. Childress, Y. Liu, T.R. Tiersch, Design, alpha testing, and beta testing of a 3-D printed open-hardware portable cryopreservation device for aquatic species, J. Appl. Aquac. 35 (2023) 213–236. [Google Scholar]
Y. Liu, M. Eskridge, A. Guitreau, J. Beckham, M. Chesnut, L. Torres, T.R. Tiersch, W.T. Monroe, Development of an open hardware 3-D printed conveyor device for continuous cryopreservation of non-batched samples, Aquac. Eng. 95 (2021) 102202. [Google Scholar]
A. Amini, R.M. Guijt, T. Themelis, J. De Vos, S. Eeltink, Recent developments in digital light processing 3D-printing techniques for microfluidic analytical devices, J. Chromatogr. A 463842 (2023). [Google Scholar]
M.J. Schwing, Y. Liu, J.A. Belgodere, W.T. Monroe, T.R. Tiersch, A. Abdelmoneim, Initial assessment of the toxicologic effects of leachates from 3-dimensional (3-D) printed objects on sperm quality in two model fish species, Aquat. Toxicol. 256 (2023) 106400. [Google Scholar]
N.C. Zuchowicz, J.A. Belgodere, Y. Liu, I. Semmes, W.T. Monroe, T.R. Tiersch, Low-cost resin 3-D printing for rapid prototyping of microdevices: opportunities for supporting aquatic germplasm repositories, G 7 (2022), https://doi.org/10.3390/fishes7010049. [Google Scholar]
R. Zhou, M. Versace, B. Boisnard, M. Gomez-Castano, C. Viana, J.-L. Polleux, A.- L. Billabert, Thermal stability of SU-8 low-loss optical coupling interconnects at 850 nm, IEEE Photon. Technol. Lett. 36 (2023) 159–162. [Google Scholar]
D. Qin, Y. Xia, G. Whitesides, Soft lithography for micro- and nanoscale patterning, Nat. Protoc. 5 (2010) 491–502. [Google Scholar]
S.M. Montgomery, F. Demoly, K. Zhou, H.J. Qi, Pixel-level grayscale manipulation to improve accuracy in digital light processing 3D printing, Adv. Funct. Mater. 2213252 (2023). [Google Scholar]
S. Aati, Z. Akram, B. Shrestha, J. Patel, B. Shih, K. Shearston, H. Ngo, A. Fawzy, Effect of post-curing light exposure time on the physico–mechanical properties and cytotoxicity of 3D-printed denture base material, Dent. Mater. 38 (2022) 57–67. [Google Scholar]
D. Wu, Z. Zhao, Q. Zhang, H.J. Qi, D. Fang, Mechanics of shape distortion of DLP 3D printed structures during UV post-curing, Soft Matter 15 (2019) 6151–6159. [Google Scholar]
A.R. Renner, E. Winer, Exploring print setting tradeoffs to improve part quality using a visual thermal process simulation, Adv. Eng. Softw. 173 (2022) 103243. [Google Scholar]
B.N. Dhanunjayarao, N.V.S. Naidu, Assessment of dimensional accuracy of 3D printed part using resin 3D printing technique, Materials Today: Proceedings 59 (2022) 1608–1614. [Google Scholar]
S. Martínez-Pellitero, M.A. Castro, A.I. Fernandez-Abia, S. Gonzalez, E. Cuesta, Analysis of influence factors on part quality in micro-SLA technology, Procedia Manuf. 13 (2017) 856–863. [Google Scholar]
J.S. Shim, J.-E. Kim, S.H. Jeong, Y.J. Choi, J.J. Ryu, Printing accuracy, mechanical properties, surface characteristics, and microbial adhesion of 3D-printed resins with various printing orientations, J. Prosthet. Dent. 124 (2020) 468–475. [Google Scholar]
H.-Y. Chang, W.-L. Kao, Y.-W. You, Y.-H. Chu, K.-J. Chu, P.-J. Chen, C.-Y. Wu, Y.- H. Lee, J.-J. Shyue, Effect of surface potential on epithelial cell adhesion, proliferation and morphology, Colloids Surf. B Biointerfaces 141 (2016) 179–186. [Google Scholar]
C.A. Graham, H. Shamkhalichenar, V.E. Browning, V.J. Byrd, Y. Liu, M. T. Gutierrez-Wing, N. Novelo, J.-W. Choi, T.R. Tierschc, A practical evaluation of machine learning for classification of ultrasound images of ovarian development in channel catfish (Ictalurus punctatus), Aquaculture 552 (2022), https://doi.org/10.1016/j.aquaculture.2022.738039. [Google Scholar]

Reversible electrochemical pH modulation in thin-layer compartments using poly(aniline-co-o-aminophenol)

January 16, 2026July 15, 2024 by CadWorks3DAdmin

Academic Article

Reversible electrochemical pH modulation in thin-layer compartments using poly(aniline-co-o-aminophenol)

by Alexander Wiorek, Chen Chen, María Cuartero and Gastón A. Crespo

Abstract: The analysis of many environmental and clinical samples requires the modification of the original pH, which is conventionally carried out by manual/automatic addition of acid, base, or buffering reagents. In the case of decentralized measurements, often, this approach is not plausible. Instead, reagentless alternatives, such as electrochemically activated in-situ pH adjustments, are suitable. Herein, we present a method for electrochemical, reversible pH modulation of thin-layer samples (<100 µm thickness) using the co-polymer poly(aniline-co-o-aminophenol) (PANOA). The PANOA’s electropolymerization strategy was optimized considering the proton exchange properties in the final material. Thus, limiting the maximum anodic potential to 0.85 V and with the number of cyclic scans being ≤150), the optimal pH modulation capabilities were observed. The reversible proton exchange properties of PANOA were quantified by monitoring the pH inside the thin-layer sample (volume of 0.6 µL), which was defined by a 3D-printed microfluidic cell and a pH-sensor placed in a face planar configuration to the PANOA film. A pH value in the range of 2–4 can repeatably be reached in the samples in 3 min, purely by an electrochemical means and without the addition of external reagents. The concept has been demonstrated to acidify samples at environmental pH (artificial samples and Seawater). The outcomes suggest that the family of polyaniline-co-polymers are interesting to be explored and utilized for electrochemically based pH modulation strategies, if careful considerations are taken regarding their electropolymerization process. Overall, such materials could contribute to the development of continuous, decentralized measuring devices requiring acidification for the formal detection of environmental markers, such as nutrients, carbon species speciation and alkalinity, among others.

Keywords: poly(aniline-co-o-aminophenol); polyaniline; PH-modulation; thin-layer electrochemistry; microfluidics

We kindly thank the researchers at KTH Royal Institute of Technology for this collaboration, and for sharing the results obtained with their system.

1. Introduction

The modulation of sample pH is essential in many applications, ranging from analysis and separation of aminoacids [1] to the potentiometric determination of water hardness [2]. In environmental analysis, the use of acid to lower sample pH is necessary before the detection of relevant markers. For example, the detection of phosphate is inherently reliant on the formation of a complex with molybdate (phosphomolybdate), which can be detected optically and/or electrochemically at pH 2 or lower [3], [4]. Another environmentally relevant parameter that requires acid addition (i.e., acid-base titration) for its quantification is alkalinity (typically a pH of 3–4.5 is needed) [5]. Overall, the requirement for acid addition restricts these measurements to centralized laboratories, although on-site or in-situ operation using automatic analyzers has become popular over recent years. While these have helped increasing the frequency of data acquisition for phosphate and alkalinity [6], [7], [8], [9], macronutrients [10] and formaldehyde detection [11], they require waste storage tanks and sometimes, sample dilution before/during the operation, which may cause additional uncertainties in the provided outcomes. To circumvent these drawbacks, light- or electrochemical-driven methodologies for pH modulations could be beneficial instead of reagents additions.

It is possible that the most explored method for pH modulation is the one based on water-splitting at an electrode surface, generating either protons or hydroxide ions [12]. Utilizing a constant applied current or potential, which tends to be very high (ca. 1 mA/cm² and 2 V), in a thin-layer sample (thickness <100 µm), the pH can be exhaustively shifted in the entire volume. The concept was demonstrated by Van der Schoot et al., showing titrations where the generated charge was used to calculate the moles of acid/base delivered to the sample [12], [13]. Later on Steininger et al. reported on the determination of the sample buffer capacity by measuring the dynamic pH change in the sample generated from water splitting at both the working and counter electrodes, using chemical imaging [14]. Overall, water splitting is indeed an efficient method for adjusting sample pH, but the drastic conditions it requires could lead to undesired side reactions at the electrode surface.

Solid materials presenting proton-coupled redox reactions have proven certain advantages over water splitting for acidification purposes because proton delivery to the sample is possible at low potentials. Balakrishnan et al. showed that 4-aminothiolphenol covalently attached to an electrode surface could shift the pH under electrochemical control in nL volumes for up to 100 reversible cycles [15]. The oxidation of the compound involves a proton exchange: amines are converted into imines in the potential window from 0.65 to 0.8 V. Yet, the proton release capacity shown by this approach is expected to be enough only for nL-volume samples, because of the finite amount of active compound that can be attached to the electrode’s surface. Our group has demonstrated that polyaniline (PANI) is an excellent material for electrochemical pH modulation at even softer potentials than amines (0.2–0.4 V) [16], [17]. This has been used for thin-layer samples acidification coupled to the detection of alkalinity and phosphate [17], [18], [19], as well as chemical imaging of buffer capacity [20], dissolved inorganic carbon (DIC), and carbonate alkalinity [21] in environmental samples. In these works PANI showed suitability for being used for several weeks as long as it is regenerated in acidic solution (10 mM H₂SO₄) between each acidification usage [22]. Notably, the redox activity and proton release from PANI is not reversible above pH 4–5.

Other PANI-like materials (i.e., co-polymers or polymers containing aniline backbone) could be also of interest, which may be electroactive at higher pH. For example, poly(aniline-co-o-aminophenol) (PANOA), a co-polymer of aniline and o-aminophenol, is an electrochemically active material presenting reversibility at environmental pH values. Mu and coworkers were pioneers investigating the PANOA electropolymerization [23], reporting the structure and redox mechanism as presented in Scheme 1a. The o-aminophenol unit in the PANOA structure allows for reversible redox and proton exchange in solutions up to a pH of 9–10 [23], [24]. In addition, it exhibits anion-exchange properties [25], [26]. The reversibility of PANOA has been attributed to the conversion between the phenolic group (reduced state) and the quinone group (oxidized state) in its backbone. Another suggestion of the PANOA structure and its redox mechanism found in the literature is presented in Scheme 1b, where each o-aminophenol unit is separated by larger PANI-like segments [24], [27]. Here the proton exchange is additionally associated to these segments in the co-polymer. However, because of the poor reversibility of the proton exchange from the amine-imine redox reaction, less efficient proton exchange reversibility at environmental pH is expected than that of the phenol-quinone reaction.

Interestingly, PANOA’s redox activity at physiological and environmental pH values has been crucial for its implementation as a transducer in biosensors and heavy metal sensors [28], [29], among others. In these cases, PANOA was claimed to be superior to PANI in terms of redox activity at physiological pH. However, to the best of our knowledge, the use of PANOA for pH modulation of samples has not been investigated yet, in contrast to PANI [16], [17], [18], [19], [20], [21]. Herein, the use PANOA for reversible proton exchange in thin-layer samples is investigated. PANOA was characterized with both spectroelectrochemistry and thin-layer electrochemistry coupled to in-situ pH sensing. Our experiments revealed the analytical potential regarding further sensing in artificial and real samples that needs for acidification prior to such a detection.

2. Experimental section

2.1. The microfluidic cell and experimental setup of pH modulation in thin-layer samples

The microfluidic thin-layer cell was designed in AutoCAD 2022 (Autodesk) and printed using a Profluidics 285D 3D-printer and Clear Microfluidics Resin V7.0a (CADworks3D). The electrode configuration is presented in Fig. 1a, with a cell schematic shown in Fig. 1b. The cell has an inlet and outlet to allow sample exchange. The outlet additionally contains a pseudo reference/counter electrode (Ag/AgCl wire, RE₁/CE₁) that is to be connected to the potentiostat. The cell includes a second reference electrode (Ag/AgCl wire, RE₂) connected to the potentiometer. Such electrode was inserted into a separate opening present in the cell (the hole with the wire inside was sealed with the 3D-printing resin by curing it with UV-light for 30 s). The center of the cell had a 10-mm-diameter hole to allocate the two electrodes functioning as the acidification actuator (WE₁: PANOA) and the pH sensor (WE₂: PANI), creating a thin-layer gap sandwiched between them (<100 µm in thickness). Thus, two Au electrodes with a diameter of 3 mm (model 6.09395.034, Metrohm Nordic) were differently modified with PANOA (the working electrode, WE₁, in the potentiostat) and PANI (the working electrode, WE₂, in the potentiometer), being positioned in such a way that their areas faced each other, being separated by a 90-µm-thick double adhesive tape (RS Online, stock no: 555–033) that was placed on the edges of the electrodes.

Apparatus Used

Clear Microfluidic Resin

Learn More

ProFluidics 285D

Learn More

As the PANOA never exceeded a thickness of 10 µm (lower limit of the caliber used for the measurement) this spacer was used for all experiments involving PANOA-based acidifications, providing a configuration with a sample volume of ca. 0.6 µL. The optimized PANOA (as described below; Table 1) was electropolymerized over three different potential windows in succession for cyclic voltammetry (CV); –0.2–1.1 V, –0.2–0.9 V and –0.2–0.8 V, all with a scan rate of 60 mV/s. The PANI-based pH sensor was prepared as optimized elsewhere (–0.05–1.05 V for 10 scans at 100 mV/s) [17]. The pH sensor was calibrated using standards of pH 9.0–1.8 (0.1 M NaCl as background electrolyte; details in the supporting information) inside the microfluidic cell. A typical calibration profile is shown in Figure S1.

Notably, the sample to be acidified is sandwiched between the two electrodes, being confined to a thin-layer domain. This guarantees no mass transport limitation along the sample thickness [30]. In this context, our group has published a model based on the finite element approach to describe the electrochemically controlled release of ions (e.g., H⁺) from a redox-active film (such as PANOA or PANI) into a sample confined to a thin-layer spatial domain [19]. Calculations were found to rather agree with the experimental results regarding the sample thickness influence on the mass transport regime. On the other hand, the pH achieved in the sample plug after the acidification (i.e., once the needed applied potential stops) may be affected by the lateral diffusion of the rest of the sample contained in the microfluidic system. To confirm that this was not the case, an extra step consisting of the pH recording for some time after acidification ceased was added to the experimental protocol (see below). It is here anticipated that we observed that the pH value achieved through the acidification was maintained.

2.2. Instrumentation

All the electrochemical experiments were performed using a PGSTAT204 Autolab potentiostat (Metrohm Nordic AB) and the Nova 2.1.6 software. The pH sensor was operated by measuring the electromotive force (emf) with a high input impedance (10¹⁵Ω), Lawson labs EMF16 Interface (Lawson Laboratories, Inc.). In the spectroelectrochemistry experiments, absorbance spectra were collected using an Avantes ULS2048CL spectrometer with AvaLight-DHc as light source (Avantes) coupled with fiber optics (M92L01, Thorlabs). The pH of the standards used for the calibration of the pH sensor were adjusted using 1 M HCl or 1 M NaOH and a 914 pH/Conductometer from Metrohm (6.0228.000).

3. Results and discussion

The concept and working mechanism herein investigated for PANOA-based acidification of thin-layer samples is illustrated in Fig. 2. The principle is based on a thin-layer sample sandwiched between the source of protons (PANOA) and a potentiometric pH sensor, everything configured in a microfluidic cell. The sample is introduced by means of a peristaltic pump. When the sample plug enters the thin-layer space between the PANOA and the pH sensor, the pump is stopped. The pH of the sample is monitored by the sensor, providing a value representing the initial sample pH. The pH is expected to be stable and determined by the buffer(s) concentration(s) (Fig. 2, left). Then, when PANOA is electrochemically activated by an applied potential, it converts into its higher oxidation state, which involves transforming the phenolic groups of the o-aminophenol units in the co-polymer backbone into quinones [23], [28]. Additionally, some conversion of amines into imines may also occur [24]. Both of these structural changes trigger a release of protons from the PANOA to the thin-layer sample, which converts any base (B^–) into its conjugated acid (HB), breaking first the buffer capacity and resulting later in the decrease of the sample pH (Fig. 2, right).

Because of the confined space for the sample, the protons will be largely retained as they can only laterally diffuse in the thin layer, consequently maintaining the lowered sample pH for extended times even after the potential step is finished. Then, by applying a negative potential step to the PANOA, protons in the sample are expected to be re-inserted into the polymer backbone based on its reversible redox mechanism. Thus, in an ideal case, proton exchange with the sample is achievable over numerous cycles. This process, which involves changes in the sample pH, can be followed by the pH sensor placed on the opposite side of the thin layer sample and facing the PANOA. Importantly, once the working mechanism underlying the reversible PANOA-based sample acidification is demonstrated, the pH sensor can be replaced with another sensor (i.e., electrochemical or optical) capable of measuring a pH sensitive analyte, such as a CO₂ optode for DIC detection [21], voltammetric sensor for phosphate detection [18], and potentiometric sensors for anions [31], among others.

According to previous findings by Mu et al. and Holze et al., PANOA’s final structure depends largely on the ratio of monomers (aniline and o-aminophenol) involved in its synthesis [23], [24], [32], [33]. As such, if the aniline:o-aminophenol ratio is too high, a more PANI-like structure is expected because of low accessibility of o-aminophenol in the monomer solution [32]. However, when the aniline:o-aminophenol concentration ratio becomes too low, it inhibits the growth of PANOA [23]. Herein, a ratio of 20:1 aniline:o-aminophenol (200:10 mM) in 0.6 M H₂SO₄ was used for the PANOA electropolymerization, which has been demonstrated to be an adequate condition for producing films that are electroactive at environmental pH values [23]. First, we set out to characterize PANOA using spectroelectrochemical studies, followed by tuning the voltammetric parameters according to its proton exchange properties in the thin-layer cell.

3.1. Spectroelectrochemistry investigation of PANOA

To verify the formation of the PANOA, this was electropolymerized on a transparent ITO electrode while simultaneously recording the absorbance. For this purpose, the experimental setup was based on the spectroelectrochemical cell used in our previous works [16], [34]. Initially, the electropolymerization was performed by CV (from –0.2–1.1 V at 60 mV s^–1 for 20 scans). Some selected scans are presented in Fig. 3a. Notably, the results are analogous to previous studies reporting PANOA formation at the same potential window [23], [35], implying the successful formation of the co-polymer.

Several waves can be observed in the anodic part. Two peaks at 0.25 and 0.38 V, which have been ascribed to the first oxidation state of the polymer chain during its growth [23], [36], including anion insertion into the film [26]. Then, a small peak is observed at ca 0.7 V during the first scan, which corresponds to the oxidation of the phenolic group in acidic conditions [23]. Also, two partly overlapping peaks are found at 0.8–0.9 V, which are ascribed to the second oxidation state of the polymer [26] as well as oxidations of the monomers [23], [26]. After eight scans, two peaks appeared at 0.55 and 0.63 V. Despite these being observed in previous works [23], [37], the origin is not clear yet, resulting in incomplete explanations. Notably, by analogy to PANI and considering that both polymers (PANI and PANOA) present a similar structure, this peak likely originates from degradation products in the hydrolysis of imines in the polymer backbone [38]. In the cathodic part, in the potential window from 0.8 to 0.9 V (i.e., in the region of the second oxidation state of the polymer), there is a small peak at approximately 0.65 V. Then, the first oxidation state relates to three peaks in the potential range from –0.1–0.3 V, instead of the two peaks presented in the anodic part.

The spectra connected to the anodic part of the final scan of the electropolymerization process are presented in Fig. 3b. Different absorbance bands are absorbed in the region from 380 to 530 nm, with small increases in magnitude with the applied potential. This effect is in accordance with previous results about PANOA electropolymerization [33], confirming the formation of the co-polymer. To further analyze the spectroelectrochemical results, the change in absorbance at the absorbance maximum (420 nm) during the anodic scans at PANOA’s reduced state (0 V) and its fully oxidized state (1.1 V) versus the number of scans are presented in Fig. 3c. It was observed that the absorbance at 420 nm increased until the 10th scan, whereafter it remained almost constant. The full spectra during the growth of the polymer are provided in Figure S2. Additionally for each individual scan, the oxidized state always presented a higher absorbance. This behavior contrasts with that found for the current in the voltammetric peaks, where all peaks gradually increased over the 20 scans. Overall, the result in the absorbance suggested that further growth of the molecular structure corresponding to the 420 nm band does not occur after the 10th scan. Interestingly, this ceased increase in absorbance coincides with the emergence of the anodic peak at 0.55 V.

3.2. Optimization of PANOA fabrication via electropolymerization

After the spectroelectrochemical measurements and for further experiments, the ITO electrode substrate was replaced by the Au electrode tip to improve the mechanical stability of the created film. The total number of scans of in the electropolymerization process was increased from 20 (for ITO) to 100–250 for the Au-electrodes. Thus, thicker films were expected in the Au than in the ITO substrate, aiming for a more efficient acidification strategy (i.e., the film will contain a higher number of protons to be delivered from the PANOA to the sample) considering the proof of concept in real water samples [16], [17]. Additionally, having identified in the spectroelectrochemistry results that there is a peak at 0.55 V surely related to the polymer degradation, an even more improved efficiency for the PANOA-sample proton exchange was expected by decreasing the maximum anodic potential used in the CV during electropolymerization. Effectively, thanks to preliminary experiments based on PANI electropolymerization by decreasing the upper limit of the CV potential window from 1.2 to 0.75 V (Figure S3a), it was found out that lowering the anodic potential below 0.9 V translated into the disappearance of the peak at 0.55 V with consecutive scans (Figure S3b).

Next, a systematic study was performed to understand if avoiding the degradation peak at 0.55 V led to an improved electrochemical performance for PANOA films. In essence, we investigated the effect of the upper potential in the growth window for PANOA electropolymerization while keeping constant the scan rate (60 mV/s) and the initial potential (–0.2 V) on the obtained CV (i.e., number of peaks and the related current). The experimental setup used was based on a three-electrode configuration in a beaker. The electropolymerization conditions are listed in Table 1 and the generated PANOA films were classified as PANOA types I, II and III. As justified below, the synthesis of PANOA type II included an initial nucleation step and that for PANOA type III an additional intermediate step that led to improved film growth. For PANOA type III, the number of scans in the growth part was changed, giving rise to the subclasses 1, 2 and 3. Overall, the CVs on Au shared similar peaks as those on the ITO-electrode (Fig. 4).

The trend over the first 45 scans of PANOA type I is illustrated in Fig. 4a. The following differences with the results observed for the ITO electrode. Were found. The first peak shifted to slightly lower potentials (0.23 V) and is initially lower in current magnitude than the second peak at 0.38 V. After 30 scans, the peak at 0.23 V becomes the most prominent. The second oxidation state associated to the two peaks at 0.63 V and 0.75 V, which also has been attributed to PANOA growth [23], exhibited a higher current than the first oxidation state (peaks at 0.23 and 0.38 V) until the 35th scan. The peak at 0.55 V did not appear until the 45th scan, indicating that no degradation occured before this point.

Fig. 4b presents the growth of PANOA type I from the 50th to the 200th scan. After the 65th scan the peaks at 0.23 and 0.38 V start to overlap and are no longer distinguishable, while the peaks at 0.63 and 0.75 V start to overlap with the peak at 0.55 V. Notably, PANOA of type I did not exhibit an increase in peak currents after 150 scans, whereafter the peaks shifted towards higher potentials. This can be likely scribed to two origins; i) an increase in film thickness providing additional resistance [17], and/or ii) the fact that the peak at 0.55 V becomes more pronounced between the 50th and 200th scans may result in film degradation and hence, impaired film conductivity [38], [39].

For PANOA of type II, the anodic limit was lowered to 0.9 V compared to type I, attempting to avoid the peak at 0.55 V. However, the peaks’ currents were found to increase very slowly (data not shown) using this potential window alone, indicating a slow film growth. Thus, a nucleation step of 20 scans between –0.2 and 1.1 V was implemented prior to the regular CV protocol. In such a case, the peak at 0.55 V was not observed (Figure S4a). Subsequently 180 scans between –0.2 and 0.9 V were adapted (Fig. 4c) for a total of 200 scans considering the entire procedure. This was still not sufficient to remove the peak at 0.55 V at the end of electropolymerization. However, PANOA II accumulated a final charge of 21.0 mC, which was more than a three-time-increase compared to PANOA Type I (6.1 mC). Thus, although the potential window was decreased, the amount of charge inserted into PANOA was increased.

Then, the potential window for the CV was fixed from –0.2–0.8 V for PANOA III-1, III-2, and III-3 after the established nucleation (from –0.2–1.1 V, 20 scans). Notably, preliminary tests applying this protocol provided a very slow growth of PANOA (i.e., slow current increase with subsequent scans). Thus, an intermediate step was implemented: from –0.2–0.9 V for 40 scans. Within the 40 scans, the peak at 0.55 V has not appeared yet (Figure S4b). Thus, the third strategy for the PANOA synthesis comprised three steps: i) nucleation step (CV from –0.2–1.1 V, 20 scans), ii) intermediate step (CV from –0.2–0.9 V, 40 scans), and iii) growth step (CV from –0.2–0.8 V for 190, 90 or 40 scans to obtain PANOA III-1, III-2, and III-3, respectively). The progression of the final step is presented in Fig. 4d. The main differences considering PANOA I and II (Fig. 4a-c) is that the peak for the second oxidation state (0.74 V) is still visible at the end of the electropolymerization.

This becomes more evident when comparing the last CV in the growth part for PANOA I, II and III (of any subclass), which are displayed in Fig. 4a-d and Figure S4c. It can be observed that PANOA III-1 (with the higher number of scans in the growth part) presented the peak at 0.74 V corresponding to the second oxidation state, and the peak at 0.33 V (Fig. 4d) did not shift as much as it does in PANOA I and II, which suggests a lower degree of degradation in PANOA III. Lowering the upper potential in the growth window further could perhaps be an option to avoid the peak at 0.55 V completely but would also decrease the rate of growth.

Comparing PANOA III-1, III-2 and III-3, the one that presented final current levels much closer to that displayed for PANOA I and II was PANOA III-1. Moreover, an extension in the number of scans from 200 to 250 for PANOA III-1 was considered because, unlike PANOA I and II, the voltammetric peaks were still increasing for each scan after 200 scans. This increase in current magnitude for each scan was suggesting that the film thickness was still growing. On the other hand, PANOA Type III-2 and III-3 presented lower peak currents at the end of their formation, but the shape of the CV are more similar to that claimed as characteristic for PANOA [23], [35], with the peak at 0.55 V being far less pronounced than in the case of all the other PANOA types.

Overall, it can be concluded that limiting the upper potential for the electropolymerization window avoids PANOA degradation to some extent but, at the same time, it restricts the rate of growth (considering the increase in peak currents). Accordingly, the optimal conditions to be selected are expected to be a compromise between degradation and growth effect providing the best proton exchange capacity that can be held by PANOA (i.e., sites to store protons in the co-polymer backbone).

3.3. Investigation of PANOA acidification capacity

To quantify the proton exchange properties of the different PANOA types and the reversibility of the process, we introduced the corresponding PANOA-Au electrode into the microfluidic thin-layer sample cell (Fig. 1), where the sample pH could be continuously monitored by the potentiometric pH sensor. The experimental protocol to induce electrochemical pH modulations in the sample together with the expected readout from the pH sensor are presented in Fig. 5a and b respectively. This consists of: (1) initial reading of the open circuit potential (OCP) by the potentiostat and the potentiometer with the pH sensor; (2) acidification step by applying the +0.4 V for 300 s to the PANOA electrode; (3) passive monitoring step at the OCP for 60 s; and (4) PANOA regeneration step at –0.2 V for 600 s. All these steps were performed in the same sample plug (i.e., with the pump turned off). Importantly, the acidification potential was selected to be 0.4 V to avoid being closer to potentials inducing secondary processes beyond proton release that may contribute to increase the charge released from PANOA and even its degradation. A similar strategy was followed in our previous papers involving PANI [16], [17].

Figure 5. Illustration of the protocol and outcomes for electrochemically modulated acidification and regeneration with the expected outcomes. (a) The protocol for the potentiostat. (b) The potential readout from the potentiometer. (c) The dynamic pH changes in the sample. In essence, the following steps and readouts apply to a general case: (Step 1, ca.60 s) OCP measurement with no change in the pH (i.e., constant EMF of the pH sensor). (Step 2, 300 s, acidification) Application of a constant positive potential with the simultaneous monitoring of the decreasing pH (i.e., increasing EMF in the pH sensor). (Step 3, 60 s, holding of the acidified pH in the solution) OCP measurement (ideally, the pH reached in step 2 is maintained). (Step 4, 600 s, regeneration) application of the regeneration negative potential for a double time of that in the acidification step, and with the simultaneous monitoring of increasing pH (i.e., decreasing EMF in the pH sensor), ideally increasing up to the initial pH of the sample.

