CadWorks3DAdmin

Applied tutorial for the design and fabrication of biomicrofluidic devices by resin 3D printing

by

Applied tutorial for the design and fabrication of biomicrofluidic devices by resin 3D printing

by Hannah Musgrove, Megan Catterto and Rebecca Pompano

Abstract: Stereolithographic (SL) 3D printing, especially digital light processing (DLP) printing, is a promising rapid fabrication method for bio-microfluidic applications such as clinical tests, lab-on-a-chip devices, and sensor integrated devices. The benefits of 3D printing lead many to believe this fabrication method will accelerate the use of microfluidics, but there are a number of potential obstacles to overcome for bioanalytical labs to fully utilize this technology. For commercially available printing materials, this includes challenges in producing prints with the print resolution and mechanical stability required for a particular design, along with cytotoxic components within many SL resins and low optical compatibility for imaging experiments. Potential solutions to these problems are scattered throughout the literature and rarely available in head-to-head comparisons. Therefore, we present here a concise guide to the principles of resin 3D printing most relevant for fabrication of bioanalytical microfluidic devices. Intended to quickly orient labs that are new to 3D printing, the tutorial includes the results of selected systematic tests to inform resin selection, strategies for design optimization, and improvement of biocompatibility of resin 3D printed biomicrofluidic devices.

Keywords: digital light processing, stereolithography, SLA, photopolymerizable resins, microfluidic fabrication, cell culture

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

1. Introduction

In microfluidic device development, a recurring theme is to complete bioanalytical assays at a fraction of the time and cost required for macroscale methods. This aspiration makes rapid and accessible fabrication of microfluidic devices a key goal. Historically, microfluidic fabrication relied heavily on soft lithography methods such as casting polydimethylsiloxane (PDMS) on hard micropatterned masters.1,2 Soft lithography is primarily a 2D fabrication method, in which multi-layer devices are generated by tedious or challenging manual alignment and bonding. This process is sensitive to any dust that falls onto the layers during aligning and bonding, especially if conducted outside a cleanroom environment. Thus, in order to produce complex 3-dimentional devices monolithically and without a clean room, many groups have turned to 3D printing as a simplified workflow for rapid fabrication in the laboratory.3–6

Since the early 2010s, stereolithographic (SL) 3D printing has emerged as a promising technique for fabricating microfluidic devices.5,7,8 Briefly, this technique works by curing photopolymerizable resins with a UV or visible light source in sequential layers that build on top of one another. Stereolithographic apparatus (SLA) printers were some of the first SL printers and utilize a UV laser guided by mirrors, curing resin point-by-point in a scanning manner in the x and y directions. Direct light processing (DLP) printers were developed later and utilize UV projectors that allow an entire layer to be cured simultaneously from a direct light path. Because of their direct light path, DLP printers tend to have slightly better resolution compared to the similar SLA printers.5,9 The development of higher resolution DLP printing, along with other SL printers, has increased the use of 3D printing in fields of dentistry, audiology, medicine, and microfabrication.

Compared to traditional fabrication methods, 3D printing reduces cost and fabrication time while increasing product customization.5,10 Having the ability to readily customize microfluidics also expediates the “fabrication-application-results” process for devices.1,11 Indeed, many bioanalytical 3D printed devices have been well documented in reviews.12–16 Applications have ranged widely from bioreactors and probes made for direct contact with a range of cell lines,17–20 to approaches that utilized 3D-printed molds to cast microfluidic devices in other materials.21–23

Like all fabrication methods, 3D printing requires compromises between desirable features, e.g. resolution and cytocompatibility, and the past 5 years have seen an explosion of papers seeking to address key limitations. The ability to print leak-free devices, small internal microchannels (<1000 μm), and biocompatible, optically-clear devices depends on several factors that will be discussed further below, including aspects that affect mechanical integrity and the rate and resolution of photopolymerization (Fig.1).24–26 For commercial resins, strategies to diminish resin cytotoxicity have started to emerge recently,27–30 along with solutions addressing print resolution,31–36 imaging compatibility,33,37–39 and surface modification and bonding.40,41 Sifting through the plethora of literature for best practices can be challenging for researchers who are new to this rapidly growing field.

Figure 1. Schematic of a DLP 3D printer, highlighting the design and mechanical factors as well photopolymerization parameters which influence print resolution. Digital light processing 3D printers project UV or violet light through optically clear sheets (usually Teflon) into a vat of photopolymerizable resin (pink). In locations where the light is projected, the resin crosslinks to form a solid structure. Exposure and crosslinking are performed layer by layer on the base plate, which lifts up as each concurrent layer is formed. Production of a clean print is dependent on instrumental, environmental, chemical, and design elements that impact either the print surface (base plate), mechanics (print orientation, design specifics), or chemical reaction (resin composition, light source, exposure settings, temperature, or humidity).
Figure 1. Schematic of a DLP 3D printer, highlighting the design and mechanical factors as well photopolymerization parameters which influence print resolution. Digital light processing 3D printers project UV or violet light through optically clear sheets (usually Teflon) into a vat of photopolymerizable resin (pink). In locations where the light is projected, the resin crosslinks to form a solid structure. Exposure and crosslinking are performed layer by layer on the base plate, which lifts up as each concurrent layer is formed. Production of a clean print is dependent on instrumental, environmental, chemical, and design elements that impact either the print surface (base plate), mechanics (print orientation, design specifics), or chemical reaction (resin composition, light source, exposure settings, temperature, or humidity).

Here we present an applied, data-supported guide to the key factors that should be considered for design and fabrication of SL 3D printed biomicrofluidic devices, written especially for groups that are new to this growing field. Types of instrumentation and resin formulations have been reviewed and characterized in depth recently, and will not be presented in detail here.8,16,42–44 This tutorial follows the order of a typical workflow by first considering resin selection and then demonstrating how to increase print resolution by using simple changes in feature design and printer settings. Following printing, cytotoxicity of materials is addressed, particularly the extent and effectiveness of post-treatment strategies for applications involving contact with primary cells. Finally, preparation and considerations for use with fluorescent microscopy is outlined with data displaying autofluorescence and optical clarity of materials. We include the results of systematic, head-to-head comparisons of printing and post-treatment conditions designed to optimize the integrity of a printed piece, the resolution of interior channels, and the biocompatibility of the part. It is our hope that this work will streamline the adoption of 3D printed devices by more specialized biomedical research fields, bioanalytical laboratories, and others new to 3D printing.

2. Materials and Methods

2.1. 3D Design and Printing

All printed pieces used for this work were designed either in Autodesk Fusion 360 2020 or Autodesk Inventor 2018 and exported as an .stl file. DWG files of prints shown in figures can be found in the supplementary information. Files were opened in the MiiCraft Utility Version 6.3.0t3 software, where pieces would be converted into sliced files with appropriate layer heights. The files were converted to the correct file format to include settings optimized for either the CADworks3D M50−405 printer (MiiCraft, CADworks3D, Canada), which had a 405 nm light source, or for the CadWorks 3D Printer P110Y, which had a 385 nm source. All prints were printed in one of three resins: FormLabs BioMed Clear V1 (FormLabs, USA), FormLabs Clear V4 resin (FormLabs, USA) or MiiCraft BV007a Clear resin (CADworks3D, Canada). These resins were chosen as representative of those formulated for biocompatibility (FL BioMed), standard clear printing (FL Clear), or high-resolution microfluidics (MC BV007a).

Apparatus Used

A Bottle of the Clear Microfluidic Resin

Clear Microfluidic Resin

Learn More

The CADworks3D Pr110 3D Printer with a 385nm wavelength projector

PR110
3D Printer

Legacy

After printing, all materials were rinsed with isopropyl alcohol using the FormWash from FormLabs, following the wash suggestions from the FormWash online guide.45 Alternatively, materials were placed in a container with IPA and placed on a rocker for longer periods, extending times by 5 minutes for low viscosity materials and by 1 hour for more viscous resins. Once residual resin was rinsed from the prints, the pieces were dried with compressed air and post-cured with additional UV dosages using the FormCure (FormLabs, USA). Specific print and post-print settings for each resin and printer can be found in Table S1.

2.2. Viability testing with primary murine lymphocytes

2.2.1 Primary Cell Preparation

Animal work was approved by the Institutional Animal Care and Use Committee at the University of Virginia under protocol #4042 and was conducted in compliance with guidelines from the University of Virginia Animal Care and Use Committee and the Office of Laboratory Animal Welfare at the National Institutes of Health (United States). Following isoflurane anesthesia and cervical dislocation, spleens were harvested from female and male C57BL/6 mice between 8-12 weeks old. The spleens were collected into complete media consisting of RPMI (Lonza, Walkersville, MD, USA) supplemented with 10% FBS (VWR, Seradigm USDA approved, Radnor, PA, USA), 1× l-glutamine (Gibco Life Technologies, Gaithersburg, MD, USA), 50 U/mL Pen/Strep (Gibco, MD, USA), 50 μM beta-mercaptoethanol (Gibco, MD, USA), 1 mM sodium pyruvate (Hyclone, Logan, UT, USA), 1× non-essential amino acids (Hyclone, UT, USA), and 20 mM HEPES (VWR, PA, USA).

To produce a splenocyte suspension, harvested spleens were crushed through a 70-μm Nylon mesh filter (Thermo Fisher, Pittsburgh, PA, USA) into 10 mL of complete media. The cells were then centrifuged for 5 minutes at 400 xg. The pellet was resuspended into 2 mL of ACK lysis buffer, which consisted of 4.15 g NH4CL (Sigma-Aldrich, St. Louis, MO, USA), 0.5 g KHCO4 (Sigma, MO, USA), 18.7 g Na2EDTA (Sigma, MO, USA) into 0.5 L MiliQ H2O (Millipore Sigma, Burlington, MA, USA). The cells were lysed for 1 minute before being quenched by bringing up the solution to 10 mL with complete media and immediately centrifuging again. The pellet was resuspended into 10 mL of complete media, and density determined by trypan blue exclusion. To prepare for cell culture, the suspensions were diluted with complete media to a concentration of 1×106 cells/mL of media.

2.2.2 Print preparation for biocompatibility studies

Disks with a diameter of 15 mm and a height of 1 mm, designed to fit snugly against the base a 24-well plate, were printed in all three representative resins (BioMed, Clear, and BV007a) following the print settings outlined in Table S1. These pieces were divided into “non-treated,” with no further post-treatment, or “treated”. The latter prints were post-treated by soaking in sterile 1x phosphate-buffered saline without calcium and magnesium (PBS, Prod. No. 17-516F, Lonza, USA) for 24 hours at 37°C for BV007a or at 50°C (BioMed and Clear resins) to mitigate cytotoxicity, with similar treatments shown to be effective in previous works.46

To compare post-treatment strategies for the BioMed resin, disks were printed as above. The post-treatments included a 24-hr PBS soak at room temperature within a biosafety cabinet; 24-hr incubation in a 37°C cell culture incubator while dry or soaked in PBS; or autoclaving for 30 minutes at 120°C gravity cycle. In all cases, both treated and untreated pieces were rinsed again with IPA, dried, and UV sanitized for an additional 10 minutes before use.

2.2.3 Analysis of cell viability

Aliquots of suspended splenocytes (1 mL, 106 cells/mL) were added to two 24-well plates containing samples of either treated resins or non-treated resins as previously outlined in section 2.2.2. Wells that did not contain any resins were reserved for plate controls. The cell cultures were incubated for 4 hours at 37°C with 5% CO2. Following the culture period, the viability of the splenocytes was assessed by flow cytometry using a previously established protocol.47 Concisely, 500 μL of the cultured samples were stained using Calcein AM (eBioscience, San Diego, CA, USA) at 67 nM in 1x PBS for 20 min at 37°C. The stained samples were centrifuged at 400 xg for 5 min and resuspended in flow buffer (1 x PBS with 2% FBS), after which 4 μL of 1 mg/mL 7-AAD (AAT Bioquest, Sunnyvale, CA, USA) was added. Calcein-AM single stains were prepared using live cells, and 7-AAD single stains were prepared using cells pre-treated for 20 min with 70% ethanol added in a 1:1 v/v ratio to the culture. Additional controls included unstained cells and an ethanol-treated double-stained control. All samples and controls were run on a Guava 4-color cytometer (6-2L) and analyzed with Guava® InCyte™ Software. Live cells were defined as being high in Calcein-AM and low in 7-AAD signal, while dead cells were defined as the inverse.

2.2.4 Analysis of the viability after direct and indirect contact with treated resin

Indirect contact was defined as cell culture in media that had been conditioned by incubation with printed resin, whereas direct contact was defined as cell incubation in physical contact with the printed resins. To test indirect viability, treated BioMed disks were prepared for cell culture as noted in section 2.2.2. Following treatment, the disks were added to a 24-well plate and incubated in complete media for 24 hours at 37□C. After incubation, 1 mL of suspended splenocytes at 106 cells per mL were spun down and brought back up in 500 μL of resin-conditioned media. All samples were then cultured for 45 minutes, 4 hours, and 24 hours. Viability was analyzed as in section 2.2.3 to determine the percent of live cells present for each sample.

Direct viability was tested in a similar manner. Treated BioMed disks were added to a 24-well plate, and 1 mL aliquots of suspended splenocytes at 106 cells per mL in fresh complete media were added to sample and control wells. Viability was analyzed after 45 minutes, 4 hours, and 24 hours.

2.3. Characterization of material properties of printed pieces

2.3.1 Autoclave compatibility and heat tolerance

To test heat stability of printed pieces, small pieces with square channels (as used for print resolution tests, .DWG files are included in the supplement) were 3D printed in each of the three resin types using settings from Table S1 and autoclaved at 120°C for a 30-minute gravity cycle. Following autoclaving, the pieces were visually evaluated for cracks, delamination, or other alterations to the original design. Similar tests were conducted by leaving the prints overnight in ovens at 37°C, 70°C, and 120°C and then visually assessing the prints for discrepancies after 1, 3, and 7 days.

2.3.2 Autofluorescence

Disks with a diameter of 15 mm and a height of 1 mm were printed in each representative resin. A square piece of PDMS was used as a control. All images were collected on a Zeiss AxioZoom macroscope (Carl Zeiss Microscopy, Germany) with Zeiss filter cube channels including Cy5 (Ex 640/30, Em 690/50, Zeiss filter set #64), Rhodamine (Ex 550/25, Em 605/70, #43), EGFP (Ex 470/40, Em 525/50, #38), and DAPI (Ex ~320-385 nm, Em 445/50, #49). A 500 ms exposure time was used for all images. Following imaging, analysis was performed using Image J v1.530 (imagej.nih.gov). On each image, three 1 × 1 in2 regions were analyzed for mean gray value in each channel. Background regions were also measured from the borders in each image (outside of the printed parts) and subtracted from each sample measurement individually. The mean gray intensity was calculated for each resin piece and the PDMS control; higher mean gray intensities represented higher autofluorescence of the pieces.

2.3.3 Optical clarity

Disks with a diameter of 15 mm and a height of 5 mm were printed in the FormLabs Clear resin following the print settings listed in Table S1. Several post-processing methods were compared to determine which had the greatest improvement on optical clarity of printed devices. These included printing on glass, a nail polish coating, a resin coating, sanding, and buffing the pieces. A non-processed piece was used as a control, and a glass slide (0.17 mm thick) was used as a benchmark for optimal material clarity.

To prepare the printed-on-glass piece, a print was set up as previously described by Folch, et al.38 First, small drops of resin were applied to the baseplate using a transfer pipette. A large cover glass with dimensions of 1.42” × 2.36” and a thickness of 0.13-0.17 mm (Ted Pella, Inc. USA) was attached to the baseplate by lightly pressing the slide over the resin then using a UV flashlight to quickly cure the resin between the slide and the baseplate. In the printer software, the initial layer height (“gap height”) was increased by the thickness of the slide (1.7 mm) prior to printing to account for the change in the z-position of the first layer. After printing, the piece was removed from the glass slide with a razor blade and post-cured typically. The glass slide and adherent resin drops were also easily removed with a razor blade.

For acrylate coating, a baseplate-printed piece was coated with generic clear nail polish from a convenience store. The top was coated using the polish applicator, allowed to dry for ~15 minutes, and the process was repeated on the bottom of the piece. Similarly, a pipette tip was used to apply a thin layer of FormLabs Clear resin to a separate piece on both the top and bottom, with both sides being UV cured for 10 minutes.

For the sanding method, 3M WetorDry Micron Graded Polishing Paper (ZONA, USA) was used. The piece was sanded on both sides starting at a 30 μm grit paper and followed by 15 μm, 9 μm, 3μm, and finally 1 μm grit. Moderate pressure was used to press the piece into the polishing paper in a circular motion to smooth the surface of the piece. Similarly, a generic 4-sided nail buffer (similar products, Walmart, USA) was used to evaluate the impact buffing could have on the printed piece.

Following post-processing, all pieces were imaged to determine optical clarity. All images were collected in the brightfield under transmitted light on a Zeiss AxioZoom macroscope (Carl Zeiss Microscopy). The intensity of light that passed through each piece was measured using Image J for three 1 × 1 inch2 sections on each image. The average intensity and standard deviation were recorded for n = 3 regions per sample, with the background subtracted from each measurement individually. The relative transmittance, T, of each sample was calculated according to Equation 1,

 

Embedded Image

 

where I is defined as the average mean gray intensity of the sample, and I0 is defined as the mean gray intensity of a glass slide. Error was propagated using Equation 2,

 

Embedded Image

 

where δ is the standard deviation.

3. Results and Discussion

3.1. Selecting a resin based on materials properties

Design of a successful 3D printed bioanalytical tool begins with selection of a suitable resin, a process that currently requires compromises. Ideally, the resins used to 3D print bioanalytical microfluidic devices would be compatible with all cell types, be able to produce milli- and microfluidic sized internal features without mechanical defects and meet imaging requirements of having low background fluorescence and high optical clarity when needed. There has yet to be a commercial resin, however, that integrates all of these ideal properties. Custom resin formulations may offer improved performance, at least for laboratories prepared to produce them consistently and tweak them for the intended instrument and application.31,38,48 Nevertheless, focusing on enhancing one feature (e.g. cytocompatibility) usually results in compromising on another (e.g. print resolution). Because of this, it is useful to understand the key components of resins before choosing a material to work with.

We have found that many polymeric resins suitable for microfluidics fall into one of three categories based on the use case for which they were designed: biocompatible, optically clear, or high microscale resolution. Though specific resin components differ, resins in the same category often have similar material properties. Therefore, for this work, one test resin from each of these categories was chosen as a case study, and these are listed in Table 1 along with important intrinsic properties. FormLabs BioMed Clear V1 (FormLabs, USA) is a representative “biocompatible” resin with a USP Class VI biocompatibility rating, where it is approved for contact with live mucosal membranes and skin tissue for >30 days.49 The FormLabs Clear V4 resin (FormLabs, USA) is representative of a material that offers increased optical clarity. MiiCraft BV007a Clear resin (CADworks3D, Canada) is a low-viscosity resin designed for high print resolution specifically for microfluidic devices. Researchers using one of the dozens of other available resins may use these properties to identify the extent to which it falls into one of these categories and thus predict performance of the resin for the intended use. When protected from ambient light, all resins tested were stable in open vats at room temperature without noticeable variation in volume or viscosity for at least two months.

Table 1. Properties of representative SL resins that inform material choice.

Resins for SL printing are comprised of a photocrosslinkable polymer base, photoinitiators to initiate crosslinking, and (optionally) additives such as photoabsorbers, dyes, and plasticizers.54–56 Common photoinitiators such as Irgacure compounds or lithium phenyl-2,4,6-trimethylbenzoylphosphinate (LAP) are activated by UV (365 or 385 nm) or violet (405 nm) light.55 While commercial photocrosslinkable resins for SL printing usually keep their exact compositions as a trade secret, the fundamental chemistry can be found in MSDS documentation. All three test resins in this work contained monomers and oligomers of an acrylate or methacrylate polymer base.50,51,53 Many SL resins for microfluidics are fairly similar in their basic composition and can be used across different printers, especially with printers that allow for exposure setting adjustments.

Resins may be formulated with additives to achieve a large increase in print resolution or other desired properties. For example, addition of photoabsorbers reduces the effect of scattered light between layers and dramatically improves z-resolution.32 Addition of plasticizers lowers the viscosity of the pre-cured resin, which improves print resolution of hollow internal features by facilitating drainage of uncured resin from the feature during printing and cleaning steps. MiiCraft’s microfluidic BV007a resin, with the highest percent of additives, has a manufacturer-reported viscosity about 10-fold lower than that of the other listed resins (Table 1), and superior resolution for microscale channels. In highly viscous materials like the BioMed resin, undrained resin is easily retained inside the internal features, where it may be crosslinked by light that has been transmitted or scattered in excess, especially as subsequent layers of resin are cured above the hollow features.

On the other hand, photoinitiators and plasticizers, as well as acrylate monomers, can increase cytotoxicity with various cell types due to factors including oxidative stress, enzymatic inhibition, and lipophilic reactions with cell membranes.56–59 Deliberately leaching these toxic components from the printed materials after post-curing can increase cytocompatibility.29,30 However, we have found that in some cases it also decreases material stability, causing cracking or a decrease in material strength and flexibility as the plasticizers are removed. This problem was particularly prevalent with BV007a prints, which peeled apart when leached in PBS for longer than 48 hours, presumably due removal of plasticizers that were essential to the structural stability of the print. For this reason, selecting a more biocompatible material (e.g. with fewer leachable toxic additives) to begin with may better address this problem when working with sensitive cells.48

As heat stability is an important factor for chips that will be subject to autoclave sanitation or extended cell culture, we tested the heat stability of each resin (Table 1). The FL BioMed material was autoclavable and also stable overnight at 120°C, allowing for thorough sterilization should any prints need to be reused or prepped for use with live biomaterials. In contrast, MC BV007a withstood mild sanitation procedures, e.g. alcohol rinses or UV sanitation, but high heat (>50°C) delaminated the material over time, although it was stable for 7-day incubation at 37°C when dry, or for 48-hrs in PBS. The FL Clear resin was found to withstand conditions heating conditions up to 70°C for 7 days dry or in solution and also mild sanitation procedures (i.e. UV curing, solvent rinses).

3.2. Physiochemical and environmental factors that influence print resolution

In addition to resin composition, the functionality and printability of a piece are also influenced by other factors that affect photopolymerization and thus the resolution and the mechanical integrity of the part (Fig.1).

3.2.1. Choosing wavelength and exposure settings to mitigate light scattering and bleed-through

Internal features such as enclosed microchannels are imperative to many 3D printed microfluidic devices. Most commercial resin materials yield features at the scale of millimeters or hundreds of micrometers on standard printers with 30-40 μm pixel size, and it is possible to improve the print resolution of internal features through strategic selection of resins, light sources and exposure settings, and number of repeat exposures.32 The central requirement is to avoid unintentional crosslinking of uncured resin inside the feature, which would lead to blocked features.

Light source wavelength, intensity, and exposure time have a large impact on resolution by modulating on the rate and extent of crosslinking.31,32,60 Common light sources include a laser, LED, or UV lamp, and typically emit at 385 nm or 405 nm in commercial printers.61 Though most commercial resins can be printed at either wavelength, more efficient reactions are achieved by matching the wavelength of the light source with the excitation and absorbance peaks of the photoinitiators and photoabsorbers, respectively. Doing so decreases the exposure time and intensity required to achieve crosslinking and reduces unwanted scattered and transmitted light through print layers, as documented by Nordin and Wooley32 as well as Pontoni.61 To briefly demonstrate the impact of wavelength on resolution of internal features, test pieces containing six internal channels of decreasing square cross-section (0.2 – 1.2 mm side-length; Fig. 2A) were printed in FL Clear resin, using either a 405-nm or 385-nm printer (Fig. 2B vs 2C). The X,Y-resolution (effective pixel resolution) was similar between the two printers, at 30 and 40 μm for the 405- and 385-nm printers, respectively. In this resin, the crosslinking reaction was more efficient with the 385 nm light source, enabling reduced light dosage (shorter times and lower intensities; Table S1), which assisted in diminishing bleed-through light allowing uncured resin to drain more easily from channels. Consistent with this, channels were printable at sizes ~0.2 mm smaller with the 385 nm printer (Fig. 2C) versus the 405 nm source (Fig. 2B).

Figure 2. The resolution of internal features can be increased by making changes to the light source, resin viscosity, and print layer height. (A) A test piece had six internal channels, 9-mm long with 0.75 mm diameter inlets and varied channel cross-sections as noted. It was printed with (B-D) (i) 50 μm layer height or (ii) 100 μm layer height, as follows: FormLabs Clear resin was printed with a (B) 405 nm or (C) 385 nm light source; (D) for comparison, a low viscosity resin, BV007a, was printed at 385 nm. Other settings were left unchanged, with the exception of increasing exposure times slightly for some 100-μm prints (Table S1). Resolution was determined visually by observing the smallest channel cross-section that could be printed fully (pink arrows) or partially open (purple arrows).
Figure 2. The resolution of internal features can be increased by making changes to the light source, resin viscosity, and print layer height. (A) A test piece had six internal channels, 9-mm long with 0.75 mm diameter inlets and varied channel cross-sections as noted. It was printed with (B-D) (i) 50 μm layer height or (ii) 100 μm layer height, as follows: FormLabs Clear resin was printed with a (B) 405 nm or (C) 385 nm light source; (D) for comparison, a low viscosity resin, BV007a, was printed at 385 nm. Other settings were left unchanged, with the exception of increasing exposure times slightly for some 100-μm prints (Table S1). Resolution was determined visually by observing the smallest channel cross-section that could be printed fully (pink arrows) or partially open (purple arrows).

An additional print setting that affects resolution is layer height, which sets the thickness and number of layers that must cured directly above hollow features, known as overhang layers. Each overhang layer, though required to close off the top of the feature, is a chance for light to unintentionally penetrate or scatter into the uncured resin that is trapped in the hollow space, potentially crosslinking it. Increasing the layer height increases layer thickness and reduces the number of overhang layers mitigating some bleed-through curing. This setting can be modified on most printers during the file slicing step when converting a design file into a printable file and works well for designs without strong diagonal features in the z-direction.63 Using the FL Clear resin, doubling from 50-μm (Fig. 2i) to 100-μm layer height (Fig. 2ii) improved the print resolution of interior channels in the test piece to partially open the next smaller channel (an improvement of <0.2 mm, Fig. 2ii, purple arrows). Therefore, simply decreasing the number of overhang layers decreased the degree of overexposure or bleed-through light and improved print resolution, though not as much as changing the light source.

3.2.2. Lower viscosity improves drainage from internal channels

As noted in Section 3.1, the need to drain uncured resin out of hollow features during both printing and cleaning means that resin viscosity has a major impact on the resolution of internal features. To demonstrate this, we compared the resolution of FL Clear resin (Fig. 2C) to MC BV007a (Fig. 2D), which have viscosities of ~900 mPa s and ~100 mPa s, respectively (Table 1), using the 385 nm light source. The FL Clear resin retained uncured resin in the channels during printing (visible when the device was removed from the printer), and produced open channels only down to 0.6 mm under these conditions. In contrast, no residual BV007a resin was observed prior to rinsing the channels, and the print yielded a resolution of 0.2 mm, the smallest size tested, thus confirming the significant benefit of low viscosity for channel resolution.

In summary, with its ability to print efficiently at 385 nm and to drain easily with low viscosity, BV007a provided the best print resolution, with 0.2-mm channels printing cleanly at a 50-μm layer height. Printing with the Clear resin, however, achieved nearly 0.4 mm channels if used with a 385 nm light source and 100-μm layer height. While the specifics of printability will change for each design, we found that the light source, material viscosity, and layer height each provide opportunities to increase print resolution of interior microchannels.

Apparatus Used

A Bottle of the Clear Microfluidic Resin

Clear Microfluidic Resin

Learn More

The CADworks3D Pr110 3D Printer with a 385nm wavelength projector

PR110
3D Printer

Legacy

3.2.3. Influences of print environment on photopolymerization

Finally, the humidity and temperature within or surrounding the printer can also impact how well the photopolymerization reactions take place. Many resins are recommended to be printed at relatively low humidity (20-40%), as photopolymers are often hygroscopic.64 We have observed that when the humidity increases (averaging 45% in the summer in our laboratory, sometimes higher than 50-60%; versus ~30-35% in winter), print failures became more common: print layers delaminated or base layers did not remain attached to the baseplate. Keeping printers and resins in an environment with regulated humidity should be considered, e.g. dehumidification. We also found that increasing the power of the light source by 5-10% partially compensated for humidity increases.

Ambient temperature can also influence how well a material prints as it directly impacts the viscosity of the material and reaction times. This topic was explored in depth by Steyrer, et al. in an evaluation of hot vs. room temperature lithography.65 In general, many resins tend to produce better print resolution in a warmer environment (above 20°C) due to lower viscosity and more efficient polymerization.65,66 For this reason, some printers allow for control over vat temperature. For printers without such control, we have found that decreases in ambient temperature (e.g. from 23°C to 20°C) can be compensated by increasing exposure time, usually by a few seconds depending on the resin and other factors, to overcome the decreased reaction efficiency. Care should be taken, though, as increasing exposure times may lead to overcuring.65

3.3. Print and design considerations for common microfluidic features

In this section, we discuss design principles for troubleshooting common print failures and improving the resolution of 3D printed microfluidic channels, along with the role of light exposure and printer settings in controlling print quality.

3.3.1. Selecting a print orientation

The build orientation and design of a part impact how the print experiences gravitational strain and other mechanical stress during printing. Print orientation techniques are a common topic for 3D printing blogs and education resources.67–69 In general, considerations for choosing a suitable build orientation include baseplate adhesion, location of part functionalities, the use of build supports, and fabrication time. Baseplate adhesion is usually strongest when the largest surface area of a design can be placed flush against the baseplate (Fig. 3A–D, i), which increases the overall stability of piece during the printing process. A slightly rough surface allows prints to adhere better to the plate while printing with lower base layer curing times, although too much adhesion makes removal difficult and can lead to breakages. Location of part features also influence how to orient a print. In this example, orienting the reservoir sideways (ii, iv) caused distortion of its walls or damage from supports, and inhibited resin drainage from the channel. Surface roughness or clarity were also influenced by the direction in which the layers were built: clarity and smoothness are maximized by orienting so that the face of the chip is printed in one layer (i, iii), and reduced by gravitational strain (Fig. 3C, iii) or too many print supports (Fig. 3D, iii). The latter can also contribute to overexposure and light scattering, causing features to fill in (Fig. 3D, ii-iv). Finally, as each additional layer adds cure time to the print, and supports add post-processing time, it is useful to minimize the use of supports while printing if possible, and to select print orientations with fewer layers to create a more efficient printing process.

Figure 3. Orientation of prints on a build plate impacts the print resolution and optical clarity. A microfluidic design containing a reservoir, channel, and chamber is shown in the MiiCraft Utility slicing software in four orientations (A) without supports and (B) with supports: (i) base-down, (ii) vertical, (iii) top-down, and (iv) horizontal. Blue grid represents the base plate. Supports were added to all areas using default settings in the slicing software. (C, D) Photos of pieces printed in BV007a resin with the MiiCraft Ultra 50 printer, (C) printed without supports, and (D) printed with supports, with 2 minutes of support removal. Surface roughness and clarity variation is evident between the 4 orientations. Defects were noted as follows: In C: (i) no defects, (ii) overhang drooping, (iii) delamination (purple) and stress fracturing (pink), (iv) blocked channel (purple), overhang drooping (pink); in D: (i) no defects, (ii) stress fracturing, (iii) damage from supports (purple) and filled chamber (pink), (iv) filled channel (purple) and support damage (pink).
Figure 3. Orientation of prints on a build plate impacts the print resolution and optical clarity. A microfluidic design containing a reservoir, channel, and chamber is shown in the MiiCraft Utility slicing software in four orientations (A) without supports and (B) with supports: (i) base-down, (ii) vertical, (iii) top-down, and (iv) horizontal. Blue grid represents the base plate. Supports were added to all areas using default settings in the slicing software. (C, D) Photos of pieces printed in BV007a resin with the MiiCraft Ultra 50 printer, (C) printed without supports, and (D) printed with supports, with 2 minutes of support removal. Surface roughness and clarity variation is evident between the 4 orientations. Defects were noted as follows: In C: (i) no defects, (ii) overhang drooping, (iii) delamination (purple) and stress fracturing (pink), (iv) blocked channel (purple), overhang drooping (pink); in D: (i) no defects, (ii) stress fracturing, (iii) damage from supports (purple) and filled chamber (pink), (iv) filled channel (purple) and support damage (pink).

3.3.2. Reducing mechanical strain in wells, cups, or ports

In microfluidic design, particularly for bioassays, wells or cup-like features are often used as reservoirs or ports. If not properly designed, these wells often fall victim to print failure.70–72 Like other photocrosslinkable polymers, many SL resins shrink 1-3% by volume upon crosslinking which may induce mechanical strain.55,71,73 If the walls of a print are thinner at some points and thicker at others, the thicker regions experience greater shrinkage and may generate defects.71 Weak structural points may form when thin walls, sharp corners, and sharp edges (~90° angles) are used in a design, and these may not hold up well in the printing process.74,75 Wells or cup-like features, for example, may experience “cupping,” which occurs when hollow features become damaged during printing due to the formation of a pressure differential (Fig. 4). This effect is due to a low pressure region, or suction, formed within the feature as the print is peeled or pulled away from the Teflon sheet after each layer is exposed, leading to cracks and/or holes at weaker structural points in the design as it caves inward under the surrounding pressure (Fig. 4B).70 Such defects may cause leaking when the printed well is later filled with fluid.

Figure 4. Schematic of “cupping” damage during the printing of a hollow, cup-like feature. (A) During printing, a UV light source (LED) forms a new crosslinked layer of resin flush against the Teflon sheet. This design is an inverted bowl shape; supports are not shown for clarity. (B) After a layer is finished printing, the print is peeled away from the Teflon sheet, e.g. by pulling the vat down and/or the baseplate up. This process creates a region of suction within the hollow cup feature; the surrounding pressure, now higher than the pressure within piece, may form a stress fracture on the print.
Figure 4. Schematic of “cupping” damage during the printing of a hollow, cup-like feature. (A) During printing, a UV light source (LED) forms a new crosslinked layer of resin flush against the Teflon sheet. This design is an inverted bowl shape; supports are not shown for clarity. (B) After a layer is finished printing, the print is peeled away from the Teflon sheet, e.g. by pulling the vat down and/or the baseplate up. This process creates a region of suction within the hollow cup feature; the surrounding pressure, now higher than the pressure within piece, may form a stress fracture on the print.

As a demonstration, a hollow well printed in a square base with 90° square angles and thin walls broke routinely under the pressure build up from cupping (Fig. 5A). We tested the impact of various strategies to reduce mechanical strain in the design, drawing upon engineering principles.74,75 The square exterior, with its varied wall thickness around the radially symmetric well, contributes to an uneven stress distribution during resin shrinkage. Increasing the thickness of the base and wall and filleting the connecting edge at the base of the well reduced the risk of cracking the base of the print, but not the appearance of small holes at the base of the well (Fig. 5B). Making the thickness of the walls more uniform around the well-like feature, by rounding the exterior corners of the feature either partially (Fig. 5C) or fully (Fig. 5D) reduced the strain unequal shrinkage as expected, with full rounding producing a well feature with no holes or other leaks.

Figure 5. Strengthening the design around well-like features decreases the impacts of cupping and resin shrinkage. (i) Illustrated computer-aid designs, and (ii) photos of the top and bottom views of the corresponding print. All pieces were printed in FormLabs Clear resin. Well A had thin walls, thin base, and strain from 90□ connections at the bases and the sides. Well B had thicker surrounding walls and a thicker, filleted base, but retained 90□ outer corners. Wells C and D further reduced the strain by rounding out the external edges. All pieces were qualitatively evaluated, with the absence of cracks (arrowheads) and pinholes (shown by arrows) indicating a strong design.
Figure 5. Strengthening the design around well-like features decreases the impacts of cupping and resin shrinkage. (i) Illustrated computer-aid designs, and (ii) photos of the top and bottom views of the corresponding print. All pieces were printed in FormLabs Clear resin. Well A had thin walls, thin base, and strain from 90□ connections at the bases and the sides. Well B had thicker surrounding walls and a thicker, filleted base, but retained 90□ outer corners. Wells C and D further reduced the strain by rounding out the external edges. All pieces were qualitatively evaluated, with the absence of cracks (arrowheads) and pinholes (shown by arrows) indicating a strong design.

3.3.3. Improving print resolution of interior channels via device design

Similar to the effect of layer height during printing, we predicted that changing the device design itself to reduce the number of UV exposures over the channel or otherwise facilitate drainage of uncured resin would improve resolution. To demonstrate, we again selected the FL Clear resin due to its high viscosity and transparency, properties that result in frequent bleed-through curing. The number of overhanging layers was controlled by repositioning the channels in the z-direction (Fig. 6), using a fixed layer height of 50 μm. Square channels that were printed near the base of the print, with 1.5 mm thick overhangs and thus 30 overhang layers, printed cleanly only down to 1.0 mm width (Fig. 6B). Decreasing the overhang thickness to 0.5 mm (10 overhang layers) improved the resolution by ~0.2 mm (Fig. 6C), consistent with the prediction. Additionally, adding drainage holes further improved print resolution of long channels that otherwise did not drain (Figure S1).