Initially, when no potential is applied in step 1, a constant readout from the potentiometer was expected, which corresponds to the initial pH of the sample. Then in step 2, the PANOA is activated at the +0.4 V for 300 s, triggering the release of protons from the film to the sample. This causes a response from the pH sensor because of the pH shift in the thin-layer sample. Fig. 5c additionally illustrates the dynamic change in pH expected in the sample: from the initial pH to an acidified pH and finally coming back to the initial pH because of the regeneration step. In step 3, the applied potential was switched off and the pH readout was recorded in the thin-layer sample for 60 s, expecting the pH to be constant and equivalent to the level of acidification achieved by the actuator (always that there are not diffusion related uniformities in the process). In step 4, the PANOA is regenerated in the same sample plug: proton uptake from the acidified sample thanks to the application of –0.2 V for 600 s. This step causes the sample pH to return to its original value. Regarding the time of 600 s, according to the pH monitoring in numerous experiments, it was found that shorter times did not allow for a complete regeneration of the PANOA (meaning that the initial sample pH was not recovered), and longer times did not improve the regeneration efficiency. Then, after the regeneration, the peristaltic pump was turned on for 5 – 15 minutes to exchange the sample plug for further experiments.

The repeatability of the results provided by this protocol was first tested on PANOA type I in 0.1 M NaCl sample solution, accomplishing five consecutive cycles of pH modulation. Fig. 6a depicts the dynamic pH that was observed. The first three cycles reached an acidified pH of 3.11±0.11; whereafter, a decrease in the acidification capacity was observed (pH of 3.77 and 4.91 for the fourth and fifth acidifications respectively). Notably, this final pH was calculated as the average pH value shown during step 3 (i.e., no applied potential, just measuring the pH for 60 s after acidification, when the sample just holds the acidified pH). This criterion was used through the paper. The described behavior coincided with a progressive change in the current profiles associated to each proton release (Figure S5a, charge of 3.28±0.84 mC). Moreover, the regeneration step for taking up protons was also found to become less efficient in consecutive cycles (Figure S5b, charges of –3.84±0.95). These trends also manifested in an overall decrease in charge for both the delivery and regeneration steps (Figures S5c-d): for the second to fifth cycle of acidification, the charge was decreased from 4.25 mC to 1.80 mC and the decrease was 57.6 %; for the first to fifth cycles of proton uptake process, the charge was decreased from –4.82 mC to –2.2 mC, decrease in 54.3 %. Inspecting the pH acidification with the corresponding current profile for the first and fourth cycles, which are those displaying the biggest differences, some conclusions can be established. As observed in Fig. 6b, the first current transient displayed a Cottrell-like profile, while the fourth one displayed an initial fast decay and then the current slowly increased, therefore displaying a peak. For the first acidification, the decrease in pH was initiated within the first 15 s after activating the potential step. On the other hand, the fourth acidification displayed a considerable delay (ca. 100 s) before a decrease in pH was observed, which interestingly coincided with the peak in the current profile. This may imply that such a peak is closely related to the process of releasing protons from PANOA type I. However, because of the poor repeatability in both the acidification and charge delivery profile (i.e., electrochemical performance), we averted from further studies into this process, because PANOA type I was concluded not suitable for reversible pH modulations.

Figure 6. (a) Consecutive pH modulations using PANOA type I in 0.1 M NaCl solutions. The gray areas represent the times of acidification. (b) Chronoamperometric curves with overlapping pH-time profiles for the 1st and 4th acidification cycles. pH modulations were performed by applying +0.4 V for 300 s, followed by a 60 s waiting period where the pH was passively monitored and then, a regeneration step of –0.2 V for 600 s. The steps of the experimental protocol as described in Fig. 5 are indicated. Notably, the regeneration part has been shortened up for simplicity.

The results for the repeatability study of PANOA type II are presented in Fig. 7a, revealing ∆pH=2.82±0.06 for eight cycles (ΔpH=2.96±0.18 for the first 3 cycles). Effectively, the overall electrochemical performance was found to improve with respect to PANOA type I, but again showing some differences within increasing number of cycles (Figure S6). As observed in Figure S6a, the current for the first acidification is lower than subsequent pH modulations and displays no peak in the dynamic current profile. The increase in the current magnitude from the first acidification and the latter ones can be explained from a decrease observed in the OCP: 0.080 V for the first and –0.139±0.012 V for the subsequent cycles. In essence, because the potential step is larger from –0.139 V to 0.4 V than from 0.080, more charge is expected to be generated in the PANOA. Despite the mentioned differences in the current profiles, only small variations were found in the corresponding charges (4.69±0.13 mC, excluding the first acidification; Figure S6b). In addition, the pH measured in the regeneration step was found to always return to a pH very close to the initial sample pH, also displaying very similar current profiles and acceptable reproducibility in terms of charge (–4.89±0.20 mC, Figures S6c and S6d).

Figure 7. Successive cycles for sample acidification based on different types of PANOA. (a) PANOA type II in 0.1 M NaCl. (b) PANOA type III-1 in 0.1 M NaCl. (c) PANOA type III-2 in 0.5 mM NaHCO3 with 0.1 M NaCl as background electrolyte. pH modulations were performed by applying +0.4 V for 300 s, followed by a 60 s waiting period where the pH was passively monitored and then, a regeneration step of −0.2 V for 600 s. The gray areas represent the times of acidification.

With an acceptable electrochemical and pH modulating performance confirmed for 0.1 M NaCl solution, PANOA type II was further tested in a buffered solution (0.5 mM NaHCO₃/0.1 M NaCl). The operational performance in higher, buffered pH is interesting to be evaluated in terms of PANOA usability in environmental waters and physiological conditions. However, it was found that the pH modulation using PANOA type II was not reversible under these conditions (Figure S7). Triplicate measurements revealed a successful first pH modulation, shifting the pH from the initial value of 7.6 to below 4 at the end of the acidification. But, in subsequent pH modulations, the pH was unable to be shifted below 6.5. This motivated the development of the PANOA type III and its subclasses, to provide an electrochemically induced pH actuator that is reversible at higher pH values.

The repeatability of the pH modulation induced by PANOA type III-1 in 0.1 M NaCl was found to be higher and more efficient than PANOA type II when tested in unbuffered conditions. An acidification of ∆pH=3.12±0.19 over seven cycles (Fig. 7b), with charge deliveries of 6.43±0.53 mC and –6.82±0.49 mC for the acidifications and regenerations (Figure S8) were revealed. Moreover, this proper performance was additionally accompanied by an improved repeatability in 0.5 mM NaHCO₃ (with 0.1 M NaCl as background electrolyte), reaching final pH values of 3.4, 3.7 and 4.1 in consecutive acidifications (initial pH=7.5; Figure S9a). Yet, the decreasing trend within each subsequent acidification in unbuffered conditions motivated the development of the PANOA type III-2 and type III-3 subclasses, which showed excellent repeatability in buffered media (Fig. 7c and Figure S9b for type III-2 and type III-3, with average charges of 2.66±0.17 mC and 2.92±0.13 mC, respectively).

The overall improvement over all tested PANOA types is summarized in Fig. 8 for all tested samples; 0.1 M NaCl (blue), 0.5 mM NaCO₃^–, and seawater (discussed in detail in the next section). The corresponding pH values are provided in Table S1 in the Supporting Information. The horizontal lines indicate the average starting pH over all measurements for the corresponding samples, the bars extending from the starting pH give the magnitude of the acidification and their average final pH, and the error bars provided the standard deviations for the measurements. Indeed, the subclasses of both PANOA type III-2 and III-3 behaved roughly the same, where both provided improved acidification-capacities and repeatability compared to types I, II and III-1 for both unbuffered (Fig. 8, blue bars) and buffered samples (Fig. 8, orange bars). By comparing the preparation protocols (Table 1) and the CVs in Fig. 4 with the acidification results, the following can be concluded. By limiting the formation of the degradation peak at 0.55 V, the acidification capacity (i.e., decrease in pH) increases, and the repeatability (standard deviation) decreases indicating an improvement in reversibility of the proton release from the PANOA film.

3.4. pH modulation of seawater samples using PANOA as an electrochemically driven actuator

Considering the superior efficiency and reversibility of the proton exchange from the PANOA Type III-2 and Type III-3 compared to the other PANOA types, we further tested their performance in a seawater sample (collected at Torrevieja, Spain, Supporting Information). Three subsequent acidifications/regenerations were tested. Notably, the applied potential was changed to 0.45 V for acidification and –0.15 V for regeneration to adjust for the higher chloride concentration (ca. 0.6 M) in seawater, which shifts the reference potential by approximately 50 mV compared to the artificial samples containing 0.1 M NaCl as the background electrolyte (with or without buffer). The sample’s pHs measured before and after acidification are presented in Fig. 9. The dynamic pH-time profiles are additionally provided in Figure S10. As observed, the acidified sample presented a pH value that increased with the number of cycles, while the pH obtained after the regeneration decreased. Averages values of 3.19±0.67 and 3.67±0.35 were respectively found for PANOA Type III-2 and PANOA Type III-3 (yellow bars Fig. 8), indicating that PANOA III-2 has a slightly higher capacity for acidification than PANOA III-3. Overall, the proton exchange efficiency was mitigated with the number of cycles, an effect that was not noticed in the buffered samples used in the previous section. The higher complexity of the sample matrix together with the higher buffer capacity (alkalinity=2.47±0.04 mM, details in the Supporting Information) may indeed affect the PANOA’s proton exchange properties.

To extend the number of uses of the PANOA film, a more drastic regeneration procedure was investigated. Thus, after the three consecutive pulses, an acid solution (10 mM H₂SO₄/0.1 M NaCl) was pumped into the cell and a potential of –0.2 V was applied for 600 s. Thereafter, the seawater sample was re-introduced in the microfluidic cell, and four consecutive acidification-regeneration cycles were performed. The regeneration step in acid solution was found to significantly increase the acidification capacity of the PANOA films, which was more pronounced for PANOA Type III-2 than PANOA Type III-3, with final pHs of 2.2, 2.6, 2.9 and 3.2 versus 3.1, 3.5, 3.7 and 4.1.

These results are indeed relevant in view of the further application of the PANOA as an acidification actuator for environmental monitoring purposes. When acid-based regenerations are possible to implement, the use of PANOA Type III-2 is preferred over PANOA III-3 because of its higher acidification capacity. Moreover, it is possible to acidify the sample, run a sensor-based measurement (i.e., with the microfluidic cell integrating an analytical sensor instead of the pH one, which only means to monitoring the acidification-regeneration process in this study), and a regeneration step using the same sample for many successive cycles (at least 4). This option drastically reduces the need for acid solutions compared to traditional manual/automatized acid additions, empowering the greener perspective of the developed concept.

The frequency selected for introducing the acid for regeneration will then depend on the pH threshold required for the analytical application. Other option, which could be adopted depending on the final pH desired in the sample after acidification, is the sole use of the acidified sample for the regeneration. Some pH values to be considered as examples would be 4.5 to detect dissolved inorganic carbon [21], and 4.8 for the total sulfide detection [40], both attainable without using acids in the regeneration step. Although PANOA Type III-2 was found to be capable of lowering the pH more than PANOA Type III-3, the latter generates more charge, as obtained in the integration of the current-time curves (Figure S11). The higher charge output is likely because of that the thinner film avoids the degradation peak (0.55 V), which allows the polymer to maintain its conductivity and capacitance compared to thicker films. Additionally, it is likely that all charge generated does not correspond to protons, but also other processes such as anion-insertion into the PANOA [25], and different relaxation processes of the polymer [41].

3.5. Performance comparison between PANOA- and PANI-based acidification

The use of PANI has been recently demonstrated for the successful reagentless acidification of environmental samples. Thus, we performed a series of additional experiments to compare the performance of PANOA and PANI, investigating the reversibility and efficiency of the proton delivery/uptake from PANI. The PANI film was prepared with previously optimized voltammetric parameters (1.1 V for 10 s, followed by –0.35–0.85 V at 100 mV/s) [17], [18], [19], [20], [21], with 200 scans of electropolymerization (Figure S3b) resulting in a film thickness of ca. 350 µm. The spacer in the electrochemical cell was adjusted to assure that the thin-layer thickness would remain approximately constant and like that used for the PANOA films (details in Section 3 in the Supporting Information). Then, compared to already published investigations with PANI, the acid-based regeneration step normally conducted in 10 mM H₂SO₄ [17], [18], [19], [20], [21], was substituted using the acidified sample plug, to be comparable with the experimental conditions herein established for PANOA.

The results for 5 acidification-regeneration cycles are displayed in Fig. 10. A poor reversibility of the proton exchange was observed in 0.1 M NaCl sample solution, with a dramatic decrease in the acidification capacity over the five scans (final pH values of 3.3, 4.4, 4.8, 5.0 and 5.2). Accordingly, PANI exhibited excellent reversibility when regenerated in an acidic solution with a pH lower than 2 (10 mM H₂SO₄ solution) [16], [17]. The regeneration pH significantly influences the conversion of PANI to its protonated state. In this experiment, using a non-acid-based regeneration step, the lowest achieved pH was 3.3. This pH level was not sufficient to fully convert PANI to its protonated state, demonstrating that non-acid-based regeneration is not suitable to ensure successive acidification. On the other hand, the charge delivered by PANI was found to be ca. 10 times higher than for PANOA, and quite constant over all the pH modulation cycles (26.8±1.4 mC for the acidifications and –32.2±0.35 mC for the regenerations), as observed in Figure S12. This suggested that PANI would indeed be capable of delivering a superior charge compared to PANOA, although all of this charge is not expected to be correlated to the release of protons but also anion-exchange processes [17], [42]. But, if more protons are needed for the specific application, PANI can be grown thicker by increasing the number of scans or subsequently adjusting the potential window during its electropolymerization without observing the degradation peak observed for PANOA (0.55 V) [17], [21]. However, all these strategies will always accompanied by an acid-based regeneration step in contrast to PANOA.

Apparatus Used

Clear Microfluidic Resin

Learn More

ProFluidics 285D

Learn More

4. Conclusion

An electrochemical method for reversible pH modulations in thin layer samples (<100 µm thickness) was herein presented using the electropolymerized copolymer between o-aminophenol and aniline PANOA. The outcomes demonstrated the relevance of the electropolymerization strategy towards achieving optimal proton-coupled redox properties from the material. Specifically, by limiting the maximum anodic potential to 0.85 V and number of cyclic scans to ≤150, optimal pH modulation capabilities were observed, as quantified by in-situ potentiometric measurements of the pH inside the thin-layer sample (volume of 0.6 µL; defined by a 3D-printed microfluidic cell). It was found that the optimized PANOA film can consistently acidify samples (artificial and real seawater) to pH 2–4 within 3 min by purely electrochemical means and without addition of reagents to the sample. Such a pH is indeed suitable for the sensing of certain environmental markers, such as dissolved inorganic carbon and dissolved inorganic phosphate.

Supplementary material

Download

References

T. Ueda, R. Mitchell, F. Kitamura, T. Metcalf, T. Kuwana, A. Nakamoto, Separation of naphthalene-2,3-dicarboxaldehyde-labeled amino acids by high-performance capillary electrophoresis with laser-induced fluorescence detection, J. Chromatogr. A 593 (1) (1992) 265–274, https://doi.org/10.1016/0021-9673(92)80295-6.
M. Müller, M. Rouilly, B. Rusterholz, M. Maj-Zurawska, ˙ Z. Hu, W. Simon, Magnesium selective electrodes for blood serum studies and water hardness measurement, Microchim. Acta 96 (1) (1988) 283–290, https://doi.org/10.1007/BF01236112.
C. Warwick, A. Guerreiro, A. Soares, Sensing and analysis of soluble phosphates in environmental samples: a review, Biosens. Bioelectron. 41 (2013) 1–11, https://doi.org/10.1016/j.bios.2012.07.012.
J. Jonca, ´ V. Leon´ Fernandez, ´ D. Thouron, A. Paulmier, M. Graco, V. Garçon, Phosphate determination in seawater: Toward an autonomous electrochemical method, Talanta 87 (2011) 161–167, https://doi.org/10.1016/j.talanta.2011.09.056.
T. Michałowski, A.G. Asuero, New approaches in modeling carbonate alkalinity and total alkalinity, Crit. Rev. Anal. Chem. 42 (3) (2012) 220–244.
M.Z. Bieroza, A.L. Heathwaite, Seasonal variation in phosphorus concentration–discharge hysteresis inferred from high-frequency in situ monitoring, J. Hydrol. 524 (2015) 333–347, https://doi.org/10.1016/j.jhydrol.2015.02.036.
L. Qiu, Q. Li, D. Yuan, J. Chen, J. Xie, K. Jiang, L. Guo, G. Zhong, B. Yang, E. P. Achterberg, High-precision in situ total alkalinity analyzer capable of monthlong observations in seawaters, ACS Sens. (2023), https://doi.org/10.1021/acssensors.3c00552.
C. Sonnichsen, D. Atamanchuk, A. Hendricks, S. Morgan, J. Smith, I. Grundke, E. Luy, V.J. Sieben, An automated microfluidic analyzer for in situ monitoring of total alkalinity, ACS Sens. 8 (1) (2023) 344–352, https://doi.org/10.1021/acssensors.2c02343.
R.S. Spaulding, M.D. DeGrandpre, J.C. Beck, R.D. Hart, B. Peterson, E.H. De Carlo, P.S. Drupp, T.R. Hammar, Autonomous in situ measurements of seawater alkalinity, Environ. Sci. Technol. 48 (2014) 9573–9581, dx.doi.org/10.1021/es501615x.
M. Cuartero, G.A. Crespo, T. Cherubini, N.C.F. Pankratova, F. Massa, M.L.A. M. Tercier-Waeber, J. Scha¨fer, E. Bakker, In situ detection of macronutrients and chloride in seawater by submersible electrochemical sensors, Anal. Chem. 90 (2018) 4702–4710, https://doi.org/10.1021/acs.analchem.7b05299.
I.-Y. Eom, Q. Li, J. Li, P.K. Dasgupta, Robust hybrid flow analyzer for formaldehyde, Environ. Sci. Technol. 42 (4) (2008) 1221–1226, https://doi.org/10.1021/es071472h.
B. Van der Schoot, P. Bergveld, An IFSET-based microlitre titrator integration of a chemical sensor-actuator system. Sens. Actuators 8 (1985) 11–22, https://doi.org/10.1016/0250-6874(85)80020-2.
B. van der Schoot, P. van der Wal, N. de Rooij, S. West, Titration-on-a-chip, chemical sensor–actuator systems from idea to commercial product, Sens. Actuators B 105 (2005) 88–95.
F. Steininger, S.E. Zieger, K. Koren, Dynamic sensor concept combining electrochemical pH manipulation and optical sensing of buffer capacity, Anal. Chem. 93 (2021) 3822–3829, https://doi.org/10.1021/acs.analchem.0c04326.
D. Balakrishnan, J. El Maiss, W. Olthuis, C. Pascual García, Miniaturized control of acidity in multiplexed microreactors, ACS Omega 8 (8) (2023) 7587–7594, https://doi.org/10.1021/acsomega.2c06897.
A. Wiorek, M. Cuartero, R. De Marco, G.A. Crespo, Polyaniline films as electrochemical-proton pump for acidification of thin layer samples, Anal. Chem. 91 (2019) 14951–14959, https://doi.org/10.1021/acs.analchem.9b03402.
A. Wiorek, G. Hussain, A.F. Molina-Osorio, M. Cuartero, G.A. Crespo, Reagentless acid− base titration for alkalinity detection in seawater, Anal. Chem. 93 (2021) 14130–14137, https://doi.org/10.1021/acs.analchem.1c02545.
C. Chen, A. Wiorek, A. Gomis-Berenguer, G.A. Crespo, M. Cuartero, Portable all-inone electrochemical actuator-sensor system for the detection of dissolved inorganic phosphorus in seawater, Anal. Chem. 95 (8) (2023) 4180–4189, https://doi.org/10.1021/acs.analchem.2c05307.
A.F. Molina-Osorio, A. Wiorek, G. Hussain, M. Cuartero, G.A. Crespo, Modelling electrochemical modulation of ion release in thin-layer samples, J. Electroanal. Chem. 903 (2021) 115851, https://doi.org/10.1016/j.jelechem.2021.115851.
F. Steininger, A. Wiorek, G.A. Crespo, K. Koren, M. Cuartero, Imaging sample acidification triggered by electrochemically activated polyaniline, Anal. Chem. 94 (40) (2022) 13647–13651, https://doi.org/10.1021/acs.analchem.2c03409.
A. Wiorek, F. Steininger, G.A. Crespo, M. Cuartero, K. Koren, Imaging of CO2 and dissolved inorganic carbon via electrochemical acidification–optode tandem, ACS Sens. (2023), https://doi.org/10.1021/acssensors.3c00790.
E.M. Genies, A. Boyle, M. Lapkowski, C. Tsintavis, Polyaniline: a historical survey, Synth. Met. 36 (1990) 139–182.
S. Mu, Electrochemical copolymerization of aniline and o-aminophenol, Synth. Met. 143 (3) (2004) 259–268, https://doi.org/10.1016/j.synthmet.2003.12.008.
J. Zhang, D. Shan, S. Mu, Chemical synthesis and electric properties of the conducting copolymer of aniline and o-aminophenol, J. Polym. Sci. Part A: Polym. Chem. 45 (23) (2007) 5573–5582, https://doi.org/10.1002/pola.22303. DOI: https://doi.org/10.1002/pola.22303 (acccessed 2023/07/03).
Y. Zhang, S. Mu, B. Deng, J. Zheng, Electrochemical removal and release of perchlorate using poly(aniline-co-o-aminophenol), J. Electroanal. Chem. 641 (1) (2010) 1–6, https://doi.org/10.1016/j.jelechem.2010.01.021.
M. Liu, M. Ye, Q. Yang, Y. Zhang, Q. Xie, S. Yao, A new method for characterizing the growth and properties of polyaniline and poly(aniline-co-o-aminophenol) films with the combination of EQCM and in situ FTIR spectroelectrochemisty, Electrochim. Acta 52 (1) (2006) 342–352, https://doi.org/10.1016/j.electacta.2006.05.013.
Y. Zhang, F. Wen, Y. Jiang, L. Wang, C. Zhou, H. Wang, Layer-by-layer construction of caterpillar-like reduced graphene oxide–poly(aniline-co-o-aminophenol)–Pd nanofiber on glassy carbon electrode and its application as a bromate sensor, Electrochim. Acta 115 (2014) 504–510, https://doi.org/10.1016/j.electacta.2013.10.143.
S. Mu, Direct determination of arsenate based on its electrocatalytic reduction at the poly(aniline-co-o-aminophenol) electrode, Electrochem. Commun. 11 (7) (2009) 1519–1522, https://doi.org/10.1016/j.elecom.2009.05.050.
J. Zhang, D. Shan, S. Mu, Improvement in selectivity and storage stability of a choline biosensor fabricated from poly(aniline-co-o-aminophenol), FBL 12 (2) (2007) 783–790, https://doi.org/10.2741/2101.
M. Cuartero, G.A. Crespo, E. Bakker, Thin layer samples controlled by dynamic electrochemistry, Chimia 69 (4) (2015) 203, https://doi.org/10.2533/chimia.2015.203 (acccessed 2024/07/01).
N. Pankratova, M.G. Afshar, D. Yuan, G.A. Crespo, E. Bakker, Local acidification of membrane surfaces for potentiometric sensing of anions in environmental samples, ACS Sens. 1 (2016) 48–54.
A.-u-H.A. Shah, R. Holze, Spectroelectrochemistry of aniline-o-aminophenol copolymers, Electrochim. Acta 52 (3) (2006) 1374–1382, https://doi.org/10.1016/j.electacta.2006.07.040.
A.-u-H.A. Shah, R. Holze, In situ UV–vis spectroelectrochemical studies of the copolymerization of o-aminophenol and aniline, Synth. Met. 156 (7) (2006) 566–575, https://doi.org/10.1016/j.synthmet.2006.03.001.
Y. Liu, A. Wiorek, G.A. Crespo, M. Cuartero, Spectroelectrochemical evidence of interconnected charge and ion transfer in ultrathin membranes modulated by a redox conducting polymer, Anal. Chem. 92 (2020) 14085–14093, https://doi.org/10.1021/acs.analchem.0c03124.
S. Mu, Rechargeable batteries based on poly(aniline-co-o-aminophenol) and the protonation of the copolymer, Synth. Met. 143 (3) (2004) 269–275, https://doi.org/10.1016/j.synthmet.2003.12.009.
S. Mu, Poly(aniline-co-o-aminophenol) nanostructured network: electrochemical controllable synthesis and electrocatalysis, Electrochim. Acta 51 (17) (2006) 3434–3440, https://doi.org/10.1016/j.electacta.2005.09.039.
L.H. Mascaro, A.N. Berton, L. Micaroni, Electrochemical synthesis of polyaniline/poly-o-aminophenol copolymers in chloride medium, Int. J. Electrochem. 2011 (2011) 292581, https://doi.org/10.4061/2011/292581.
W.C.W.T.C. Chen, A. Gopalan, The inductive behavior derived from hydrolysis of polyaniline, Electrochem. Acta 47 (26) (2002) 4195–4206.
H. Zhang, H. Li, J. Wang, Capacitance fading induced by degradation of polyaniline: cyclic voltammetry and SEM study, Adv. Mater. Res. (2012), https://doi.org/10.4028/www.scientific.net/AMR.535-537.1205.
C. Szabo, A timeline of hydrogen sulfide (H2S) research: from environmental toxin to biological mediator, Biochem. Pharmacol. 149 (2018) 5–19, https://doi.org/10.1016/j.bcp.2017.09.010.
T.F. Otero, H. Grande, J. Rodríguez, A new model for electrochemical oxidation of polypyrrole under conformational relaxation control, J. Electroanal. Chem. 394 (1) (1995) 211–216, https://doi.org/10.1016/0022-0728(95)04033-K.
M.R. Nateghi, B. Savabieh, Study of polyaniline oxidation kinetics and conformational relaxation in aqueous acidic solutions, Electrochim. Acta 121 (2014) 128–135, https://doi.org/10.1016/j.electacta.2013.12.111.

On the Development of a New Flexible Pressure Sensor

January 16, 2026June 25, 2024 by CadWorks3DAdmin

On the Development of a New Flexible Pressure Sensor

by Florian Pistriţu ,2ORCID,Marin Gheorghe ,Marian Ion ,Oana Brincoveanu,Cosmin Romanitan ,Mirela Petruta Suchea Paul Schiopu andOctavian Narcis Ionescu

Abstract The rapid advancement of the Internet of Things (IoT) serves as a significant driving force behind the development of innovative sensors and actuators. This technological progression has created a substantial demand for new flexible pressure sensors, essential for a variety of applications ranging from wearable devices to smart home systems. In response to this growing need, our laboratory has developed a novel flexible pressure sensor, designed to offer an improved performance and adaptability. This study aims to present our newly developed sensor, detailing the comprehensive investigations we conducted to understand how different parameters affect its behaviour. Specifically, we examined the influence of the resistive layer thickness and the elastomeric substrate on the sensor’s performance. The resistive layer, a critical component of the sensor, directly impacts its sensitivity and accuracy. By experimenting with varying thicknesses, we aimed to identify the optimal configuration that maximizes sensor efficiency. Similarly, the elastomeric substrate, which provides the sensor’s flexibility, was scrutinized to determine how its properties affect the sensor’s overall functionality. Our findings highlight the delicate balance required between the resistive layer and the elastomeric substrate to achieve a sensor that is both highly sensitive and durable. This research contributes valuable insights into the design and optimization of flexible pressure sensors, paving the way for more advanced IoT applications.

Keywords: pressure sensors; flexible electronics; elasticity; wearability

We kindly thank the researchers at IMT Bucharest for this collaboration, and for sharing the results obtained with their system.

1. Introduction

The need for wearable pressure sensors has significantly increased in recent years due to their extensive application in the medical and Internet of Things (IoT) domains. These advanced sensors are crucial in monitoring and providing valuable data regarding human motion and physiological signals. They play an essential role in capturing human movement dynamics, including intricate muscle activities [1] and specific joint movements such as arm motion [2] and knee bending [2,3]. Additionally, these sensors are essential for monitoring vital signs such as pulse rate [4,5,6,7], respiration [1], and even phonation detection [2,8], which is the ability to detect and analyze sounds produced by the vocal cords. The information obtained from wearable pressure sensors is invaluable in numerous applications. In the medical field, these sensors contribute to more accurate diagnosis, patient monitoring, and the development of personalized treatment plans. For instance, tracking muscle movement and joint activities can aid in the rehabilitation process, allowing for real-time feedback and adjustments to therapy protocols. Monitoring pulse rate and respiration provides critical data for managing cardiovascular and respiratory conditions, ensuring timely interventions. Furthermore, in the realm of the IoT, these sensors enable enhanced human–machine interactions and smart environments, offering a seamless integration of technology into daily life. The manufacturing technology for flexible electronics, which underpins the development of wearable pressure sensors, is generally based on advanced printing techniques. Inkjet printing [9,10,11,12] is widely used due to its precision and ability to create the fine patterns necessary for high-performance sensor components. This method allows for the deposition of functional materials onto flexible substrates, forming the essential layers of the sensors. Screen printing [12,13,14,15], another prevalent technique, is known for its cost-effectiveness and scalability. It involves transferring ink through a mesh screen to create thicker and more durable sensor layers, suitable for various flexible electronic applications. The combination of these advanced manufacturing technologies ensures the production of reliable, flexible, and high-performance wearable pressure sensors. By leveraging inkjet and screen printing methods, manufacturers can achieve the necessary precision, efficiency, and adaptability required to meet the growing demands in medical and IoT applications. As a result, these sensors continue to evolve, providing increasingly sophisticated solutions for monitoring and enhancing human health and interaction with technology.