Figure 6. Reducing the number of overhang layers in the chip design improved the resolution of internal features. (A) Schematic of the test piece with six internal channels, in (i) top view and (ii) side view. (B,C) Channels were printed with (B) 1.5 mm or (C) 0.5 mm overhang thickness, and imaged from (i) top and (ii) side. The square channel is traced with dashed outlines; the features visible above the channel in the side view are inlets and outlets. All pieces were printed in FL Clear resin with a 405 nm light source (settings in Table S1). Resolution was determined visually by observing the smallest channel cross-section that could be printed fully (pink arrows).
Figure 6. Reducing the number of overhang layers in the chip design improved the resolution of internal features. (A) Schematic of the test piece with six internal channels, in (i) top view and (ii) side view. (B,C) Channels were printed with (B) 1.5 mm or (C) 0.5 mm overhang thickness, and imaged from (i) top and (ii) side. The square channel is traced with dashed outlines; the features visible above the channel in the side view are inlets and outlets. All pieces were printed in FL Clear resin with a 405 nm light source (settings in Table S1). Resolution was determined visually by observing the smallest channel cross-section that could be printed fully (pink arrows).

3.4. Cytocompatibility of resins with primary cells

Within this section, we discuss actionable steps towards improving resin cytocompatibility, as it is typically the most challenging issue for 3D printing microfluidics to be used with live cultures and tissues. Many recent publications have addressed this issue, with most solutions focused on preventing potentially toxic resin components from reacting with cell and tissue cultures. Methods include coating the prints to reduce direct contact with cells;24 minimizing unbound toxins via autoclaving, overcuring, or pre-leaching toxic substances;29 or producing resins with more biocompatible photoabsorbers.38,76,77 Leaching in particular is convenient and consistent, especially for treating internal print features that may not be directly accessible to over-curing. Therefore, here we first compared biocompatibility across the representative commercial resins, with and without a standardized, leaching-based post-treatment in heated saline (see Methods; Fig. 7A).18,29 Primary, naïve murine splenocytes were chosen for the viability tests as they are typically more sensitive to small changes in their culture environment than hardier, immortalized cell lines.78,79

Figure 7. Viability of primary murine splenocytes in contact with 3D printed materials. Primary splenocytes from male and female mice (Nmice = 2 per experiment) were evaluated by flow cytometry after live/dead staining with Calcein-AM/7AAD. (A) Cell viability after 4 hr of direct physical contact with PBS+incubation treated (T) or non-treated (N) resins compared with well plate control (C). Treated BioMed and Clear prints maintained relatively high viability compared to the well plate, while BV007 did not. Bars show mean ± SD. One-way ANOVA per data set, ns > 0.06, *p > 0.01, **p = 0.01, ****p < 0.0001. (B) Multiple post-treatment methods (PBS, incubation at 37 °C, PBS+incubation, and autoclavation) were evaluated using BioMed resin and culturing for 4 hours with direct contact. Bars show mean ± SD. One-way ANOVA, ns > 0.2, *p > 0.02. ****p < 0.0001. (C) Viability of direct contact (i.e. culture with resin) and indirect contact (i.e. culture with resin leachate) of cells with treated BioMed prints at 45 min, 4 hours, and 24 hours showed a decrease in viability over time. Values show mean ± SD. One-way ANOVA at final time point, ****p < 0.0001.
Figure 7. Viability of primary murine splenocytes in contact with 3D printed materials. Primary splenocytes from male and female mice (Nmice = 2 per experiment) were evaluated by flow cytometry after live/dead staining with Calcein-AM/7AAD. (A) Cell viability after 4 hr of direct physical contact with PBS+incubation treated (T) or non-treated (N) resins compared with well plate control (C). Treated BioMed and Clear prints maintained relatively high viability compared to the well plate, while BV007 did not. Bars show mean ± SD. One-way ANOVA per data set, ns > 0.06, *p > 0.01, **p = 0.01, ****p 0.2, *p > 0.02. ****p < 0.0001. (C) Viability of direct contact (i.e. culture with resin) and indirect contact (i.e. culture with resin leachate) of cells with treated BioMed prints at 45 min, 4 hours, and 24 hours showed a decrease in viability over time. Values show mean ± SD. One-way ANOVA at final time point, ****p < 0.0001.

We found that for experiments under 4 hours, overnight heat and saline leaching treatment was sufficient to increase viability of primary murine splenocytes in contact with 3D printed materials (Fig. 7A). Splenocytes cultured in direct physical contact with either of the treated FL resins retained viability high enough (>60%) after 4 hr to enable short on-chip experiments, whereas untreated materials resulted in lower viability on average. For the MC BV007a resin, the treatment did improve viability over the non-treated resin, but the resin was still largely cytotoxic for these cells at 4 hours. The higher cytotoxicity is consistent with the greater quantity of potentially toxic additives in BV007a compared to other resins (10-15%, Table 1). Furthermore, BV007a had lower heat stability and could not be leached at the same temperature as the FL resins without mechanical damage (Section 3.1). Therefore, for short experiments with primary cells in suspension, we concluded that the Biomed or Clear resins are more suitable than BV007a and should be pre-treated to minimize release of toxic components into the culture.

Next, we compared a number of common leaching treatments head-to-head: heat (37°C incubation, 24 hr), saline (sterile PBS soak at room temperature, 24 hr), a combination of the two (sterile PBS soak at 37°C, 24 hr), or autoclaving (gravity, 120°C, 30 min) to leach unbound toxins out of solid prints (Fig. 7B). Using the BioMed resin, we found all overnight saline (PBS) and incubation (37°C heat) treatments increased splenocyte viability at 4 hours, with non-significant differences between the live plate control and treated materials. Autoclaving was slightly less effective, and though this treatment takes less time (~45 minutes), it is not recommended, at least for this resin.

Finally, as many experiments require longer term culture than just 4 hr, we tested splenocyte viability over time when cultured in direct physical contact with treated FL BioMed resin or in “indirect contact,” i.e. cultured in contact with media that was pre-conditioned by incubation with the resin. Direct contact with resin is likely to occur in a fully integrated bioanalytical microchip with on-chip cell culture, while indirect contact may occur when microdevices are used to prepare media or drug solutions (e.g. mixers or droplet microfluidics) or to deliver media components to cultures (e.g. flow channels or bubble traps). We again found that splenocyte viability was high for both contact conditions within 0-4 hr, but decreased overnight (24 hrs) for both contact conditions compared to plate controls (Fig. 7B). Others have reported longer culture times for hardier cell lines.18,25,27–29,33,38,80,81 This result indicates that at least for primary murine splenocytes and perhaps for other fragile cells, more work is needed to identify best practices for post-print treatments and/or more biocompatible resin formulations.

3.5. Optical components of resins

Microfluidics is frequently integrated with on-chip imaging. Autofluorescence is a potential limitation of polymeric chips, whereas PDMS is often praised for its optical compatibility. Therefore, we quantified the autofluorescence of each of the representative resins in four standard fluorescence channels. In the Cy5 and Rhodamine channels, all three resins showed relatively low autofluorescence and were comparable to PDMS (Fig. 8). However, UV excitation (DAPI channel) elicited high levels of autofluorescence from the two FormLabs resins. The Clear resin also showed moderate autofluorescence in the GFP channel. Autofluorescence at short wavelengths is consistent with the use of photoabsorbers intended to absorb at short wavelengths.76 On the other hand, MC BV007a had negligible autofluorescence in all channels, similar to PDMS. Thus, optical compatibility in the intended fluorescence channels should inform material choice for microfluidics devices.

Figure 8. Autofluorescence depended on resin composition and fluorescence filter set. BioMed, Clear, and BV007a prints were evaluated in comparison to PDMS for autofluorescence in Cy5, Rhodamine, EGFP, and DAPI fluorescent channels. Background subtracted mean grey values were analyzed with ImageJ and used to determine the fluorescent intensity of each material. Saturation values were at 65,000 AU. Bars show mean ± SD, N=3 intensity measurements per print. Two-way ANOVA, comparisons shown between resin results versus PDMS for each respective fluorescent channel, ****p
Figure 8. Autofluorescence depended on resin composition and fluorescence filter set. BioMed, Clear, and BV007a prints were evaluated in comparison to PDMS for autofluorescence in Cy5, Rhodamine, EGFP, and DAPI fluorescent channels. Background subtracted mean grey values were analyzed with ImageJ and used to determine the fluorescent intensity of each material. Saturation values were at 65,000 AU. Bars show mean ± SD, N=3 intensity measurements per print. Two-way ANOVA, comparisons shown between resin results versus PDMS for each respective fluorescent channel, ****p<0.0001.

Optical clarity (transparency) is also important for microscopy and imaging. Various approaches for improving this property have been suggested in published papers as well as on vendor and hobbyist websites (r\3Dprinting, All3DP, etc.).37–39 Here we compared 5 of these methods head to head, using the FL Clear resin as a base, to determine which would be best for increasing the transparency of a printed piece (Fig. 9). All methods were compared to a glass slide as a positive control. Pieces that were printed flush against the rough base plate (i.e. without the use of supports or rafts) came off the base plate slightly cloudy in appearance (Fig. 9VI). This cloudiness was intensified when attempting to buff both the base end and the vat end of the print with a typical nail file (Fig. 9VII). Since the nail file did not have a fine sanding grain, it ended up doing more harm than good by producing new scratches on the surface of the print.

Figure 9. Optical clarity of clear resins was enhanced with post-treatments that achieved material transparency similar to glass. Round prints of FormLabs Clear resin were printed with the MiiCraft Ultra 50 405 printer and post-treated as listed in the legend, labels II -VII. (A) The test pieces were positioned over the Pompano Lab logo for qualitative, visual comparison (captured with phone camera). (B) Transmittance was evaluated using an upright microscope. Relative transmittance was compared to the average light transmittance through a 0.17 mm thick glass slide (I). Bars show mean ± standard deviation, n = 3. One-way ANOVA, Tukey’s post-hoc test; *p = 0.0284, **p = 0.0067, and ns > 0.05. (C) To show reproducibility for different resins, technique II was applied to similar round prints in FL BioMed, FL Clear, and MC BV007a resins and evaluated similarly to panel B. All samples were statistically similar compared to the glass slide control. Two-way ANOVA, Tukey’s multiple comparison test; ***p=0.0003, ****p0.8.
Figure 9. Optical clarity of clear resins was enhanced with post-treatments that achieved material transparency similar to glass. Round prints of FormLabs Clear resin were printed with the MiiCraft Ultra 50 405 printer and post-treated as listed in the legend, labels II -VII. (A) The test pieces were positioned over the Pompano Lab logo for qualitative, visual comparison (captured with phone camera). (B) Transmittance was evaluated using an upright microscope. Relative transmittance was compared to the average light transmittance through a 0.17 mm thick glass slide (I). Bars show mean ± standard deviation, n = 3. One-way ANOVA, Tukey’s post-hoc test; *p = 0.0284, **p = 0.0067, and ns > 0.05. (C) To show reproducibility for different resins, technique II was applied to similar round prints in FL BioMed, FL Clear, and MC BV007a resins and evaluated similarly to panel B. All samples were statistically similar compared to the glass slide control. Two-way ANOVA, Tukey’s multiple comparison test; ***p=0.0003, ****p0.8.

Several methods proved successful in improving optical clarity. The quickest method that produced glass-like clarity was the nail polish coating of both the base and vat sides of the print (Fig. 9II). The nail polish only took a few seconds to apply and approximately 10 minutes to dry on each side, which introduced a challenge to keep dust out of the polish while drying on an open countertop. Sanding the piece with micron grain sandpaper avoided the issue of dust and the pieces emerged smooth (Fig. 9III) instead of scratched like the nail-filed (“buffed”) piece (Fig. 9VII). Sanding the printed part on at least the baseplate-side is especially recommended if it will not disrupt any print features, as it more easily maintained a uniform surface than the other methods. Printing directly on glass38 (Fig. 9IV) had similar impacts to sanding and is recommended for smaller prints that would be easier to remove from a glass slide. This approach requires some caution, though, as attaching glass to a baseplate can cause increased wear and tear to a printing vat. It also requires resetting the initial layer height so that the printer does not lower the baseplate too far into the vat. Resin coating (Fig. 9V) was also helpful for increasing transparency, but it was difficult to achieve without overcuring the rest of the print or under-curing the additional coat, which was also prone to capturing dust. Relative to the glass slide control, most of the post-treated 3D prints were determined to have acceptable transparency. The nail polish technique (Fig. 9II) was tested on all three representative resins and consistently improved the relative transparency of each resin (Fig. 9C). We expect that each of methods I-V will provide improved optical clarity to most transparent or semi-transparent resins.

4. Conclusion

In using 3D printing for production of microfluidic devices, compromises and strategic design choices are often required to best match the material and design to the required experiment. After identifying priorities based on the planned experiments, a resin should be chosen that best fits the requirements of print resolution, mechanical stability, cytocompatibility, and optical compatibility, informed by a foundational understanding of material components. If needed, printer settings and device designs can be modified to increase the integrity of printed parts and resolution of interior channels, and post-treatment methods can be used to increase the cytocompatibility and optical clarity of a printed piece. Print stability can be improved by reducing mechanical stress in the design of a piece, and internal feature resolution can be increased by ensuring adequate resin drainage and minimizing the photoexposure of trapped resin, e.g. by reducing the number of overhanging layers. Viability can be improved upon by leaching toxins out of prints prior to application with cells, though there is still a need for more biocompatible options, especially for sensitive primary cells. Optical clarity of parts printed with clear resins can be improved via polishing methods to achieve glass-like transparency. In the future, resins that are high-resolution, cytocompatible, and optically clear will certainly be an area of continued commercial development, and promising PEG-DA based resin formulas have been reported that can be made in the laboratory.38,76,77 Meanwhile, the considerations and best practices recorded here can help researchers begin to integrate SL 3D printing fabrication with commercially available products into their microfluidics research. We envision that this guide and its head-to-head comparison of conditions will help streamline the fabrication workflow for researchers who are new to 3D printing within the biomicrofluidic community.

A Low-Cost Microfluidic Method for Microplastics Identification: Towards Continuous Recognition

by

A Low-Cost Microfluidic Method for Microplastics Identification: Towards Continuous Recognition

Pedro Mesquita, Liyuan Gong and Yang Lin

Abstract Plastic pollution has emerged as a growing concern worldwide. In particular, the most abundant plastic debris, microplastics, has necessitated the development of rapid and effective identification methods to track down the stages and evidence of the pollution. In this paper, we combine low-cost plastic staining technologies using Nile Red with the continuous feature offered by microfluidics to propose a low-cost 3D printed device for the identification of microplastics. It is observed that the microfluidic devices indicate comparable staining and identification performance compared to conventional Nile Red staining processes while offering the advantages of continuous recognition for long-term environmental monitoring. The results also show that concentration, temperature, and residency time possess strong effects on the identification performance. Finally, various microplastics have been applied to further demonstrate the effectiveness of the proposed devices. It is found that, among different types of microplastics, non-spherical microplastics show the maximal fluorescence level. Meanwhile, natural fibers indicate better staining quality when compared to synthetic ones.

Key words:  microfluidics; microplastics; continuous identification; low-cost; 3D printing

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

Introduction

It has been estimated that the ongoing COVID-19 pandemic has exceedingly increased the demand for the use of plastics [1]. As of 23 August 2021, more than 8.4 million tons of pandemic-associated plastic debris were released into the oceans [2]. Among them, most of the plastic debris was microplastics with a size smaller than 5 mm [3,4], and this can be classified into primary and secondary microplastics. Generally, primary microplastics are the microscopic plastics that were intentionally made small (e.g., microbeads used in cosmetics) [5,6,7], while secondary microplastics are particles resulting from the breakdown of macroscopic pieces due to the conjoint environmental effects (i.e., photo-oxidation, hydrolysis, microorganism degradation, mechanical shear, etc.) [8,9]. Despite the fact that a variety of plastic types have been identified in microplastics, most of the microplastics in seawater originate from packaging materials (e.g., polyethylene, and polypropylene) [10]. Owing to their small density, these microplastics tend to float on the surface of the water, thus they can spread worldwide (in opposition to denser plastics that tend to settle down) and are more difficult to remove [11,12].

Nevertheless, the current understanding of plastic pollution in terms of the quantity, type, lifetime, and associated health effects largely remains unknown. As a result, microplastic separation and identification serve as an important approach to providing evidence and metrics of the pressing environmental issues caused by plastic pollution [13]. For example, worldwide microplastic assessment is possible to identify hot pollution spots and determine the historical trends which may lead to novel strategies for fighting debris spread [14,15,16]. At present, quite a few identification techniques have been explored for microplastic identification. Among them, common methods include visual inspection, Fourier-transform infrared spectroscopy (FTIR), Raman spectroscopy and scanning electron microscopy (SEM) [17,18]. Despite their effectiveness, these techniques, except visual inspection, rely on expensive apparatus and time-consuming detection methods that are limited to trained personnel, thus hindering the expansion of these methods for high-throughput detection [19,20,21].

As a result, visual inspection, though not as effective as other sophisticated counterparts, is still widely applied for faster recognition [22,23]. Currently, a variety of sampling and identification technologies have been used to improve the performance of visual inspection, of which a commonly used one is the combination of filtration and staining [24,25]. However, filtration often leads to false positives due to potential interference from organic matters in the samples [26,27]. More importantly, its performance is highly reliant on the size of the filters, thus limiting its capabilities in sampling small microplastics. In addition, particles are also prone to adhere to the filters, resulting in ineffective separation of the microplastics for identification [28,29]. In addition, staining of the microplastics relies on staining agents that turn microplastics into prominently visible particles [30,31]. Currently, it is unsurprising that quite a few staining agents (e.g., Rhodamine B, Rose Bengal, Trypan Blue, etc.) have been explored for this purpose. Among them, Nile Red is a hydrophobic fluorophore and was reported to be one of the most effective agents due to its favorable binding performance with lipophilic substances [30,32,33].

Nevertheless, the focus of current studies on microplastic staining and identification has been largely given to batch-by-batch or case-by-case analysis [25,34,35,36]. Therefore, sample collection is still inevitable and remains a time-consuming step in the whole sampling process. Moreover, temporal information is hardly achievable. Given the need of acquiring in-depth studies of microplastic pollution in oceans and other water bodies, long-term monitoring or continuous monitoring is essential and a low-cost, simple and effective method should be developed. Thanks to the burgeoning developments of microfluidic technologies over the past decade, microfluidic devices can be a promising solution to address this need due to their powerful particle control capabilities and the ease of integration in modern electronic systems [37,38]. For example, microfluidics has been used for long-term monitoring of algae in the past [39]. Tumor responses to hypoxia conditions were also analyzed continuously in a microfluidic platform [40]. Indeed, microplastic identification is also not a new research area for microfluidics [41,42], yet to the best knowledge of the authors, microfluidics has not been applied for long-term microplastics assessment and we believe the combination of low-cost Nile Red staining and microfluidic fluid control would provide a novel venue to confront the ever-deteriorating plastic issues without complicated analysis and costly instrumentations. Low-cost fabrication methods such as 3D printing and molding can be applied to further minimize the cost associated with this method, and the miniaturized devices may also be integrated into the monitoring stations near seashores, along with data collection of other water quality metrics on a continuous basis.