Flexible substrate-based pressure sensors have been extensively studied. For illustration, Congcong Yang et al. designed a film with a double surface structure to create a sandwich resistive pressure sensor, demonstrating excellent parameters including a high sensitivity (77.78 kPa−1, 24 Pa minimum detection), a wide detection range (0.024–230 kPa), a fast response time (30 ms), and a high reliability over 5000 repetitive cycles. Mohammed Mohammed Ali et al. [16] developed a printed strain sensor by screen printing a silver nanowire (Ag NW)/silver (Ag) flake composite on a flexible and stretchable thermoplastic polyurethane (TPU) substrate. The sensor was tested in two configurations, as follows: a straight line and a wavy line. The average resistance changes over 100 cycles were 104.8%, 177.3%, and 238.9%, and over 200 cycles were 46.8%, 141.4%, and 243.6% for elongations of 1 mm, 2 mm, and 3 mm, respectively. This sensor is intended for biomedical and civil infrastructure applications. Daniel Gräbner et al. [17] examined the influence of electrode geometry on the performance and cross-sensitivity to strain in screen-printed pressure sensors. Their findings indicate that pressure sensitivity increases with the number of interdigital electrodes, while temperature cross-sensitivity remains unaffected by electrode configuration. Dinesh Maddipatla et al. [18] developed a carbon nanotube (CNT)-based pressure sensor for pressure monitoring applications. The flexible capacitive pressure sensor was fabricated using screen printing technology, employing CNT ink for the top and bottom electrodes and PDMS for the non-conductive dielectric layer. The sensor exhibited an 8.2% change in capacitance at a maximum detectable pressure of 337 kPa, a 0.021% change in capacitance per kPa, and a correlation coefficient of 0.9971. Potential applications include sports, military, robotics, automotive, and biomedical fields. Sepehr Emamian et al. [19] successfully fabricated a fully printed piezoelectric-based touch sensor device on flexible polyethylene terephthalate (PET) and paper substrates using screen printing. The device utilized silver (Ag) ink and sandwiched a screen-printed polyvinylidene fluoride (PVDF) piezoelectric layer between the printed Ag top and bottom electrodes. Sensitivities of 1.2 V/N and 0.3 V/N, with correlation coefficients of 0.9954 and 0.9859, were achieved for the PET- and paper-based sensors, respectively. These sensors are suitable for both touch- and force-based applications. As shown by these examples, pressure sensors fabricated using this technology feature flexible PDMS substrates [20,21], contrasting with the rigid substrates used in traditional electronics. Screen printing remains the most prevalent method for printed electronics, as highlighted in the review by Saleem Khan et al. [22].

A new pressure sensor developed in our laboratories features a novel architecture, as illustrated in Figure 1. To enhance sensitivity, the substrate consists of two layers of PDMS with micro-pyramids, topped with a sensitive layer printed on Kapton (Goodfellow, Coraopolis, PA, USA).

Figure 1. The architecture of the proposed pressure sensor and details. (A) The substrate and (B) the KAPTON Foil with the printed sensor.

2. Materials and Methods

The micro-structured substrate used in the newly developed flexible pressure sensor was made of PDMS Sylgard 184 (DOW Chemical Company, Midland, MI, USA), to which was added 10% powder of aerogel (Powder aerogel < 0.125 mm–Green Earth Aerogels, Barcelona, Spain). The micro-structured substrate designed was of the micro-pyramidal type, with the pyramids having a base of 1500 µm and a height of 1060 µm. The distance between the pyramids was equal to the length of the pyramid base. To obtain the micro-structured substrate, the moulding technique has been used. The moulds were created via 3D printing using the µMicrofluidics M50 3D printer (CADWORKS3D, Toronto, ON, Canada), together with the Utility 6.0 software. The resin used for mould production was Master Mold for PDMS Devices—3D Printing Resin Photopolymer Resin (Composition: methacrylated oligomer, methacrylated monomer, photoinitiator, and additives, (CADWORKS3D).

The procedure of moulding the micro-structured substrate started with the mixing of the polymer Sylgard 184 with the hardening agent in a ratio of 10 to 1. Then, 10% of powder aerogel was added and the process of mixing continued for 10 min. The mixture was poured on the mould and was introduced into a vacuum oven VO400 (MEMMERT, Schwabach, Germany): 49 L; + 20… + 200 °C; 10…1100 mbar). The hardening treatment was conducted at a temperature of 100 °C for 50 min.

The sensing material deposition was performed by printing with the semi-automatic screen printer device LC-TA-250 Model (LC Printing Machine Factory Limited, Guangzhou-Foshan, China) on Kapton. The screen printing process was conducted by using two types of ink. The first type of screen printing ink (Ink1) is composed of 50% carbon micro-powder (graphite and C black), 10% polymer (binder), and the remaining 40% is a xylene-based solvent. The second version of the screen printing ink (Ink2) is composed of 50% carbon micro-powder (graphite and C black), 10% polymer (binder), 15% TiO2, and the remaining 25% is a xylene-based solvent (Chemical materials were sourced from Goodfellow, Coraopolis, PA, USA).

To ensure that the ink components were properly integrated in the printed layers, SEM was used to examine meander resistor prints on a flexible substrate and to analyze their morphology. For this, the Nova NanoSEM 630 device (FEI Company, Hillsboro, OR, USA) was used. The materials used in the inks and the printed resistor were investigated.

Figure 2 below shows some examples of SEM images of nanomaterials used in the ink formulations—Figure 2a–c, as well as the printed layers’ surfaces in Figure 2d,e. As one can see, the printed surfaces are uniform and smooth.

Figure 2. SEM images of nanomaterials used in the ink formulations, as well as the printed layers’ surfaces. (a) graphite; (b) Amorphous carbon; (c) TiO2; (d) Ink 1 on polyamide; (e) Ink 2 on Kapton substrate. Scale 10 µm.

The crystalline structure of nanomaterials used in ink formulations, as well as the printed resistor, were verified using X-ray diffraction analysis (XRD—X-ray Thin-film Diffraction System (XRD)/SmartLab (FN2670N)/ from Rigaku Corporation, Osaka, Japan) and it was observed that their structure is preserved in the printing process.

The pattern of piezoresistive sensors was kept identical for all types of flexible substrates used in printing, varying only the thickness of the ink layer. Thus, we were able to study both the influence of the ink layer thickness and the influence of the flexible substrate on the behaviour of the flexible piezoresistive sensor. After printing the piezo-resistor on the flexible substrate, a measurement of the resistor value was performed to establish the resistance value for each sample prepared. Following the resistance recording, the next step was to connect the terminals to the sensors. The process was carried out by using silver paste LOCTITE ABLESTIK 84-1LMI, (Henkel, Düsseldorf, Germany) and AWG 26 wires (Alpha Wire, Elizabeth, NJ, USA). The polymerization was finalized by applying a heat treatment at 125 °C for 30 min.

Method of Testing

Figure 3 presented below shows the measurement scheme that was used to test the resistive film assembly with the elastomeric substrate, for compression.

Figure 3. Diagram for tests of the resistive film assembly with elastomeric substrate, for compression.

The device used for the compression tests is the Mecmesin MultiTest 2.5I (Mecmesin, Slinfold, West Sussex, UK), together with the Emperor™ Force software v1.18 (Mecmesin, Slinfold, West Sussex, UK). MECMESIN is a single-column computerized tensile testing machine suitable both for metal and non-metallic materials testing. This machine adopts a mechanical and electrical integration design, mainly composed of a force sensor, a transmitter, a microprocessor, a load driving mechanism, and a computer. It has a wide and accurate range of loading speed and force measurement, the ability to measure and control the load, and the displacement has a high precision and sensitivity, but it can also carry out the automatic control test of constant speed loading and constant speed displacement.

Using Mecmesin MultiTest 2.5I, one can test various materials, semi-finished products, and finished products for their tensile strength, compressive strength, and elongation—elongation can be used for peeling, tearing, bending, compression, and other tests; these tests are suitable for metal, plastic, rubber, textiles, synthetic chemicals, wire and cable, leather, and other industries. Due to its capabilities to be programmed in a flexible manner, it could be used easily for cycle testing the pressure sensors. The measurements of the screen-printed resistors were performed with the FLUKE 8846A 6-1/2 Digit multimeter (Fluke, Everett, WA, USA), plus the Pomona 6730–Wide Jaw Kelvin Lead Set test leads (Pomona Electronics, Everett, WA, USA,). The power supply used is Agilent E3648A (Agilent, Santa Clara, CA, USA), a dual output power supply—two 8 V, 5 A or 20 V, 2.5 A. The resistor used in the electrical scheme for testing has a value of 6750 Ω and a tolerance of ±5%.

3. Results and Discussion

Several variants (replications) of meander resistor assemblies with flexible substrates have been produced. It was investigated whether the heat treatment at 125 °C for soldering the terminal wires has an influence on the value of the meander resistor or not.

The values obtained before applying the thermal treatment (named in the table column ‘Initial Value’) for the bonding of the terminal wires and the value obtained after the thermal treatment (named in the table column ‘Final Value’) are given in Table 1.

The thermal treatment at 125 °C was found to affect the resistors by reducing the thickness of the resistive layer, which, in turn, increased the resistance values. This may be a consequence of solvent evaporation. Therefore, it is recommended to use a silver paste that requires a significantly lower heat treatment.

The subsequent investigation focused on examining how bending affects the meander resistor, specifically at 45° and 90° angles. The results indicated that ABS (R3) exhibited the highest variation in resistance during bending, followed by PET (R2) and Kapton (R1). Table 2 provides detailed data, where “Material” denotes the substrate material of the meander resistor, “Initial Value” is the resistance value before bending, “Bending 45°” and “Bending 90°” indicate the resistance values at 45° and 90° bends, respectively, and “Total Variation” shows the percentage change in resistance at a 90° bend.

Table 2. Detailed data of resistance values at bending for resistors onto various substrates.

Three models were tested for assembling a resistive film with an elastomeric substrate: Model I: The resistive film was placed on top of the elastomeric substrate with the micro-pyramidal structure oriented tip up (see Figure 4a). Model II: The resistive film was placed above the elastomeric substrate with the micro-pyramidal structure oriented tip down (see Figure 4b). Model III: The resistive film was placed on an elastomeric substrate featuring an interlocked, paired pyramid structure (see Figure 4c).

Apparatus &Materials

Master Mold Resin

Learn More

M Series

Figure 4. (a) Model I assembly. (b) Model II assembly. (c) Model III assembly.

At the outset of the study, we evaluated the Model I assembly. The meander resistor printed on ABS and PET exhibited a non-linear response (see Figure 5a,b), leading to its exclusion from subsequent tests. This may be due to the thermal degradation of the polymers. However, the response of the meander resistor printed on Kapton is depicted in Figure 5c.

Figure 5. The non-linear response of the meander resistor. (a) Model R2 v2. (b) Model R3 v2. (c) Response given by three replicas of meander resistors printed on Kapton.

Cyclic compression tests were performed, 20 cycles each, on the meander resistors printed on Kapton (model R1 v2); the assembly mode used was Model I, and the tests were performed at a compression force of 50 N/cm2 and a speed compression of 1 mm/min. An example of the answer to these tests can be seen in Figure 6.

Figure 6. Piezoresistive response of R1 v2 tested in Model I assembly mode, 50 N/cm2 and 1 mm/min.

Compression tests were performed on the printed meander resistor on Kapton (model R1 v3), assembled in all three assembly modes, to study the influence of the assembly mode on the piezoresistive response of the printed meander resistor. The test conditions were as follows: resistor model R1 v3, 50 N/cm2, 1 mm/min, 20 cycles, maximal load applied for 1 s. The results can be seen in Figure 7.

Figure 7. (a) Piezoresistive response of resistor R1 v3, assembly mode Model I, 50 N/cm2, 1 mm/min. (b) Piezoresistive response of resistor R1 v3, assembly mode Model II, 50 N/cm2, 1 mm/min. (c) Piezoresistive response of resistor R1 v3, assembly mode Model III, 50 N/cm2, 1 mm/min.

The piezoresistive response of the meander resistor model R1 v3, to compression, is shown in Figure 8, under the following conditions: 50 N/cm2, 1 mm/min, for all three assembly models—Model I, Model II, and Model III.

Figure 8. The response given by the resistor R1 v3, for the assembly modes—Model I, Model II, and Model III.

After this comparative testing of the three assembly models, it turns out that the assembly models Model II and Model III offer the best piezoresistive response. Further on, the tests will be performed only for these two assembly models.

We optimized our technology by substituting the silver paste used for soldering wires. The original paste required a curing temperature of 125 °C, whereas the new paste cures at just 70 °C. We then compared the piezoresistive response of the meander resistor model R1 before and after this optimization. The tests were conducted on resistor model R1 v3 under conditions of 50 N/cm2, at a speed of 1 mm/min, using assembly models II and III. The results are shown in Figure 9a,b.

Figure 9. (a) Response of meander resistor printed on Kapton R1 v3, before optimization and after optimization, for the Model II assembly model. (b) Response of meander resistor printed on Kapton R1 v3, before optimization and after optimization, for the Model III assembly model.

In this study, we observed that optimization significantly improved the linearity of the piezoresistive response. Post-optimization, the Model III assembly exhibited superior piezoresistive characteristics. We also examined the piezoresistive response of a meander resistor printed on a flexible polyimide substrate. The tests revealed a completely non-linear response, leading to its exclusion from further studies.

The final phase of the research focused on the impact of the resistivity of screen printing inks on the piezoresistive response of meander resistors printed on Kapton R1 v3. We conducted comparative tests using two different inks—Ink1 and Ink2—under identical conditions, as follows: resistors printed on Kapton R1 v3, subjected to 50 N/cm2 pressure, and a 1 mm/min test speed, using both Model II and Model III assembly models, as shown in Figure 10. The results indicated that the sensors printed with Ink 1 and Ink 2 on Kapton exhibited nearly identical piezoresistive responses.

Figure 10. The piezoresistive response given by the meander resistor R1 v3, with Ink 1 and Ink 2, respectively. (a) Assembly Model II and (b) Model III.

Figure 11 show the piezoresistive response of the resistor printed on Kapton R1 v3, with Ink 2, to cyclic tests of 20 cycles, 50 N/cm2, 5 mm/min, for the two assembly models Model II and Model III, respectively.

Figure 11. Piezoresistive response of R1 v3, Ink 2, (a) Model II, (b) Model III assembly model at 50 N/cm2, 20 cycles, 5 mm/min.

The optimization process enhanced sensor performance by improving linearity. While the polyimide substrate was unsuitable due to non-linear responses, the Kapton R1 v3 substrate with both tested inks performed reliably, ensuring consistent piezoresistive responses under the specified test conditions.

4. Conclusions

In this study, the impact of various flexible substrates on the performance of a meander resistor and the effect of elastomeric substrates on a proposed flexible pressure sensor were explored. The findings of this study revealed that the meander resistor, when deposited on a flexible Kapton substrate, demonstrated superior linearity in response. Further investigations indicated that the optimal elastomeric substrate combination involved the Model II and Model III assembly configurations, both of which exhibited nearly identical responses during cyclic testing. Optimization efforts led to significant improvements in response under low compression pressures, specifically for tests conducted up to 50 N/cm2. The introduction of Ink 2 did not yield substantial enhancements; however, it resulted in a higher value of the printed meander resistors. In conclusion, the optimal configuration was determined to be the meander resistor on a Kapton substrate with the Model II assembly model. Both Ink 1 and Ink 2 were deemed suitable for use. For optimal performance, the terminal wires should be soldered with silver paste and cured at 70 °C to maintain the integrity of the Ink 1 ink. Future work will involve developing a series of these flexible pressure sensors and evaluating their performance in a Wheatstone bridge configuration to further enhance their applicability and performance metrics. Sensor Performance

Linearity of Response: The Kapton substrate significantly improved the linearity of the meander resistor’s response, making it the preferred choice for the sensor’s base material.

Elastomeric Substrate Optimization: Model II and Model III assembly models provided the best results, showing a consistent and reliable performance under cyclic loading.

Pressure Response: The optimized sensor showed enhanced sensitivity to low compression pressures, up to 50 N/cm2, ensuring accurate pressure measurements in this range.

Ink Performance: While Ink 2 increased the resistor’s value, it did not notably improve overall performance, indicating that both Ink 1 and Ink 2 are viable options for the sensor.

Assembly and Soldering: The use of silver paste for soldering at 70 °C was crucial to preserving the behaviour and performance of Ink 1, highlighting the importance of the assembly process in sensor fabrication.

By focusing on these optimized configurations and materials, we aim to enhance the performance and reliability of flexible pressure sensors, paving the way for their application in various fields. The next stage of development will involve integrating these sensors into a Wheatstone bridge configuration to further test and refine their performance characteristics.

All-optical complex field imaging using diffractive processors

January 16, 2026May 28, 2024 by CadWorks3DAdmin

All-optical complex field imaging using diffractive processors

Jingxi Li, Yuhang Li, Tianyi Gan, Che-Yung Shen, Mona Jarrahi & Aydogan Ozcan

Complex field imaging, which captures both the amplitude and phase information of input optical fields or objects, can offer rich structural insights into samples, such as their absorption and refractive index distributions. However, conventional image sensors are intensity-based and inherently lack the capability to directly measure the phase distribution of a field. This limitation can be overcome using interferometric or holographic methods, often supplemented by iterative phase retrieval algorithms, leading to a considerable increase in hardware complexity and computational demand. Here, we present a complex field imager design that enables snapshot imaging of both the amplitude and quantitative phase information of input fields using an intensity-based sensor array without any digital processing. Our design utilizes successive deep learning-optimized diffractive surfaces that are structured to collectively modulate the input complex field, forming two independent imaging channels that perform amplitude-to-amplitude and phase-to-intensity transformations between the input and output planes within a compact optical design, axially spanning ~100 wavelengths. The intensity distributions of the output fields at these two channels on the sensor plane directly correspond to the amplitude and quantitative phase profiles of the input complex field, eliminating the need for any digital image reconstruction algorithms. We experimentally validated the efficacy of our complex field diffractive imager designs through 3D-printed prototypes operating at the terahertz spectrum, with the output amplitude and phase channel images closely aligning with our numerical simulations. We envision that this complex field imager will have various applications in security, biomedical imaging, sensing and material science, among others.

We kindly thank the researchers at University of California for this collaboration, and for sharing the results obtained with their system.

Introduction

Optical imaging can characterize diverse properties of light, including amplitude, phase, wavelength, and polarization, which provides abundant information about samples, such as their morphology and composition. However, conventional image sensors and focal plane arrays, based on e.g., Complementary Metal-Oxide-Semiconductor (CMOS) or Charge-Coupled Device (CCD) technologies, are inherently constrained to detecting only the intensity of the optical field impinging on their active area. Measuring the phase information of a complex field presents challenges, which require indirect encoding through interferometric or holographic detection systems1,2,3. Some of the traditional examples of phase imaging techniques include Zernike phase contrast microscopy and interferometric microscopy. Remarkedly, interferometry-based techniques4,5,6 exhibit very high accuracy in phase measurements (exceeding 1/100 of a wavelength) and allow one to directly obtain wavefront aberrations at very large apertures; however, the relative complexity of decoding interferograms and the sensitivity of the measurement equipment to vibrations hinder their wide applications. Subsequent developments in quantitative phase imaging (QPI) have enabled high-precision characterization of phase information; advances such as the Fourier phase microscopy7, Hilbert phase microscopy8 and digital holographic microscopy9,10,11,12,13,14,15 have also emerged, making QPI a potent label-free optical measurement technique. Nevertheless, these QPI methods often necessitate relatively bulky experimental setups and rely on iterative algorithms based on multiple measurements to digitally reconstruct the desired phase information, leading to slow imaging speeds. In parallel, Shack-Hartmann sensors16,17 can provide phase information by analyzing wavefront distortions inferred from the displacement of light spots created by a perforated mask. This approach, while eliminating the need for a reference wavefront or field, faces challenges in detection resolution due to its discrete measurement nature and still demands data processing for precise phase or wavefront reconstructions. In addition to these, there are other methods that decompose the wavefront into discrete elements, such as Zernike polynomials, offering a direct path to understanding aberrations in wavefronts exhibiting small phase variations18,19,20,21. However, these approaches are less suitable for investigating phase objects with complex structures and high spatial frequencies, underscoring a limitation in their broad applicability.

Recently, fueled by the advances made in deep learning, the application of deep neural networks has been adopted for accurate and rapid reconstruction of phase information in complex fields through a single feed-forward operation22,23,24,25,26,27,28,29. While these deep learning-based approaches offer considerable benefits, they typically demand intensive computational resources for network inference, requiring the use of graphics processing units (GPUs). Simultaneously, the progress in micro- and nano-fabrication technologies has facilitated the development of metasurfaces30,31,32,33,34,35,36,37,38 and thin-film optical components39,40 for QPI applications. However, the functionality of these devices still relies on indirect encoding processes that generate intensity variations on the sensor plane, such as diffused speckles30 or polarized interference patterns38. These existing solutions, therefore, necessitate digital computation for image reconstruction and, in certain cases, also require the incorporation of additional hardware along the optical path, such as polarizers and polarization cameras.

In this work, we demonstrate the design of a complex field imager that can directly capture the amplitude and phase distributions of an incoming field using an intensity-only image sensor array. As shown in Fig. 1a, this complex field imager is composed of a series of spatially engineered diffractive surfaces (layers), which are jointly optimized using supervised deep learning algorithms to successively perform the modulation of incoming complex fields. This diffractive architecture, known as a diffractive optical neural network, has previously been explored for all-optical information processing covering various applications, including all-optical image classification41,42, space-to-spectrum encoding42,43, logic operations44,45,46, optical phase conjugation47, among others48,49,50,51,52,53,54,55,56,57,58,59,60. Within the architecture of our diffractive complex field imager, the diffractive surfaces are trained to simultaneously perform two tasks: (1) an amplitude-to-amplitude (A → A) transformation and (2) a phase-to-intensity (P → I) transformation. Here, the first task involves mapping the amplitude of the incoming complex field to a specific output field of view (FOV) that is solely dedicated to amplitude imaging, independent of the input wave’s phase profile. The second task, on the other hand, aims to approximate a nonlinear transformation by converting the phase of the incoming wave into an intensity pattern at another output FOV, exclusively used for quantitative phase imaging, QPI. Therefore, by placing complex objects or feeding complex fields into the input FOV of the diffractive complex field imager and measuring the intensity distributions at its output FOVs, the amplitude and phase information of the input complex objects/fields can be directly obtained within a single intensity-only image recording step, eliminating the necessity for any form of image reconstruction algorithms. In addition to this spatially multiplexed design of the diffractive complex field imager, termed design I (Fig. 1a), we also explored additional complex field imager designs by incorporating wavelength multiplexing. As illustrated in Fig. 1b, c, these wavelength multiplexed designs, termed designs II and III, operate by detecting the output amplitude and phase signals at two distinct wavelengths ( and , respectively). The difference between designs II and III is that design II utilizes a shared FOV for both the amplitude and phase channel outputs, while design III maintains two spatially separated FOVs, each dedicated to either the output amplitude or phase images.

Schematics for different designs of our diffractive complex field imager. a Illustration showing a spatially multiplexed design for the diffractive complex field imager (design I), which performs imaging of the amplitude and phase distributions of the input complex object simultaneously by channeling the output amplitude and phase images onto two spatially separate FOVs at the output plane, i.e., the amplitude and phase FOVs (or FOVAmp and FOVPhase). b Illustration for an alternative design of the diffractive complex field imager using wavelength multiplexing (design II), wherein the output amplitude and phase profiles are directly measured using a common output FOV but at different wavelengths, i.e., and , respectively. c, Illustration for design III of the diffractive complex field imager, wherein the output amplitude and phase images are measured using two spatially separate FOVs (FOVAmp and FOVPhase) and also at different wavelengths ( and , respectively)

After completing the training of these different designs, we blindly tested the performance and generalization capabilities of our trained diffractive complex field imager models. We quantified their imaging errors using thousands of examples of input complex fields, each composed of independent information channels encoded in the amplitude and phase of the input field. The results demonstrated that our diffractive models could successfully generalize to new, unseen complex test fields, including those with structural features distinctly different from the training objects. Through numerical simulations, we further analyzed the spatial resolution and sensitivity of both the amplitude and phase channels of our diffractive complex field imagers. These analyses revealed that our designs could resolve amplitude features with a linewidth of ≥1.5 and phase features with a linewidth of ≥3 , where represents the mean wavelength. Furthermore, our studies showed that by integrating an additional diffraction efficiency-related loss term into the training function, one could achieve diffractive imager models with enhanced output power efficiencies with minimal compromise in imaging performance.

Apart from these numerical analyses, we also conducted an experimental proof-of-concept demonstration of our diffractive complex field imagers using the terahertz part of the spectrum by fabricating the resulting diffractive layers using 3D printing. For our experiments, we constructed test objects (never seen during the training) with spatially structured amplitude or phase distributions through 3D printing and surface coating techniques. Our experimental results successfully reconstructed the amplitude and phase images of the test objects, closely matching our numerical simulations and the ground truth, validating the effectiveness of our diffractive complex field imager designs. While our experimental demonstrations were conducted in the terahertz spectrum, our designs are scalable and can be adapted to other spectral bands by scaling their dimensions proportional to the wavelength of operation. The compact size of our diffractive designs, with an axial span of ~100 × , facilitates easy integration into existing optical imaging systems and focal plane arrays that operate at different parts of the electromagnetic spectrum, including the visible spectrum. For operation within the visible band, our design can be physically implemented through various nano- and microfabrication techniques, such as two-photon polymerization-based nanolithography50,61. Furthermore, this complex field imager design also does not include any components that are sensitive to the polarization of light, maintaining its amplitude and phase imaging function regardless of the input polarization distribution of the input field. Given all these advantages, including the small footprint, speed of all-optical computation and low-power operation, we believe that this all-optical complex field imaging approach will find broad applications in e.g., defense/security, biomedical imaging, sensing and material science.

Materials

Pr Series

Results

Designs of diffractive complex field imagers

Figure 1a illustrates a spatially multiplexed design of our diffractive complex field imager, termed the design I. This diffractive imager is composed of 5 diffractive layers (i.e., L1, L2, …, L5), where each of these layers is spatially coded with 200 × 200 diffractive features, with a lateral dimension of approximately half of the illumination wavelength, i.e., ~ /2. These diffractive layers are positioned in a cascaded manner along the optical axis, resulting in a total axial length of 150 for the entire design. A complex input object, , illuminated at is placed at the input plane in front of the diffractive layers. This complex object field exhibits an amplitude distribution that has a value range of [ADC, 1], along with a phase distribution ranging within [0, απ]. Here, ADC denotes the minimum amplitude value of the input complex field, and α is the phase contrast parameter of the input complex field. Without loss of generality, we selected default values of ADC and α as 0.2 and 1, respectively, for our numerical demonstrations. Note that it’s essential to work with ADC ≠ 0 since otherwise the phase would become undefined. After the input complex fields are collectively modulated by these diffractive layers L1-L5, the resulting optical fields at the output plane are measured by the detectors within two spatially separated output FOVs, i.e., FOVPhase and FOVAmp, which produce intensity distributions and that correspond to the phase and amplitude patterns of each input complex field, respectively. In addition, we also defined a reference signal region at the periphery of the FOVPhase, wherein the average measured intensity across is used as the reference signal for normalizing the quantitative phase signal . This normalization process is essential to ensure that the detected phase information is independent of the input light intensity fluctuations, yielding a quantitative phase image , regardless of the diffracted output power. Overall, the objective of our training process is to have the phase image channel output approximate the ground truth phase distribution of the input complex field, i.e., , demonstrating an effective phase-to-intensity (P → I) transformation. Concurrently, the training of the diffractive layers also aims to have the diffractive output image in the amplitude channel, i.e., , proportionally match the ground truth amplitude distribution of the input complex field after subtracting the amplitude DC component ADC, i.e., , thereby achieving a successful amplitude-to-amplitude (A → A) transformation performed by the diffractive processor. Note that phase-to-intensity transformation is inherently a nonlinear function58. In the phase imaging channel of our diffractive complex-field imager, the amplitude-squared operation as part of the intensity measurement at the sensor plane represents the only occurrence of nonlinearity within the processing pipeline.

In addition to the spatially multiplexed design I described above, we also created an alternative complex field imager design named design II by incorporating wavelength multiplexing to construct the amplitude and phase imaging channels. As illustrated in Fig. 1b, this approach utilizes a dual-color scheme, where the amplitude and phase of the input images are captured separately at two distinct wavelengths, with dedicated to the phase imaging channel and dedicated to the amplitude imaging channel. As an empirical parameter, without loss of generality, we selected = × 1.28 and + = 2 for our numerical diffractive designs. With this wavelength multiplexing strategy in design II, the amplitude and phase imaging FOVs can be combined into a single FOV – as opposed to 2 spatially separated FOVs as employed by design I shown in Fig. 1a. Consequently, the output amplitude and phase images, i.e., and , can be recorded by the same group of sensor pixels.

As illustrated in Fig. 1c, we also developed an additional complex field imager design, referred to as design III, which integrates both space and wavelength multiplexing strategies in constructing the amplitude and phase imaging channels. Specifically, design III incorporates two FOVs that are spatially separated at the output plane (similar to design I) for amplitude and phase imaging, also utilizing two different wavelength channels (akin to design II) to encode the output amplitude/phase images separately.