Herein, we further explored the staining capabilities of Nile Red through a microfluidic device capable of continuously staining microplastics for rapid identification. The proposed device has two inlets for respective Nile Red and sample injection (Figure 1A), and a serpentine channel that allows for sufficient mixing of the staining agent and the sample. In this paper, we studied the effects of dominating parameters on the identification performance, including Nile Red concentration, temperature, and residency time. Note that prior to performing microfluidic studies, static studies that resemble traditional Nile Red staining processes were adopted and served as a baseline for comparison.

2. Materials and Methods

In this paper, the process of static microplastic identification using Nile Red was carried out without a filter. Specifically, the staining Nile Red solution was added directly into an Eppendorf tube containing the microplastics sample and placed inside an oven (Figure 1B). On the other hand, microfluidic experiments followed a similar procedure: mixing Nile Red and microplastics in the device (which was placed inside an oven). Since the mixing process is passively induced without human operation, this process holds promise for continuous staining.

2.1. Nile Red Preparation
The staining solution was prepared by dissolving Nile Red (technical grade, N3013, Sigma-Aldrich, St. Louis, MO, USA) in methanol to different concentrations. We have considered the limit of solubility of Nile Red in methanol (1 mg/mL) to be the stock solution for further dilution, from which the solution was diluted into 50X, 100X, 250X, 500X, and 1000X samples.

2.2. Microplastics Sample Preparation
 In this paper, lab-prepared and commercially available microplastics were adopted in lieu of naturally formed microplastics. More specifically, microspheres made of polyethylene (PE), ranging from 10–45 µm (Cospheric, Inc., Santa Barbara, CA, USA) were applied to determine the optimal parameters for staining. Other plastics including the microspheres made of polystyrene (PS) with sizes from 9.5–11.5 µm (Cospheric, Inc., Santa Barbara, CA, USA), cotton and acrylic fabric acquired from clothing, polypropylene (PP) and non-spherical PE prepared from plastic storage containers were also applied to test the versability of the proposed method. All the samples were mixed with deionized (DI) water prior to staining. Note that commercial microspheres were diluted in a concentration of 10 mg/mL, while the other samples were diluted to 1 mg/mL, which is because the commercial particles were more available than the ones obtained from other sources.

2.3. Static Experiments
To prepare the samples for static experiments, 100 µL of Nile Red solution was thoroughly mixed with 100 µL of PE microplastic solution inside an Eppendorf tube, followed by baking inside an oven (Quincy Lab, model 10, Burr Ridge, IL, USA). On the other hand, all static experiments were performed using PE microspheres. To investigate the effect of Nile Red concentration on staining performance, different concentrations were tested: 100X, 250X, 500X, and 1000X; in addition, different temperatures (i.e., 25, 40, 50, 60, 70, and 80 °C) were applied to study the effects of temperature. All the samples were placed inside the oven for 10 min, and analysis was conducted immediately after baking.

2.4. Microfluidic Experiments
To create the microfluidic devices, soft lithography, a commonly used method in microfluidics, was applied. Specifically, a 3D printer (CADWorks 3D, µMicrofluidics edition, Toronto, ON, Canada) was used to create the molds for casting polydimethylsiloxane (PDMS) to obtain the final devices. After curing the PDMS mixture in an oven overnight at 65 °C, a corona treater (BD-20AC Laboratory Corona Treater, Electro-Technic Products, Chicago, IL, USA) was used to permanently bond the device onto a glass slide. Finally, the device was placed inside the oven and a syringe pump (Fusion 200, Chemyx Inc., Stafford, TX, USA) was used to run the samples as well as the staining agents inside the device.

2.5. Sample Observation
For both static and microfluidic staining, an inverted microscope (Zeiss Axio Vert.A1) was used. To visualize the fluorescent signal from the samples, an illumination system (X-Cite mini+, Excelitas, Waltham, MA) with a wavelength of 365 nm was used. All the images were recorded using a camera attached to the microscope (VEO E310L, Phantom, Wayne, NJ, USA). ImageJ (https://imagej.nih.gov/ij/, accessed on 10 November 2021) was used to analyze and quantify the results. Each experiment was performed four times for statistical analysis. We have not filtered the particles prior to observation, instead, we have directly placed a droplet of the diluted sample on top of a glass slide.

3. Results

3.1. Static Results

It is worth mentioning that high concentration Nile Red can lead to undesired aggregation [43,44], which may clog microfluidic channels and mask signals from stained microplastics. Moreover, the aggregation may destroy the samples into unrealistic microplastics (once aggregated the original size and shape are lost) and induce misleading conclusions [45,46]. We have observed that aggregations occurred for Nile Red solutions diluted up to 50X. Therefore, Nile Red solutions diluted to a minimum of 100X were used in our experiments to guarantee that no induced aggregations would happen. Figure 2 illustrates how aggregation occurs over time. Specifically, 50X Nile Red solution was placed onto a glass slide containing PE microspheres, and the aggregation process was recorded at 3000 fps. Figure 2A shows the initial frame (0.0003 s), it is possible to observe that particles are separated. The other images show the subsequent frames (from 0.0006 s to 0.0013 s), where the aggregation is shown. In this image, it is possible to see how fast aggregations are induced in microplastics due to the excess of Nile Red. Furthermore, the original features of the particles are lost, if someone were to study the size distribution or the shape of this sample, the outcome would certainly not be accurate due to the aggregation.

Once the threshold for the Nile Red concentration was defined, static experiments were conducted to determine the effects of Nile Red concentration, and temperature on staining efficiency. It was already known that temperature, residency time and ambient lights were important for the staining quality, however, no systematic study was available [30,47]. We have observed that for an infinitely long time (72 h) the highest pixel intensity of a sample containing 100X Nile Red at 25 °C is 150, thus we defined this intensity to be the reference for results normalization (all results shown in this paper are normalized with respect to this result). Figure 3 shows the results for the concentration and temperature static analysis, indicating that higher concentrations associated with higher temperatures provide better staining results, which is in accordance with the results from other groups [47,48]. However, it is difficult to identify relevant fluorescent signals at 25 °C, thus we have added arrows to indicate the particle positions.

Following the concentration and temperature experiments, we determined the effect of time on the staining quality (Figure 4). To do so, samples were kept inside the oven at a fixed temperature and Nile Red concentration, varying only the time. Since the previous results indicate that 100X and 250X Nile Red solutions at 80 °C are the most prominent combinations, thus these parameters were chosen along with variation in time: 5, 6, 7, 8, 9, 10, 11, and 12 min. As shown in Figure 4A, after 10 min, no significant changes in fluorescence level were observed, which means that this is enough time to extract the maximum performance from the staining agent. Figure 4B,C show the differences between the minimum and maximum staining time, where it is possible to observe that more time produces a stronger fluorescence signal in the particles.

Materials

Master Mold Resin

Learn More

H Series

M50-405

3.2. Microfluidic Results

As aforementioned, microfluidics hold great potential in providing continuous monitoring of microplastics in various water bodies. In this section, we applied the parameters under optimal conditions obtained from static experiments to explore the possibilities of using microfluidics for continuous microplastic identification.

As aforementioned, concentration and temperature are important parameters, thus their optimized values were adopted for the microfluidic device. When it comes to the flowing conditions, residency time becomes another important parameter that is subject to the external devices (i.e., syringe pump). In this paper, the total microchannel length was 400 mm, and its cross-sectional area was 2 × 2 mm. Using this design, we could achieve 5, 6, 7, 8, 9, 10, 11, and 12 min of residency time by applying corresponding flow rates of 7.82, 6.52, 5.58, 4.89, 4.34, 3.91, 3.55, and 3.26 µL/min, respectively. Note that the microfluidic device was placed inside the oven while the syringe pump was kept outside. The input and output hoses were long enough to enable sample collection and syringe manipulation outside the oven. Figure 5 shows the set-up arrangement.

From the information acquired during the static experiments, we have performed the microfluidic experiments with the most promising configurations with respect to concentration and temperature (i.e., 100X and 250X; at 80 °C). Different flow rates were tested to compare the performance of static and microfluidic staining regarding the residency time. As expected, lower flow rates provided better results, which is in accordance with the static experiments [46,49]. Nonetheless, it is possible to observe that for the lowest flow rate (and highest residency time) the static staining had superior fluorescence levels (~37% higher). This behavior could be attributed to the lower mixing quality governed by diffusion inside the device since the static samples were actively shaken prior to oven insertion [50,51,52]. Even though the microfluidic results exhibited lower fluorescence levels compared to the static experiments, it provides passive mixing and staining without tedious and time-consuming manual sample preparation. Nonetheless, it is worth mentioning that for higher flow rates, identification becomes difficult due to low fluorescence levels arising from short residency time. Figure 6 shows the results for microfluidic staining of the PE microspheres.

Besides PE microspheres, we further demonstrated the capabilities of our device for identifying other types of plastic. In this regard, multiple types of microplastics were applied, including microspheres (PS), fibers (cotton and acrylic), plastic parts scratched from storage containers (PP and PE). Moreover, yeast was adopted as a model of potential organic particles in seawater. Figure 7 shows the results of microfluidic staining for these samples. Note that PS microspheres showed better results when compared to the PE microspheres stained by the microfluidic device. Amongst the fibers, cotton indicated stronger fluorescence levels compared to acrylic, yet both were identifiable. Surprisingly, we found that all results obtained using PP and PE samples indicated the highest pixel intensity (i.e., 255, though larger than the threshold, it is indeed a strong indicator). However, as a recognized downside of staining identification, our method still suffers from the incapability of distinguishing microplastics from other natural particles, which can be seen from the results obtained using yeasts. It showed comparable fluorescence levels with respect to the plastics, highlighting the necessity for eliminating organic matter prior to sample analysis. Nevertheless, our results have demonstrated that continuous staining is achievable in microfluidic devices.

4.Discussion

In this paper, we have presented a novel microfluidic identification method for the continuous recognition of microplastics in water. Our method combines the Nile Red staining protocols with the high-throughput advantages imposed by microfluidics [45,47,51]. We acknowledge that the flow rates used must be small in order to achieve reasonable residency time, which has a negative effect on the throughput; however, the use of multiple (parallel) devices is feasible (especially due to its miniaturized size) which can enhance the throughput significantly [53,54]. In addition, the devices could be further improved and integrated into water monitoring stations in the future for continuous sampling and identification. According to the results obtained, the best staining quality is at the lowest flow rate (3.26 µL/min), which was expected since the static experiments showed that the lowest residency time performed the best.

In addition, though microfluidic results are still not as good as the static ones, future improvements can be carried out by adopting a better mixing strategy for Nile Red and samples [50,51,52]. Currently, a myriad of mixing methods has been developed for microfluidic devices, including both passive and active mixing. For example, better mixing performance could be addressed by adding pillars inside the channels [55,56]. Active mixers such as acoustofluidic mixers are alternatives and often provide more rapid mixing due to their superior particle control abilities [57].

The device can be further improved by coupling an on-chip heater, eliminating the need for an oven [58], thus reducing costs and enhancing its integrability. Once fully miniaturized, the device could be used for in situ analysis of water samples [47,48]. In situ analysis could also benefit from the use of smartphones, possibly for both identification and for device operation (pump and active mixers control) [59,60,61].

Note that the concentration of microplastics in seawater samples varies widely, being less concentrated off-shore (down to 8 particles/m³) [62]. In addition, global plastic distribution also changes significantly from one place to another, thereby a rapid and continuous identification prior to in-depth analysis would be beneficial. Though the staining method is not capable of distinguishing different types of microplastics, including other particles such as marine organisms, it indeed provides a simple, low-cost and effective method to confirm the presence of microplastics prior to more in-depth analysis including type differentiation [63]. Moreover, compared to regular visual inspection that bypasses the fluorescence staining, this proposed method turns microplastics into more prominent particles for better identification [64].

5. Conclusions

Overall, we have suggested the adoption of a microfluidic device for the continuous analysis and further detection of microplastics. Nile Red has proven to be effective for the identification of microplastics. Static experiments were performed to systematically assess the influence of staining agent concentration, temperature, and residency time. Based on the results, the microfluidic configuration for continuous staining was optimized, leading to the best fluorescence results among the tested configurations. Our method demonstrated to be feasible for the identification of different types of microplastics with the advantage of continuous staining and with the possibility of future integration for in situ identification along with higher throughputs. This platform demonstrated to successfully identify microplastics in a continuous manner, representing a valuable option for environmental management.

Microscale impeller pump for recirculating flow in organs-on-chip and microreactors

by

Microscale impeller pump for recirculating flow in organs-on-chip and microreactors

Sophie R. Cook,1 Hannah B. Musgrove,1 Amy L. Throckmorton, Ph.D.,2 and Rebecca R. Pompano, Ph.D.1 Author information Copyright and License information PMC Disclaimer

Fluid flow is an integral part of microfluidic and organ-on-chip technology, ideally providing biomimetic fluid, cell, and nutrient exchange as well as physiological or pathological shear stress. Currently, many of the pumps that actively perfuse fluid at biomimetic flow rates are incompatible with use inside cell culture incubators, require many tubing connections, or are too large to run many devices in a confined space. To address these issues, we developed a user-friendly impeller pump that uses a 3D-printed device and impeller to recirculate fluid and cells on-chip. Impeller rotation was driven by a rotating magnetic field generated by magnets mounted on a computer fan; this pump platform required no tubing connections and could accommodate up to 36 devices at once in a standard cell culture incubator. A computational model was used to predict shear stress, velocity, and changes in pressure throughout the device. The impeller pump generated biomimetic fluid velocities (50-6400 μm/s) controllable by tuning channel and inlet dimensions and the rotational speed of the impeller, which were comparable to the order of magnitude of the velocities predicted by the computational model. Predicted shear stress was in the physiological range throughout the microchannel and over the majority of the impeller. The impeller pump successfully recirculated primary murine splenocytes for 1 hr and Jurkat T cells for 24 hr with no impact on cell viability, showing the impeller pump’s feasibility for white blood cell recirculation on-chip. In the future, we envision that this pump will be integrated into single- or multi-tissue platforms to study communication between organs.

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

INTRODUCTION

Organ-on-chip (OOC) devices aim to mimic a tissue’s native environment by integrating single or multiple tissues in tandem into biomimetic perfusion systems.1,2 One feature that is critical to the function of these devices is directing fluid flow in a highly controlled manner. In vivo, there is constant fluid flow at varying velocities.3 Slower physiological fluid flow rates are found in the interstitium (0.1-1 μm/s)4 and within lymphatic capillaries (1.4-20.4 μm/s),5 while faster fluid flow rates are found in the blood vessel capillaries (80-180 μm/s),6 lymphatic vessels (870 μm/s, with a peak of 2200-9000 μm/s)7, veins (15,000-71,000 μm/s),8 and aortic artery (1,000,000 μm/s).9 As fluid moves, it provides nutrient and waste exchange as well as communication between organs through recirculation of cells, signaling molecules, exosomes, and so on. Flow also applies shear stress that impacts cellular function and viability and can result in cellular adhesion, activation, and extravasation.10–12 Thus, flow control systems for organs-on-chip must generate flow in a range of physiological and pathological flow rates, while ideally enabling transport of blood-borne cells between organs without damage. In addition to controllable flow rates, additional desirable qualities for flow control systems within OOC platforms include multiplexing capabilities, compatibility with cell culture incubators in terms of temperature output, and ability to recirculate media to enable cell circulation and communication between tissues.

Current technology provides a variety of methods to achieve biologically-relevant fluid flow rates on a microfluidic device, but these remain challenging for use when running many organ-on-chip devices simultaneously with fine control over flow rate, particularly for fluid recirculation. External, motorized fluid control systems such as syringe pumps13,14 and peristaltic pumps15,16 provide precise fluid control at physiological flow rates, but they can be expensive, bulky, and require many tubes or wires if running multiple devices at once. Furthermore, these pumps may emit heat, making them incompatible with use inside of an incubator for long-term culture. Recently, an elegant in-plane peristaltic pump was developed that is more compact than commercially-available peristaltic pumps and is compatible with incubators.17 With the ability to switch between multiple fluid inputs, this pump was designed primarily for rapid drug testing rather than for continuous media recirculation within OOC models. Alternatively, on-chip pneumatic peristaltic pumps use changes in pneumatic pressure to drive fluid flow, e.g. by serial compression of microfluidic channel.18–21 While powerful, this type of pump requires at least three tubing connections per device to drive fluid flow, which introduces complexity in handling for high throughput applications as well as sites for potential leaks.14,16,19–22 To avoid these issues, passive gravity-driven flow through a microfluidic device greatly simplifies handling by minimizing fluid or pneumatic connections, in exchange for less fine control over the flow rate.23–26 While most gravity-driven systems provide alternating or pulsatile unidirectional flow,23,24 cleverly-designed fluidics have also enabled continuous recirculation that is unidirectional through a single channel.25 However, actively controlled fluid recirculation for organs-on-chip remains a challenge.

A promising alternative means of active flow control uses rotating external magnets and an on-chip stir bar or impeller to drive fluid flow through a microfluidic chip.27–29 In prior reports, this approach elegantly reduced the need for tubing connections and allowed for controllable flow rates within the device.27,28,30 However, magnetic flow control has not been widely adopted, likely because the magnetic element within these devices was powered by commercially-available stir plates,27,28,30 most of which are large and lack precise rotational control or quantification. This limitation makes it challenging to run many devices simultaneously, especially within a culture incubator, and also to achieve consistent flow rates. In addition, the prior example of a rotating stir bar-based pump for OOC applications relied on manual PDMS-based fabrication and yielded a narrow range of flow rates recirculating through the device.28,30 This system was recently extended to perfuse media between two tissue models in an integrated polystyrene microfluidic plate fabricated by injection molding and laser fusion.29

Here, we present a magnetically-driven microscale impeller pump platform for recirculating fluid flow that is inexpensive, easy to fabricate and use, has low heat output, and has multiplexing capabilities. We designed and fabricated a prototype impeller pump and tested its ability to achieve a range of physiologically-relevant flow rates by varying pump and device features. We conducted computational modeling of the impeller pump geometry to assess fluid flow performance and scalar stress present within the device. As a proof-of-concept, we tested the cytocompatibility of the pump components with primary murine splenocytes and Jurkat T cells, models of recirculating white blood cells, and finally demonstrated the impeller pump’s ability to circulate cells across a range of flow rate regimes without loss of cell viability.

EXPERIMENTAL

3D-printed device fabrication

The microfluidic device and impeller piece were designed using Fusion 360. The device consisted of a large well with a micro channel loop intersecting the well tangentially. Channels had a square cross-section and were 0.5 or 1 mm in width. The microfluidic devices were printed using a CADWorks3D MiiCraft Ultra 50 DLP printer (CADWorks3D, Toronto, Canada) and a CADWorks3D MiiCraft P110Y DLP Printer (CADWorks3D, Toronto, Canada) using BV007a (MiiCraft, Jena, Germany) and Clear v.1 (FormLabs, Massachusetts, USA) resins, both recommended for use with microfluidics by their manufacturers. Designs for the device and impeller are provided in the Supporting Information. Drain ports were added in the channels when printing in the Clear resin to enable uncured resin to drain out of the channel during printing. For both resins, printer settings included a 0.10 mm gap adjustment with a slow peel speed. For BV007a in the Ultra 50 printer, a cure time of 1.15 s was used with a base cure time of 9 s for a single base layer with 1 buffer layer. All layers were printed at 50 μm and the light intensity was set to 75% power (9 mW/cm2) at 405 nm. For the P110Y printer, a cure time of 1.20 s was used with a base cure time of 25 s for a single base layer with 2 buffer layers. The light intensity was set to 100% power (5 mW/cm2) at 385 nm. For the Clear resin in the Ultra 50 printer, a 5 s cure time was used with a single base layer at a 6 s cure time, with a single buffer layer. Pieces with 1 mm channels used 50-μm layers with an 80% power setting (9.6 mW/cm2) at 405 nm; pieces with 0.5 mm channels used 100-μm layers with 75% power (9 mW/cm2) and 405 nm. For the P110Y printer, all pieces printed in the Clear resin had 100-μm layers with a 3.75 s cure time with a base cure time of 8.75 s for 4 base layers with 12 buffer layers. The light intensity was set to 70% power (3.5 mW/cm2) at 385 nm.

For the hand-washed conditions, all printed parts were rinsed with a spray bottle for 2 min with isopropyl alcohol (IPA) after coming out of the printer to remove any excess resin; parts printed in Clear resin were subsequently soaked in IPA for an additional 10 minutes. For the machine-washed conditions, all printed parts were submerged in IPA within a Form Wash (FormLabs, Massachusetts, USA) for 5 min (BV007a resin) or 8 min (Clear resin). After cleaning with alcohol, the pieces were dried thoroughly with nitrogen and placed in a high-intensity UV light box, either the CureZone (ResinWorks3D, Ontario, CA) UV light box (60 mW/cm2) or the Form Cure (FormLabs, Massachusetts, USA) UV light box (10 mW/cm2) for post curing. BV007a pieces were post-cured for 30 s in the CureZone or 1 min in the Form Cure, and Clear pieces were post-cured for 1 hour in both the CureZone or the Form Cure. After post-curing, Teflon-encapsulated magnetic stir bars (3 x 10 mm, Thomas Scientific, New Jersey, USA) were inserted into printed impeller pieces and glued in place using super glue (Loctite, Düsseldorf, Germany).

Assembly of the impeller pump external platform

Within an ABS plastic Universal Project Enclosure (200 x 120 x 56 mm, uxcell, Hong Kong, China), two 3-pin sleeve bearing computer fans (80 mm, Cooler Master, Taipei, Taiwan) were mounted on 4 screws that were glued to the base of the enclosure, termed the fan project box. Each computer fan was connected to a mini digital DC voltmeter (2.5 – 30 V, MakerFocus, Hong Kong, China) that was mounted within the enclosure so it was visible through the transparent top of the box. Two magnets were glued to the center of each computer fan. The initial prototype used 17.5 mm ceramic ferrite industrial magnets (Clout Science), which were later replaced with 6 mm brushed nickel magnets with a strength of 0.08 T (FINDMAG). The strength of the magnets used in the final prototype was measured using a Bell 610 Gaussmeter (F.W. Bell, Oregon, USA). On the outside of the fan project box, a 3D-printed chip holder (BV007a) was glued above the computer fan to hold the device in place. This project box, which resided within the cell-culture incubator during experimentation, was connected to an ABS plastic IP65 Hinged Junction box (150 x 100 x 70 mm, LMioEtool), which housed the PWM low voltage DC potentiometer (ALDECO) and 12 V DC female power connector (Chanzon), termed the power box. The power connector plugged into the 12 V AC DC power supply adapter wall plug (EWETON) that provides power to the entire pump platform. As the power project box is housed outside of the incubator, it allows for voltage and power control while an experiment is running. While the fan boxes were usually built with two fans, one pump box (Pump 7) was built with a single fan housed alongside a potentiometer; this pump box was not used for cell culture. All wiring was connected using a tin-lead rosin-core solder wire (ICESPRING) and wrapped in heat shrink tubing (Eventronic, Kommanditgesellschaft, Germany).

Assessment of the external platform

A digital laser photo tachometer (AGPtek, Brooklyn, New York, USA) was used to measure the revolutions per minute (RPM) of the magnetic impeller as it rotated. All RPMs reported were conducted at the onset of each experiment unless stated otherwise, and are the average of three RPM measurements made at a consistent voltage. The impeller pump stability was tested by measuring the impeller RPM over a period of 90 hr at a constant voltage. To monitor heat emission of the impeller pump platform, the single fan pump platform was run with no device for 24 hr in an insulated Styrofoam box at >10 V. For comparison, a peristaltic pump (BT100-1F-B, Langer Instruments, Boonton, New Jersey, USA) was run in the same box at 10 μL/min. Next, to monitor the impact of the pumps on temperature within a cell culture incubator, six external pump platforms with no devices (>10 V) were run for 24 hr in a cell culture incubator that was either off or on, as noted. Temperature was recorded from two locations inside the incubator: at the pump location (front of the top shelf) and at the back of the bottom shelf; the self-reported incubator temperature was also recorded when the incubator was on.

Computational modeling

To investigate the design of the fluid circuit, numerical modeling using computational fluid dynamics (CFD) studies was performed. ANSYS 15.0 CFX (ANSYS Inc., Canonsburg, PA, USA) was employed to mesh the geometry of the fluid circuit, including the impeller. Each fluid circuit consisted of three separate domains: 1) fluid channel; 2) top region of the pump well; and 3) the lower region of the pump well, which included the rotating impeller. Two fluid channel widths of 1 mm and 0.5 mm, with square cross-section, were considered. Each of the regions were connected via fluid-fluid interfaces. The fluid channel and top region of the pump well were specified to be in the stationary reference frame, while the lower region of the pump well with the impeller was defined to be in the rotating reference frame. A frozen rotor interface connected the top and lower regions of differing reference frames and maintained flow properties without circumferential averaging.

Each domain required separate meshes. The final mesh density for each channel width model was found using a standard grid independence study. Five separate meshes (5x105, 1x106, 5x106, 7.5x106, 10x106 element numbers), were created for each of the channel widths; the velocity values at multiple locations, pressure drop across the fluid channel, and mass flow rates in fluid channel varied by less than 5% for mesh densities greater than 5x106. The final number of mesh elements for the two channel width models were 5,758,350 and 5,947,380, respectively.

A turbulence modeling approach was employed due to the strong rotational fluid dynamics in the tank reservoir and fluid velocity in the channel. All simulations were performed under steady state, with a no-slip boundary condition on surfaces and a high-resolution advection scheme. In accordance with prior profilometry measurements of 3D printed materials,31 a surface roughness was specified at 3.5 μm on the internal fluid contacting surfaces; all walls were treated as rigid. To account for the effect of surface roughness, we further utilized a k-ω turbulence model where the y+ criterion (y+ < 1) was a design requirement for the mesh construct along the surface walls. Inflation layers were utilized to ensure achievement of the mesh y+ criterion. We verified that the low-Reynolds k-ω turbulence model requirement of a y+ mesh value of less than 1 was satisfied along all of the surfaces and walls of both models. Mesh quality was confirmed using standard mesh metrics including aspect ratio, Jacobian ratio, skewness and an ANSYS metric called element quality. A hybrid mesh of tetrahedral and/or hexahedral elements defining its volume was created for each region. The grid structure was designed to satisfy standard quality metrics, including the skewness and aspect ratio.32

Mesh quality metrics for all of the models met target goals: 1) aspect ratios less than 100, 2) Jacobian ratio less than 10, 3) skewness less than 0.25 and 4) element quality measure greater than 0.75. Convergence was achieved when the residual calculation error for the state variables reached less than 10−4. In line with experimental measurements in this study, water was indicated as the fluid media with Newtonian properties of a dynamic viscosity of 0.001 kg/m*s and density of 1000 kg/m3. Rotational speeds of 500 to 900 RPM were modeled.

Simulation results were assessed qualitatively and quantitively. Pressure losses, average velocity profiles, and mass flow rates in the fluid channels were determined. Each plane for analysis was created as a cross-sectional slice of the flow domain. Scalar fluid stress was estimated using the six components of the stress tensor (Equation 1). This approach estimates the 3D flow field and calculates a scalar stress (σ) as representative of the level of stress experienced by the fluid traveling through the entire model.33,34

Experimental fluid flow characterization

To measure the maximum velocity of the fluid flow within the device, a drop of blue food coloring (McCormick Culinary Food Color) was inserted into a reservoir in the device and tracked using a Dino-Lite Edge 3.0 digital microscope (SunriseDino, Torrance, CA, USA). Images were collected over time, and the distance the food coloring front moved over time was measured using DinoXcope software (SunriseDino, Torrance, CA, USA) to determine the fluid velocity. In preliminary experiments, we found that in situ measurement was preferable to addition of an external in-line flow meter, as the latter offered too high of a flow resistance and slowed the flow rate through the device.

For the experiments comparing channel size, the BV007a resin was used and the devices were printed using the MiiCraft Ultra 50 printer, hand washed, and cured in the CureZone. We later acquired a new, higher capacity printer and automated washer, which were used for subsequent experiments. Therefore, when comparing the changes in inlet size, the devices were printed using BV007a resin on the MiiCraft P110Y printer, washed using the Form Wash, and cured using the Form Cure.

The flow resistance (R) in each condition was calculated via an approximation for resistance in a square channel, where η is viscosity, L is channel length, and w is channel width (Equation 2):35

Primary murine splenocyte preparation

 All animal work was approved by the Institutional Animal Care and Use Committee at the University of Virginia under protocol #4042, and was conducted in compliance with guidelines from the University of Virginia Animal Care and Use Committee and the Office of Laboratory Animal Welfare at the National Institutes of Health (United States). Spleens were harvested from female and male C57BL/6 mice (Jackson Laboratory, USA) under the age of 6 months following isoflurane anesthesia and cervical dislocation. The spleens were collected into “complete RPMI” media consisting of RPMI (Lonza, Walkersville, MD, USA) supplemented with 10% FBS (VWR, Seradigm USDA approved, Radnor, PA, USA), 1× l-glutamine (Gibco Life Technologies, Gaithersburg, MD, USA), 50 U/mL Pen/Strep (Gibco, MD, USA), 50 μM beta-mercaptoethanol (Gibco, MD, USA), 1 mM sodium pyruvate (Hyclone, Logan, UT, USA), 1× non-essential amino acids (Hyclone, UT, USA), and 20 mM HEPES (VWR, PA, USA). Spleens were crushed through a 70-μm Nylon mesh filter (Thermo Fisher, Pittsburgh, PA, USA) with 10 mL of complete media, then centrifuged for 5 minutes at 400 x g. To lyse red blood cells, the pellet was resuspended in 2 mL of ACK lysis buffer prepared from 4.15 g NH4Cl (Sigma-Aldrich, St. Louis, MO, USA), 0.5 g KHCO4 (Sigma, MO, USA), and 18.7 g Na2EDTA (Sigma, MO, USA) in 0.5 L MilliQ water (Millipore Sigma, Burlington, MA, USA). Cells were lysed for 1 minute, then quenched with 10 mL of complete media, and centrifuged again. The pellet was resuspended in complete media, producing a splenocyte suspension with the density determined by trypan blue exclusion. The suspensions were diluted with complete media to a concentration of 1 x 106 cells/mL in preparation for culture.

Jurkat T cell preparation

For 24 hr long culture on 3D-printed devices, human Jurkat T lymphoblast cells (Clone E6-1 TIB-152, ATCC, VA, USA) were obtained from the laboratory of Ken Hsu at the University of Virginia. The cell line was cultured in media consisting of RPMI (Lonza, Walkersville, MD, USA) supplemented with 10% FBS (VWR, Seradigm USDA approved, Radnor, PA, USA), 1× l-glutamine (Gibco Life Technologies, Gaithersburg, MD, USA) and 50 U/mL Pen/Strep (Gibco, MD, USA). Before on-chip culture, the cells were centrifuged at 200 x g for 5 min and resuspended in 10 mL of media, producing a cell suspension whose density was determined by trypan blue exclusion. The suspensions were diluted with additional media to a density of 0.8 – 1 x 106 cells/mL in preparation for culture.

Analysis of cell viability

Prior to culture, all prints were post-treated as outlined above, and then subjected to an additional leaching process. For initial 1- and 4-hr tests of resin cytotoxicity and pump biocompatibility, the prints were soaked in 1x PBS (Lonza, MD, USA) for 24 hr at 37°C (BV007a prints) or 70°C (Clear prints) to mitigate cytotoxicity.36 Primary splenocytes (3.5 mL, 106 cells/mL in complete media) were aliquoted into the pump well of a 3D printed device or into a 12-well polystyrene non-treated tissue culture plate (VWR, Radnor, PA, USA) that served as a control. The wells within the tissue culture plate (23 mm diameter) were similar in diameter to the 3D printed pump wells (26 mm diameter), and the volume of media used was the same, so oxygenation and dilution of secreted factors is expected to be similar in the two systems. Cells were cultured at 37°C with 5% CO2 for either 4 hours for material cytotoxicity testing or 1 hour for “pump-on” viability testing.

In preparation for 24-hr cell culture on Clear prints, a longer leach step was performed, in which the devices were soaked in 1x PBS for 8 days and in complete media for 2 days at 37°C.37 Jurkat T cells or primary murine splenocytes were cultured within the printed and leached devices. The Jurkat cells and splenocytes (3.5 mL, 106 cells/mL in Jurkat-specific media) were aliquoted into the pump well of a 3D printed device or into a 12-well polystyrene non-treated tissue culture plate (VWR, Radnor, PA, USA) as a control. The cells were cultured at 37°C with 5% CO2 for 24 hr.

Following the culture period, the cell viability was assessed by flow cytometry using a previously established protocol.38 Briefly, 500 μL samples at 106 cells/mL were stained with Calcein AM (eBioscience, San Diego, CA, USA) at 67 nM for splenocytes and 95 nM for Jurkat T cells in 1x PBS for 20 minutes at 37°C. The stained samples were washed by centrifugation at 400 x g for splenocytes and 200 x g for Jurkat T cells for 5 min, then resuspended in flow buffer (1x PBS with 2% FBS). Following the wash step, 4 μL of 1 mg/mL 7-AAD (AAT Bioquest, Sunnyvale, CA, USA) was added to the cell suspension. Single-stain compensation controls were prepared using plate control cells (Calcein AM) or cells treated with a 1:1 v/v mix of media and 70% ethanol for 20 minutes (7-AAD); single-stains were mixed 1:1 v/v with unstained cells for analysis. All samples and controls were run on a Guava 4-color cytometer (6-2L) and analyzed with Guava® InCyte™ Software. 7-AADlow was defined as Live, and 7-AADhigh was defined as Dead.

The shear stress for each condition was approximated by calculating the fluid shear stress (FSS), where η is viscosity, Q is volumetric flow rate, h is channel height, and w is channel width (Equation 2).39

Imaging cell recirculation on-chip

Primary splenocytes were suspended at 3x106 cells/mL in 1x PBS and labelled with 3 μM Calcein AM for 20 minutes at 37°C. The labelled samples were washed by centrifugation at 400 x g for 5 min, then resuspended in 1x PBS at 3x106 cells/mL. Cell circulation was monitored in real time using a Zeiss AxioZoom macroscope (Carl Zeiss Microscopy, Germany) with an Axiocam 506 Mono camera and a filter cube for GFP (Zeiss filter set #38). 300 μL of the cell suspension was pipetted into the pump well of a 1x PBS-filled device, with the impeller off. After 15 minutes, initial images were captured within the pump well and near the reservoir in the channel. The pump was then turned on (~6.20 V), and images were captured of cells moving through the channel and reservoir.

Materials

Clear Microfluidics Resin V7.0a

Learn More

H Series

Learn More

RESULTS AND DISCUSSION

Concept of the microscale impeller pump

When designing the impeller pump platform, the major design goals included 1) small overall size, 2) user-friendly interface, 3) easily expandable to run multiple devices at once, 4) low heat output to be compatible with use inside cell culture incubators, and 5) ability to achieve controllable recirculating fluid flow at physiologically relevant flow rates. To address many of these design goals, we selected a fluid pumping mechanism inspired by that of the centrifugal water pump.40 The centrifugal water pump consists of a circular chamber with a rotary impeller that has curved vanes to generate a suction force, converting rotational energy into hydrodynamic energy and inducing fluid flow. Historically, large-scale centrifugal pumps were designed to recirculate and oxygenate water for fish and live bait on a fishing boat, similar to the purpose of media perfusion on a microfluidic device.40 This design and related waterwheel-based systems have been integrated into smaller-scale pumps for use as an artificial heart pump41 as well as in microfluidic technology.27,28,30,42–44

To adapt the centrifugal pump design to be compatible with recirculating fluid flow in a microfluidic device, spinning magnets were used to drive impeller rotation. Unlike rotating the impeller using an external motor, the use of a rotating magnetic field to drive fluid flow resulted in a simple set-up that was readily compatible with microfabrication techniques. The magnetic impeller rested within a large fluid-filled well, similar to the main chamber in a centrifugal water pump, and its rotation was used to drive fluid recirculation through an attached microchannel loop (Figure 1). Consistent with the vorticity of the impeller-driven flow in the well, preliminary tests showed that a tangential intersection provided better flow of fluid into the channel loop than a perpendicular intersection (data not shown). The direction of the recirculating fluid flow was determined by the rotational direction of the magnetic impeller (Figure S1), which in turn is controlled by the direction of the rotating magnets on the external pump platform.

Figure 1 . Design and prototype of the impeller pump and external pump platform.

(a) The impeller pump consisted of a magnetic stir bar inserted into a 3D-printed impeller placed within a large well on a 3D-printed microfluidic device. (b) The device was placed on top of an external pump platform, where the rotation of magnets on a computer fan caused the impeller to rotate, moving fluid through the device. The computer fan voltage was controlled using a potentiometer (POT), with voltage readout from a voltmeter. (c) An image of the device in place on the external pump platform. A reflective material was placed on half of the impeller piece to allow for RPM detection using a digital laser photo tachometer. Channel width was 1 mm. (d) Time-lapse images of recirculating fluid flow. The impeller rotated counter-clockwise (white arrow). Blue dye was inserted into the pump well, and over time, the dye exited the well and traveled through the channel (dye front marked with red arrow). Channel width was 1 mm.