Following these design configurations (I, II and III) depicted above, we performed their numerical modeling and conducted the training of our diffractive imager models. For this training, we constructed an image dataset comprising 55,000 images of EMNIST handwritten English capital letters, and within each training epoch, we randomly grouped these images in pairs – one representing the amplitude image and another representing the phase image – thereby forming 27,500 training input complex fields. The phase contrast parameter αtr used for constructing these training input complex fields was set as 1. We utilized deep learning-based optimization with stochastic gradient descent to optimize the thickness values of the diffractive features on the diffractive layers. This training was targeted at minimizing a custom-designed loss function defined by the mean squared error (MSE) between the diffractive imager output amplitude and phase images with respect to their corresponding ground truth. More information about the structural parameters of the diffractive complex field imagers, the specific loss functions employed, and additional aspects of the training methodology can be found in the Methods section.

Numerical results and quantitative performance analysis of diffractive complex field imagers

After the training phase, the resulting diffractive layers of our complex field imager models following designs I, II and III are visualized in Supplementary Figs. S1a, S2a and Fig. 2a, respectively, showing their thickness value distributions. To evaluate and quantitatively compare the complex field imaging performances of these diffractive processors, we first conducted blind testing by selecting 10,000 test images from the EMNIST handwritten letter dataset that were never used in the training set and randomly grouped them in pairs to synthesize 5000 complex test objects. To compare the structural fidelity of the resulting output amplitude and phase images (i.e., and ) produced by our diffractive complex field imager models, we quantified the peak signal-to-noise ratio (PSNR) metrics between these diffractive output images and their corresponding ground truth (i.e., and ). Our results revealed that, for the diffractive imager model using design I that performs space-multiplexed complex field imaging, the amplitude and phase imaging channels provided PSNR values of 16.47 ± 0.96 and 14.90 ± 1.60, respectively, demonstrating a decent imaging performance. Additionally, for the diffractive imager models using designs II (and III), these performance metrics became 16.46 ± 1.02 and 14.98 ± 1.51 (17.04 ± 1.06 and 15.06 ± 1.63), respectively. Therefore, design III demonstrated a notable performance advantage over the other two models in both phase and amplitude imaging channels when both the space and wavelength multiplexing strategies were used. Apart from these quantitative results, we also presented exemplary diffractive output images for the three models of designs I, II and III in Supplementary Figs. S1b, S2b and Fig. 2b, respectively. These visualization results clearly show that our diffractive output images in both amplitude and phase channels present structural similarity to their input ground truth, even though these input complex fields were never seen by our diffractive models before. These analyses demonstrate the internal generalization of our diffractive complex field imagers, indicating their capability to process new complex fields that have similar statistical distributions to the training dataset.

Blind testing results of the diffractive complex field imager using design III. a Thickness profile of the trained layers of the diffractive complex field imager following the design III in Fig. 1c. The layout of the amplitude and phase FOVs in comparison to the size of a diffractive layer is also provided. b Exemplary blind testing input complex objects never seen by the diffractive imager model during its training, along with their corresponding output amplitude and phase images. c, d Same as b, except that the testing images are taken from the MNIST and QuickDraw datasets, respectively, demonstrating external generalization to image datasets with different structural distributions

We also conducted blind testing of these diffractive complex field imager designs by synthesizing input fields from other datasets where the complex images exhibit distinctly different morphological features compared to the training complex field images. For this purpose, we selected the MNIST handwritten digits62 and the QuickDraw image63 datasets, and for each dataset, we synthesized 5000 input complex fields to test our diffractive models blindly. When using the MNIST-based complex field images, the amplitude and phase PSNR values of our diffractive complex field imager models using designs I, II and III were quantified as (16.59 ± 0.71, 15.42 ± 1.28), (16.40 ± 0.68, 15.53 ± 1.25) and (17.05 ± 0.78, 15.59 ± 1.32), respectively. The corresponding diffractive output images for these results are also exemplified in Supplementary Figs. S1c, S2c and Fig. 2c. When testing using input complex fields synthesized from the QuickDraw images, these PSNR values revealed (14.42 ± 0.94, 13.34 ± 1.10), (14.17 ± 1.01, 13.54 ± 1.61) and (14.72 ± 0.97, 13.46 ± 1.13), with exemplary diffractive output images visualized in Supplementary Figs. S1d, S2d and Fig. 2d, respectively. Once again, these PSNR values, along with the visualization results of the output patterns, demonstrate that all our diffractive models (following designs I, II and III) achieved successful reconstructions of the amplitude and phase channel information of the input complex fields, wherein the design III model presented slightly improved performance over the other two designs. Importantly, these analyses demonstrate the external generalization capabilities of our diffractive imagers, positioning them as general-purpose complex field imagers that can handle input complex field distributions markedly distinct from those encountered during their training stage.

Next, we quantified the complex field imaging performance of our diffractive models as a function of the phase contrast and spatial resolution of the incoming complex fields. For this analysis, we selected various grating patterns to form our test images, which have different linewidths and are oriented in either horizontal or vertical directions. We first considered using these grating patterns encoded within either the phase or the amplitude channels of the input complex fields, forming phase-only or amplitude-only grating test objects. To be more specific, the phase-only input fields were set to have a uniform distribution within their amplitude channel, while the amplitude-only input fields were set to have their phase channel values set as zero/constant. For both kinds of gratings, we selected their linewidths as 1.5 or 3 to generate the grating patterns, and tested the spatial resolution for the amplitude and phase imaging channels using our diffractive models; here = for the design I model and = ( + )/2 for the designs II and III. For phase-only gratings with linewidths of 3 , we also used different phase contrast parameters {0.25, 0.5, 1} to form grating patterns with different phase contrast so that we can evaluate the sensitivity of phase imaging by our diffractive complex field processors. To better quantify the performance of our diffractive complex field imagers for these test grating patterns, we used a grating image contrast (Q) as our evaluation metric, defined as:

The results of using these amplitude- or phase-only grating patterns as input fields to our diffractive models using designs I, II and III are provided in Supplementary Figs. S3a, b, S4a, b and Fig. 3a, b, respectively. Through visual inspection and quantification of grating image contrast Q values, our diffractive imager models were found to resolve most of the amplitude-only grating objects of different linewidths and orientations, with quantified Q values consistently above 0.17. The only exception is that the diffractive model using design II fell short in resolving the horizontal grating patterns with 1.5 linewidth, achieving Q < 0.1. For the phase-only grating inputs, all three diffractive models succeeded in resolving the gratings with {0.5, 1} and linewidths of 3 , presenting Q values consistently over 0.19. However, when using the phase-only grating inputs with 0.25 and linewidths of 3 or those with 1 and linewidths of 1.5 , all of our diffractive models struggle to provide consistently clear grating images, exhibiting relatively poor Q values of ≤0.1. These findings reveal that our diffractive imager models exhibit similar performance in imaging resolution and phase sensitivity, providing an amplitude imaging resolution of >1.5 for amplitude-only objects and a phase imaging resolution of ≥3 for phase-only objects with ≥ 0.5. We also calculated the average Q values for different diffractive models using these amplitude- or phase-only grating inputs; the design III model emerges as the most competitive one, presenting average Q values of 0.418 and 0.181 for the amplitude and phase channels, respectively. The suboptimal performance of the design II model, we believe, can primarily be attributed to its utilization of the same output FOV for both the phase and amplitude image formation. This strategy results in the overlap of the diffractive features to serve the two imaging channels, thereby not fully utilizing the degrees of freedom provided by the diffractive layers. This is also corroborated by the visualization of the diffractive layer designs shown in Supplementary Fig. S2a: compared to designs I and III, the areas with significant modulation patterns in the design II layers are significantly smaller and more concentrated in the central region, indicating a less efficient utilization of the diffractive degrees of freedom available for optimization, consequently limiting its imaging performance.

Performance analysis of the diffractive complex field imager model shown in Fig. 2. a Imaging results using phase-only gratings as input fields. The binary phase grating patterns encoded within the phase channel of the input objects are shown and compared with the resulting output amplitude and phase images produced by our diffractive imager (i.e., and ). For each diffractive output image, the grating image contrast Q and SCR values were quantified and shown in red and blue numbers, respectively. b Same as in a, except that the amplitude-only gratings are used as input fields. c Imaging results using complex grating objects as input fields. These grating test objects include ones with the same grating patterns encoded in both the amplitude and phase channels (top), as well as ones where horizontal and vertical gratings are orthogonally placed, with one encoded in the phase channel of the input field and the other encoded in the amplitude channel (bottom)

n addition to the analyses of spatial resolution and phase sensitivity, we also utilized amplitude- and phase-only grating images to investigate the crosstalk between the amplitude and phase imaging channels of our diffractive complex field imagers. Since the amplitude-only grating inputs have constant/zero phase distributions, the ground truth of their corresponding diffractive output images in the phase channel should have zero intensities, where the residual represents the crosstalk coming from the amplitude channel. Similarly, for the phase-only grating inputs that have a uniform amplitude distribution (ADC), their diffractive output images in the amplitude channel should reveal no intensity distributions, with the residual representing the crosstalk coming from the phase channel. As shown by the diffractive output images in Supplementary Figs. S3a, b, S4a, b and Fig. 3a, b, we observe some crosstalk components in the amplitude and phase channel imaging results. To provide a quantitative evaluation of this crosstalk, we used the signal-to-crosstalk ratio (SCR) metric, defined as:

where and denote the resulting output phase image when encoding the same grating pattern within the phase and amplitude channels of the input complex field, respectively; the first term represents the true signal, and the latter represents the crosstalk term in Eq. (2). Similarly, and denote the resulting output amplitude image when encoding the same grating pattern within the amplitude and phase channels of the input complex field, respectively. denotes the intensity summation operation across all the pixels. Following these definitions, we quantified the and values for all the grating imaging outputs in Supplementary Figs. S3a, b, S4a, b and Fig. 3a, b. These SCR analyses reveal that, for all the diffractive imager models, the grating inputs with 1.5 linewidth and 1 present a ~30% lower and a ~53% lower when compared to their counterparts with 3 linewidth, revealing that imaging of finer, higher-resolution patterns is more susceptible to crosstalk. Furthermore, we found that an increase in the input phase contrast ( leads to more crosstalk in the output amplitude channel, which results in a lower value; for example, from >3.5 for 0.25 down to 2.5-3 for 1. Additionally, we calculated the average and values across these grating images for different diffractive imager models; for the diffractive models using designs I, II and III, the average values are 2.805, 3.178 and 3.155, respectively, and the average values are 2.331, 2.262 and 2.252, respectively.

These analyses were performed based on amplitude and phase-only grating objects. Beyond that, we also used complex-valued gratings to further inspect the imaging performance of our diffractive models. Specifically, we created complex test fields that have the same grating patterns encoded in both the amplitude and phase channels. The results reported in the top row of the Supplementary Figs. S3c, S4c and Fig. 3c revealed that all our diffractive imager models are capable of distinctly resolving complex gratings with 3 linewidth, while being largely able to resolve those with 1.5 linewidth, albeit with occasional failure. We further created complex fields by orthogonally placing horizontal and vertical gratings, with one of these gratings encoded in the phase channel of the input field and the other encoded in the amplitude channel. As evidenced by the bottom row of Supplementary Figs. S3c, S4c and Fig. 3c, our diffractive models could successfully reconstruct the amplitude and phase patterns of the input complex fields with a grating linewidth of 3 .

Impact of input phase contrast on the performance of diffractive complex field imagers

In the analyses conducted so far, all the input fields fed into our diffractive model maintained a consistent phase contrast with an value of 1, i.e., , regardless of the training and testing phases. Next, we investigated the impact of greater input phase contrast on the performance of diffractive complex field imagers. For this analysis, we utilized the same diffractive design III model shown in Fig. 2 and tested its imaging performance using the same set of complex test objects used in Fig. 2b, but with an increased object phase contrast, , chosen within a range between 1 and 1.999 (i.e., ∈ [0, 2π)). The corresponding results are shown as the blue curve in Fig. 4a, illustrating a degradation in the imaging performance of the diffractive model as increases. This degradation is relatively minor in the PSNR results for the amplitude channel but more pronounced for the phase channel. Specifically, as increases from 1 to 1.5, the average amplitude PSNR value slightly drops from 17.04 to 15.82, while the phase PSNR falls from 15.06 to 11.96. When approaches 2, the average amplitude and phase PSNR values further decrease to 15.34 and 9.87, respectively. The visual examples in Fig. 4b and Supplementary Fig. S5a, which correspond to the cases of = 1.5 and 1.25, respectively, reveal that the amplitude channel of the diffractive model can consistently resolve the amplitude patterns of the objects, which were never encountered during the training phase. However, in the phase channel, despite the patterns being distinguishable and very well matching the ground truth, their intensities were lower than the correct level, leading to incorrect quantitative phase values and, thus, a drop in the phase PSNR values.

To address this limitation, we explored training two additional diffractive models using higher phase contrast values of 1.25 and 1.5, respectively. The quantitative evaluation results for these models are presented in Fig. 4a as orange and green curves, respectively. The findings indicate that, compared to the original model trained with = 1, the diffractive model employing = 1.5 exhibits a significantly enhanced imaging performance in the phase channel for > 1.25. For instance, at = 1.5, the phase PSNR improved from 11.96 to 14.42, while the amplitude PSNR remained almost identical across various values. A similar improvement was also observed with the other diffractive model trained using = 1.25. These findings are further confirmed by the exemplary visualization results shown in Fig. 4c and Supplementary Fig. S5b; the quantitative phase signals in the phase channel of these new diffractive models markedly ameliorate the issues encountered by the original model. Nonetheless, it is also noted that these new models exhibited some minor disadvantages; for example, the diffractive model trained using = 1.5 demonstrated slightly inferior phase PSNR results at < 1.25 compared to the original model. This suggests a propensity of our diffractive imager models to achieve better imaging performance on complex test objects with phase dynamic ranges akin to those encountered during the training.

It is also important to note that the current design of our diffractive complex field imager is specifically tailored for imaging thin complex objects. For such thin objects, their phase variations are relatively slower. In contrast, substantially thicker objects can exhibit rapidly varying phase distributions, leading to phase wrapping issues. These complexities would pose a challenge for the diffractive network-based complex field imager to perform effective processing and image formation. To potentially address this challenge, incorporating a wavelength-multiplexing strategy (similar to that used in certain QPI methods) into the diffractive imager framework could be a potential path forward. Such an approach, which is left as future work, may involve leveraging the diffractive network for all-optical processing across multiple wavelength channels, followed by minimal digital post-processing to accurately reconstruct quantitative phase signals from wrapped phase information acquired at different wavelengths.

Output power efficiency of diffractive complex field imagers

To quantify the output diffraction efficiencies of our complex field imagers, we utilized 5000 test complex fields created from the EMNIST image dataset, and calculated the average diffraction efficiencies of our diffractive complex field imager models. By integrating an additional loss term into our training loss function to balance the complex field imaging performance along with the output diffraction efficiency, we demonstrated the feasibility of increased power efficiency for all three designs (I, II and III), with minimal compromise in the output image quality. The added loss term, denoted as , is specifically designed to control and improve the output diffraction power efficiency, with its definition given by:

where and denote the output diffraction power efficiency within FOVPhase and FOVAmp, respectively, with their detailed definition provided in the Methods section. refers to the target diffraction efficiency threshold for . By minimizing the loss function that incorporates the term, we trained 6 diffractive imager models for each design (I, II and III). For each of these models, we set at distinct levels: 0.1%, 0.2%, 0.4%, 0.8%, 1.6% and 3.2%, and trained the respective model to satisfy the specified . Note that all these new diffractive complex field imager models maintain the same physical architecture as the designs illustrated in Fig. 1, and they were trained using the same EMNIST-based complex image dataset. A performance comparison for these models is provided in Fig. 5, where their amplitude and phase average PSNR values were calculated across the test set and shown as a function of their average diffraction efficiency values. Taking the architecture of design III as an example, one of our complex field imager designs achieved an output power efficiency of ~0.2% in both amplitude and phase channels, resulting in average PSNR values of 14.72 ± 1.47 and 16.64 ± 1.03 for the two corresponding channels, respectively. An additional model, optimized with a heightened emphasis on the output power efficiency, demonstrated the capability of performing complex field imaging with >0.8% diffraction efficiency in both the phase and amplitude channels, while achieving average amplitude and phase PSNR values of 13.51 ± 1.32 and 16.74 ± 1.05, respectively. A similar trend was also observed for the other models using designs I and II, where a significant increase in the output diffraction efficiency could be achieved with a modest trade-off in the output image quality. Moreover, a comparative assessment of the three different designs under various output diffraction efficiencies reaffirms the overall performance advantage of design III: it presents remarkable advantages over design I in phase imaging while outperforming design II in amplitude imaging. Overall, Fig. 4 serves as a “designer rule plot”, which offers guidance in selecting suitable diffractive complex field imager models by balancing the phase/amplitude imaging fidelity with output power efficiency according to specific application requirements.

Experimental validation of diffractive complex field imagers

We performed experimental validation of our diffractive complex field imagers using the terahertz part of the spectrum, specifically employing the design II configuration as illustrated in Fig. 1b; we used = 0.75 mm and = 0.8 mm for the phase and amplitude imaging channels, respectively. We used three diffractive layers for our experimental design, each layer containing 120 × 120 learnable diffractive features with a lateral size of ~0.516 (dictated by the resolution of our 3D printer). The axial spacing between any two adjacent layers (including the diffractive layers and the input/output planes) was chosen as ~25.8 (20 mm), resulting in a total axial length of ~103.2 for the entire design. As a proof of concept, we designed two experimental models that use different input phase contrast parameters, and 0.5. These experimental models were trained using a dataset composed of phase-only and amplitude-only objects, which feature randomly generated spatial patterns with binary phase values of {0, } or amplitude values of {0, 1}. In these proof of concept experiments, we did not employ input objects with spatial distributions in both the amplitude and phase channels due to the fabrication challenges of such objects; however, the amplitude-only or phase-only objects used here still share a single common input FOV and are processed by the same diffractive imager. Therefore, this experimental demonstration serves as an effective proof of our all-optical complex field imaging framework, which has never been demonstrated before in prior works.

After the training, the resulting layer thickness profiles of the diffractive models with = 1 and 0.5 are visualized in Fig. 6a, d, respectively. These diffractive layers were fabricated using 3D printing, with their corresponding photographs showcased in Fig. 6b, e. Additionally, we constructed phase-only or amplitude-only test objects, which were never seen by the trained diffractive models. The phase-only test objects were fabricated by 3D printing layers with spatially varying height profiles representing the phase distributions, and the amplitude-only objects were created by padding aluminum foils onto 3D-printed flat layers to delineate the amplitude patterns. In our proof-of-concept experiments, these objects were designed to have 5 × 5 pixels, each featuring a size of 4.8 mm (~6.19 ). As shown in Fig. 7b, the printed diffractive layers and input complex objects were assembled using a custom 3D-printed holder to ensure that their relative positions follow our numerical design. In our experiments, we employed a THz source operating at = 0.75 mm and = 0.8 mm, and used a detector to measure the intensity distribution at the output plane, yielding the output amplitude and phase images. The photograph and schematic of our experimental setup are provided in Fig. 7a and c, respectively. Further details related to the experiment are provided in the Methods section.

The experimental results for these two models are shown in Fig. 6c, f, where the output amplitude and phase images present a good agreement with their numerically simulated counterparts, also aligning well with the input ground truth images. These experimental results demonstrate the feasibility of our 3D fabricated diffractive complex field imager to accurately image the amplitude and phase distributions of the input objects; these results also represent the first demonstration of all-optical complex field imaging achieved through a single diffractive processor.

Discussion

The numerical analyses and experimental validation presented in our work showcased a compact complex field imager design through deep learning-based optimization of diffractive surfaces. We explored three variants of this design strategy, with comparative analyses indicating that the design employing spatial and wavelength multiplexing (design III) achieves the best balance between the complex field imaging performance and diffraction efficiency, albeit with a minor increase in hardware complexity. Leveraging the all-optical information processing capabilities of multiple spatially engineered diffractive layers, diffractive complex field imagers reconstruct the amplitude and phase distributions of the input complex field in a complete end-to-end manner, without any digital image recovery algorithm, setting it apart from other designs in the existing literature for similar applications. This capability enables direct recording of the amplitude and phase information in a single snapshot using an intensity-only sensor array, which obviates the need for additional computational processing in the back-end, thereby significantly enhancing the frame rate and reducing the latency of the imaging process. Furthermore, our diffractive imager designs feature a remarkably compact form factor, with dimensions of ~100λ in both the axial and lateral directions, offering a substantial volumetric advantage. In contrast, conventional methods based on interferometry and holography often involve relatively bulky optical components and necessitate multiple measurements, leading to optical and mechanical configurations that require a large physical footprint. While some of the recent single-shot complex amplitude imaging efforts using metasurfaces have aimed for greater compactness, they typically require metalenses with large lateral sizes of >1000λ32,37,38. Moreover, achieving a similar FOV (covering several tens of wavelengths) as in our work would require imaging path lengths of thousands of wavelengths.

In our previous research, we developed diffractive processor designs tailored for imaging either amplitude distributions of amplitude-only objects51 or phase distributions of phase-only objects53,59,60. However, these designs would become ineffective for imaging complex objects with independent and non-uniform distributions in the amplitude and phase channels. In this work, we have overcome this limitation by training our diffractive imager designs using complex objects with random combinations of amplitude and phase patterns, thus allowing a single imager device to effectively generalize to complex optical fields with various distributions in the amplitude and phase channels.

The diffractive complex field imager designs that we presented also exhibit certain limitations. Our results revealed residual errors in their targeted operations, particularly manifesting as crosstalk coming from the amplitude channel into the phase channel. This suggests that the actual phase-to-intensity transformation represented by our diffractive imager, while effective, is an approximation with errors that are dependent on the object amplitude distribution. The mitigation approach for this limitation might involve further enhancement of the information processing capacity of our diffractive imagers, which can be achieved through employing a larger number of diffractive layers (forming a deeper diffractive architecture), thus increasing the overall number of diffractive features/neurons that are efficiently utilized64. Additionally, we believe another performance improvement strategy could be to increase the lateral distance between the two output FOVs dedicated to the phase and amplitude channels, thereby allowing the trainable diffractive features to better specialize for the individual tasks of phase/amplitude imaging; this approach, however, would increase the size of the output FOV of the focal plane array and also demand larger diffractive layers.

Moreover, in our experimental results, we observed the emergence of noise patterns within certain regions, which did not exist in our numerical simulations. This discrepancy can be attributed to potential misalignments and fabrication imperfections in the diffractive layers that are assembled. A mitigation strategy could be to perform “vaccination” of these diffractive imager models, which involves modeling these errors as random variables and incorporating them into the physical forward model during the training process42,50,65. This has been proven effective in providing substantial resilience against misalignment errors for diffractive processors, exhibiting a noticeably better match between the numerical and experimental results42,50,65.

Leveraging its unique attributes, our presented complex field imaging system can open up various practical applications across diverse fields. For biomedical applications, it can be seamlessly integrated into endoscopic devices66 and miniature microscopes67,68,69 to enable real-time, non-invasive quantitative imaging of tissues and cells, which might also be useful for, e.g., point-of-care diagnostics with its compactness and efficiency. This might potentially pave the way for their use in intraoperative imaging, providing surgeons with critical, high-resolution insights during a medical procedure70,71. For environmental monitoring, as another example, the presented system may facilitate the development of portable lab-on-a-chip sensors capable of quickly identifying microorganisms and pollutants, streamlining on-site quantitative analysis without delicate and tedious sample preparation steps72,73,74,75. Furthermore, the portability and compactness of these diffractive designs can make them a valuable tool for rapid inspection of materials in industrial settings76,77. Overall, this compact and efficient complex field imager design could be used in various settings, opening new avenues in scientific research and expanding the measurement capabilities for practical, real-world applications.

Materials and methods

Numerical forward model of a diffractive complex field imager

In our numerical implementation, the transmissive layers within the diffractive complex field imager were modeled as thin dielectric optical modulation elements with spatially varying thickness profiles. For the the diffractive layer, the complex-valued transmission coefficient of its the feature at a spatial location was defined depending on the illumination wavelength ( ):

where and denote the amplitude and phase coefficients, respectively. The free-space propagation of complex fields between diffractive layers was modeled through the Rayleigh–Sommerfeld diffraction equation41:

where represents the complex field at the ith diffractive feature of the lth layer at location . and . Based on Eq. (8), can be viewed as a secondary wave generated from the source at . As a result, the optical field modulated by the ith diffractive feature of the lth layer ( ≥ 1, treating the input object plane as the 0th layer), , can be written as:

where denotes the number of diffractive features on the (l − 1)th diffractive layer and represents the location of the layer in the z direction parallel to the optical axis. The amplitude and phase components of the complex transmittance of the th feature of diffractive layer , i.e., and in Eq. (7), were defined as a function of the material thickness over the region of that diffractive feature, , as follows:

Here the parameters and represent the refractive index and the extinction coefficient of the diffractive layer material, respectively. These parameters correspond to the real and imaginary parts of the complex-valued refractive index, denoted as , such that . We determined the values of and through experimental characterization of the dispersion properties of the diffractive layer materials, and their values are visualized in Supplementary Fig. S6. The trainable thickness values of the diffractive features were limited within the range of [ , ], representing the learnable parameters of our diffractive complex field imagers. For training the diffractive imager models used for numerical analyses, the values of , were selected as 0.2 and 1.2 mm, respectively. For training the diffractive imager models used for experimental validation, the values of , were selected as 0.4 and 1.4 mm, respectively.

Training loss functions and quantification metrics

The total loss function for the training of our diffractive complex field imagers was defined as:

Here, stands for the loss term for the quantitative phase imaging function, and is defined as:

where is the reference signal measured within the reference signal region , which is a frame region surrounding the FOVPhase with a width of one image pixel. represents the loss term for the amplitude imaging function, which is defined as:

where represents the intensity averaging operation across all the spatial pixels of the image. The definition of is provided in Eqs. (4)–(6). In Eqs. (5) and (6), both and were defined using the following Equation:

where represents the total power of the optical field calculated within the output FOV that is either FOVPhase or FOVAmp, i.e., , and represents the total power of the complex field within the input FOV, i.e., .

The hyperparameters, and , in Eq. (12) refer to the weight coefficients associated with the amplitude imaging and output diffraction efficiency penalty-related loss terms, respectively. When training our diffractive imager models shown in Fig. 2a, Supplementary Figs. S1a and S2a, the values of and were set as 3 and 0, respectively, i.e., no diffraction efficiency-related penalty was applied for these models. During the training of the diffractive imager designs exhibiting different output diffraction efficiencies for the analysis in Fig. 4, the values of and were set as 3 and 100, respectively. When training the experimental designs shown in Fig. 6a, d, the values of and were set as 1 and 100, respectively, and were selected as 4%.

For analyzing the imaging performance of our diffractive complex field imagers, we used PSNR as our evaluation metric. The definition of the PSNR value between an image A and an image B is given by:

where N is the total number of pixels within the image. Another metric employed to quantify the performance of our diffractive complex field imaging is the grating image contrast (Q), and its definition has been provided in Eq. (1). In Eq. (1), is determined by taking the average intensity of the grating images along the grating orientation and, finding the maximum intensity values within the bar regions, and is computed in a way similar to but by locating the minimum values.

Implementation details of diffractive complex field imagers

For the diffractive imager models used for numerical analyses in this manuscript, we used a minimum sampling period of 0.3 mm for simulating the complex optical fields (i.e., 0.375 for = 0.8 mm). The lateral size of each feature on the diffractive layers is also selected as 0.3 mm. Both the input and output FOVs, including FOVPhase and FOVAmp, were set to 18 mm × 18 mm. These fields were discretized into arrays of 15 × 15 pixels, with each pixel measuring 1.2 mm (i.e., ~1.5 ).

For simulating the diffractive imager models used for experimental validation, both the sampling period for the optical fields and the lateral dimensions of the diffractive features were set at 0.4 mm (i.e., ~0.516 for = 0.775 mm). The input and output FOVs in these models were 24 mm × 24 mm (i.e., ~30.97 × 30.97 ). These fields were discretized into arrays of 5 × 5 pixels, with each pixel measuring 4.8 mm (i.e., ~6.19 ).

For training our diffractive imager models, we randomly extracted 55,000 handwritten English capital letter images from the EMNIST Letters dataset to form our training set. During the training stage, we also implemented an image augmentation technique to enhance the generalization capabilities of the diffractive models. This involves randomly flipping the input images vertically and horizontally. These flipping operations were set to be performed with a probability of 50%. For testing our diffractive models, we used a testing image dataset of 10,000 handwritten English capital letter images, which were also randomly extracted from the EMNIST Letters dataset while ensuring no overlap with the training set. In addition, for preparing the blind test images used for evaluating the external generalization capabilities of our models, we used 10,000 handwritten digit images from the MNIST testing dataset and 10,000 QuickDraw images from the QuickDraw dataset63. Before being fed into our diffractive models, all these training and testing images further underwent a bilinear downsampling and normalization process to match the corresponding dimensions and value ranges of the input amplitude or phase images.

Our diffractive models presented in this paper were implemented using Python and PyTorch. In the training phase, each mini-batch was set to consist of 64 randomly selected EMNIST handwritten letter images from the EMNIST dataset78. Subsequently, these images were randomly grouped in pairs to synthesize complex fields. Within each training iteration, the loss value was calculated, and the resulting gradients were back-propagated accordingly to update the thickness profiles of each diffractive layer using the Adam optimizer79 with a learning rate of 10−3. The entire training process lasted for 100 epochs, which took ~6 h to complete using a workstation equipped with a GeForce RTX 3090 GPU.