Design and fabrication of 3D-printed microfluidic device and impeller insert

To generate a microscale impeller pump for use with microfluidic technology, digital light processing (DLP) 3D printing was used to generate the microdevice and impeller-like part (Figure 1c). DLP 3D printing produces highly reproducible devices in a shorter timescale than traditional soft lithography, and the open-source nature of this fabrication allows for easy translation to collaborators.45,46 Furthermore, this fabrication technique provided access to the required complex architecture, which would be challenging to produce by standard planar fabrication methods. To test the pump, we fabricated a simple chip that contained a large well that was connected to a loop of internal microchannels for recirculating fluid flow (Figure 1a,,c).c). To modulate the flow resistance through the microchannel and thus access a wide range of flow rates driven by the impeller pump, we designed two versions of the device with 0.5-mm or 1-mm square channel dimensions. Optically transparent resins were selected to facilitate imaging of flow through the microchannel (Figure 1d). Each device print took only 1 hr with an overall materials cost of $1-3 per device, making it easy and relatively inexpensive to produce a large number of chips for experiments. The impeller took 15 min to print at a materials cost of $0.03-0.12 per impeller, and was designed to hold a 10 mm magnetic stir bar in a rectangular hole in its center (Figure 1a).

When designing the dimensions of the pump well, impeller, and entry points of the microchannel, we reasoned that the hydrodynamic energy produced in the pump well by the rotating impeller would decrease further from the impeller, due to viscous energy losses. Consistent with this principle, in preliminary work, we found that if the impeller filled the majority of the cross-sectional area of the well, then the hydrodynamic energy generated fluid velocities that were able to reach as high as 33,000 μm/s (Figure S2), sufficient to model venous or arterial fluid flow.8,9 As we were primarily interested in slower capillary and lymphatic vessel flow velocities, a low impeller/well ratio was chosen, with a large well diameter (26 mm) relative to the smaller impeller piece (11.5 mm). Similarly, the intersection height of the channels approximately two-thirds up the side of the well (8.5 mm from the base of a 12-mm deep well) was optimized to slow the flow rates through the corresponding channel compared to a lower intersection height (Figure S2). In this manner, a 3-dimensional architecture for the device and impeller was achieved and optimized through rapid 3D printing, enabling the pump to attain biologically-relevant fluid flow regimes.

Design and fabrication of the external pump platform

The primary purpose of the external pump platform was to rotate magnets at a controlled and user-selectable rotational speed. Initially, we used a small DC motor that rotated two magnets with an Arduino board to control the rotational speed. However, we found that the DC motor emitted heat over time (data not shown), so we replaced it with a computer fan, which we hypothesized would emit less heat. The two magnets were glued to the center of the fan to control the rotation of the impeller within the microfluidic device (Figure 1b, ,2a).2a). For simplicity, a DC power adaptor was used that could be plugged into a wall outlet, and a potentiometer (POT) and voltmeter were used to provide voltage control and readout, respectively. The use of a POT and voltmeter minimized the complexity of the pump by removing the need for pump-computer interfacing to select the speeds, and allowed for the external platform to be condensed into a small overall size. To use the pump platform in a humidified incubator, the electronics needed to be sealed within an air-tight enclosure to prevent damage from the water vapor. As a result, the computer fan, POT, and voltmeter were mounted within an air-tight plastic project box (Figure 2a). No fluidic or pneumatic tubing was required for pump function in this device, which resulted in a pump platform that was easy to use that avoids leaks and further complications. Furthermore, the use of common, inexpensive items such as small magnetic stirrers and computer fans resulted in an overall materials cost of $50-75, and it took approximately 2 hr to assemble each pump platform. These qualities made the external impeller pump platform easy to replicate to be able to run many devices simultaneously.

Figure 2.  Impeller pump control and external platform expansion..

(a) An image of an external pump platform with a single computer fan in a project box with a single voltmeter and POT. (b) As the voltage increased, the RPMs increased for pumps 1-7. The black line represents the line of best fit for the entire data set (y = 281(x) - 1221, R2 = 0.9522). (c) Impeller RPM was measured at three voltages in devices with varied channel dimensions. Results were compared using a one-way ANOVA (n = 3); ns indicates p > 0.9. (d) The distance between the center of the magnets dictated the stability of the magnetic stirrer within the impeller. When the distance (d1) was greater than the length of the stirrer, the impeller rotation is unstable, resulting in no fluid flow. When the distance (d2) was equal to the length of the stirrer, the impeller rotation was stable. (e) Stability of impeller rotation speed over a period of 90 hr. (f) An image of the multiplexed external pump platform with two computer fans per project box, each with their respective voltmeter. Each fan was connected to an individual POT, which were all housed in a shared hinged project box. (g) Six devices placed on a single shelf of a cell culture incubator.

Finally, to secure the location of the device relative to the location of the spinning magnets, a 3D-printed chip holder was designed and mounted on the top of the project box (Figure 1c). The holder was positioned such that the center of the pump well was aligned over the spinning magnets, thus ensuring stable impeller rotation. The height of the fan and magnet assembly was also fixed relative to the chip holder, thus ensuring consistent magnetic engagement.

Tunability and stability of impeller rotation

We tested the ability to control the speed of impeller rotation by tuning the voltage provided by the potentiometer to the computer fan. A digital laser photo tachometer was used to measure the rotations per minute (RPM) of the impeller by placing a reflective material across half of the impeller (Figure 1c). As the impeller rotated within a water-filled well, the tachometer counted the number of times the reflective material passed through the laser emitted and provided a numerical value of the RPMs. As expected, the speed of rotation increased linearly with the voltage each computer fan received, which was read out using the voltmeter within the external pump platform (Figure 2b). There were variations between the RPMs across seven copies of the pump platform (root-mean-square deviation from line of best fit = 89 RPM), perhaps due to differences in the response of the computer fans to voltage supply. Since this variation was minor, the pump platforms were treated as equivalent. As expected, there was not a significant difference in impeller RPM when used in microchips with different channel dimensions at the same voltage (Figure 2c).

As the impeller rotated within the pump well, a vortex of fluid was generated to drive fluid through the channel. In an initial design with large magnets, we observed that unstable impeller rotation failed to form a fluidic vortex and did not drive stable fluid flow (illustrated in Figure 2d). In this design, the distance between the center points of the magnets (d1 = 17.5 mm) was greater than the length of the 10 mm stir bar. To resolve this issue, smaller magnets were used to matched the center-to-center distance to the length of the stir bar (d2 = 10 mm) (Figure 2d).47 With this design, impeller rotation was stable for at least 90 hr (Figure 2e). The steady computer fan rotation provided nonpulsatile unidirectional fluid flow by design; if pulsatile or bidirectional fluid flow were desired, additional electronics such as Arduino could be added to modulate the voltage the computer fan receives, or a bi-directional, low-heat motor could be used instead of the fan.

Scale up to a multiplexed pump platform

The impeller pump was designed to be used with OOCs, for which simultaneous use of many devices is crucial to reduce the experiment time and increase the power of biological experiments. Therefore, we tested the scale up of the external platform to run multiple micropumps simultaneously. To scale up the prototype external platform, two computer fans were mounted within a single project box. This design was easily scaled up by producing 3 project boxes (Figure 2f) to run 6 devices on a single shelf of a cell culture incubator (Figure 2g), with total space for up to 36 devices (18 boxes) in the incubator if needed. In the future, use of smaller fans or low-heat motors may enable further miniaturization of the pump platform to fit additional pumps into the incubator, but this was not explored here. We chose a design in which each computer fan was individually connected to a potentiometer so that each fan could run at different speeds. The potentiometers for all six fans were kept in a separate hinged project box that remained outside of the cell culture incubator to increase the available space within the incubator (Figure 2g). This feature also reduced the potential for damage of the electrical parts at 37 °C and high humidity. For simplicity, the entire multiplexed system shared a single power adapter, with the voltage split between each fan.

Finally, we tested the variation in impeller rotation across all six of the external pump platforms. Similar to the slight variation in the resulting RPMs at different voltages (Figure 2b), we observed variation (26% CV) in the measured fluid velocity at a fixed voltage between copies of the hand-built external pump platform (Figure S4). We anticipate that in experiments where accuracy of flow rate is critical, each pump platform will be screened in a quality control step to verify that it provides flow in a specified range. Alternatively, in the future, automation of production of the platforms would likely reduce variation.

Negligible heat emission for long-term culture

A major design goal for the impeller pump was minimal heat emission to allow for extended cell culture within an incubator. Stable temperatures (±1°C) are critical to maintain viable cell cultures,48 and we previously found that a peristaltic pump rapidly raised the temperature inside a culture incubator if not countered with cooling packs.16 To test this feature of the impeller pump, the heat emission was first measured by placing the pump platform at a high rotational speed (> 10 V) or a peristaltic pump at 10 μL/min within an insulated Styrofoam box for 24 hr (Figure 3a). After 24 hr, the temperature within the box had increased modestly to 29.0°C with the impeller pump (temperature with no pump, 21.0°C), versus a drastic increase to 52.7°C with the peristaltic pump (Figure 3a).s simultaneously.

Figure 3.  Heat emission of the impeller pump

 (a) The temperature within a closed Styrofoam box was measured with no pump, a single impeller pump platform (>10 V), and a peristaltic pump (10 μL/min) over a period of 24 hr. (b) A schematic of a cell culture incubator during the experiment. When measuring the temperature within the incubator, the six pumps were placed on the front of the top shelf to the right. The temperature was measured at the pumps on the top shelf to the right (P1) as well as in the back of the bottom shelf to the left (P2). (c,d) The temperature within a cell culture incubator was measured at positions P1 and P2 while six pumps were run (>10 V) over a period of 24 hr while the incubator was (c) off or (d) on. Black lines indicate room temperature (c) and the temperature readout from the incubator when it was on (d). (Single column)

As the pumps will primarily be used within cell culture incubators, we also tested how use of the impeller pump affected the internal temperature of an enclosed incubator. For a rigorous test, we ran six pump platforms simultaneously at a high rotational speed (>10 V) for 24 hr in the incubator. When this test was conducted with the incubator shut down, i.e. without any built-in temperature control, there was a 4°C increase in temperature near the pumps (P1), while the temperature far from the pumps remained comparable to room temperature (P2) (Figure 3b,,c).c). Conducting the same test with the incubator on, i.e. to replicate cell culture circumstances, resulted in only a 0.4°C increase at the pumps (P1), which is within the ±1°C acceptable temperature window,48 and the temperature far from the pumps remained relatively constant (Figure 3d). Furthermore, there was no change in the temperature reported by the cell culture incubator (37.0°C), whose sensor is located near the pumps at the top of the incubator. Collectively, these data indicated that the multiplexed external pump platform did not emit a noticeable amount of heat and was compatible with extended use inside a cell culture incubator.48

Predicted fluid flow using a computational model of the impeller pump and microfluidic chip.

To model the low (~10-100 μm/s)6 and high (1-10 mm/s)7 fluid velocities found within blood and lymphatic vessels, the impeller pump platform must be able to control the flow rate over several orders of magnitude. To understand the factors that controlled impeller-driven fluid flow through the microfluidic device, a computational model was developed (Figure 4a). The frozen rotor method was used to model the spinning impeller,41 with one rotating domain and one stationary domain to avoid discontinuities at the entry to the microchannel. The stationary domain consisted of the top of the pump well and the connecting channel (Figure 4bi), while the rotating domain consisted of the base of the pump well and the impeller piece (Figure 4bii). The mesh density for each fluid domain was determined using a grid independence study to elucidate when the physics being modeled was no longer dependent upon the mesh resolution (see Experimental). Briefly, to establish mesh independence for each channel geometry, we examined value fluctuations for key study parameters: pressure drop across the channel, mass flow rate in the channel, and fluid velocities on the channel inflow and outflow. The analysis revealed higher value fluctuations (7.6-15.3%) for coarser densities and then leveled to fluctuations of less than 5% for mesh sizes greater than 5 million elements (Figure 4b). A turbulence modeling approach was employed due to the strong rotational fluid dynamics in the tank reservoir and fluid velocity in the channel. To account for the effect of surface roughness, a low-Reynolds k-ω turbulence model was adopted (see Experimental). The y+ turbulence mesh parameter was also verified. Mesh quality was confirmed using standard mesh metrics (i.e., aspect ratio, skewness, element quality).

Figure 4.  Simulated flow control with the microscale impeller pump.

 (a) Top view of the fluid circuit model showing the narrow fluid channel, the impeller, and the pump well. Arrows show direction of fluid flow. (b) Side view of the fluid circuit model showing the (i) top domain mesh that contained the top of the pump well and connecting channel and (ii) the bottom domain mesh that contained the lower region of the pump well and the rotating impeller. (c) As the impeller rotated in the simulation, fluid left the pump well and crossed Plane 1, passed through the remaining channel domain, and crossed Plane 2 prior to re-entering the pump well. Flow rate and pressure drops were determined at these two locations. (d) Predicted average velocity across plane 1 and (e) predicted pressure loss between plane 1 and 2 increased with the RPM of the impeller, for both the 0.5 mm and 1 mm channel size.

The computational model was used to predict the trends in velocity and the pressure drop across a transverse plane in the microfluidic channel as a function of these parameters (Figure 4c). As expected, increased rotational speed of the impeller drove increased rates of flow through the microchannels, and the larger channel size resulted in higher average velocities (580-4900 μm/s) than the smaller channel size (9-64 μm/s) (Figure 4d).6,7 There was a more substantial pressure drop for the 1 mm channel, as compared to the 0.5 mm channel (Figure 4e); while initially counter-intuitive, this result is consistent with Hagen-Poiseuille’s law, as the increase in velocity for a larger channel size exceeded the decrease in flow resistance. Thus, these computational findings demonstrated that the flow regime can be selected by altering microfluidic circuit geometry and fine-tuned by varying the RPM.

Experimental fluid velocity control over two orders of magnitude

After the computational model was used to predict trends in velocity within the device, we proceeded to experimentally evaluate the velocity in the prototype experimental system across a variety of different conditions. To measure the velocity within the 3D-printed device, reservoirs were added near the pump well for dye insertion. An equilibration period was needed to achieve a consistent velocity after initially starting the impeller rotation, which was observed to be approximately 3-5 minutes. After this time, a drop of dye was inserted into the reservoir, and images were collected as the front of dye moved over time (Figure 5a). The maximum velocity was measured at the center point of the parabolic flow of the moving front of dye within the channels (Figure S3, ,5a).5a). This in situ measurement method ensured that no additional pressure drop was introduced by adding an external flow rate sensor.

Figure 5.  Experimental flow control with the microscale impeller pump.

 (a) 3D rendering and photos of a 3D printed device (0.5 mm channel size), showing colored dye moving through the channel over time. The white arrow corresponds to the dye front as it moves away from the reservoir over time. (b) Experimentally measured maximum velocity in a 0.75-mm channel without a constriction (0.75-mm inlet; n = 3) and with a constriction (0.5-mm inlet; n = 3) increased as RPM increased. The constriction occupied 16% of the 91-mm total channel length. (c) Experimentally measured maximum velocity in 0.5- or 1-mm channels with varied channel length. (n = 3, except for 1 mm, 93 mm long channel where n = 4).

We reasoned that as the impeller circulated fluid within the well, a portion of the rotating fluid was pushed through the intersecting channel inlet, driving fluid into the channel. Therefore, the volumetric flow rate (mass flow) at the channel inlet should be driven by both the rotational velocity and the cross-sectional area of the entryway to the channel, with a larger cross-sectional area allowing more fluid entry. As expected, and consistent with the simulated data, the velocity within the channels increased with the rotational speed of the impeller (Figure 5b,,c).c). To test the effect of the entry area on flow rate, the inlet size was varied while keeping the cross-sectional area of the downstream channel (square cross-section) the same. Interestingly, constricting the inlet from 0.75 mm square (11.8 Pa*s/mm3 flow resistance) to 0.5 mm square (19.8 Pa*s/mm3) decreased the velocity measured in the channel downstream by over 6-fold (Figure 5b), despite the channel itself retaining the same dimensions. As the short constriction only increases flow resistance by < 2-fold, the increased resistance cannot fully account for the significant drop in velocity, so we predict that the reduced inlet area also played a role. The dependence on the cross-sectional area of the entry point in this open system is different from the behavior of pressure-driven flow in a fully closed system, which must maintain a constant volumetric flow rate.

Next, we explicitly tested the prediction that increasing the resistance within the microfluidic loop, e.g. with longer or narrower channels, would result in a lower velocity regime. The larger channel width, 1 mm (2.5 Pa*s/mm3), resulted in a higher velocity range (1680-6400 μm/s at 61 mm) that was comparable to lymphatic vessels in vivo (Figure 5c).7 Reducing the channel width to 0.5 mm (39.8 Pa*s/mm3) yielded a lower velocity range (130-800 μm/s at 61 mm) that was comparable to blood vessel capillaries in vivo (Figure 5c).6 This trend matched that of the computational model and combines an increase in flow resistance with a decrease in entry area. Lengthening the total channel from 61 mm to 93 mm (60.5 Pa*s/mm3 for 0.5 mm channel, 3.8 Pa*s/mm3 for 1 mm channel), which changes flow resistance alone, resulted in a slower velocity as expected, though interestingly only for higher impeller rotational speeds (Figure 5c), again potentially attributable to the open system.

These results confirm that changes in the resistance of the microfluidic network impact the velocity of impeller-driven flow, and also indicate that experimental calibration of flow rate versus impeller rotation speed must be performed for each microdevice design. While the experimental trends were similar to the computational model, the magnitudes of the experimental and predicted velocities differed quantitatively, especially for the smaller channel size. We speculate that the experimental system is subject to additional forces not yet captured by the model, such as the generation of vortex flow within the pump well upon impeller rotation,49 surface tension and wetting at the air-water interface in the pump well, varied surface roughness, and small variations in channel dimensions. While the model will be further refined in the future, changes in resins, print quality, and device architecture also would be expected to impact difficult-to-control parameters such as surface roughness and shrinkage, which alter the precise dimensions of the channels. Therefore, calibration of the velocity across a range of impeller rotational speeds should be performed for quantitative flow rate control, similar to calibration of peristaltic pumps.

Shear stress approximation across device

Having validated the trends of the computational model, we used it to predict the levels of shear stress within the device during impeller-driven fluid flow. Shear stress is a major consideration for cell recirculation, as high shear stress can damage the cells and diminish viability. Physiological shear stress spans 0.6-12 dyn/cm2 in lymphatic vessels and 0.35-70 dyn/cm2 in normal blood vasculature.7,50 100 dyne/cm2 is sometimes considered the threshold for pathological shear, which reaches >1500 dyn/cm2 in diseased or stenotic vessels.50 Using the computational model, fluid shear stress levels during high impeller rotational speeds (900 RPM) were estimated at various regions within the device. Looking at the impeller surface, 93.2% of the surface was < 100 dynes/cm2 (Figure 6a), i.e. within the physiological range. The highest shear stress, 400 dynes/cm2, was found along the edges of the impeller; we reasoned that the cells suspended in the circulating media would rarely contact the impeller surface or edges due to centrifugal forces and the large volume of the pump well. Within the channels, the surface shear stress approximations were much lower and well within the physiological range: 0.04-0.10 dynes/cm2 in the 0.5 mm channel (Figure 6b), and 0.40-1.22 dynes/cm2 in the 1 mm channel (Figure 6c), with the highest stress in the corners of the channel. Based off of these results, we predicted that the impeller rotation would not have a significant impact on the viability of circulating cells.