Experimental terahertz set-up

For our proof-of-concept experiments, we fabricated both the diffractive layers and the test objects using a 3D printer (PR110, CADworks3D). The phase objects were fabricated with spatially varying thickness profiles to define their phase distributions. The amplitude objects were printed to have a uniform thickness and then manually coated with aluminum foil to define the light-blocking areas, while the uncoated sections formed the transmission areas, resulting in the creation of the desired amplitude profiles for test objects. Additionally, we 3D-printed a holder using the same 3D printer, which facilitated the assembly of the printed diffractive layers and input objects to align with their relative positions as specified in our numerical design. To more precisely control the beam profile for the illumination of the complex input objects, we 3D printed a square-shaped aperture of 5 × 5 mm and padded the area around it with aluminum foil. The pinhole was positioned 120 mm away from the object plane in our experiments. This pinhole serves as an input spatial filter to clean the beam originating from the source.

To test our fabricated diffractive complex field design, we employed a THz continuous-wave scanning system, with its schematic presented in Fig. 7c. To generate the incident terahertz wave, we used a WR2.2 modular amplifier/multiplier chain (AMC) followed by a compatible diagonal horn antenna (Virginia Diode Inc.) as the source. Each time, we transmitted a 10 dBm sinusoidal signal at frequencies of 11.111 or 10.417 GHz (fRF1) to the source, which was then multiplied 36 times to generate output radiation at continuous-wave (CW) radiation at frequencies of 0.4 or 0.375 THz, respectively, corresponding to the illumination wavelengths of 0.75 and 0.8 mm used for the phase and amplitude imaging tasks, respectively. The AMC output was also modulated with a 1 kHz square wave for lock-in detection. We positioned the source antenna to be very close to the 3D-printed spatial pinhole filter, such that the illumination power input to the system could be maximized. Next, using a single-pixel detector with an aperture size of ~0.1 mm, we scanned the resulting diffraction patterns at the output plane of the diffractive complex field imager at a step size of 0.8 mm. This detector was mounted on an XY positioning stage constructed from linear motorized stages (Thorlabs NRT100) and aligned perpendicularly for precise control of the detector’s position. For illumination at = 0.75 mm or = 0.8 mm, a 10-dBm sinusoidal signal was also generated at 11.083 or 10.389 GHz (fRF2), respectively, as a local oscillator and sent to the detector to down-convert the output signal to 1 GHz. The resulting signal was then channeled into a low-noise amplifier (Mini-Circuits ZRL-1150-LN+) with an 80 dBm gain, followed by a bandpass filter at 1 GHz (±10 MHz) (KL Electronics 3C40-1000/T10-O/O), effectively mitigating noise from undesired frequency bands. Subsequently, the signal passed through a tunable attenuator (HP 8495B) for linear calibration before being directed to a low-noise power detector (Mini-Circuits ZX47-60). The voltage output from the detector was measured using a lock-in amplifier (Stanford Research SR830), which utilized a 1 kHz square wave as the reference signal. The readings from the lock-in amplifier were then calibrated into a linear scale. In our post-processing, we further applied linear interpolation to each intensity field measurement to align with the pixel size of the output FOV used in the design phase. This process finally resulted in the output measurement images presented in Fig. 6c, f.

Open-Top Patterned Hydrogel-Laden 3D Glioma Cell Cultures for Creation of Dynamic Chemotactic Gradients to Direct Cell Migration

January 16, 2026April 23, 2024 by CadWorks3DAdmin

Academic Article

Open-top patterned hydrogel-laden 3D Glioma cell cultures for creation of dynamic chemotactic gradients to direct cell migration

by Aditya Rane, Steven Tate, Jenna L. Sumey, Qing Zhong, Hui Zong, Benjamin Purow, Steven R. Caliari and Nathan S. Swami

Abstract: The laminar flow profiles in microfluidic systems coupled to rapid diffusion at flow streamlines have been widely utilized to create well-controlled chemical gradients in cell cultures for spatially directing cell migration. However, within hydrogelbased closed microfluidic systems of limited depth (≤0.1 mm), the biomechanical cues for the cell culture are dominated by cell interactions with channel surfaces rather than with the hydrogel microenvironment. Also, leaching of poly(dimethylsiloxane) (PDMS) constituents in closed systems and the adsorption of small molecules to PDMS alter chemotactic profiles. To address these limitations, we present the patterning and integration of a PDMS-free open fluidic system, wherein the cell-laden hydrogel directly adjoins longitudinal channels that are designed to create chemotactic gradients across the 3D culture width, while maintaining uniformity across its ∼1 mm depth to enhance cell−biomaterial interactions. This hydrogel-based open fluidic system is assessed for its ability to direct migration of U87 glioma cells using a hybrid hydrogel that includes hyaluronic acid (HA) to mimic the brain tumor microenvironment and gelatin methacrylate (GelMA) to offer the adhesion motifs for promoting cell migration. Chemotactic gradients to induce cell migration across the hydrogel width are assessed using the chemokine CXCL12, and its inhibition by AMD3100 is validated. This open-top hydrogel-based fluidic system to deliver chemoattractant cues over square-centimeter-scale areas and millimeter-scale depths can potentially serve as a robust screening platform to assess emerging glioma models and chemotherapeutic agents to eradicate them.

Keywords: hydrogel; microfluidics; tumor microenvironment; cell migration; glioma

We kindly thank the researchers at University of Virginia for this collaboration, and for sharing the results obtained with their system.

Glioblastoma (GBM) is the most common and aggressive type of primary brain cancer in adults (1) that remains incurable and recurs frequently, (2) highlighting the need for in vitro patient-specific models that can predict disease outcomes. Recent work has correlated the infiltrative nature of the disease to glioma cell migration characteristics, (3) but recapitulation of the chemical and biomechanical cues that affect cell migration in the tumor microenvironment (TME) remains a major challenge within these models. (4) Based on hyaluronan, which is a primary extracellular matrix component in the brain, hyaluronic acid hydrogel networks are known to induce dose-dependent alterations to markers of glioma malignancy. (5) The inclusion of hyaluronic acid-based hydrogel as a matrix, with soluble CXCL12 as a chemoattractant for CXCR4-expressing GBM cells (6) and with chemotherapeutic agents to eradicate them, (7) is being investigated as a less invasive and more targeted pathway to remove residual glioma cells from the TME. To complement prior work on coupling laminar flow profiles to rapid diffusion at flow streamlines in closed microfluidic systems, (8) we present an open-top microfluidic platform to create chemotactic gradients for spatiotemporal control of glioma cell migration in a patterned hydrogel of millimeter-scale depth that is designed to enhance cell–biomaterial interactions.

Pressure-based microfluidic flow control in closed microchannels for cell culture within poly(dimethylsiloxane) (PDMS) backing layers on one side to enable fluidic access and bonded to a glass coverslip to enable access to live-cell imaging, is commonly used to investigate cellular processes. However, to create 3D cultures with biomechanical cues from the hydrogel-based cell microenvironment, without being limited by cell–cell interactions or cell–surface interactions, the cultures must be patterned over several cell layers (∼mm-scale depths) between the flows that deliver the chemotactic cues. The chief challenge to maintaining such 3D cultures in closed microfluidic systems over several days for directing cell migration (9) is the leaching of PDMS monomers into the culture medium, which affects cell proliferation, cell adhesion, and differentiation of cells in pharmacokinetic studies. (10−12) Furthermore, the adsorption of small hydrophobic molecules by PDMS over the long term of the cell culture affects their transport to the culture, thereby altering dose response and drug gradients. (13,14) Also, alterations in oxygen permeability with PDMS bonding (15) and its high permeability to water vapor (16,17) can cause culture media evaporation, drying, and bubble formation that are detrimental to the establishment of chemotactic gradients. (18) Alternate substrates for 3D cell culture, such as PMMA, COC, and adhesive tapes, usually involve cumbersome fabrication and assembly steps that limit rapid prototyping of device designs and assays. (19,20) The availability of open-top hydrogel-laden 3D cultures devoid of PDMS interfaces would mitigate many of these limitations, but current reports (21) do not yet integrate fluidic control operations for dynamic modulation of gradients that provide chemotactic cues to direct cell migration.

The patterning of cell-laden hydrogels for 3D culture in closed microfluidic systems is usually accomplished with arrays of microposts or pillars that use the surface tension of the hydrogel material to confine it between the adjoining media channels. To prevent the hydrogel from spilling into the adjoining fluidic channel, the injection pressure must be carefully controlled, (22−24) which depends on the viscosity, wettability, and other material properties of the hydrogel. High-aspect-ratio microfabrication is needed to create posts that extend several micrometers in the lateral scale and up to millimeters in the depth scale to contain cell-laden hydrogels for 3D culture. This can lead to discontinuities in the interface between the hydrogel and the perfused medium, thereby subjecting the cells to altered biochemical cues. (22) Recent approaches to create continuous interfaces of the hydrogel to the adjoining fluid in closed channels have emerged, such as the use of phase guides, stepped height channels, recoverable elastic barriers, and alignment of core–shells, but these require multilayer fabrication and assembly. (23−27) Rail-based capillary-pinning approaches for the patterning of open-top 3D cell cultures have been reported (21,28,29) but are static and without the fluidic control needed to deliver chemical gradients for dose/drug response assays or directed migration studies. Microfluidic probes (MFPs) to deliver gradients to open-top cell cultures by using injection and aspiration flows (30−38) require careful optimization of the geometry and flow rates of the probes. (35−40) Also, these are impractical for use in patterned 3D hydrogel cultures due to limited depth control of the confined fluid and depletion of the medium that immerses the 3D culture. To address these issues, we present a PDMS-free open microfluidic system (Figure 1A) integrating the patterned cell-laden hydrogel (∼1 mm depth) with adjoining longitudinal microchannels for dynamic flow control to create chemotactic gradients across the 3D culture width to direct glioma cell migration.

Figure 1. (A) Patterned cell-laden hydrogel adjoining fluidic channels. (i) A silanized glass substrate treated for adhesion to the cross-linked hydrogel is (ii) reversibly bonded to a PDMS mold that is then filled with the cell-laden hydrogel and (iii) photo-cross-linked to create the patterned hydrogel on glass. (iv) The PDMS mold is removed to leave open fluidic channels that directly adjoin the patterned hydrogel. (v) The structure is surrounded with culture medium to maintain cell viability and prevent hydrogel shrinkage. (vi) An example of the patterned hydrogel with addressable open fluidic channels through which a FITC gradient was established. (B) (i) Microfluidic flow control setup. (C) (i) 3D-printed holder for fluidic integration with the patterned hydrogel and (ii) image of the patterned hydrogel with tubing to the 3D-printed holder and channel with yellow dye.

While a reversibly bonded PDMS mold to glass is used as a receptacle to fill in and cure the hydrogel to enable its patterning, this PDMS layer is carefully detached after hydrogel curing, leaving behind only the patterned hydrogel layer adjoining the addressable fluidic channels. Another essential feature of our design is the integration of flow control (Figure 1B) over centimeter-scale lengths in the longitudinal open fluidic channels to deliver the culture medium and remove the waste products, while enabling dynamic modulation of the gradient of chemotactic molecules across the cell-laden hydrogel. This is accomplished through injection flow at one end and aspiration flow on the other end, which are placed in a 3D-printed construct to adjust the height of the microfluidic probes at the injection side to be below the hydrogel height level (Figure 1C(i,ii)), and the aspiration tubing to be just above the hydrogel height level, with surface tension confining the culture medium within the channels over the flow length. Rather than perfusing the culture with a peristaltic pump that creates pulsatile flow, which is difficult to monitor and control over the 48 h culture period, a pressure pump is used to deliver injection fluid at a continuous flow rate, and a vacuum line is used for the aspiration flow. In this manner, a set of integrated flow sensors can continually monitor any alterations and correct them by modulating the injection or aspiration flow rates, thereby maintaining an open cell culture within an incubator jacket, without drying of the hydrogel. Using a flow rate of 7.5 μL/min, which is close to the level reported for interstitial flow in brain tissues, (41−43) we present the ability to tune chemical gradients to the cells within the patterned hydrogel culture. This open-top system is validated using a patterned culture of U87 glioma cells laden within a hyaluronic acid (HA)–gelatin methacrylate (GelMA) hybrid hydrogel that is optimized to maintain cell viability, while including the adhesive groups necessary for integrin-mediated cell migration. A chemotactic gradient of CXCL12 delivered to the open-top cell-laden hydrogel culture is used to assay migration cues in the presence and absence of AMD3100, an inhibitor to CXCR4-expressing GBM cells. We envision utilization of this open-top integrated system to deliver chemotactic gradients and serve as a drug testing platform for micropatterned cell-laden hydrogel models.

Results and Discussion

Optimizing the Hydrogel Composition for Maintaining Cell Viability and Adhesion Motifs for Migration

Hyaluronan is a primary extracellular matrix component in the brain, leading to the interest in utilization of HA hydrogels to mimic the glioma microenvironment. However, while it can interact with cells through cell surface markers such as CD44, it lacks the adhesion motifs necessary for integrin-mediated cell migration. (44) Hence, we optimized a hybrid photo-cross-linked hydrogel composed of GelMA, which includes adhesive groups such as RGD, with HA to mimic the brain TME. As shown in Figure 2A, this is accomplished using lithium phenyl-2,4,6-trimethylbenzoylphosphinate (LAP) as a common photoinitiator for free-radical-initiated cross-linking of norbornene-modified HA (NorHA) and GelMA hydrogels. While GelMA hydrogels have higher stiffness, (45) HA hydrogels have been used to mimic the stiffness of native brain tissue. (46) As shown in Figure 2B, this photopatterned hybrid hydrogel, consisting of 2% GelMA and 1% NorHA, exhibits a Young’s modulus that mimics the brain’s white matter and gray matter components. (47) Based on propidium iodide staining (Figure 2C), GelMA hydrogels (10%) support U87 cells to a viability level of only 60% over the 48 h culture period, whereas the hybrid hydrogel supports ∼75% cell viability, which is closer to the 90% viability levels observed within 1% NorHA hydrogels.

Figure 2. (A) Individual and hybrid hydrogels. Molecular structure before and after cross-linking of (i) norbornene-modified hyaluronic acid (norHA), (ii) gelatin methacrylate (GelMA), and (iii) a hybrid hydrogel of NorHA and GelMA. (B) Young’s modulus of the hybrid hydrogel formulation (2% HA, 1% GelMA) compared to that of brain tissue. The hybrid hydrogel recapitulates the reported stiffness of white and gray matter. (C) Cell viability as a function of hydrogel composition. (D) Cell mobility based on cell migration through the hydrogel at the indicated time points (i)–(iii). (E) Measured cell migration speed as a function of hydrogel composition. Significance was determined by one-way ANOVA with Tukey’s post-hoc test, represented as ns, p ≥ 0.05; *, p ≤ 0.05; **, p ≤ 0.01; ***, p ≤ 0.001; ****, p ≤ 0.0001.

Differences in pore size between the respective hydrogels can affect cell viability, with 1% NorHA hydrogels reported at a theoretical mesh size of ∼85 nm, (46) while mesh sizes of 10% GelMA are reported as ∼20 nm.(48) However, this difference in mesh size, as reported for fully swollen hydrogels, is likely less pronounced in our study, since the hydrogels are confined in the device channels and are somewhat restricted from swelling. Instead, GelMA hydrogels that are formed by chain-growth cross-linking are inhibited by oxygen and require a high concentration of free radicals for initiation of polymerization, (49) leading to a loss in viability. On the other hand, step-growth-polymerized NorHA hydrogels require lower free radical concentrations and are not inhibited by oxygen, (50) thereby resulting in faster cross-linking and improved cytocompatibility. (51) This hybrid hydrogel also supports viability of a 3D culture of highly migratory and malignant oligodendrocyte progenitor cells (OPCs) that are progenitors in glioma (52,53) (Figure S1). Based on time-lapse images of U87 cell alignment and migration (Figure 2D(i–iii)), the cell speeds in the hybrid hydrogel resemble those observed within the GelMA hydrogel, rather than the HA hydrogel (Figure 2D(iv) and Movie S1). This indicates the optimization of the hybrid hydrogel for its high viability and migration abilities.

Open-Top Microfluidics for Spatiotemporal Control of Chemotactic Gradients Across Patterned Hydrogel

The open-top 3D culture was utilized with longitudinal injection and aspiration flows to deliver and control chemotactic gradients across the patterned hydrogel width. Using fluorescein isothiocyanate (FITC)-labeled dextran (FITC-dextran) with a 400 Da molecular weight to mimic small-molecule drugs, the images (Figure 3A(i,ii)) show development of the gradient across the hydrogel width. Using FITC-dextran with a 10 kDa molecular weight to mimic proteins or cytokines, fluorescence levels measured from time-lapse images show molecular diffusion profiles at different widths across the patterned hydrogel, indicating the ability of the open-top microfluidic system to induce gradual flattening of the initial diffusion profile. It is apparent that the 10 kDa FITC-dextran takes well over 12 h to reach a steady-state profile (Figure 3B(ii)), whereas the 400 Da FITC-dextran reaches deep into the hydrogel within an hour (Figure 3B(i)). This chemotactic gradient can be created across the complete depth of a 1 mm thick hydrogel, as apparent from similar Z-stack FITC levels (Figures 3C and S2 and Movie S2) at the top and bottom of the hydrogel (normalized to FITC level at the center). Also, this gradient can be sustained across a large width of the hydrogel (∼2 mm), with an ∼15 mm long uninterrupted and continuous interface between the fluid and hydrogel, whereas prior work had used posts. Hence, the open microfluidic system can create well-defined chemical gradients to viable glioma cells over the long durations needed to assay cell responses (48 h).

Figure 3. (A) Bright-field and FITC overlay of the gradient across the hydrogel at various time points with 400 Da FITC-dextran at (i) T = 0 h and (ii) T = 1 h. (B) Temporal development of the gradient across the open hydrogel for (i) 400 Da FITC-dextran, with the gradient developing rapidly to approach steady state within 1 h, and (ii) 10000 Da FITC-dextran, with the gradient developing slowly over 12 h. (C) The chemical gradient develops over the entire hydrogel depth (1 mm) based on the similar FITC levels at the top, bottom, and center of the hydrogel, and its invariance over the hydrogel width at steady state (24 h) for 10000 Da FITC-dextran. Significance is based on one-way ANOVA with Tukey’s post-hoc test, represented as ns, p ≥ 0.05; *, p ≤ 0.05.

Migration Cues to Glioma Cells in the Patterned Hydrogel Using Open-Top Microfluidics

To assess the ability of the open-top microfluidic system to deliver chemotactic gradients for cues to cells across the hydrogel width, we utilized the chemokine: CXCL12, which induces migration of CXCR4-expressing glioma cells, and AMD3100, which inhibits this signaling pathway. (54) As shown in the schematic in Figure 4A(i), gradients of CXCL12 induce calcium influx into the cell upon binding to its cell membrane receptor. Hence, glioma cells labeled with a calcium signaling probe that fluoresces upon calcium binding were used to quantify effect of the chemotactic gradient on cells across the hydrogel width. Fluorescence image analysis of a static culture of glioma cells was used to determine the CXCL12 level in the medium that is needed for signal rise above the baseline. Similarly, inhibition of CXCL12 stimulation at this level was validated using cells pretreated with 1 μg/mL AMD3100. (54) Based on fluorescence signal plots (Figure 4A(ii)) and images (Figure 4A(iii,iv)), we infer that 66 ng/mL CXCL12 in the medium is sufficient to cause signal rise within 5 min of stimulation, and this signal rise is effectively inhibited for cells pretreated with 1 μg/mL AMD3100. Beyond a critical time in the static culture, there is a gradual signal dropoff to the baseline level, similar to that under no CXCL12 stimulation. This is attributed to CXCL12 diffusional limitations, since this is not apparent in the dynamic 3D culture that constantly is replenished with CXCL12. When 66 ng/mL CXCL12 is used in the medium under dynamic 3D culture (7.5 μL/min perfusion) to create a gradient across the width of the cell-laden hydrogel, the fluorescence signal (Figures 4B(i) and S3) develops rapidly at proximal channel widths (0.25 mm) while taking longer to extend to 1 and 1.5 mm widths. With cells pretreated with 1 μg/mL AMD3100, the fluorescence level is diminished at the same widths (Figure 4B(ii)). Comparison of the normalized fluorescence signal with CXCL12 shows a 1.8-fold increase over the control, while the level remains close to the control after inhibitor pretreatment (Figure 4B(iii)). This validates ability of the open-top microfluidic system to deliver well-controlled cues from the CXC12 gradient to cells across the hydrogel width and cause its inhibition with AMD3100.

Figure 4. (A) (i) CXCL12 stimulation through calcium ion influx into U87 cells by binding to its receptor to cause fluorescence upon labeling with signaling probe. (ii) CXCL12 stimulation (66 ng/mL) in a static 2D culture causes fluorescence signal rise for ∼5 min, while the minimal signal rise for AMD3100 (1 μg/mL)-pretreated cells validates inhibition of this stimulation. (iii, iv) Fluorescence images with CXCL12 stimulation (scale bar = 500 μm) at indicated time points of (iii) maximum and (iv) baseline or control level with no stimulation. (B) (i) The temporal fluorescence signal from dynamic (7.5 μL/min) 3D cultures of U87 cell-laden hydrogel with 66 ng/mL CXCL12 in the medium rises sharply for cells at a hydrogel width of 0.25 mm, in proximity to the chemoattractant channel boundary, while widths farther from this boundary show less signal increase. (ii) The signal at the same hydrogel widths is inhibited after pretreatment with AMD3100 (1 μg/mL). (iii) Comparison of CXCL12-stimulated cells at the hydrogel width of 0.25 mm, without and with AMD3100 pretreatment, validates inhibition of CXCL12 stimulation. Significance of the differences between cells in the hydrogel under stimulation vs inhibition at the same time point were determined by one-way ANOVA with Tukey’s post-hoc test: *, p ≤ 0.05; **, p ≤ 0.01; ***, p ≤ 0.001; ****, p ≤ 0.0001.

Conclusions and Outlook

We present patterning of open-top 3D cultures that are integrated with microfluidic flow control to deliver biochemical cues, such as chemotactic gradients, for the purpose of assaying dynamic cell responses, such as cell migration. This open-top microfluidic approach using longitudinal aspiration and injection flows over 15 mm of the hydrogel/fluidic channel boundary, creates a sustained gradient across the hydrogel width (2 mm) that extends invariantly over the entire hydrogel depth (1 mm). This long and continuous hydrogel to fluidic channel interface is used to pattern the chemotactic gradient and sustain glioma cells over the 48 h culture period, whereas the posts used in prior reports within closed microfluidic systems create discontinuous interfaces, while challenges to maintaining a high aspect ratio limit the hydrogel depth to <0.1 mm. A GelMA–NorHA hybrid hydrogel was used to support both glioma cell viability and migration. The improved cytocompatibility of glioma cells in NorHA versus GelMA hydrogels is attributed to their rapid cross-linking ability due to the lower free radical levels needed to initiate step-growth polymerization, whereas the GelMA hydrogels formed by chain-growth cross-linking require a higher concentration of free radicals for polymerization initiation, leading to a loss in viability. Formation of chemotactic gradients was quantified using FITC-dextran of different molecular weights. The effect of chemotactic gradients across the hydrogel width on glioma cell responses, such as migration, was quantified by the fluorescence signal due to CXCL12, with signal inhibition through AMD3100 pretreatment. Improvements in spatial resolution would enable accurate determination of the specific hydrogel width regions that present significant differences in migration cues for modeling brain microenvironments. This open-top hydrogel microfluidic system requires release of the PDMS master mold from the cell-laden hydrogel for patterning adjoining fluidic channels. While release of the flexible silicone-like material is less likely to disrupt the adjoining UV-cross-linked cell-laden hydrogel pattern, inappropriate release can deform edges of the hydrogel region that interfaces with the fluid, thereby disrupting the ability to form long (∼cm scale) uninterrupted and continuous interfaces between the fluid and hydrogel. Hence, we optimized the UV cross-linking time to create well-cured hydrogels, while fixing the UV intensity at levels that maintain cell viability. Furthermore, the length to width (∼2 mm) and length to height (∼1 mm) aspect ratios for the hydrogel and channel patterns were maintained at 5 or less to ensure reproducible release of the PDMS master mold, without deformation to the edges of the hydrogel region that interfaces with the fluid. Given the challenges that closed microfluidic systems present to biocompatibility and growth factor loss from molecular adsorption, this open-top microfluidic system with longitudinal aspiration and injection flows can serve as an alternative platform for creating transverse chemotactic gradients across the 3D culture width, while extending invariantly over its length (∼15 mm) and depth (∼1 mm). This ability to create 3D cultures and chemotactic gradients over large lateral areas of approximately millimeter-scale depth that resemble the tissue microenvironment will be essential for in vitro disease modeling and emerging drug screening assays.

Apparatus Used

Master Mold for PDMS

Learn More

Clear Microfluidic Resin

Learn More

PR110
3D Printer

Legacy

Methods

Materials

HA with a 20–28% degree of norbornene functionalization was synthesized by the Caliari group. (46) LAP (Sigma), DTT (Sigma), and GelMA (300 g, 60%, Sigma) were used for the patterning of the hydrogel. For the temporal quantification of developed gradients, 0.33 mg/mL solutions of 10 kDa and 400 Da FITC-dextran (Sigma) were used. To study the migration response of U87 cells, cells were labeled with Fluo-4-AM (Thermo Fisher) and a CXCL12 (Biolegend) gradient in the presence or absence of pretreatment with AMD3100 (Sigma). 3-(Trimethoxysilyl)propyl methacrylate (Sigma) was used for surface modification of the glass slide.

Cell Culture

U87 cells were cultured in 1× MEM (Gibco) supplemented with 10% FBS (Gibco) and pen-strep (100 units/mL penicillin + 100 μg/mL streptavidin) at 37 °C in a humidified incubator. For harvest upon confluency (80%), the medium was removed, and cells were washed in 1× PBS (Thermo Fisher), followed by a 0.25% trypsin-EDTA treatment (Gibco) for 5 min, after which complete medium was added. Cells were then pelleted at 300g for 10 min. For migration assays, cells were loaded with Fluo-4-AM dye (2.5 μM), and CXCL12 (66 ng/mL) was added to the dynamic 3D culture to generate a gradient across the cell-laden hydrogel, containing cells without or with AMD3100 pretreatment (54) (1 μg/mL at 37 °C for 30 min).

Hydrogel Composition

For experiments in which cells were encapsulated in the hybrid hydrogel, ∼2–3 × 10⁶ cells were pelleted and resuspended in 100–200 μL of 1× PBS. Hydrogel precursor solution was added to the cell solution, such that the final solution contained 2% GelMA, 1% NorHA, 0.23 mg/mL DTT, and 0.0328 wt % LAP and the U87 cells at a concentration of ∼4 × 10⁶ cells/ml in 1× PBS.

Nanoindentation

A displacement-controlled nanoindenter (Optics 11 Piuma) was utilized to measure the elastic modulus of the photo-cross-linked hydrogel. After calibration of the nanoindenter, a hydrogel sample was loaded. Three measurements were taken for each hydrogel. An array of indentations were made at each measurement site to collect the data necessary for the analysis. Using the Hertzian contact mechanics model and assuming a Poission’s ratio of 0.5, the Young’s modulus was determined through the loading portion of the force versus distance indentation curve generated by the nanoindenter software. The elasticity data were analyzed and plotted using MATLAB.

3D Printing for Sample Holder and PDMS Molding

To pattern the hydrogel on the glass slide, a temporary PDMS mold was used. This mold was made using soft lithography techniques by casting 10:1 PDMS (Dow) on a 3D-printed master mold overnight at 60 C. The 3D-printed master mold was designed such that the patterned hydrogel and fluidic channel would have a total depth of ∼1 mm, the fluidic channel would have a width of 2 mm and be ∼15 mm long, and the hydrogel would be 2 mm wide in the center and have a total length of ∼22 mm. The PDMS master mold was printed using a Cadworks3D PR series printer in a Master Mold for PDMS Resin. The 3D-printed holder for interfacing the patterned hydrogel with fluidics was printed in Clear Microfluidics Resin (Cadworks3D) to enable transmitted light imaging. The holder was designed to have two holes on each side, with the same diameter as 1/16 in. microfluidic tubing through which the tubing is threaded, to serve as ports for the injection and aspiration. The holder was machined to rest on the edge of the Petri dish, with the ports designed such that the injection tubing can deliver fluid directly into the channel, while the aspiration tubing can be set to the height of the hydrogel.

Open-Top Hydrogel Patterning

Glass slides were methacrylate-silanized using 5 mL of 1:100 3-(trimethoxysilyl)propyl methacrylate in ethanol, with 150 μL of 1:10 diluted glacial acetic acid in water added and mixed in. This solution was pipetted onto fully coated plasma-cleaned glass slides (Tergeo Plasma Cleaner) and left for 3 h at room temperature. Glass slides were then rinsed three times in ethanol, followed by three rinses in distilled water. Slides were allowed to dry and were sterilized under a UV lamp for 6 h. Cured PDMS was detached from the 3D-printed PDMS mold carefully, and an inlet and outlet were drilled using a biopsy punch, followed by cleaning of the PDMS with compressed nitrogen to remove any dust, and rinsing of the PDMS mold in water. The PDMS molds were then sterilized under a UV lamp for 6 h. Prior to patterning, the PDMS mold was immersed in a 2% BSA solution (in 1× PBS) for 45 min, after which the PDMS mold was allowed to air-dry. The PDMS mold was brought into conformal contact with the silanized glass slide, leading to a reversible bond. Through the inlet, a cell-laden hydrogel precursor solution was filled into the mold. Photopolymerization of the hydrogel was carried out using 365 nm UV light at 5 mW/cm² for 120 s (Omnicure S2000). After 3 min, the PDMS mold was gently peeled away from the glass slide, with the patterned hydrogel adhered to the glass. The cell-laden hydrogel was carefully washed in 1× PBS, and complete medium was added such that the medium level was in line with the top surface of the hydrogel.

Fluidic Interfacing to Open-Top Patterned Hydrogels

The patterned hydrogel was moved to a microscope stage, and the 3D-printed fluidic holder was placed on top of it, with injection tubing to deliver fluid into the fluidic channel and aspiration tubing set on the top surface of the hydrogel. Two channels of an MFCS-EZ pump (Fluigent) were used for fluid injection: one was connected to a reservoir containing culture medium or 1× PBS, while the other was connected to a reservoir containing either FITC-dextran (for profiles in Figure 3) or CXCL12, in the absence or presence of AMD3100 (for profiles in Figure 4). Tubing from the injection reservoirs was connected to Flow-EZ flow sensors (Fluigent) and set up using the control software to deliver a continuous steady flow rate of 7.5 μL/min to the hydrogel channels. Tubing was routed from the flow sensors through the on-stage incubator into the hydrogel through the 3D-printed holder. Aspiration tubing from the 3D-printed holder was connected to a reservoir that was under negative pressure using a vacuum pump (Cellix). Vacuum levels were modulated by using an in-line air valve. All microfluidic connections were made using 1/16 in. outer diameter microfluidic tubing.

Time-Lapse Imaging

Z-stack time-lapse images through the depth of the hydrogel were acquired using an EVOS 620 microscope at 10× magnification with an on-stage incubator (ThermoFisher) every 15 min for 24 h. The on-stage incubator was set to maintain a humidified environment at 37 °C, with 5% CO₂ for maintaining cell viability. Microfluidic tubing was routed through the machined holes on the on-stage incubator for delivery of medium and chemoattractant, as well as for aspiration.

Image Analysis

ImageJ analysis was performed to capture the fluorescence signal from FITC-dextran and from cells cultured in monoculture, as well as cells encapsulated in 3D hydrogels. The grayscale intensity levels of individual cells were measured in each field of view to calculate the change in signal intensity over time due to stimulation. The distance from the hydrogel channel was measured based on the scale of the captured images. Cells were measured within similar corresponding distances from the channel boundary to quantify the spatiotemporal signal gradient. All signal intensities were normalized to a common control and plotted by utilizing MATLAB. For quantification of migration velocity in the hydrogels, migrating cells were identified at random and tracked from an origin point. Based on the scale, the distances were measured between each time point. The velocities between each time point were averaged over the duration of the experiment and plotted using MATLAB.

Statistical Analysis

Statistical analysis was performed in MATLAB and presented as the respective mean ± standard deviation for each data point in Figures 2–4, based on at least three experiments conducted on the hydrogel-integrated fluidic system. Significance was calculated by one-way ANOVA with Tukey’s post-hoc test, with p ≤ 0.05 considered as significant.

Apparatus Used

Master Mold for PDMS

Learn More

Clear Microfluidic Resin

Learn More

PR110
3D Printer

Legacy

Supplementary Materials

Download:

Supplemental Information.
Movie 1: showing the migration activity of U87 cells in the hybrid hydrogel.
Movie 2: showing the temporal evolution of the concentration gradients of 10 000 Da FITC-dextran across the hydrogel width.

References

Ostrom, Q. T.; Gittleman, H.; Truitt, G.; Boscia, A.; Kruchko, C.; Barnholtz-Sloan, J. S. CBTRUS statistical report: primary brain and other central nervous system tumors diagnosed in the United States in 2011–2015. Neuro-oncology 2018, 20 (suppl_4), iv1– iv86, DOI: 10.1093/neuonc/noy131 [Google Scholar]
Chaichana, K. L.; Zadnik, P.; Weingart, J. D.; Olivi, A.; Gallia, G. L.; Blakeley, J.; Lim, M.; Brem, H.; Quiñones-Hinojosa, A. Multiple resections for patients with glioblastoma: prolonging survival. J. Neurosurg. 2013, 118 (4), 812– 820, DOI: 10.3171/2012.9.JNS1277 [Google Scholar]
Wong, B. S.; Shah, S. R.; Yankaskas, C. L.; Bajpai, V. K.; Wu, P.-H.; Chin, D.; Ifemembi, B.; ReFaey, K.; Schiapparelli, P.; Zheng, X.A microfluidic cell-migration assay for the prediction of progression-free survival and recurrence time of patients with glioblastoma. Nat. Biomed. Eng. 2021, 5 (1), 26– 40, DOI: 10.1038/s41551-020-00621-9 [Google Scholar]
Logun, M.; Zhao, W.; Mao, L.; Karumbaiah, L. Microfluidics in malignant glioma research and precision medicine. Adv. Biosyst. 2018, 2 (5), 1700221, DOI: 10.1002/adbi.201700221 [Google Scholar]
Pedron, S.; Becka, E.; Harley, B. A. Regulation of glioma cell phenotype in 3D matrices by hyaluronic acid. Biomaterials 2013, 34 (30), 7408– 7417, DOI: 10.1016/j.biomaterials.2013.06.024 [Google Scholar]
Giarra, S.; Ierano, C.; Biondi, M.; Napolitano, M.; Campani, V.; Pacelli, R.; Scala, S.; De Rosa, G.; Mayol, L. Engineering of thermoresponsive gels as a fake metastatic niche. Carbohydr. Polym. 2018, 191, 112– 118, DOI: 10.1016/j.carbpol.2018.03.016 [Google Scholar]
Kasapidou, P. M.; de Montullé, E. L.; Dembélé, K.-P.; Mutel, A.; Desrues, L.; Gubala, V.; Castel, H. Hyaluronic acid-based hydrogels loaded with chemoattractant and anticancer drug–new formulation for attracting and tackling glioma cells. Soft Matter 2021, 17 (48), 10846– 10861, DOI: 10.1039/D1SM01003D [Google Scholar]
Young, E. W.; Beebe, D. J. Fundamentals of microfluidic cell culture in controlled microenvironments. Chem. Soc. Rev. 2010, 39 (3), 1036– 1048, DOI: 10.1039/b909900j [Google Scholar]
Varma, S.; Voldman, J. Caring for cells in microsystems: principles and practices of cell-safe device design and operation. Lab Chip 2018, 18 (22), 3333– 3352, DOI: 10.1039/C8LC00746B [Google Scholar]
Monteduro, A. G.; Rizzato, S.; Caragnano, G.; Trapani, A.; Giannelli, G.; Maruccio, G. Organs-on-chips technologies – A guide from disease models to opportunities for drug development. Biosens. Bioelectron. 2023, 231, 115271, DOI: 10.1016/j.bios.2023.115271 [Google Scholar]
Carter, S.-S. D.; Atif, A.-R.; Kadekar, S.; Lanekoff, I.; Engqvist, H.; Varghese, O. P.; Tenje, M.; Mestres, G. PDMS leaching and its implications for on-chip studies focusing on bone regeneration applications. Organs-on-a-Chip 2020, 2, 100004, DOI: 10.1016/j.ooc.2020.100004 [Google Scholar]
Castiaux, A. D.; Spence, D. M.; Martin, R. S. Review of 3D Cell Culture with Analysis in Microfluidic Systems. Anal. Methods 2019, 11 (33), 4220– 4232, DOI: 10.1039/C9AY01328H [Google Scholar]
van Meer, B. J.; de Vries, H.; Firth, K. S. A.; van Weerd, J.; Tertoolen, L. G. J.; Karperien, H. B. J.; Jonkheijm, P.; Denning, C.; Ijzerman, A. P.; Mummery, C. L. Small molecule absorption by PDMS in the context of drug response bioassays. Biochem. Biophys. Res. Commun. 2017, 482 (2), 323– 328, DOI: 10.1016/j.bbrc.2016.11.062 [Google Scholar]
Toepke, M. W.; Beebe, D. J. PDMS absorption of small molecules and consequences in microfluidic applications. Lab Chip 2006, 6 (12), 1484– 1486, DOI: 10.1039/b612140c

[Google Scholar]
Markov, D. A.; Lillie, E. M.; Garbett, S. P.; McCawley, L. J. Variation in diffusion of gases through PDMS due to plasma surface treatment and storage conditions. Biomed. Microdevices 2014, 16 (1), 91– 6, DOI: 10.1007/s10544-013-9808-2 [Google Scholar]
Heo, Y. S.; Cabrera, L. M.; Song, J. W.; Futai, N.; Tung, Y.-C.; Smith, G. D.; Takayama, S. Characterization and Resolution of Evaporation-Mediated Osmolality Shifts That Constrain Microfluidic Cell Culture in Poly(dimethylsiloxane) Devices. Anal. Chem. 2007, 79 (3), 1126– 1134, DOI: 10.1021/ac061990v [Google Scholar]
Hosic, S.; Bindas, A. J.; Puzan, M. L.; Lake, W.; Soucy, J. R.; Zhou, F.; Koppes, R. A.; Breault, D. T.; Murthy, S. K.; Koppes, A. N. Rapid Prototyping of Multilayer Microphysiological Systems. ACS Biomater. Sci. Eng. 2021, 7 (7), 2949– 2963, DOI: 10.1021/acsbiomaterials.0c00190 [Google Scholar]
Torino, S.; Corrado, B.; Iodice, M.; Coppola, G. PDMS-Based Microfluidic Devices for Cell Culture. Inventions 2018, 3 (3), 65, DOI: 10.3390/inventions3030065 [Google Scholar]
Sackmann, E. K.; Fulton, A. L.; Beebe, D. J. The present and future role of microfluidics in biomedical research. Nature 2014, 507 (7491), 181– 189, DOI: 10.1038/nature13118 [Google Scholar]
Cao, U. M. N.; Zhang, Y.; Chen, J.; Sayson, D.; Pillai, S.; Tran, S. D. Microfluidic Organ-on-A-chip: A Guide to Biomaterial Choice and Fabrication. Int. J. Mol. Sci. 2023, 24 (4), 3232, DOI: 10.3390/ijms24043232 [Google Scholar]
Lee, U. N.; Day, J. H.; Haack, A. J.; Bretherton, R. C.; Lu, W.; DeForest, C. A.; Theberge, A. B.; Berthier, E. Layer-by-layer fabrication of 3D hydrogel structures using open microfluidics. Lab Chip 2020, 20 (3), 525– 536, DOI: 10.1039/C9LC00621D [Google Scholar]
Su, C.; Chuah, Y. J.; Ong, H. B.; Tay, H. M.; Dalan, R.; Hou, H. W. A Facile and Scalable Hydrogel Patterning Method for Microfluidic 3D Cell Culture and Spheroid-in-Gel Culture Array. Biosensors 2021, 11 (12), 509, DOI: 10.3390/bios11120509 [Google Scholar]
Pei, J.; Sun, Q.; Yi, Z.; Li, Q.; Wang, X. Recoverable elastic barrier for robust hydrogel patterning with uniform flow profile for organ-on-a-chip applications. J. Micromech. Microeng. 2020, 30 (3), 035005, DOI: 10.1088/1361-6439/ab68b2 [Google Scholar]
Menon, N. V.; Tay, H. M.; Wee, S. N.; Li, K. H. H.; Hou, H. W. Micro-engineered perfusable 3D vasculatures for cardiovascular diseases. Lab Chip 2017, 17 (17), 2960– 2968, DOI: 10.1039/C7LC00607A [Google Scholar]
Trietsch, S. J.; Israëls, G. D.; Joore, J.; Hankemeier, T.; Vulto, P. Microfluidic titer plate for stratified 3D cell culture. Lab Chip 2013, 13 (18), 3548– 3554, DOI: 10.1039/c3lc50210d [Google Scholar]
Varhue, W. B.; Rane, A.; Castellanos-Sanchez, R.; Peirce, S. M.; Christ, G.; Swami, N. S. Perfusable cell-laden micropatterned hydrogels for delivery of spatiotemporal vascular-like cues to tissues. Organs-on-a-Chip 2022, 4, 100017, DOI: 10.1016/j.ooc.2022.100017 [Google Scholar]
Su, C.; Menon, N. V.; Xu, X.; Teo, Y. R.; Cao, H.; Dalan, R.; Tay, C. Y.; Hou, H. W. A novel human arterial wall-on-a-chip to study endothelial inflammation and vascular smooth muscle cell migration in early atherosclerosis. Lab Chip 2021, 21 (12), 2359– 2371, DOI: 10.1039/D1LC00131K [Google Scholar]
Park, D.; Lee, J.; Lee, Y.; Son, K.; Choi, J. W.; Jeang, W. J.; Choi, H.; Hwang, Y.; Kim, H.-Y.; Jeon, N. L. Aspiration-mediated hydrogel micropatterning using rail-based open microfluidic devices for high-throughput 3D cell culture. Sci. Rep. 2021, 11 (1), 19986, DOI: 10.1038/s41598-021-99387-6 [Google Scholar]
Park, D.; Son, K.; Hwang, Y.; Ko, J.; Lee, Y.; Doh, J.; Jeon, N. L. High-Throughput Microfluidic 3D Cytotoxicity Assay for Cancer Immunotherapy (CACI-IMPACT Platform). Front. Immunol. 2019, 10, 1133, DOI: 10.3389/fimmu.2019.01133 [Google Scholar]
Qasaimeh, M. A.; Ricoult, S. G.; Juncker, D. Microfluidic probes for use in life sciences and medicine. Lab Chip 2013, 13 (1), 40– 50, DOI: 10.1039/C2LC40898H [Google Scholar]
Brimmo, A.; Goyette, P.-A.; Alnemari, R.; Gervais, T.; Qasaimeh, M. A. 3D Printed Microfluidic Probes. Sci. Rep. 2018, 8 (1), 10995, DOI: 10.1038/s41598-018-29304-x [Google Scholar]
Qasaimeh, M. A.; Pyzik, M.; Astolfi, M.; Vidal, S. M.; Juncker, D. Neutrophil Chemotaxis in Moving Gradients. Adv. Biosyst. 2018, 2 (7), 1700243, DOI: 10.1002/adbi.201700243 [Google Scholar]
Juncker, D.; Schmid, H.; Delamarche, E. Multipurpose microfluidic probe. Nat. Mater. 2005, 4 (8), 622– 628, DOI: 10.1038/nmat1435 [Google Scholar]
Cors, J. F.; Lovchik, R. D.; Delamarche, E.; Kaigala, G. V. A compact and versatile microfluidic probe for local processing of tissue sections and biological specimens. Rev. Sci. Instrum. 2014, 85 (3), 034301, DOI: 10.1063/1.4866976 [Google Scholar]
Shinha, K.; Nihei, W.; Kimura, H. A Microfluidic Probe Integrated Device for Spatiotemporal 3D Chemical Stimulation in Cells. Micromachines 2020, 11 (7), 691, DOI: 10.3390/mi11070691 [Google Scholar]
Zhang, Q.; Mao, S.; Khan, M.; Feng, S.; Zhang, W.; Li, W.; Lin, J.-M. In Situ Partial Treatment of Single Cells by Laminar Flow in the “Open Space. Anal. Chem. 2019, 91 (2), 1644– 1650, DOI: 10.1021/acs.analchem.8b05313 [Google Scholar]
Lovchik, R. D.; Kaigala, G. V.; Georgiadis, M.; Delamarche, E. Micro-immunohistochemistry using a microfluidic probe. Lab Chip 2012, 12 (6), 1040– 1043, DOI: 10.1039/c2lc21016a [Google Scholar]
Queval, A.; Ghattamaneni, N. R.; Perrault, C. M.; Gill, R.; Mirzaei, M.; McKinney, R. A.; Juncker, D. Chamber and microfluidic probe for microperfusion of organotypic brain slices. Lab Chip 2010, 10 (3), 326– 334, DOI: 10.1039/B916669F [Google Scholar]
Qasaimeh, M. A.; Gervais, T.; Juncker, D. Microfluidic quadrupole and floating concentration gradient. Nat. Commun. 2011, 2 (1), 464, DOI: 10.1038/ncomms1471 [Google Scholar]
Safavieh, M.; Qasaimeh, M. A.; Vakil, A.; Juncker, D.; Gervais, T. Two-Aperture Microfluidic Probes as Flow Dipoles: Theory and Applications. Sci. Rep. 2015, 5 (1), 11943, DOI: 10.1038/srep11943 [Google Scholar]
Xie, L.; Kang, H.; Xu, Q.; Chen, M. J.; Liao, Y.; Thiyagarajan, M.; O’Donnell, J.; Christensen, D. J.; Nicholson, C.; Iliff, J. J.; Takano, T.Sleep drives metabolite clearance from the adult brain. Science 2013, 342 (6156), 373– 377, DOI: 10.1126/science.1241224 [Google Scholar]
Ray, L. A.; Heys, J. J. Fluid flow and mass transport in brain tissue. Fluids 2019, 4 (4), 196, DOI: 10.3390/fluids4040196 [Google Scholar]
Groothuis, D. R.; Vavra, M. W.; Schlageter, K. E.; Kang, E. W.; Itskovich, A. C.; Hertzler, S.; Allen, C. V.; Lipton, H. L. Efflux of drugs and solutes from brain: the interactive roles of diffusional transcapillary transport, bulk flow and capillary transporters. J. Cereb. Blood Flow Metab. 2007, 27 (1), 43– 56, DOI: 10.1038/sj.jcbfm.9600315 [Google Scholar]
Goodarzi, K.; Rao, S. S. Hyaluronic acid-based hydrogels to study cancer cell behaviors. J. Mater. Chem. B 2021, 9 (31), 6103– 6115, DOI: 10.1039/D1TB00963J [Google Scholar]
Yue, K.; Trujillo-de Santiago, G.; Alvarez, M. M.; Tamayol, A.; Annabi, N.; Khademhosseini, A. Synthesis, properties, and biomedical applications of gelatin methacryloyl (GelMA) hydrogels. Biomaterials 2015, 73, 254– 71, DOI: 10.1016/j.biomaterials.2015.08.045 [Google Scholar]
Unal, D. B.; Caliari, S. R.; Lampe, K. J. 3D Hyaluronic Acid Hydrogels for Modeling Oligodendrocyte Progenitor Cell Behavior as a Function of Matrix Stiffness. Biomacromolecules 2020, 21 (12), 4962– 4971, DOI: 10.1021/acs.biomac.0c01164 [Google Scholar]
Budday, S.; Nay, R.; de Rooij, R.; Steinmann, P.; Wyrobek, T.; Ovaert, T. C.; Kuhl, E. Mechanical properties of gray and white matter brain tissue by indentation. J. Mech. Behav. Biomed. Mater. 2015, 46, 318– 30, DOI: 10.1016/j.jmbbm.2015.02.024 [Google Scholar]
Vigata, M.; Meinert, C.; Bock, N.; Dargaville, B. L.; Hutmacher, D. W. Deciphering the molecular mechanism of water interaction with gelatin methacryloyl hydrogels: Role of ionic strength, ph, drug loading and hydrogel network characteristics. Biomedicines 2021, 9 (5), 574, DOI: 10.3390/biomedicines9050574 [Google Scholar]
Lin, C. C.; Sawicki, S. M.; Metters, A. T. Free-Radical-Mediated Protein Inactivation and Recovery during Protein Photoencapsulation. Biomacromolecules 2008, 9 (1), 75– 83, DOI: 10.1021/bm700782c [Google Scholar]
Fairbanks, B. D.; Schwartz, M. P.; Halevi, A. E.; Nuttelman, C. R.; Bowman, C. N.; Anseth, K. S. A Versatile Synthetic Extracellular Matrix Mimic via Thiol-Norbornene Photopolymerization. Adv. Mater. 2009, 21 (48), 5005– 5010, DOI: 10.1002/adma.200901808 [Google Scholar]
Mũnoz, Z.; Shih, H.; Lin, C. C. Gelatin hydrogels formed by orthogonal thiol–norbornene photochemistry for cell encapsulation. Biomater. Sci. 2014, 2 (8), 1063– 1072, DOI: 10.1039/C4BM00070F [Google Scholar]
Liu, C.; Sage, J. C.; Miller, M. R.; Verhaak, R. G. W.; Hippenmeyer, S.; Vogel, H.; Foreman, O.; Bronson, R. T.; Nishiyama, A.; Luo, L.; Zong, H. Mosaic Analysis with Double Markers Reveals Tumor Cell-of-Origin in Glioma. Cell 2011, 146, 209– 221, DOI: 10.1016/j.cell.2011.06.014 [Google Scholar]
Ledur, P.; Liu, C.; He, H.; Harris, A.; Minussi, D.; Zhou, H.; Shaffrey, M.; Asthagiri, A.; Lopes, M.; Schiff, D.; Lu, Y.; Mandell, J.; Lenz, G.; Zong, H. Culture conditions tailored to the cell of origin are critical for maintaining native properties and tumorigenicity of glioma cells. Neuro-Oncology 2016, 18 (10), 1413– 1424, DOI: 10.1093/neuonc/now062 [Google Scholar]
Hatse, S.; Princen, K.; Bridger, G.; De Clercq, E.; Schols, D. Chemokine receptor inhibition by AMD3100 is strictly confined to CXCR4. FEBS Lett. 2002, 527 (1–3), 255– 262, DOI: 10.1016/S0014-5793(02)03143-5 [Google Scholar]

Emergence of preferential flow paths and intermittent dynamics in emulsion transport in porous media

January 16, 2026March 1, 2024 by CadWorks3DAdmin

Emergence of preferential flow paths and intermittent dynamics in emulsion transport in porous media

Michael Izaguirre and Shima Parsa

We investigate the dynamics of emulsions within a two-dimensional porous medium using an integrated experimental approach that combines pore-level dynamics of single emulsions and bulk transport properties of the medium. Using an on-chip microfluidic drop-maker, we precisely control the concentration and sizes of emulsions injected into the medium. The dynamics of emulsion droplets are highly intermittent despite a small average velocity over the trajectory of an individual emulsion. At low concentrations, emulsions predominantly flow through pores with higher local velocities including pores smaller than the size of emulsion droplets, leading to trapping of emulsions and a decrease in medium porosity. Preferential pathways for the emulsions emerge within the medium once the porosity of the medium decreases significantly, from 55% to 36%. At constant injection flow rates and low concentrations of monodisperse emulsions, these pathways remain the only paths of transport of emulsions within the medium. Introducing a slight polydispersity in emulsion sizes unveiled additional transport pathways. Our pore-level measurements reveal that the average velocity of emulsions scales with the inverse residence time of an emulsion, and this scaling separates the emulsions into distinct groups along the emergent preferential pathways.

We kindly thank the researchers at University of Rochester for this collaboration, and for sharing the results obtained with their system.

Introduction

Transport of emulsions in porous media is a subject of significant interest in industrial, medical, and environmental applications including many food products, drug delivery, and immiscible displacement.1–6 The diversity and heterogeneity of most natural and environmental porous materials lead to heterogeneous flow distribution which significantly impacts the transport of droplets of emulsions in a medium.7 Furthermore, the transport properties of porous media can undergo dynamic alterations as a result of the flow and retention of materials inside the pores.8–10 Growth of biofilms in filters,2,11 transport of water-based emulsion in personal care product,12,13 or oil recovery3,8,9,14 are some of the examples in which the properties of the medium change in response to the flow of an immiscible phase. Earlier research shows that although the changes in bulk transport properties such as medium permeability and interstitial flow velocity are not considerably large upon the flow of individual droplets, the local and pore-scale flow can change dramatically leading to anomalous flow behavior locally.8,9,15,16 The transport properties of a single droplet of emulsion in porous media are dictated by the droplet sizes and network properties such as pore size distribution and medium wettability.1,4,15,17,18 Hence, the dynamics of a droplet can be described by the balance of the viscous, interfacial, and drag forces. Only two non-dimensional numbers Capillary number (ratio of viscous to interfacial forces) and Weber number (ratio of drag to interfacial forces) are used to describe the dynamics of droplets with small deformations.1,19–22 However, the collective dynamics of a group of emulsions in a complex network of pores are affected by the fluctuations in local flow due to the droplet–droplet and droplet–pore structure interactions.18 The collective transport of high concentration of emulsions in a medium with random pinning sites shows that dynamics of the droplets sharply transition from a creeping regime to flow along smectic rivers and in groups.23 The deformation of droplets in these experiments was negligible and the majority of the droplets never squeeze through small pores and only pin on the surfaces. Furthermore, measurements of bulk transport of large quantities of polydisperse droplets stabilized by a surfactant and injected into a three-dimensional porous medium show that mostly small droplets appear in effluent and large droplets remain trapped in the medium due to the large pressure required to deform the large emulsion droplets.15,17,18,24 Nevertheless, the pore-level and collective dynamics of droplets in a network of pores, and the impact of trapping and re-mobilization of droplets on pore-level and macro-scale transport properties remain to be examined at the pore-scale. One of the challenges in accurate experimental investigation is tracking and precise object detection in an environment where the interfaces of droplets are in contact and droplets deform based on pore sizes.

In this paper, we quantify the pore-level dynamics of monodisperse emulsions flowing through a two-dimensional (2D) porous medium experimentally. By incorporating a microfluidics drop-maker on the same chip as the 2D porous medium, we control the concentration and sizes of the injected emulsions precisely. In these experiments, we track individual droplets as they flow into the medium using optical microscopy and a long-range recording mode while monitoring the bulk pressure gradient across the medium. By employing advanced image analysis and object tracking, we track individual emulsions as they flow through the medium. We show that at low concentrations, emulsions flow through pores with higher local velocities without being selective about the size of the pores they encounter, and this lack of selectivity can lead to the emulsions becoming trapped. Once a significant number of pores are filled with droplets, newly injected emulsions continuously flow through a few remaining open paths. We show that the average velocity of the droplets that flow through the medium scale with the inverse of the total time of residence in the medium and is proportional to the path lengths of the droplets independent of the distribution of sizes of the emulsions.

Materials

Master Mold Resin

Learn More

Pr Series

Experimental method

We generate emulsions and characterize the dynamics of emulsions in 2D micromodel of porous media using microfluidics, fluorescent microscopy, and bulk transport properties of the medium. One of the challenges in studying emulsions in porous media is to control the size, concentration, and injection frequency of emulsions.18,25 This is mainly due to the density contrast between the dispersed phase (emulsions) and the continuous phase. To overcome this challenge, we leverage the capabilities of microfluidics in producing well-controlled monodisperse emulsions.21,26–28 We design an on-chip drop-maker in series with a 2D porous medium as shown in Fig. 1a. This design allows us to control the injection frequency and concentration of emulsions in a porous medium.

2.1 Microfluidics 2D porous media

To generate monodisperse droplets and inject them into a porous medium in a laminar flow condition, we design the microfluidic drop-maker to operate in the dripping regime.27 The drop-maker consists of an inlet for the dispersed fluid (water and 0.1 w% fluorescein sodium salt) at the center, and two inlets for the continuous phase on either side. The continuous phase is a fluorinated oil HFE750 (engineering fluid by 3 M) with 5 w% surfactant FSH oil (by Krytox). The interfacial tension between the dispersed phase and the continuous phase is γ = 26 mN m−1. In the dripping regime, the droplet sizes are proportional to the inlet geometry.13 At equilibrium, where the inner phase fluid is protruding out of the inlet and into the outer phase, the pressure inside the droplet Pd is balanced by the pressure in the outer fluid (P0) and the capillary pressure, image file: d3sm01465g-t1.tif. Here, Rd is the radius of the droplet. The droplet snaps off once the pressure inside the droplet exceeds the outer pressure. The radius of the droplet is Rd > 2R, in a channel with radius R and circular cross-section.29–33

Here, the water inlet is a rectangular channel with dimensions of 84 μm × 200 μm, entering an area measuring 1050 μm × 200 μm, as shown in Fig. 1a. The two oil inlets each have dimensions of 115 μm × 200 μm. The entire channel spans 1100 μm in length and 200 μm in height, tapering down to a 325 μm wide channel before entering the porous medium. We use a syringe pump to inject the continuous phase at a constant flow rate of 5 mL h−1. However, to control the generation of individual droplets precisely, we use a pneumatic pump as shown in the schematic of the experimental setup in Fig. 1b. The viscous pressure of the flow of the continuous phase is balanced by a constant pressure, provided by the hydrostatic pressure of the closed water column. The pneumatic pump applies an additional pulse of pressure to the closed water column at 174 kPa for a duration of 200 ms. This method robustly produces monodisperse emulsion droplets with an average diameter of 295 ± 7 μm with a narrow distribution as shown in Fig. 2. The corresponding capillary and Weber number of the dropmaker in these experiments are Ca = 3 × 10−3 and We = 1.5 × 10−3. The radii of the emulsions match our prediction of Rd > 2R. The snap-off and monodispersity of the droplets in our experiments are assisted by the hydrophobic coating (aquapel) of all surfaces.34 However, small variations in pulses result in a slightly more polydisperse distribution of droplet sizes. For example, we find that multiple consecutive pulses result in a wider distribution of droplet sizes of 350 ± 10 μm as shown in Fig. 2. We continuously monitor the pressure gradient across the medium using a pressure transducer (Omega-PX409) and apply variational mode decomposition to the signal to eliminate the high-frequency noise of the transducer.35

Fig. 2 Probability distribution function of the sizes of the emulsions in monodisperse (Exp1: blue) and polydisperse (Exp2: red) experiments. The total number of emulsions in Exp1 is 1334, and in Exp2 is 1666.