Figure 6.  Predicted shear stress within the device.

The scalar stress was approximated using the computational model across the surface of (a) the impeller, (b) the 0.5 mm channel, (c) and the 1 mm channel. (a) Along the impeller, the highest scalar stress was present along the edges. (b,c) The scalar stress was highest at the corners of the channel in both channel designs. (Single column)

Selection of a sufficiently biocompatible resin for the 3D-printed micropump

As the impeller pump platform is intended to recirculate fluid and cells within OOCs and other biological model systems, the material used to fabricate the device and impeller must be cytocompatible for the timescale of the experiment. While 3D printing is an easy way to reproducibly fabricate microfluidic devices with complex architecture in a short period of time, the liquid photopolymer resins used for stereolithography (SLA) and digital light processing (DLP) 3D printing are often cytotoxic.51 The use of additives such as optical absorbers and plasticizers can enhance the print resolution, enabling smaller internal channel sizes and smaller port diameters, such as with the MiiCraft BV007a resin, but these may result in increased toxicity if these molecules leach out of the device.36,51–53 Some photopolymer resins designed for biomedical applications, such as FormLabs Clear, may have reduced cytotoxic additives but also reduced print resolution.53

To identify a cytocompatible resin for the microscale impeller pump, the device and impeller were printed in two different resins, MiiCraft BV007a and FormLabs Clear (Figure 7a), and tested with primary murine splenocytes as a model for circulating white blood cells. Primary splenocytes provided a rigorous cytotoxicity test because they are more susceptible to damage than immortalized cell lines. To remove cytotoxic leachates, the BV007a and Clear devices were soaked in PBS for 24 hr at 37°C and 70°C, respectively (“post-treatment”) prior to use.36,53 After 4 hr of culture in complete media in the pump well without impeller rotation, primary splenocytes cultured on the BV007a piece had significantly decreased viability compared to off-chip controls, with less than 30% viable cells (Figure 7b, Figure S5). In contrast, culture on the Clear piece yielded no significant difference in viability compared to the off-chip controls, though there was a non-significant drop (Figure 7b).

Figure 7.  Assessment of the biocompatibility of 3D-printed pump chambers with primary splenocytes and an immortalized lymphocyte cell line.

 (a) Image of devices printed in the Clear resin and BV007a resin for cytotoxicity testing. The device printed in the Clear resin required drain ports along the channel to print internal channels. (b) Primary splenocytes were cultured for 4 hr without impeller rotation in devices post-treated with a 24 hr soak, and viability was analyzed by flow cytometry. Quantification of live (7-AADlow) cells after culture off-chip, in the Clear pump well, or in the BV007a pump well. Results were compared using a one-way ANOVA with Tukey post-hoc tests (n = 3). *** indicates p <0.0006; ns indicates p > 0.08. (c) Primary splenocytes (Splen.) and Jurkat T cells were cultured for 24 hr without impeller rotation in devices printed using the Clear resin with a 10-day post-treatment ,and viability was analyzed by flow cytometry. Quantification of live (7-AADlow) cells after culture both off-chip and in resin for both cell types. Results were compared using a one-way ANOVA with Tukey post-hoc tests (n = 5, except for off-chip splenocytes where n = 4). * indicates p < 0.02; ns indicates p > 0.2.

Having identified the Clear resin as the more promising material, we further tested it for overnight cell culture using primary and immortalized cells, since the latter are more hardy. To ensure that all possible leachates were removed, devices printed in the Clear resin were soaked in PBS for 8 days and then in media for 2 days at 37°C, according to a published protocol.37 Primary murine splenocytes and Jurkat T cells were cultured within the 3D-printed microfluidic pump wells for 24 hr without impeller rotation. Whereas primary splenocytes showed a significant decrease in viability, the viability of Jurkat T cells was not significantly different compared to off-chip controls (Figure 7c). Therefore, we concluded that the Clear resin was sufficiently compatible for use in experiments of 4 hr or shorter duration with primary cells, and that the use of cell lines expanded compatibility to at least 24 hr. Biocompatibility of SLA/DLP resins for primary cell culture continues to be an area of investigation in our lab and others.

Recirculation of lymphocytes under biomimetic flow regimes

Cell recirculation is a key feature of inter-organ communication in vivo, and a new pump for organs on chip should be able to drive cell recirculation without impairing viability. Here, we tested the ability of the impeller pump to drive continuous white blood cell recirculation under fluid velocities found within lymphatic vessels and vasculature in vivo. Given the depth and size of the pump well, it was possible that cells would settle to the bottom of the pump well instead of remaining suspended for recirculation through the microfluidic channel, especially at low RPM. To address this concern, primary splenocytes were stained with Calcein AM and deliberately allowed to settle to the base of the pump well of a pre-filled device while the impeller was off (Figure 8a). Imaging at this time confirmed that the cells settled along the base of the pump well (Figure 8b) and that no cells were present within the channels (Figure 8c). Once the impeller began to rotate at a low rotational speed (6.20 V, 420 RPM), the cells resting on the base of the pump well were resuspended and began to recirculate through the channels, where they were visible entering the reservoir (Figure 8d). Cells moved much faster through the 1-mm channel, as evidenced by the blurring of fluorescently-labeled cells moving through the center of the reservoir, than through the 0.5 mm channel, consistent with the slower flow rate in narrower channels (Figure 8d). Thus, the rotation of the impeller pump successfully resuspended cells even from rest and achieved continuous cell recirculation through the device.

Figure 8. Recirculating cells under different flow regimes

 (a) Schematic of the cell recirculation procedure. Cells were labeled with Calcein-AM and inserted into the pump well of a pre-filled device with no impeller rotation. (b) Image of the fluorescently-labelled cells resting on the base of the pump well after a 15 min rest period with the pump off. (c) Image of the reservoir in the channel of the device with no impeller rotation. There are no cells present. (d) Images of cells passing through the reservoir as the impeller rotated for devices with 0.5 mm channels (left, expanded in center) and 1 mm channels (right). (e,f) Quantification by flow cytometry of live (7-AADlow) primary murine splenocytes cultured for 1 hr off-chip, on-chip with no impeller rotation, or circulated at low (6.10 V, 490 RPM) or high speed (7.45 V, 870 RPM) through (e) 0.5 mm channels or (f) 1 mm channels. Viability results were compared using a one-way ANOVA with Tukey post-hoc tests (n = 3). ns indicates p > 0.1. (g) Quantification by flow cytometry of live (7-AADlow) Jurkat T cells cultured for 24 hr off-chip, on-chip with no impeller rotation, or circulated at high impeller RPM (870 RPM, 5200 μm/s) through 1 mm channels. Viability results were compared using a one-way ANOVA with Tukey post-hoc tests (n = 6). ns indicates p > 0.2.

Next, we tested the impact of impeller rotation at various biomimetic fluid velocities on viability of primary splenocytes. We hypothesized that 1 hr would be sufficient to see any impact on cell viability from mechanical damage from impeller rotation. Cells were continuously recirculated through the chip (24-hr post-treated, FormLabs Clear) by the impeller pump for 1 hr, while the whole system was inside a cell culture incubator. Based on the results above, a 0.5 mm channel was used to achieve low velocities similar to those measured within blood capillaries in vivo (Figure 8e),5 and a 1 mm channel was used to achieve higher velocities similar to those measured within lymphatic vessels in vivo (Figure 8f).7 Compared to the off-chip control and cells in the pump well without impeller rotation, there were no significant differences in viability for the cells in fluid moving at 40 μm/s (shear stress of 0.0003 dynes/cm2), 730 μm/s (0.005 dynes/cm2), 1080 μm/s (0.03 dynes/cm2), or 5200 μm/s (0.16 dynes/cm2) (Figure 8e,,f).f). The shear stresses listed are estimated from the flow rate in the channel, which is separate from any stress imparted on cells from impeller rotation (Figure 6). We concluded the impeller-driven micropump did not cause mechanical damage to primary cells even at the higher impeller rotational speeds or flow velocities.

Finally, to model long-term white blood cell recirculation and further test the rotating impeller’s impact on cell viability, Jurkat T cells were circulated at a high speed for 24 hr using the impeller pump platform. The T cells were continuously recirculated on-chip (10-day post-treated, FormLabs Clear resin) at a high impeller rotational speed (7.45 V, 870 RPM, 5200 μm/s, 0.3 dynes/cm2) through 1-mm channels (Figure 8g). Recirculation and impeller rotation did not significantly reduce the viability of the cells compared to static culture in the pump well or off-chip controls (Figure 8g), although variability was high. Variability may be related to the inherent differences between copies of the pump platforms (Figure S4), e.g., due to slight variations in impeller RPM, though this was not tested here. In summary, the microscale impeller pump provided cell recirculation at high impeller rotational speeds for at least 24 hr, making it suitable for future use in microscale cultures and OOCs. We note that separate from compatibility of the resin, the compatibility of impeller-driven recirculation may vary as a function of cell type and impeller speed, and should be tested for each cell type and flow rate of interest.

CONCLUSION

Here, we have reported a novel and user-friendly magnetically-driven impeller pump system for recirculating fluid flow through microfluidic devices. The pump design allowed for the use of inexpensive parts to enable magnetic impeller rotation, which resulted in a simple user interface, no tubing connections, and negligible heat output, making the pump compatible with cell culture incubators. A computational model of the impeller pump was developed to predict the fluid flow through the pump and associated microfluidic chip by using a frozen rotor approach. By varying the dimensions of the channel and inlet as well as the rotational velocity of the impeller, the impeller pump achieved a wide range of physiologically-relevant flow rates, from <50 to >5200 μm/s, and the trends of flow rate as a function of channel cross-sectional area and impeller speed were comparable to the predictions of the computational model. The model predicted low shear stress in the microfluidic channels, with the highest shear at the edges of the rotating impeller where cells were not concentrated, suggesting biocompatibility of the system with recirculating cells. As a proof-of-concept, primary murine splenocytes and Jurkat T cells were recirculated through a microfluidic chip at various biomimetic fluid flow regimes while maintaining high cell viability for up to 24 hr. In the future, the impeller pump will be useful to drive recirculating fluid flow through OOC and multi-organ-on-chip platforms, to study communication and effects of cell recirculation between tissues on-chip.

Supporting Info PDF

View

CAD file of impeller

View

CAD file of device, 0.5 mm channel

View

CAD file of device, 1 mm channel

View

Rapid Fabrication by Digital Light Processing 3D Printing of a SlipChip with Movable Ports for Local Delivery to Ex Vivo Organ Cultures

by

Rapid Fabrication by Digital Light Processing 3D Printing of a SlipChip with Movable Ports for Local Delivery to Ex Vivo Organ Cultures

Megan A Catterton, Alexander G Ball, and Rebecca R Pompano

SlipChips are two-part microfluidic devices that can be reconfigured to change fluidic pathways for a wide range of functions, including tissue stimulation. Currently, fabrication of these devices at the prototype stage requires a skilled microfluidic technician, e.g., for wet etching or alignment steps. In most cases, SlipChip functionality requires an optically clear, smooth, and flat surface that is fluorophilic and hydrophobic. Here, we tested digital light processing (DLP) 3D printing, which is rapid, reproducible, and easily shared, as a solution for fabrication of SlipChips at the prototype stage. As a case study, we sought to fabricate a SlipChip intended for local delivery to live tissue slices through a movable microfluidic port. The device was comprised of two multi-layer components: an enclosed channel with a delivery port and a culture chamber for tissue slices with a permeable support. Once the design was optimized, we demonstrated its function by locally delivering a chemical probe to slices of hydrogel and to living tissue with up to 120 µm spatial resolution. By establishing the design principles for 3D printing of SlipChip devices, this work will enhance the ability to rapidly prototype such devices at mid-scale levels of production.

Keywords: SLA printing, resin printing, tissue culture, local stimulation, two-phase microfluidics

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

Introduction

The ability to produce microchips easily and with minimal manual assembly, while retaining rapid prototyping capabilities, is highly desirable for pushing microfluidic devices past the first hand-built prototype stage [1,2,3]. Scaled-up fabrication is critical to conducting experiments at moderate scale (dozens of devices) and for propagating such technology to collaborators. In particular, this scale of fabrication would be useful for SlipChips, which are two-phase, reconfigurable microfluidic devices [4,5,6,7,8,9]. SlipChips usually comprise two planar components that can be “slipped” relative to one another, contain recessed features to hold droplets or streams of aqueous solution, and are separated by a thin layer of oil [4]. SlipChip devices were first developed in the Ismagilov lab [4] as a new technology to perform in low-resource settings [5,6,7]. The first SlipChips were fabricated from glass plates, which offer ideal surface properties and optical clarity but require wet etching with HF, a hazardous procedure that requires a skilled technician [4,10]. Since then, many different Slip-based designs have evolved, including rotational Slipdisc and paper-based SlipPADs, to perform a wide range of laboratory processes such as PCR, cell culture and local delivery to tissue slices [8,9,11,12,13,14,15,16,17]. Fabrication is especially challenging for novel slip-based devices that have multiple layers per component [9,17]. Although injection molding can simplify fabrication at large scale [18], an alternative method is needed to fabricate SlipChips at a moderate scale, while retaining the ability to rapidly prototype.

Any fabrication system for SlipChips must be able to meet four platform requirements, in addition to producing the specific features needed for the intended application. To prevent the aqueous phase from spreading into the oil-filled gap between components, high capillary pressure at the oil–water interface must be maintained. Therefore, the surfaces in contact with the oil layer must be flat and smooth enough to create a gap height of ~1–10 µm across the entire face of the chip [5]. Furthermore, these surfaces must be hydrophobic; if a fluorinated oil is used [4], then a fluorophilic surface is preferred. Finally, for SlipChips that rely on visual alignment or optical detection, the layers must be optically transparent.

Considering these requirements, we reasoned that digital light projection (DLP) 3D printing, which uses UV or blue light to cure photocrosslinkable resins layer by layer [19,20], may facilitate SlipChip fabrication and allow for rapid prototyping. This additive method is quickly gaining popularity for fabricating small parts and microfluidic devices, because of both its high feature resolution and reproducibility and its rapid fabrication speed compared to traditional soft-lithography [3,21,22,23]. While 3D printing has not been reported previously for SlipChips, two of the four fabrication requirements are already met. We recently described a method for fluorination of a DLP-printed surface based on solvent-based deposition of a fluoroalkyl silane [24], and others have demonstrated optically transparent parts by printing clear resins on a glass surface to reduce light scattering [25].

As a case study for fabrication of a SlipChip by 3D resin printing, we considered a microfluidic movable port device (MP device) previously developed by our lab for local stimulation of ex vivo organ slices at user-selected locations [9]. The MP device is a SlipChip that is comprised of two multilayer components: a bottom component containing a simple enclosed microchannel that terminates in a single, vertical delivery port (delivery component), and a top component featuring a semipermeable tissue culture well (chamber component) (Figure 1a). A bolus of aqueous solution is pumped into a specific region of a tissue slice by aligning the delivery port to a port in the culture well (Figure 1b). Local delivery devices like this one have been used to study intrinsic tissue properties and to screen for potential drugs [9,26,27,28,29,30]. Compared to a device with stationary ports, the SlipChip functionality of the MP device lessens the amount of user handling of a tissue slice and allows more flexible on-demand selection of the delivery region. However, in the original hand-built prototype, an extensive fabrication process limited the accessibility and distribution of the MP device to other labs and collaborators [9].

Here, we established an approach to fabricate a 3D printed SlipChip for the first time, using the MP device as a case study. First, we validated the selection of a DLP resin designed for microfluidic devices to meet the optical transparency, surface roughness, surface chemistry, and biocompatibility requirements of the tissue-specific movable port device. Next, the device design was optimized to maximize the functionality of the required ports and channels while minimizing the fabrication time complexity with DLP printing. The ability of the assembled device to deliver aqueous solutions without leaks into the gap was tested, and finally, we tested the ability to stimulate live organ cultures locally and with the position selected on demand.

2. Materials and Methods

2.1. Device Design, 3D Printing, and Laser Etching

All 3D printed parts were designed using Autodesk Inventor 2018 (Mill Valley, CA, USA). The CAD files (in Supporting Information) were sliced at 50 µm intervals using MII Utility Shortcut V 3.27 and printed using a CADworks3D M50-405 printer (30 µm xy-resolution, CADworks3D, Toronto, ON, Canada) in BV-007A resin (MiiCraft, via CADworks 3D). The printer setting for the BV-007A resin at a 50 µm slice height was a slow peeling speed, 0.1 mm gap adjustment (unless printing on glass which was 0.27 mm), 1.15 s curing time, 1 base layer, 9.0 s base curing time, 1 buffer layer, and 75% light power. To print parts on glass, a cover glass slide, 36 mm × 60 mm with a thickness of 0.13–0.17 mm (Ted Pella, Redding, CA, USA), was attached to the baseplate by curing a thin layer of BV-007A with a 405 nm UV light (Amazon, Seattle, WA, USA) [25]. The parts were rinsed with methanol (Fisher Chemical, Waltham, MA, USA) and post-cured in a UV light box for 20 s. No additional leaching steps were applied to the printed pieces used in this work. In preliminary experiments, we found that solvent washes at varied temperatures or extended UV light exposure did not substantially improve the biocompatibility of the BV-007A resin. To complete the chamber component, an array of ports with an 80 μm diameter were laser etched (Versa Laser 3.5, Universal Laser Systems, Scottsdale, AZ, USA) into the printed BV-007A part, using a power setting of 7% and a speed of 10%.

2.2. Fluorination of Resin Surface and Contact Angle Measurements

Parts printed in BV-007A were silanized using our recently described method [24]. The parts were submerged into a 10% v/v solution of tridecafluoro-1,1,2,2-tetrahydrooctyl trichlorosilane (Gelest Inc., Morrisville PA, USA) in Fluorinert FC-40 (Sigma Aldrich, St. Louis, MO, USA) for 30 min at room temperature. The surfaces were rinsed with 95% ethanol (Koptec) and DI water and finally dried with a nitrogen gun.

Surface air–water contact angles and three-phase contact angles were measured on cubic printed pieces (8 × 8 × 15 mm3) using a ramé-hart goniometer (model 200-00, ramé-hart instrument co., Succasunna NJ, USA) and DROPimage Advanced software (ramé-hart instrument co., Succasunna, NJ, USA). For consistency, the smooth, flat face of the cube produced against the polytetrafluoroethylene (PTFE) sheet was tested in all cases; this was also the side of the print that faced the oil layer in the SlipChip. The contact angle was measured in triplicate (3 separate printed pieces per condition), by pipetting one 5 µL droplet of 1× phosphate buffered saline (PBS) (Lonza, Walkersville, MD, USA.; DPBS without calcium or magnesium) onto the printed surface. For three-phase contact angle, the printed cube with a droplet was inverted into a cuvette filled with FC-40 oil containing 0.5 mg/mL triethyleneglycol mono[1H,1H-perfluorooctyl]ether (RfOEG). RfOEG was synthesized in house as reported previously (see Supporting Methods) [9,31,32].