We design and fabricate 2D porous media using standard soft lithography and microfluidics techniques.36 To obtain a pattern of random pore size distribution, we use a 2D micrograph of a three-dimensional glass bead-pack imaged by a confocal microscope.7 We further enhance the pore size heterogeneity by imposing a gradient in pore size distribution with a larger porosity at the inlet compared to the porosity downstream. This gradient in porosity represents the heterogeneity of natural and geological porous structures.37 We quantify the porosity and pore size distribution of the 2D porous medium using a novel algorithm that utilizes Voronoi tessellation and skeletonization.38–41 The pore size distribution in the medium has an average pore size of 403 μm and varies between 150 and 1150 μm as shown in Fig. 3.

An important point to note from Eq. (3) is that the amplitude of the fundamental differs from the one of a sinewave of amplitude 2Vo by a factor of 4 π . Such a difference biases the measurements and affects the precision but can be mostly corrected (up to a couple of percents) by following an algorithm proposed by Subhan54. Another consideration is the introduction of harmonics in the circuit, which raises the noise floor of the system.

Fig. 3 Probability distribution of the pore sizes in the medium (solid line), the 1/3 entrance to the medium (light gray), and the 2/3 end of the medium (dark gray).

To ensure that the emulsion droplets are small enough to enter the medium, the physical dimensions of the porous medium are proportionally adjusted to allow some passage of the droplets. In these experiments, we utilize a microfluidics 3D printer (CADworks3D Pr110-385 nm). Using this cutting-edge resin-based 3D printer, boasting an XY resolution of 40 × 40 μm2 and a Z resolution of 5 μm, we fabricate microfluidic master-molds with a variety of dimensions. To achieve smooth surfaces on the master-mold, which is critical for the performance of our microfluidic devices, we optimize the printing settings for a commercial powder-base resin with low light dispersion. By controlling the UV-exposure and curing time, the edges and surfaces are smooth. The master-molds are then filled with polydimethylsiloxane (PDMS) and cured at 60 °C before plasma cleaning and bonding to a glass slide.

2.2 Pore-scale imaging

To quantify the dynamics of emulsion within the porous medium, we use a widefield optical microscope (Axiozoom) and a long-range-record camera (FasTec IL5). The camera is operated at 50 Hz with a resolution of 2500 × 1000 pixels at 16 bits, providing a high dynamic range. The microfluidic porous medium is illuminated with a collimated RGB backlight LED providing a high contrast image where emulsions can be identified.

We characterize the dynamics of emulsions at the pore-level and across the entire model porous medium utilizing a modified particle tracking algorithm that accounts for objects in close contact and with highly intermittent kinematics. While most particle tracking methods are optimized to identify sparse objects,42 emulsions trapped in a porous medium are in close contact with each other and are squeezed into a solid structure and can be slightly deformed, see Fig. 4. Here, we first subtract the solid background while applying a drift correction on all images to enhance the accuracy of object detection. Using a circular hough transform, we identify individual droplets within the medium as shown in Fig. 4. Once all droplets are identified, we employ a global nearest-neighbor (GNN) tracking method under Sensor Fusion and Tracking Toolbox in MATLAB R2023. The GNN tracker uses the global nearest-neighbor assignment algorithm to match its detection to identified tracks based on predicted position, velocity, and acceleration. The GNN tracker forms a cost matrix by calculating the distance between each detection and existing tracks. Using this cost matrix, it categorizes the detected objects into either assigned pairs with tracks or unassigned, subsequently updating or initializing tracks as appropriate. Since our detection method is highly accurate, we assign a high cost to new tracks created outside the spatial area in which new emulsions are introduced into the field of view. We overcome the natural challenge of tracking objects that are constantly trapped and mobilized by using an Interacting multiple-model filter. The high-resolution imaging and enhanced edge detection are crucial in successfully applying the GNN tracker to the highly intermittent dynamics of emulsion. See ESI† of Fig. S4 providing a dynamic visual representation of the emulsion transport through the porous medium.

Fig. 4 Transport of individual droplets injected into a 2D porous medium as a function of time (a) 5 s, (b) 25 s, (c) 42.5 s, (d) 62.5 s. Blue circles mark the emulsions. Scale bar is 1 mm.

Results

he dynamics of emulsions in porous media are highly intermittent despite the tendency of the droplets to travel at the center of the pores. As single droplets enter the porous medium, they flow through paths with a higher average velocity. In these experiments, we form and inject the droplets at low concentrations and distribute their points of entry into the medium in the cross-sectional direction, Fig. 4a. The low concentration of emulsion is crucial to avoid a yield stress behavior.41 The emulsions flowing into a porous medium, naturally follow the streamlines with larger velocities. However, there is no feedback mechanism that would prohibit their entry to a pore or a pore throat smaller than the diameter of the droplet. Interestingly, in a porous medium with a random distribution of pore sizes, a considerable number of high-velocity paths flow through small pores. Hence, we observe a substantial number of emulsions getting trapped in the medium during the injection of the first batches of emulsions as seen in Fig. 4b. While a few emulsions find their way to the outlet, more than 65% of the emulsions are trapped following their predecessors as seen in Fig. 4c. A droplet trapped in a pore does not completely block the flow of the continuous phase in this area and the continuous phase passes around the droplet. Consequently, the changes in the local flow within the first few seconds of these experiments do not lead to a change in the global flow, as opposed to pore blocking seen in experiments focusing on conformance control in oil recovery.9,43 Additionally, our continuous measurement of the pressure drop across the medium confirms that the bulk flow is not affected by a few trapped emulsions in the medium. Trapping of a few droplets in the medium changes the medium porosity from 55% in Fig. 4a to 49% in Fig. 4c. Despite the considerable change in porosity, the pressure gradient across the medium increases only from 1400 Pa to 1450 Pa, further confirming the presence of a flow around individual emulsions and through the pores. Further injection of emulsions into the medium results in substantial clogging of individual pores in the medium as seen in Fig. 4d. Considering that the volumetric flow rate is held constant throughout this experiment, one expects that flow should be redirected to other open pores. Once the porosity of the medium decreases to 36% and many pores are filled with emulsions, newly injected emulsions follow paths that were not explored earlier and find their way to the medium outlet. Interestingly, we find that some entire paths are filled with emulsions (seen in the center of Fig. 4d) before the flow of emulsions is diverted. Finally, a tortuous path is formed which is followed by newly injected emulsions. We do not observe clogging of the entire medium at the constant injection flow rate and the concentration of the droplets remains to be below a jamming transition.44 Moreover, the balance between viscous and capillary forces does not change dramatically to mobilize a large number of droplets.45,46

To quantify the emerging flow paths within the medium, we track individual emulsions and superimpose the paths taken by these emulsions as shown in Fig. 5. A few preferential paths are formed in the medium and the subsequently injected emulsions continue flowing along these paths. While only a few tortuous paths are established in the flow of monodisperse emulsions (Fig. 5a), additional paths are explored by introducing a slight polydispersity in the emulsion sizes (Fig. 5b). Interestingly, in the experiment with larger and polydisperse emulsions, large droplets squeeze through the pores and create small perturbations in the flow of subsequent trailing droplets. Hence, droplets are more likely to switch paths as shown in Fig. 5b.

Fig. 5 Spatial distribution of emulsions in (a) monodisperse (Exp1) and (b) polydisperse (Exp2) experiments. Heatmap represents the log-transformed time (in seconds) spent at each location, normalized to match the maximum time value of Exp2.

To quantify the variability of the velocities of the emulsions, we calculate the probability density function (PDF) of the velocities in different experiments as shown in Fig. 6a. The PDF of the magnitude of the velocities of emulsions has an exponential decay with a long stretched tail indicating the presence of rare events with very large velocities compared to the interstitial velocity. The interstitial velocity is vint = q/ϕ, where q is the volumetric flow rate per cross-sectional area and ϕ is the medium porosity. The distributions of velocities of emulsions have similarities with the PDF of the velocities of the flow of a single-phase continuous fluid, measured in identical but separate experiments using 1 μm tracer particles particle image velocimetry (PIV).7,9 However, the tail of the PDF of the velocities of droplets stretches to much larger velocities (5 × vint) than that of the single-phase flow (3 × vint).

Fig. 6 (a)–(c) Probability density function of velocities of emulsions normalized by the interstitial velocity (a) PDf of the magnitude of velocity (b) PDF of the longitudinal component of velocity (vx) (c) PDF of the transverse velocity, vy. Blue triangles represent the monodisperse emulsions, red squares represent the polydisperse emulsions, and black diamonds represent the tracer particles velocities. (d) Distribution of the deviation of location of first 100 monodisperse droplets from fluid elements for 3 time-stamps, 2 seconds after entering the medium (red), 10 seconds (blue), and by the time either object reaches the end of their path in view (black).

Comparing the PDF of velocities of droplets with a single-phase flow confirms the intermittency in the dynamics of droplets where trapping, re-mobilization, squeezing and bursts through pore throats are common. The dynamics of emulsions in these experiments exhibit unique features reminiscent of transport in a porous medium: (1) emulsions only pass through certain areas and some pores within the medium are never explored by the droplets, as seen in Fig. 5. (2) Trapping and accumulation of emulsions within the porous structure result in changes in the medium permeability, leading to an increase in the viscous forces. The latter effect, only observable in pore-level measurements,9,10,43 can significantly change the flow in neighboring pores and consequently affect the global flow. Despite the finite size of the emulsions, and an expected slower velocities than fluid elements (represented as tracers), we find that the PDF of magnitude of the velocities of emulsions has an average comparable to a single phase flow in agreement with the constant flow driven experiment.

The PDF of velocities of emulsions in the direction of the imposed flow, Fig. 6b, has a positive average, 〈vx〉 = 270 μm s−1, consistent with the direction of flow. The significant negative tail in the polydisperse experiments (Exp2) is due to the tortuous path taken by droplets in this experiment. The PDF of vy of emulsions has a slightly higher probability in the downward (vy < 0) than the upward direction, aligning with the most common paths observed in Fig. 5. The average dynamics of droplets in these experiments (Exp1: monodisperse and Exp2: polydisperse) are independent of the distribution of droplet sizes. The average velocity is dominated by the large number of droplets experiencing slow dynamics. However, the rare events with large velocities and bursts of motion are more probable in the experiments with more variable sizes of emulsions.

Additional insights into the preferential paths of the droplets can be drawn by comparing the trajectory of a droplet with a fluid element as it enters the medium. The path of a droplet is determined by the local stress (proportional to the velocity gradient) on the surface of the droplet, while the path of a fluid element is dictated by the fluid velocities. Hence, the trajectory of an emulsion droplet deviates from a fluid element due to the finite size of a droplet. The departure of the trajectory of a droplet from fluid elements increases with time as shown in Fig. 6d. We quantify the distribution of the deviation between the location of the tracers and the emulsions entering the medium at the same initial position. The locations of the tracers are determined by integrating their trajectory using the flow velocity field (from PIV) and a fourth order Runge–Kutta integration scheme.42 The emulsions closely follow the path taken by a tracer for the first few seconds but the location of the center of the droplet quickly departs from the fluid element. After only 10 seconds the distance between the location of the droplets and fluid elements is distributed evenly across the medium. The distribution of the distances shifts towards larger values and closer to the length of the medium by the time either the emulsion or the fluid element reaches the end of their paths. The distribution is converted into a smooth function using MATLAB Kernel smoothing function estimate for univariate and bivariate data.

Our understanding of emulsion transport in porous media can be further enhanced by quantifying the dependence of the average velocity of the emulsions on the time of travel through the medium, which we refer to as residence time. As shown in Fig. 7, the average velocities of all emulsions that pass through the medium scale with the inverse residence time of the emulsions, 〈v〉 ∼ 1/(resident time). We measure the residence time of each individual emulsion as it traverses the medium. Emulsions that pass through the medium quickly have a short residence time, while those that become trapped have a much longer residence time. The longest residence time recorded in our experiments is 800 seconds, comparable to the duration of the experiment, and belongs to an emulsion droplet trapped in the medium. The scaling of 〈v〉 with inverse resident time holds for all emulsions that exit the medium, represented by the light color of the symbols in Fig. 7a. The color of the symbols represents the value of the Euclidean distance along the trajectory of the emulsions, defined based on the initial and final locations of each emulsion droplet along its path. The longest Euclidean distance within the 2D porous medium corresponds to the diagonal of the medium (13.2 mm). Interestingly, the scaling of the average velocity is independent of the distribution of the sizes of the emulsions (Exp1, Exp2). Moreover, the longitudinal component of the velocity scales with the residence time similar to those with the average velocity, 〈vx〉 ∼ 1/(resident time). We attribute the 〈vx〉 scaling to the dominance of the longitudinal direction in the transport of emulsions within the medium. The transverse velocity, 〈vy〉, is an order of magnitude smaller than the longitudinal component in these experiments. The average velocities of the emulsions that are permanently trapped in the medium, or those that do not leave the medium for the duration of the experiment, are smaller than the velocities of emulsions of similar residence time that pass through the medium. Therefore, as illustrated in Fig. 7, the average velocities of the emulsions that remain within the medium consistently fall below the reference line that encompasses those that pass through it. We observe that droplets with longer Euclidean paths, or equivalently those closer to passing through the medium, are more likely to have an average velocity that approaches the population following the scaling with inverse residence time. Throughout the experiments, we extracted over 6 million positional updates and their corresponding velocities. Therefore, in Fig. 7, we aggregate numerous data points into a single symbol for better visualization. The symbol's size corresponds to the logarithmic scale of the data point count.

Fig. 7 Average velocity vs. residence time of emulsions, (a) magnitude of velocity and (b) longitudinal component of velocity. Crosses represent the monodisperse (Exp1) data and circles correspond to polydisperse (Exp2) data. Marker sizes represent the number of emulsions within each velocity-residence time bin. The colormap corresponds to the Euclidean distance along the trajectory of the emulsions.

The scaling of average velocity with inverse residence time of emulsions is described with a simple dimensional argument image file: d3sm01465g-t2.tif. We find that the corresponding length scale is the path length of the trajectory of the emulsions. Here, the emulsions are more likely to take either preferential paths identified in Fig. 5. We identify the emulsions with the paths they take and show that in the monodisperse experiments where emulsions continuously follow two distinct paths, the emulsions on the longer path have a slightly smaller average velocity. Nevertheless, the average velocities of all emulsions are distinctly split into two groups as shown in Fig. 8a. This observation is further confirmed by the location of the exit point of the emulsions as shown in Fig. 8b. Moreover, similar separation of path lengths and exit points are observed for the polydisperse emulsions as seen in Fig. 8c and d. Observation of the distinct paths and exit points in this medium provides clear evidence of the emergence of preferential paths independent of the emulsion sizes. These paths emerge as a consequence of the solid pore structure modified by the trapping of emulsions.

Fig. 8 Dependence of the average velocity of (a) monodisperse and (c) polydisperse emulsions on residence time for emulsions that exit the medium. Final exit location of (b) monodisperse and (d) polydisperse emulsions along the cross sectional direction. Blue symbols represent the path leading to the exit point on top of the medium, red corresponds to the path leading to the bottom of the medium, dashed gray line separates the two populations.

Conclusions

In the present study, we successfully investigate the pore-level dynamics of monodisperse emulsions navigating a two-dimensional porous medium. By leveraging the versatility of microfluidic techniques, we control the concentration and sizes of emulsions, in addition to the injection rate of emulsions, by integrating an on-chip drop-maker driven by an external pneumatic pulse. We find that at low concentrations, emulsions flow through pores with higher local velocities and independent of the pore sizes, leading to trapping of emulsions in pores smaller than the emulsion sizes. This leads to a 35% reduction in the porosity of the medium. Few preferential and highly tortuous flow paths emerge within the medium after this reduction in porosity, along which low-concentrations emulsions continue to flow. Our measurements of the pore-level velocities of the emulsions show a highly intermittent dynamic consisting of trapping and subsequent mobilization of emulsions within the porous structure. Nevertheless, we find that the average velocities of all emulsions that flow through the medium scale with the inverse residence time of the emulsions and is distinguished by the flow paths emulsions take within the medium. This emergent scaling holds for slightly polydisperse emulsions.

The introduction of a slight polydispersity in the emulsions enhances the transport of emulsions despite the larger sizes of the droplets revealing more fluctuations in transport paths. Independent of the distribution of droplet sizes, trapped emulsions within the porous structure play a pivotal role in defining preferential transport paths, showcasing the interaction intricacies between the droplets and the porous network. Although the current experiments are focused on the dynamics of low concentrations of emulsions in porous media at a moderately slow flow rate, corresponding to a small Reynolds number, in the laminar regime, the approach serves as a foundational method for characterizing emulsion dynamics in a variety of flow regimes. The formation and persistence of preferential flow paths and droplet–droplet interactions at higher flow rates where the local flow can be highly unstable remains to be explored. These findings and the associated experimental methodology have the potential to drive advancements in areas such as soil remediation, drug delivery, and oil spill cleanup.

CRISPR-Cas9 Extracellular Vesicles for Treating Hearing Loss

January 16, 2026December 15, 2023 by CadWorks3DAdmin

CRISPR-Cas9 Extracellular Vesicles for Treating Hearing Loss

Xiaoshu Pan , Peixin Huang ,Samantha S. Ali ,Tarun E Hutchinson

The treatment of inner ear disorders remains challenging due to the intrinsic anatomical barriers. The majority treatments and delivery approaches for accessing inner hair cells are still engaged with surgical intervention, which is highly invasive and inconsistent in terms of efficacy and safety. In order to address this challenge for crossing anatomical barriers, we report an extracellular vesicle (EVs) -based delivery approach to inner hair cells, which enables carrying CRISPR/Cas9 ribonucleoprotein (RNP)-sgRNA complex in high-throughput and high efficiency. The novel Microfluidic Droplet-based Electroporation System (µDES) is developed to efficiently load cargos into EVs via millisecond pulsed, low-voltage electroporation within flow-through droplets as enormous bioreactors in a continuous-flow and scalable manner. The observed loading efficiency of CRISPR/Cas9 RNA complex into EVs (RNP-EVs) is 10-fold higher than current bulk cuvette electroporation with hundred-fold increase of processing throughput. The low-voltage electroporation minimized the Joule heating influence on nanosized EVs, which retained the native surface membrane properties of cargo-loaded EVs. Both ex vivo and in vivo testing in Shaker-1 mice model demonstrated the high biocompatibility and biodistribution of produced RNP-EVs in the mouse cochlea penetrating inner hair cells. In contrast, the CRISPR/Cas9 RNP lipid-like nanoparticles (RNP-LNPs) control group was unable to penetrate anatomical barriers to access inner hair cells. In the Shaker-1 mouse model, DES produced RNP-EVs demonstrated much higher editing efficiency at Myo7ash1 mRNA level and showed significant hearing recovery in the Myo7aWT/Sh1 mice via Auditory Brainstem Response (ABR) testing. The report work will present a new solution to advance gene therapy in treating sensorineural hearing loss .

We kindly thank the researchers at University of Florida for this collaboration, and for sharing the results obtained with their system.

Introduction

Hearing loss is one of the most common neurodegenerative disorders with genetic causes in human affecting more than 450 million people worldwide1-2. In situ delivery of functional gene materials to cochlear hair cells is one of the most promising strategies to repair hair cells and restore hearing function in vivo1,3-6. To date, the gene therapy targeting cochlear hair cells are heavily relied on engineered AAV vectors that can transduce inner hair cells more efficiently. However, a few biosafety investigations of high dose AAV vectors in non-human primates are limiting the clinical translation5,7. On the other hand, the commonly used AAVs in gene therapy have limited capacity on cargo size (∼ ≤ 5Kb), which is unable to carry CRISPR SpCas9-gRNAs, as well as the Myo7a gene (∼100 Mb or a cDNA of ∼ 7 Kb) we studied in this work as a hearing loss causative gene in inner ear hair cells8. Although the lentiviruses have a cargo capacity of ∼10 Kb, the risk of insertional mutagenesis and severe immunogenicity are still significant concerns for clinical translation9. Alternatively, extracellular vesicle (EVs)-based delivery is emerging as a novel, safe approach for addressing such challenges employed in gene delivery10-13, owing to the intrinsic biocompatibility, low immunogenicity, tissue penetration ability, and superb tunability14-17. Although, EVs have been utilized to deliver various genes into tissues, the delivery of CRISPR/Cas9 ribonucleoprotein (RNP)-sgRNA complex has not been explored for inner ear tissue yet. The first-in-human trial using umbilical cord mesenchymal stromal cell derived EVs demonstrated their regenerative potential to attenuate inflammation-based side effects from cochlear implantation and noise trauma10,18, which indicates natural distribution of EVs across anatomical barriers in cochlear may present, making EVs more favorable in hearing loss gene therapy than its counterparts, viral vectors.

However, loading CRISPR RNP complex into EVs has been a grand challenge. Current methods suffer from low loading efficiency and not scalable. For instance, chemical transfection rate for CRISPR RNP complex is generally below < 25%, and the produced EVs are in low stability19. Utilizing cells engineered as the primary EV producer is limited with cargo type and copy numbers that can be passed to EVs for encapsulation. Although electro-transfection is more efficient in terms of transfection rate (∼ 50%), the scalability is limited with only a few milliliters of processing volume in their throughput20-21. Different from cells, EVs generally have much smaller size and higher Brownian motion. Therefore, we introduce a novel continuous-flow platform utilizing microfluidic droplet-based EV electroporation (µDES), which can handle variable cargos loaded into EVs in large throughput and high efficiency. The saturated cargo concentration in the confined uniform droplet bioreactors can maximize mass transport and electroporation efficiency. Such streamlined EV electro-transfection using continuous-flow droplets as the enormous micro-bioreactors has not been explored elsewhere. Compared to microfluidic nanoporation for EV transfection15,17,22-24, the continuous flow enables much larger scale and throughput processing (up to litter range). Only a low-voltage (∼10-30 volts) DC power is needed, which avoids Joule heating and thermal damage on nanosized EVs25. Compared to chemical transfection which introduces unpurifiable chemicals potentially toxic to in vivo system, the instant electric field application across flow-through droplets in millisecond (∼ms) minimizes perturbation of EV molecular components to retain the natural EV property and biocompatibility. We also employed FDA approved additive trehalose 26-30 in the buffer system to preserve EVs in good stability29, 31, and minimize membrane aggregation26-27 and leakage after electro-transfection as reported by other research28, which is suited in clinical settings.

In the realm of hearing loss, CRISPR/Cas9 technology are demonstrating promising editing specificity and efficiency by targeting and correcting genetic mutations responsible for various hereditary hearing disorders32. For instance, it can address mutations in TMC1Bth crucial for the development of sensory hair cells or cochlear function, potentially restoring auditory function33. More pioneer investigations exhibited that CRISPR/Cas9 technology can be applied to both congenital and acquired hearing loss, offering a multifaceted approach to treatment3,7,33-35. In this work, we target on Myo7a gene which plays an essential role in the development and maintenance of auditory hair cells. Myo7a mutation has been identified as the major causative gene (39–55% of the total cases) in Usher syndrome (USH1B), syndromic and non-syndromic hearing loss (DFNA11 and DFNB2), and age-related hearing loss36-37. Thus, timely removal of mutant myoVIIa allele could prevent progression of hearing loss. However, current Myo7a gene therapy is unattainable, due to limited options of vectors. Our work on both ex vivo and in vivo testing in Shaker-1 mouse model demonstrated the high biocompatibility and biodistribution in the inner ear tissue from our μDES produced RNP-EVs in mouse cochlea penetrating into inner hair cells. The RNP-EVs displayed much higher editing efficiency at Myo7ash1 mRNA level and showed significant hearing recovery in Myo7a WT/Sh1 mice via Auditory Brainstem Response (ABR) testing. In contrast, the CRISPR/Cas9 RNP lipid-like nanoparticles (RNP-LNPs) control group was unable to penetrate anatomical barriers to access inner hair cells. Our approach will allow the rapid loading of CRISPR into EVs for delivery of Cas9 without using a split vector, which offers the opportunity to customize sgRNAs addressing different mutant alleles within one gene, and enable customization to patient genetic heterogeneous mutation background, leading to a clinically translatable approach for overcoming current challenges in gene therapy.

RESULTS

High throughput and highly efficient EV electro-transfection via μDES platform

We developed µDES platform for enhancing EV transfection efficiency, loading capacity, and throughput. The concept of µDES platform and functionality are illustrated in Figure 1 A-E. The device composed one aqueous inlet with purified EVs and RNP cargos, one oil inlet, electroporation chamber and one droplet outlets which streamlines droplet generation with electroporation using a low voltage DC power supply. The device fabrication was detailed in supplemental materials. Continuous generation of droplets uniformly as enormous bioreactors in fast speed enables large-scale encapsulation of EVs with high concentration cargos (Figure 1F). In droplet space, the cargo transport under electric field is more efficient to cross transient pores from EV membrane via electrical mobility of cargo themselves, electric flux, and concentration gradient, which only needs milliseconds to complete in such small scale, in turn, maximizing the loading efficiency and capacity (Figure 1B). The uniform electric field distribution can be formed across each flow-through droplet for highly efficient electroporation as proved by COMSOL simulation in Figure 1C. Notwithstanding, droplet-based electroporation could also lower the dispersity of electric field in the small volume and contribute to more homogeneous electric distribution in small volume range38-40. The COMSOL simulation of μDES device was conducted which showed uniform distribution of both flow profile and electric field profile with focusing on the droplet passing through the electroporation chamber (Figure 1 D and E). The scalability of μDES was also studied by collecting droplets in a large container (Figure 1F) with uniform size of produced droplets (Figure 1G). By using fluorescently tagged 100nm polystyrene beads as the reference particles comparable in size to EVs, it showed that good encapsulation of fluorescent signal from beads in the droplets without any significant signal outside the droplets (Figure 1H). To perform the emulsion of water-in-oil, fluorinated oil FC40 and associated fluoroSurfactant was employed to generate oil phase due to their low conductivity, chemical inertness and stability, and easy removal. Pharmaceutical grade FC40 oil is considered as the highly biocompatible and low-cost recipe for droplet generation employed in pharmaceutics and in compliance with FDA 41-42. We also introduced the pharmaceutical grade trehalose as stabilization additive in the buffer during the electro-transfection, which can enhance the EV stability to minimize membrane fusion and leakage, in turn, improve the electro-transfection efficiency as documented in literatures26-30, 43. The droplet size which determines the throughput can be controlled by adjusting the pressure/flow rate of water-to-oil ratio44. For achieving high throughput, the droplets can be generated in high speed (∼700 droplets/min), which leads to ∼30 mL per hr processing throughput for each device. Note that current cuvette electroporation only handles ∼100 μL per device. The electroporation power did not alter the droplet size and quantity (Figure 1 I and J). For the efficient purification of cargo-loaded EVs from excessive Cas9 cargos, we employed Ni Sepharose high performance magnetic beads to selectively capture the His-tagged Cas9 proteins. The cargo loaded EVs in aqueous phase can be collected via centrifugation through phase separation to fully remove oil phase (Figure 1A cargo-loaded EV collection).

(A) Image of µDES device with illustration of continuous-flow droplet generation, droplet-based electroporation, and cargo-loaded EV harvesting and purification. (B) Schematic illustration of droplet-based electroporation of EVs under uniform electric field distribution as demonstrated by COMSOL simulation (C). The COMSOL simulation of continuous fluidic profile (D) and electric field profile (E) to show the uniformity for precision control. (F) Picture of large-scale collection of cargo loaded EVs in droplets. (G) Microscopic image of continuous flow generated droplets with green fluorescence stained nanobeads mimicking EVs encapsulated inside (H). The droplet size is uniform before(I) and after (J)electric field application for transfection (30 V). The insert scale bar is 1000 μm. (K) Evaluation of EV cargo loading rates among different transfection methods using fluorescence nanoparticle tracking analysis (FNTA). (L) EV recovery rate evaluated by NTA compared with conventional cuvette electro-transfection which requires 1100 volts for electroporation. (M) Quantitative measurement of CRISPR Cas9 proteins from transfected EVs normalized by EV particle number among different transfection methods. (N) The quantitative PCR analysis of transfected sgRNA copy number normalized by total EV RNAs. The electro-transfection was done by using μDES platform in 8 replicates. The native EVs and μDES prepared EVs without RNP cargo both served as the negative control groups. EVs are purified from HEI-OC1 ear hair cell culture.

Our platform can produce ∼80% transfection rate for large proteins including CRISPR Cas9/sgRNA RNP complex, which showed significantly higher efficiency than other conventional transfection methods including direct incubation, lipofection, and cuvette electro-transfection (Figure 1K). EGFP-Cas9 were pre-assembled with gRNA at 1: 2 molar ratio before the electroporation. EVs derived from HEI-OC1 cell, a putative progenitor hair cell line, were isolated and quantified for mixing with EGFP-CRISPR/Cas9 ribonucleoprotein (RNP) in electroporation low conductivity buffer. The final concentration of 1010 /mL of EVs was used for Neon cuvette electroporation as the control group, and our μDES system, as well as other chemical transfection methods. The findings revealed significantly enhanced cargo loading within EVs using μDES system, with a percentage as high as 80% for EGFP and 70% for EGFP-Cas9/sgRNA, surpassing cuvette electroporation and chemical transfection methods while maintaining minimal sample loss (Figure 1K). We also compared the recovery rate with conventional cuvette electro-transfection (Figure 1L), which showed better recovery due to continuous-flow harvesting. We quantified the RNP Cas9 protein loading amount (Figure 1M) to compare between different transfection methods, the μDES group exhibited more than 10-fold increase than other methods. The reproducibility of μDES loading was characterized using qPCR to quantify sgRNA loading amount from 8 replicates, with native EVs and µDES conditioned EVs without RNPs as control groups. Results demonstrated the good loading capacity and reproducibility, and no leakage or changes from intrinsic EV molecular components (Figure 1N). Overall, the μDES platform demonstrated the advanced performance on EV cargo loading.

Characterization of produced RNP EVs in high biocompatibility and tissue penetration

The µDES platform maintained a consistent flow rate, uniform electroporation pulse periods for each droplet and efficient cargo loading into EVs, therefore, retaining good EV properties as their natural un-treated EVs in terms of size (Figure 2A), zeta potential (Figure 2B), protein contents (Figure 2C), morphology and surface properties (Figure 2D). Results also indicated that μDES droplet oil phase with surfactant do not impose adverse influence on the produced EVs in aqueous phase due to phase separation. We tested the essential protein contents (CD81, TSG101, Alix) from µDES produced EVs, which is in line with native EVs in terms of expression level but carry significant amount of transfected Cas9 proteins (Figure 2C). The immune gold nanoparticle (AuNP) staining TEM imaging showed unnoticeable surface adsorption of Cas9/sgRNA RNPs from µDES produced EVs as compared with native EVs, which indicates RNA cargoes are loaded inside of EVs. By comparing with LNPs, we tested the cell biocompatibility (Figure 2E) and cellular uptake behavior (Figure 2F) via dosing HEI-OC1 ear hair cells with both µDES produced bone marrow mesenchymal stem cell derived EVs (RNP MSC-EVs) and HEI-OC1 hair cell derived EVs (RNP HEI-OC1 EVs). Both EV groups showed enhanced ability to promote ear hair cell growth compared with LNP group (LNP-102). The µDES produced RNP EVs did not show noticeable differences with their un-loaded native EVs. After one-hour cellular uptake, µDES produced EGFP-fused Cas9/sgRNA RNP MSC EVs exhibited higher uptake rate for cytoplasmic release and gradual entry into the nucleus (white arrow indication) compared with LNP group (Figure 2F). In order to further characterize the in vivo ear tissue biodistribution behavior, three groups of RNPEGFP LNP, µDES produced RNPEGFP MSC EVs and RNP*EGFP HEI-OC1 EVs were used via posterior semicircular canal injection into Shaker-1 mice ear individually. The confocal imaging from LNP group showed the ineffective distribution for entering into inner ear hair cells (Figure 2G and H). In contrast, both EV groups exhibited higher penetration into inner ear and uptake by both outer hair cells (OHCs) and inner hair cells (IHCs) (Figure 2 I and J). Thus, the results strongly support the feasibility of µDES transfected CRISPR RNP EVs employed in gene therapy delivering to inner ear.

(A) Characterization of EV size and zeta potential (B) after transfection using different transfection methods with the original native EVs as the control group. (C) MicroWestern Blotting analysis of essential protein contents (CD81, TSG101, Alix, Cas9) derived from µDES produced RNP MSC EVs in serial dilution, with native EVs as the control group. (D) Immune gold nanoparticle (AuNP) staining TEM imaging analysis of µDES produced RNP MSC EVs with native EVs as the control group, in terms of CD81 surface marker expression and Cas9 surface identification. Scale bar is 200 nm. (E) Cell biocompatibility analysis using MTT assay with HEI-OC1 ear hair cells (∼106) dosed with LNP group (LNP-102, Cayman Chemical, w/o RNP), MSC-EV group (w/o RNP), and HEI-OC1 EV group (w/o RNP) in ∼109 particles. (F) Confocal imaging analysis of HEI-OC1 hair cell one-hour uptake dosed with RNPEGFP LNP (LNP-102, Cayman Chemical) and RNPEGFP MSC-EV group in ∼109 particles. The white arrow indicates the cytoplasmic release and gradual entry into the nucleus. Scale bar is 10 μm. (G) schematic illustration of the biological barriers in the structure of Corti and cochlea including blood endolymph barrier (BEB), perilymph endolymph barrier (PEB), and blood perilymph barrier (BPB). Confocal tissue imaging analysis of biodistribution in the organ of Corti via posterior semicircular canal injection of (H) RNPEGFP LNP, (I) µDES produced RNPEGFP MSC EVs and (J) RNP*EGFP HEI-OC1 EVs into Shaker-1 mice ears individually. Scale bar is 30 μm. All graphs show the mean ± SEM and biological replicates.

Materials

Master Mold Resin

Learn More

M Series

CRISPR design system for allele-specific editing of pathologic Myo7ash1

The Shaker-1 mouse model has been widely used for auditory research encoded by the deafness gene Myo7a, which is expressed very early in sensory hair cell development in the inner ear45. The Myo7a mutation makes up 4.5% of cases of sensorineural hearing loss evaluated in a large human patient cohort46, which presents significant clinical populations with a high level of burden that is not addressable by current therapeutic interventions. Timely removal of mutant myo7a allele could potentially prevent progression of hearing loss. Therefore, we demonstrated an allele-specific editing system using CRISPR/Cas9 with designed gRNAs targeting G-C mutations in hearing loss Shaker-1 mouse model. For in vitro validation using ear fibroblast cells from Myo7ash1/WT mice, we screened SpCas9 and two different gRNA sets harboring sh1 mutation modified with 2’-O-Methyl and 3’-phosphorothionate bonds on the last bases on 5’ and 3’ end (Figure 3A). In each gRNA set, we designed full-length and truncated forms targeting Myo7ash1 (supplementary Table 1). The gel electrophoretic analysis shows that CRISPR/Cas9 complexes efficiently cleaved targeted Myo7a amplicons (supplementary Figure 3). The T7 endonuclease assay was used to screen the allele-specific editing in vitro in Myo7AWT/WT, Myo7Ash1/WT and Myo7Ash1/sh1. The indel percentage is reduced to ∼25% in Myo7Ash1/WT when compared to ∼45% indel percentage in Myo7Ash1/sh1(supplementary Figure 4). The similar halving reduction in indel percentage is also observed in other gRNA designs when targeting Myo7ash1/WT. The data supports a robust editing selectivity when targeting Myo7Ash1/sh1 and Myo7Ash1/WT in vitro. Sanger and next generation sequencing was further used to confirm the allelic cleavage specificity and editing efficiency. ICE (Inference of CRISPR Edits, DeskTop Genetics) analysis of sanger sequencing data showed that a good amount of indel occurred only in Myo7ash1/WT, but not in Myo7AWT/WT(supplementary Figure 4). The sequence percentage analyzed by CRISPResso247 showed that gRNA-1 and gRNA-2 have higher cleaving activity than their truncated versions respectively (Figure 3B). The indel profile revealed that the majority of CRISPR-induced variants were deletions for gRNA-1 while insertions for gRNA-2 (Figure 3C). The sequencing variations demonstrated that 94.83% of Myo7aWT is unedited in Myo7ash1/WT. The small amount of substitution in 0.33% is possibly from PCR and sequencing background errors. In contrast, Myo7ash1 sequences is greatly reduced to ∼18% and the edited sequences account for the rest 82% in Myo7ash1 allele. We then quantified the targeting specificity of gRNA designs by sorting out mutation pattern 5’-CCG-3’ and wild type pattern 5’-CGG-3’ separately when targeting Myo7ash1/WT in vitro. By designing PAM sequence of gRNAs which have closer proximity to the Myo7a mutation, 95% specificity was achieved for in vitro editing (Figure 3D) and the wild-type allele is mostly intact after the CRISPR/Cas9 editing. Taken together, full-length of gRNA-1 and gRNA-2 have the best allele specific editing property in vitro thereby suggesting the potential for further in vivo test. Based on the global analysis of all the sequences in CRISPResso2 (Figure 3F and supplementary Figure 4), the most common mutation type is either single base deletion or single base insertion causing a frame shift in the coding sequence of Myo7a protein. In summary, SpCas9-gRNA-1 and SpCas9-gRNA-2 sets were specific for the Myo7ash1 allele and robust in interfering the shaker-1 mutation at DNA level, indicating the promise for in vivo investigation.

CRISPR loaded EVs for in vivo restoring the progression of the hearing loss

The Shaker-1 mouse wild type (+/+) and heterozygotes (+/-) initially both have normal hearing. However, the mutant heterozygotes (+/-) animals will gradually progress hearing loss till 6 months of age to be completely deaf, which serves as a good model for testing gene therapy in treating hearing loss in vivo. We found that myo7a point mutation could result in the excess oxidative stress in inner hair cells48-49 compared to wild type animals due to the functional damage of hair cells, leading to develop a facile and straightforward assay for effectively evaluating hearing ability based on oxidative stress markers. As demonstrated in Figure 4, we administrated µDES produced CRISPR RNP MSC-EVs (∼ 109 particles in ∼10 μL) via the posterior semicircular canal into the ear in heterozygote (+/-) mice, with the good hearing wild type (+/+) mice and non-treated heterozygote (+/-) mice as the control groups. After monitoring at the month 3 and month 6, the Shaker-1 organ of Corti tissues were extracted for immunohistochemistry based oxidative stress analysis. The oxidative stress markers 4-Hydroxynonenal (4-HNE) and 3-Nitrotyrosine (3-NT) were stained in green fluorescence. The heterozygous (+/-) mice show high expression level of both oxidative stress markers 4-HNE and 3-NT at 3 and 6 months of age in the inner ear hair cells (Figure 4 white arrows), in contrast to normal hearing wild type (+/+) mice with no oxidative stress marker expression. Interestingly, in our treatment group of heterozygous (+/-) mice, both 4-HNE and 3-NT expression level indicating the oxidative stress was significantly reduced to the unnoticeable level at 6 months of age, indicating the recovery of hair cell auditory function and elimination of pathogenic myosin VIIa allele.

Immunohistochemistry for oxidative stress in the Shaker-1 organ of Corti. The inner hair cell is indicated by the white arrow. Primary staining with 4-Hydroxynonenal (4-HNE) 1:50, and 3-Nitrotyrosine (3-NT) 1:50. Secondary staining 1:1000, nuclei labelled with DAPI. +/+, wild type; +/-, heterozygous Shaker 1 mice. Heterozygous mice show labelling for oxidative stress markers at 3 and 6 months of age whereas wild type mice show no labelling. Heterozygous mice treated with Crispr-EVs that eliminate the pathogenic myosin VIIa allele show no evidence of oxidative stress at 6 months of age, indicating the recovery of inner hair cell auditory function. Scale bar is 150μm.

To further evaluate the gene editing efficiency at molecular precision in vivo using our µDES produced CRISPR RNP MSC-EVs, we administrated EVs (∼ 109 particles in ∼10 μL) via the posterior semicircular canal into the left ear with right ear as the untreated control group in Shaker-1 heterozygotes (Figure 5A). The tissue of Corti was extracted in week 4 after injection for mRNA sequencing (Figure 5B-C). The heterozygous Shaker 1 mutant/wild type mice hearing ability after the treatment of RNP-EVs was followed over six months of age (Figure 5D-F) based on the hearing threshold results from ABR test. In first 4 weeks of treatment, the μDES produced CRISPR EVs group already exhibited the gene changes at the mRNA level, which showed improved gene editing ability compared with RNP-LNPs and cuvette transfected RNP-EVs (Figure 5B). Such gene editing performance differences could potentially be due to the ability and efficiency of delivery carriers entering into the hair cells. The editing efficiency was amplified at mRNA level given the fact that mRNA of Myo7a was only transcribed in hair cells, although myo7a gene exists in every cell type in the organ of Corti.

A) Schematic illustration of animal testing schedule. (B) The indel% mRNA sequencing analysis of Myo7a sequence changes from extracted Corti tissue from Shaker 1 heterozygotes mice in week 4 after CRISPR EV injection, with RNP-LNPs and Cuvette transfection method as control groups. (C) Distribution of resulted sequences treated by μDES produced CRISPR MSC EVs in week 4 from mRNA sequencing analysis. The most common edited sequences were shown. (D) qPCR analysis of Myo7a fold change from extracted Corti tissue in month 6 after injection of mDES produced CRISPR MSC EVs, which showed significant reduction of mutant Myo7a in treatment group. (E) Representative auditory brainstem responses (ABRs) recorded from shaker-1 heterozygotes mice left ear treated with μDES produced CRISPR MSC EVs in month 6, and right ear without treatment. The start showed the hearing ability baseline (F) The groups of shaker-1 heterozygotes mice in p30 as the normal hearing control group, and p120 developing severe hearing loss, to compare with p120 group treated with μDES produced CRISPR MSC EVs, which indicates positive therapeutic function for preventing progression of hearing loss.

At 6 months, the pathogenic myosin VII a allele was nearly removed (Figure 5D). These mice at P30 have normal hearing serving as the control group in Figure 5F blue line, and later on developed a severe sensorineural hearing loss by six months of age in Figure 5F red line Het P120. In contrast, the treated p120 group of mice displayed significant restoration of hearing ability (black line) compared to untreated p120 group of mice (red line). There is no significant difference from treated p120 group of mice (Black) with p30 normal hearing group of mice (Blue), indicating that our developed CRISPR EVs can remove pathogenic myosin VII a allele in these heterozygous mice for restoring hearing ability.

Discussion

The development of gene therapy correcting or eradicating genetic mutations for restoring functional protein expression is essential in maintaining sensory mechanotransduction and regenerating cochlear hair cells, thereby, to improve hearing function8. It not only requires highly targeted gene editing complexed, but also most effective and cell-specific delivery platforms. Presently, more challenges with the application of viral vectors or lipid nanoparticles were found to deliver functional genetic materials into auditory sensory system, due to the low tolerance from sensitive hair cells to toxicity and the intrinsic anatomical barriers. The safety of the treatment is concerned with the transgene and constitutive expression of gene editing complexes from eukaryotic organelles, as well as the ototoxicity from chemical compounds, for future clinical translation. Using EVs to encapsulate CRISPR/Cas9 gene editing agents provides a transient way to target auditory hair cells for achieving permanent genetic alteration on the deafness gene such as shaker-1 allele. Such treatment prevents supplementary treatment to attenuate the potential ototoxicity induced by delivery vehicles or overexpression of CRISPR/Cas9 complexes in vivo. More importantly, EVs exhibited excellent tissue penetration ability to specifically access and target inner hair cells as we observed in our shaker-1 mouse model (Figure 2).

Droplet-based electro-transfection has been reported to have higher mass and heat transfer demonstrated in single cell electroporation. EVs derived from donor cells also consist of lipid bilayer and similar membrane proteins composition on cells. However, nanosized EVs contain more compact membrane curvatures and strong brownie motion. Using microfluidic droplet-based electroporation to maximize the encapsulation of cargos into EVs, in turn, maximize the gene editing efficiency in vitro and in vivo, has been demonstrated with approximately 10-fold increase on the loading amount of CRISPR/Cas9 RNP into EVs. The throughput is hundred-fold increase compared with conventional cuvette electro-transfection method. Fast continuous flow through with droplet also prevents direct contact of EVs with electrodes for retaining EV natural integrity and stability. Such platform offers an easily amenable approach for scaling up by integrating multiple chip units for future GMP grade manufacturing of cargo-loaded EVs. The rapid loading of CRISPR RNP into EVs will allow the delivery of Cas9 without using a split vector approach, which enables customization of sgRNAs for addressing different mutant alleles within one gene, thus, opening a new avenue for personalized precision gene therapy via tailoring patient genetic heterogeneous mutation background.

Materials and Methods

Materials and reagents All chemical reagents and materials were purchased from ThermoFisher unless otherwise specified. Modified gRNAs were synthesized and quantified by Synthego. NLS-spCas9-NLS, EGFP-spCas9-NLS nucleases, anti-Cas9 antibody (Clone 4A1) were purchased from Genscript. All of DNA oligos were purchased from Integrated DNA Technologies. The collagenase IV was obtained from STEMCELL Technologies. SYLGARD™ 184 Silicone Elastomer Kit was purchased from Dow Silicones Corp. Master Mold for PDMS device was obtained from CADworks3D. Q5 high-fidelity DNA polymerase was purchased from New England Biolabs. Human bone marrow-derived conditioned culture medium and human umbilical cord-derived conditioned culture medium for extracellular extraction were purchased from EriVan Bio. SM-102 LNP in ethanol was the generous gift from Dr. Fan Zhang, in the College of Pharmacy at the University of Florida. 6nm Goat anti-Mouse and anti-Rabbit, IgG, Immuno Gold reagents were purchased from AURION.

Droplet generator on chip

The 3D structure of microfluidic device was designed and drawn by SOLIDWORK CAD. The resin mold for PDMS casting containing 200 µm flow and 500 µm electrode channels, 150 µm as nozzle for droplet generation, was printed by µMicrofluidics Printer (CADworks3D, 30 µm resolution). Briefly, CAD files was opened by utility.exe that is connected to µMicrofluidics Printer. The software setting for printing as follows: 50 µm as thickness, 0.1 for grid size, 40% Power ratio. The microstructure was then sliced and ready to be launched for printing. The resulting resin mold was then soaked in 100% Ethanol or isopropanol for 10 min to remove the free resin before the final UV curing step. And then the resin mold was dried with compressed Nitrogen. The soaking-drying cycles have to be repeated several times until there is no shiny free resin on the microstructure. Each side of resin mold was then cured in the Creative Cure Zone (CADworks3D) for another 10min twice for further photopolymerization and solidification of microstructures. The resin mold was then ready for PDMS casting. PDMS was prepared using the standard 10 : 1 (base to curing agent) ratio. The PDMS mixture was stirred completely at least 3 min and then degassed for at least 30 min before being poured into the 3D-printed molds and then baked at 75°C for 3 hours. After the surface activation of molded PDMS pieces using a corona discharger and the microscope plate, PDMS molds were then assembled and bound onto the microscopic plate as the droplet-based electro-transfection device.

Electrodes with L shape were tailored to fit into electroporation chambers designed in μDES and then manually inserted into the electroporation sites to align well with each other. The 1/16 OD, 1/32 ID tubings were then inserted into the inlets and outlets in μDES. To avoid any potential pressure leakage that can result in the unstable flow rate in μDES, the additional PDMS was then added to the area surrounding the inlets, outlets and electroporation sites before use. And then the device was baked in the oven at 75°C for 30min. The resulting μDES is ready to the following electroporation.

COMSOL Simulation

The proposed microfluidic device was fine-tuned using the COMSOL Multiphysics software package. The device’s mathematical model involves fluid flow and electromagnetism. For both models a standard linear triangular extra-fine mesh was assigned to the geometry. To observe the geometric evolution of our droplets, we used the computational fluid dynamics system (CFD) module using the laminar two-phase flow. For this simulation, the oil phase material was defined as FC-40 with a density of 1850 kg/m3 and a dynamic viscosity of 0.0018 Pa/s. Microfluidic flows are defined by the Navier Stokes equation where ρ is the density of the fluid, u is the velocity of the field, t is time and P is the pressure field:

The physics of the electroporation system of the device were defined using the AC/DC module to simulate the electric field distribution in the microfluidics electroporation model. The material of the droplet was defined as electrolytic buffer with a conductivity of 1×10-4 S.m-1. The electrical conductivity of the electrode was defined by the composition of the Platinum-iridium wire as 9.43×106 S.m-1. In steady conditions, the flow of electric currents within a conducting fluid follows Ohm’s law:

Here, J represents the total current density within the material while σ represents the electrical conductivity measured in S.m-1. E is the electrical field strength measured in V.m-1 and Je represents the current density. Moreover, to mathematically describe the electric field that acts upon the droplet within the microfluidic device, we use the induced potential difference ΔΨi at a point M of the droplet membrane at any time t

The results of this simulation were then used to adjust and optimize the device’s design for the intended application. Overall, the use of the COMSOL Multiphysics software package allowed for a detailed and accurate analysis of the proposed microfluidic device, ensuring its optimal performance.

Electro-transfection of CRISPR into Extracellular Vesicles

Neon electro-transfection system

EVs in 1x PBS buffer were firstly transferred into Neon R buffer by using 30 kDa cutoff ultrafiltration column to reach the final concentration of 1010 EVs/mL. Basically, EVs were added into the pre-washed 30K cutoff column and centrifuge at 7000 xg for 8 min and then resuspend the concentrated EV (∼80 µL) in 400µL Neon R buffer to centrifuge again under the same condition. The resulting EVs in Neon R buffer (∼60µL) was placed on ice immediately for later use. To electro-transfect CRISPR/Cas9 complexes into EVs, gRNAs and EGFP-Cas9 nuclease were firstly pre-mixed together in 45µL Neon R buffer and self-assembled at room temperature for 10min and then added into EVs solution to obtain 6µM of EGFP-Cas9 and 9µM of gRNA in the ready-to-electro-transfection solution (∼106µL). To stabilize the membrane, 1250mM trehalose was prepared in PBS buffer without Ca2+ and Mg2+ and then 4.4µL added to the ready-to-electro-transfection solution to have 50mM trehalose in the final solution. The addition of trehalose increased the viscosity of electro-transfection solution therefore, to maintain the viscosity balance in Neon electroporation system, 120µL of 1250mM trehalose was added to per 3mL Neon electrolytic buffer. 1500V, 20ms, 1 pulse was used to electro-transfect CRISPR/Cas9 complexes into HEI-OC1 derived EVs with 100uL Neon platform. The resulting electro-transfected CrisprEV was gently transferred to 1.5mL protein low binding Eppendorf tube for membrane recovery at room temperature for 10min. And then 500uL PBS buffer at room temperature was added to EV solutions followed by the further membrane recovery at 37°C for 20 min. The resulting CrisprEV was stored in -20°C for the downstream purification and analysis.

High-throughput droplet-based µDES

Water-in-oil droplets were generated at the flow-focusing junction inside the µDES. As described previously, EVs were first transferred to Cytoporation® Media T (BTXpress™, USA) and then mixed with 5µM gRNA-Cas9 RNP (gRNA:Cas9 molar ratio=1.5:1) at the concentration of EVs 1010/mL. The resulting mixture was delivered into the device as the dispersed aqueous phase. The oil phase contained FC-40 mixed with 2 weight % 008-FluoroSurfactant (RAN Biotech). A microfluidic pressure flow controller (PreciGenome) was used to generate the droplets with the diameter of around 1000µm at 3.0-3.8µL/min of the aqueous solution and 1.5-2µL/min of the oil phase. The electroporation was performed ranging from 10-60V by using direct current-based power supply and the resulting emulsion was collected within an Eppendorf tube or the microplate analyzed under the inverted microscope (Cytation 5, BioTek).

Purification of CrisprEVs

The isolation of aqueous phase containing CrisprEVs was performed under the centrifuge at 2000-3000xg for 5-10min at room temperature. The aqueous phase was then collected by the pipettes and transferred to a new Eppendorf tube for the downstream purification. To remove excessive His tagged CRISPR/Cas9 complexes from CrisprEVs, 100µL Ni Sepharose high performance beads (GE Healthcare) were firstly washed with 10 volume of cold 1x PBS buffer and then pre-equilibrated in PBS buffer for 10min. The beads were then incubated with per 100µL CrisprEV at 4°C for 0.5-1 hours on the rocker. The unbound CrisprEV was then collected in 1 mL of cold 1x PBS buffer while the excessive CRISPR/Cas9 bound on the column. The purified CrisprEV was then concentrated with 30 kDa ultrafiltration column under the same condition mentioned above and stored in -80°C for further analysis.

Characterization of CrisprEV

Nanoparticle tracking analysis

The size and particle number of purified EV and CrisprEV were analyzed by nanoparticle tracking analysis (NTA) using the NanoSight NS300 instrument (Malvern Instruments, UK) supplied with a blue laser (488 nm). Briefly, 25 μL of the final EV solutions was diluted in 1:20 for NTA analysis. A solution of 400 µL was injected into the sample chamber and each trajectory measurement of every sample was repeated five times. Data analysis was performed in NTA 3.4 software (NanoSight) with the following software settings for capture and analysis: camera level=16, screen gain=1, detection threshold=18.

Surface Charge

The zeta potential of the resulting EVs was measured by dynamic laser scattering (Litesizer 500, Anton Paar). Briefly, 25 μL of the final EV solutions diluted in 1:20 with 10% PBS buffer was injected into disposable folding capillary cuvette and each zeta potential measurement was conducted five times. The final measurement is conducted with 2min for pre-equilibrium, under room temperature by selecting PBS as the referenced conductivity.

Loading efficiency of Cas9 protein

The concentration of CRISPR/EGFP-Cas9 loaded into EVs was measured by Cytation5 (BioTek). Firstly, the standard curve of fluorescent intensity of EGFP-Cas9 in PBS buffer was obtained from Cytation 5 with a serial dilution of EGFP-Cas9 in the 96 well microplate. Each dilution was conducted and measured in duplicates. 35uL of the final EV was diluted in 1:1 with PBS buffer and then added to the microplate above and the fluorescent intensity of the resulting solution measured under the same condition with the standard curve. To further quantify the loading efficiency of CRISPR/EGFP-Cas9 in single EV, 25uL of the final EV solutions diluted in 1:40 with PBS buffer was then injected into ZetaView NTA (ParticleMetrix, Germany) for the quantification of EGFP+ EV. Basically, the diluted EVs were measured in scattering and fluorescent mode by using particle number/sensitivity measurement in fluorescent NTA (fNTA) with laser 488nm. By using reference polystyrene beads with Ex 488nm, the sensitivity scale was set at 96-98 with 100% fluorescent labeled beads. And then the final percentage of EGFP+ EVs was normalized against that of the reference beads under the same sensitivity in the fluorescent mode.

Loading efficiency of gRNA

The resulting CrisprEVs were first incubated with 1U of proteinase K followed with 1X Halt™ proteinase inhibitor cocktail (Thermo, USA). The solutions were then incubated with 1U RNase A (Thermo, USA) following the manufacture’s protocol. The resulting RNP-EVs were then directly employed to extract total RNA by using Qiagen kit. The final RNA concentration for RT-qPCR was quantified with Quant-it™ RiboGreen RNA assay kit. For RT-qPCR, 1ng of total RNA was used for 20μL of AMV reverse transcriptase at 55°C for 1 hour following the manufacture’s procedure (NEB, USA). The final cDNA was aliquoted to 4 individual qPCR reactions by using the primers in Table S1.

Automated western blotting of CrisprEVs

Total protein extracts from purified CrisprEVs were prepared in the one volume of lysis solution, RIPA buffer (Thermo), followed by 5min sonication, 30s vortex. Protein concentration was then determined by the Micro BCA Protein Assay (Thermo, USA). Protein extracts (100-150ng per lane) were added and separated in the cartridge compatible with Wes Instrument (Bio-Techne, USA). Simple Western was performed and imaged according to the manufacture’s procedure.

H Series

H Series

Master Mold for PDMS

Clear Microfluidic Resin

Cyto-Clear Photopolymer Resin

Master Mold for PDMS

Clear Microfluidic Resin

Cyto-Clear Photopolymer Resin

Research Highlights

Academic Articles

Podcast: Big Ideas at μScale

Research Highlights

Academic Articles

Podcast: Big Ideas at μScale

Academic Article

A biomimetic sperm selection device for routine sperm selection

by Steven A. Vasilescu, Dale M. Goss, Kathryn H. Gurner, Rebecca L. Kelley, Maria Mazi, Fabrice K. De Bond, Jennifer Lorimer, Fabrizzio Horta, Farin Y. Parast, David K. Gardner, Reza Nosrati and Majid E. Warkiani

We kindly thank the researchers at University of Technology Sydney for this collaboration, and for sharing the results obtained with their system.

Introduction

Materials and Methods

Device fabrication

Apparatus Used

Clear Microfluidic Resin

Ultra 503D Printer

Legacy

Semen collection and patient demographics

Experimental procedure

Sperm chromatin dispersion assay and motility analysis

Statistical analysis

Results

Proof-of-concept study

Diagnostic andrology study

Site comparison

Discussion

Apparatus Used

Clear Microfluidic Resin

Ultra 503D Printer

Legacy

Supplementary Material

References

Academic Article

Offsetting Dense Particle Sedimentation in Microfluidic Systems

by Tochukwu Dubem Anyaduba and Jesus Rodriguez-Manzano

We kindly thank the researchers at the Imperial College London for this collaboration, and for sharing the results obtained with their system.

1. Introduction

Apparatus Used

ProFluidics 285D

2. Materials and Methods

2.1. Design and Fabrication of Fluid-Metering Chip

2.2. Hydrodynamic Interruption of Sedimentation

2.3. Induced Hindered Settling (i-HS)

2.4. Data Acquisition and Analysis

3. Results

3.1. Fabrication of Particle Metering Chip

3.2. Hydrodynamic Interruption of Sedimentation

3.3. Sedimentation Offset via Induced Hindered Settling

Apparatus Used

ProFluidics 285D

4. Discussion

4.1. Hydrodynamic Interruption of Particle Sedimentation

4.2. Induced Hindered Settling

5. Conclusions

Supplementary Materials

References

Academic Article

Microfabricated dynamic brain organoid cocultures to assess the effects of surface geometry on assembloid formation

by Camille Cassel de Camps, Sabra Rostami, Vanessa Xu, Chen Li, Paula Lépine, Thomas M. Durcan and Christopher Moraes

We kindly thank the researchers at the McGill University for this collaboration, and for sharing the results obtained with their system.

1. Introduction

2. Methods

2.1. Device fabrication process

Apparatus Used

Master Mold for PDMS

ProFluidics 285D

2.2. Device preparation for organoid culture

2.3. Cell culture

2.4. Organoid culture

2.5. Device loading

2.6. Insert removal

2.7. Tissue staining

Ultra 50
3D Printer

Ultra 50
3D Printer