2.3. Surface Profilometry

To assess surface roughness, the root mean square deviation of the surface height of the printed parts was measured with a Zygo optical surface profilometer (Zygo, Berwyn, PA, USA) at the Nanoscale Materials Characterization Facility at the University of Virginia. Cubes of 8 × 8 × 8 mm3 were printed, and surface roughness was measured on all sides, specifically the surfaces printed against the aluminum baseplate or printed against glass, closest to the PTFE sheet at the bottom of the vat, and the sides of the print. As a positive control, a glass microscope slide was also analyzed after plating with 30 nm of Au/Pd by a Technics sputter coater (Technics).

2.4. Measurement of Curvature of Printed Pieces

Images of the side profiles of 3D printed 30 × 30 mm2 prisms of varied height (2–5 mm) were collected using a Zeiss AxioZoom microscope (Jena, Germany). The displacement from horizontal due to curvature was manually measured in Zen 2 software (Zeiss, Jena, Germany).

2.5. Animal Work and Tissue Slice Collection

All animal work was approved by the Institutional Animal Care and Use Committee at the University of Virginia under protocol #4042, and was conducted in compliance with guidelines of the Office of Laboratory Animal Welfare at the National Institutes of Health (United States). Both male and female C57BL/6 mice aged 19–21 weeks (Jackson Laboratory, Bar Harbor, ME, USA) were housed in a vivarium and given water and food ad libitum. Lymph nodes were harvested from the mice following humane isoflurane anesthesia and cervical dislocation. The tissues were sliced according to a previously published protocol [33]. Briefly, peripheral lymph nodes were collected and embedded in 6% w/v low melting point agarose (Lonza, Walkersville, MD, USA) in 1× PBS. After the agarose had hardened, agarose blocks containing lymph nodes were extracted with a 10 mm tissue punch (World Precision Instruments, Sarasota, FL, USA). The blocks were mounted with super glue on a stage and sliced into 300 μm thick sections using a Leica VT1000S vibratome (Bannockburn, IL, USA) in ice-cold 1× PBS. The lymph nodes were sliced at a speed setting of 90 (0.17 mm/s) and frequency of 3 (30 Hz). Slices were cultured in “complete RPMI”: RPMI 1640 (Lonza, 16-167F) supplemented with 10% FBS (VWR, Seradigm USDA approved, 6 89510-186), 1× L-glutamine (Gibco Life Technologies, 25030-081, Waltham, MA, USA), 50 U/mL Pen/Strep (Gibco), 50 μM beta-mercaptoethanol (Gibco, 21985-023), 1 mM sodium pyruvate (Hyclone, GE USA), 1× non-essential amino acids (Hyclone, SH30598.01), and 20 mM HEPES (VWR, 97064–362). Slices of 6% agarose were collected in a similar manner but were stored in 1×PBS instead of complete media.

2.6. Analysis of Tissue Viability

To assess the viability of lymphoid tissue slices after a brief exposure to BV-007A, tissue slices were incubated in 1× PBS in a 3D printed culture well (30 mm × 30 mm × 5 mm printed part, with a central 10 mm-diameter well) for 15 min at room temperature. Then, the slices were moved from the printed substrate into a 24-well plate (VWR) and cultured in “complete media” for 4 h at 37 °C with 5% CO2 to allow time for any delayed effects of on-chip exposure, such as toxicity mediated by protein transcription or translation, to occur. Following a previously established protocol [33,34], the viability of live lymph node tissue slices was assessed by flow cytometry. Briefly, individual slices were crushed to generate cell suspensions. Cells were stained with 75 μL of 67 nM Calcein AM (eBioscience, San Diego, CA, USA) in 1× PBS for 20 min at 37 °C. Stained samples were washed by centrifugation at 400 g for 5 min and resuspended in 1× PBS + 2% FBS (flow buffer). 7-AAD (AAT Bioquest, Sunnyvale, CA, USA, 5 μg/mL final concentration) was then added to the cell suspension. The samples were run on a Guava easyCyte 4-color cytometer (EMD Millipore, 6-2L, Burlington, MA, USA) and analyzed using Guava® InCyte™ Software (EMD Millipore, Burlington, MA, USA). Single stain compensation controls were run on cells from crushed lymph node slices. The Calcein-AM single stain contained a 1:1 mixture of Calcein-labelled and unstained live cells. The 7-AAD single stain contained a 1:1 mixture of live and killed cells; the latter were prepared by treating cells with 35% ethanol for 10 min. Calcein positive and 7-AAD negative cells were defined as viable cells.

2.7. Assembly and Local Delivery with the 3D Printed Slipchip

Prior to assembling the SlipChip, the channel in the delivery component was filled using pressure-driven flow via a Chemyx syringe pump (Fusion 200, Houston, TX, USA). A 0.5 mg/mL solution of FITC-conjugated dextran (150 kDa and 70 kDa for agarose and tissue deliveries experiments, respectively) was flowed into the channel using a 50 μL Hamilton syringe (model 1705 RN; 26 s gauge, large hub needle) and non-shrinkable PTFE TT-30 tubing (0.012” I.D., 0.009” wall thickness, Weico Wire, Edgewood, NY, USA). Next, 500 µL of FC-40 oil containing 0.5 mg/mL RfOEG was pipetted onto the top face of the filled delivery component. The chamber component was lowered onto the delivery component, and the two components were clamped together with two binder clips, sandwiching a thin layer of oil between them. The culture chamber on the top of the chip was then filled with 1× PBS. A sample of agarose gel or tissue was placed into the chamber and weighed down using a small stainless-steel washer (10 mm O.D. and 5.3 mm I.D., Grainger, Lake Forest, IL, USA). The chamber component was manually slipped relative to the delivery component and visually aligned under a microscope to align to a desired port. To initiate a delivery, the syringe pump was turned on at the desired flow rate. After 5 s, the pump was turned off and the device was slipped away, to reposition for another delivery or to a reach a closed position. After all deliveries were complete, the sample was removed, and the chamber was flushed with 1× PBS and refilled for the next sample. All delivery experiments were performed at room temperature.

All deliveries were monitored in real time using a Zeiss AxioZoom upright microscope with a PlanNeoFluor Z 1×/0.25 FWD 56 mm objective, Axiocam 506 mono camera and HXP 200 C metal halide lamp (Zeiss, Jena, Germany), using filter cubes for GFP (Zeiss filter set #38), and Violet Chroma Filter (49021, ET-EBFP2). Images (16 bit) were collected before, during, and after delivery. During deliveries, time lapse images were collected at 1 s intervals. All images were analyzed in Zen 2 software (Zeiss, Jena, Germany).

2.8. Analysis of Delivery Widths

After alignment of the delivery port to an array port, a 5 s pulse of fluorescein (FITC)-labeled 150 kDa dextran was delivered to a 6% agarose slice at flow rates ranging from 0.2 to 1 μL min−1 (n = 3). After delivery, the device was slipped prior to imaging, to avoid the fluorescent signal from the underlying channel. Delivery width was determined from image analysis as previously described [26]. Briefly, line scans were drawn radially across the delivery region, and the background autofluorescence of the resin was subtracted. The data were fit to a Gaussian curve in GraphPad Prism version 8 (San Diego, CA, USA). The width was defined as two standard deviations of the Gaussian curve.

To fit the curve of the spread of analyte with respect to time, we used a previously published analytical model [9]. First, we assumed that the volume delivered per unit time was described by a cylinder:

where w [μm] is the width (diameter) of the delivery, h [μm] is the height of the slice, Q [μL/min] is the volumetric flow rate set by the pump, and Δt [sec] is the length of time of delivery. Solving for width gives Equation (2):

2.9. Delivery to Lymph Node Tissue

The device was assembled and a lymph node slice was placed into the chamber. A 5 s pulse of FITC-labeled 70 kDa dextran was delivered at a flow rate of 0.25 μL min−1. After the first delivery, the device was repositioned and another delivery was performed. This was repeated for four different slices and with slight variations in the number of deliveries on three separate occasions.

Results

3.1. Design Goals for a 3D Printed SlipChip with Movable Ports

The movable port device consisted of two components: a chamber to hold a tissue slice with a porous support in the form of a port array, and a delivery component with an enclosed channel with a small terminating port (Figure 1a). To assemble the device, the microchannel in the delivery component was filled with aqueous solution, and the chamber component was lowered on top while carefully sandwiching a layer of immiscible oil in between. To operate the device, the delivery port was aligned with a port in the array above, and pressure driven flow is used to deliver a short pulse of fluid into the tissue in the chamber (Figure 1b).

Before a MP device could be rapidly fabricated by DLP printing, there were two major challenges to be addressed (Figure 1c). The first challenge, applicable to any 3D printed SlipChip, was to have a small gap height between the printed parts to prevent leaking into the oil layer during delivery. To achieve a small gap height, the two surfaces closest to the oil gap must be both smooth and flat across the width of the component (30 mm) (Figure 1c). Flatness can be challenging because photocurable resins shrink when crosslinked, inducing mechanical stress that warps the print if not addressed in the print design [35]. Furthermore, an array of microscale ports and enclosed microchannel had to be integrated without disrupting the flat surface [36]. The second challenge, specific for biological applications, was biocompatibility of the printed resin with tissue slices housed in the delivery component (Figure 1c). The question of resin toxicity is of great interest to the microfluidics community and is still under active investigation [25,37,38].

3.2. Selection of Materials and Print Conditions for Transparency, Smoothness, Fluorination, and Cytocompatibility

Before designing the microfluidic device, we first selected and validated a resin for its suitability for the intended use in the SlipChip. We chose to use BV-007A resin because of its ability to generate microfluidic devices with high feature resolution [39,40]. First, we addressed the surface roughness and optical transparency of the DLP printed parts. While the polymeric surface would never be as smooth as glass, prior SlipChips have included microposts to set a defined gap height, e.g., of 2 µm, between two glass components [5]. Therefore, we hypothesized that surface roughness ≤ 2 µm would provide an acceptably small gap height. Surface roughness was expected to differ across the various faces of a printed piece, e.g., the bottom that is printed against the baseplate or against glass, the sides of the print, and the top that prints in contact with the PTFE sheet lining the vat (Figure S1a). As expected, optical profilometry showed that the surface printed against the rough aluminum baseplate and the sides of the printed piece were rough, with RMS (root mean square of surface height) > 3 µm (Figure 2a). The polymeric faces printed against glass and PTFE were much smoother, with RMS ≤ 0.3 µm (Figure 2a). For reference, glass itself had a surface roughness of 5 ± 0.5 nm (n = 3). From these data, we concluded that the print for a SlipChip must be oriented such that the surfaces intended to contact the oil gap were printed against glass or the PTFE sheet. Additionally, we also tested for optical transparency, which was desirable for visual alignment of channels and ports in the SlipChip. As previously described [25], printing against glass provided optical transparency, whereas printing against the rough aluminum baseplate yielded an opaque sample (Figure 2b).

Materials

Clear Microfluidics Resin V7.0a

Learn More

H SeriesM

To prevent spreading of aqueous solution between the components in the oil gap, the surface chemistry of the chip must be fluorophilic and hydrophobic where it contacts the oil phase. While the BV-007A resin yields parts that are moderately hydrophilic, we recently described a method for fluoroalkyl silanization for SLA resins [24]. Here, this method was applied to silanize the BV-007A, by placing the surface to be silanized in a solution of 10% fluoroalkyl silane in FC-40 oil (see Methods). We confirmed that silanization not only increased the three-phase contact angle of a 1× PBS droplet resting on the surface in air, but also when immersed in FC-40 oil (Figure 2c). The water/oil/resin contact angle of >115° indicated a highly hydrophobic surface [9].

Finally, as the MP device was intended to be used with live organ slices, we sought to identify conditions in which tissue viability was not affected by the BV-007A printed pieces. Ex vivo slices of murine lymph node tissue were used in these experiments, as we have previously characterized local delivery to such tissues [9,26]. During use of the movable port device, the tissue slice is in contact with the surface of the 3D printed chamber for only a few minutes, typically <5 min for alignment and <10 s for the delivery. In separate work, we have shown that multi-hour physical contact of murine splenocytes with parts printed in BV-007A was cytotoxic after just 4 h [41]. Therefore, we restricted this study to an exposure period of 15 min, which represents triple the expected exposure time for tissue spent in contact with the resin during use of the device. Tissue viability after 15 min exposure was comparable to that of off-chip controls (Figure 2d and Figure S2). As this viability test was based on membrane integrity and esterase activity, further tests for cell and tissue function may be appropriate depending on the intended application of the chip. We and others continue to work to identify a resin or a post-print treatment strategy that provides biocompatibility with primary tissues for longer time periods, while still maintaining with the high print resolution of BV-007A [25,37,38,42].

3.3. Optimizing the Design and Printability of the Delivery Component

Having identified the material and conditions for SlipChip function, we turned to designing the components of the movable port device. The delivery component required three key features to be printed while maintaining a smooth, flat surface: an interior channel, a delivery port, and an inlet (Figure 3a). Additionally, alignment markers (small inset wells on the top of the component) were included in the design to aid in visual alignment of the device when delivering to opaque tissues. Although it is common practice to print at angle to achieve higher resolution for interior channels (Figure S1b, angled) [43,44], the requirement for smoothness dictated that the design be printed horizontal relative to the baseplate, such that the gap-facing surface was printed against PTFE (Figure S1b, flat). In addition, the requirement for a flat profile to minimize gap height meant contending with shrinkage and associated deformation during photocrosslinking [45]. In addition to rounding the sharp corners, we found that increasing the thickness (z) of the printed part was required to minimize mechanical stress during printing for a part with a 30 × 30 mm2 footprint [45,46]. The thickness of the print was varied from 2 to 5 mm, and the delivery component required at least 5 mm thickness to prevent the part from curling (Figure 3b,c).

Having optimized the print geometry and overall dimensions of the piece, the design of the enclosed channels and ports were then optimized to minimize the channel cross-section while retaining printability (Figure 3d). We share these details to aid other researchers who are also working at the limits of the resolution of a 3D printer. To reduce blockage during printing, the channel was positioned close to the top of the part to minimize UV-exposure from subsequent layers, which is a particular issue for transparent resins. Additionally, the length of the channel was minimized (15 mm), because longer channels were more difficult to clear of uncrosslinked resin through the small terminating delivery port. To minimize reagent volume during use, we minimized the cross-sectional area of the channel. In a test piece printed with a series of 15 mm channels of varied cross-sectional size and a square or diamond shape, the minimum cross-section that remained open was 0.5 × 0.5 mm2 in both shapes (Figure 3d and Figure S3). Thus, a 0.5 × 0.5 mm2 cross-section was selected, and the square was selected over the diamond shape in order to minimize the horizontal width the channel during optical imaging of the device. The diameter of the delivery port was optimized in the same test piece, with a series of ports of varied diameter atop each channel (Figure 3d, inset). All ports with diameter ranging from 0.15 to 0.35 mm were successfully printed, with close fidelity (<10% error) to drawn dimension (Figure 3e). The smallest printable port was 0.138 ± 0.009 mm (drawn diameter 0.15 mm); ports drawn smaller failed to print (data not shown). Finally, we designed a simple press-fit female port to ensure a snug fit with the microfluidic tubing at the inlet (0.78 mm OD PTFE tubing), by printing mock inlets of 0.76–0.87 mm drawn diameter (Figure 3f). A 0.80 mm drawn diameter port was determined to give a snug fit with the tubing. In summary, the optimized design for the inlet, enclosed channel, and terminal delivery port (0.8 mm inlet, 0.5 × 0.5 mm2 square channel cross-section, 0.15 mm drawn delivery port) yielded a 3D printed delivery component that could be reproducibly printed and was sufficiently flat and smooth (Figure 3g).

3.4. Optimization to Minimize Port Size and Preserve Optical Transparency of the Chamber Component

The top component of the MP device included a chamber (12 mm diameter) to hold tissue samples in media, with a permeable support at the bottom for delivery of fluid from below. Based on our prior work, the ports in the chamber component needed to be in the range of 0.070–0.110 mm, i.e., large enough to minimize flow resistance and small enough to create a localized delivery [9]. The support needed to be transparent for visual alignment, and the requirement for smoothness meant that the bottom of the chamber component needed to be printed against glass or the Teflon vat.

We originally tested a one-step fabrication method for this component, by embedding a membrane or mesh support into the part during 3D printing or by directly printing the port array (Figure 4a,b). We found it simple to embed a nylon or metal mesh in the component by adhering it to the baseplate or glass prior to printing (Figure 4a and Figure S4). Unfortunately, due to resin shrinkage during polymerization, the mesh did not remain taut, preventing its use in the SlipChip. Next, we attempted to directly print the small ports in an array (Figure 3b), but it was challenging to meet the requirements for both small port size and transparency. Orienting the port array against glass on the baseplate proved unfeasible due to the required overexposure of the first layers of the printing, which lowered the spatial resolution in these layers. On the other hand, orienting the port array as an overhang generated ports with an acceptable diameter (~110 µm), but the unsupported overhang led to stretching and distortion, which reduced transparency (Figure 4b).

Since fabricating the port array in a single step proved challenging, we elected to use a two-step process (Figure 4c). First, the chamber component was 3D printed with the solid bottom of the chamber well (200 µm thick) oriented against the glass. Second, a CO2 laser was used to etch a port array into the bottom of the chamber. The laser-etched ports had a diameter of 0.081 ± 0.002 mm (n = 74), well within the acceptable range, and the entire array was etched in < 1min. Additionally, unlike the accumulation of melted plastic observed when laser etching acrylic [9], there was no deformation of the BV-007A polymer during laser etching on either side of the chamber components (Figure S5), thus minimizing gap height in the SlipChip. The component was sufficiently transparent for visual alignment. Thus, this straightforward fabrication strategy produced a flat, smooth, monolithic part with a well-defined port array, ready for integration into the final SlipChip (Figure 4d,e).

3.5. The Assembled 3D Printed SlipChip Delivers Fluid without Leakage into the Gap

Having fabricated both components, we assembled the 3D printed SlipChip (Figure 5a,b) and tested its ability to perform local deliveries with leakage of aqueous solution into the oil-filled gap, a critical design goal. To test that the aqueous solution did not leak into the oil gap during use, the delivery port was aligned with a port in the array, and a short pulse of fluorescent dextran solution was delivered to an agarose slice through each of three different array ports (Figure 5c). During and after each delivery, fluorescent and brightfield imaging were used to visually inspect the gap area for the appearance of an interface between aqueous and oil phases, which would indicate a leak. No such interface was observed in 3 separate chip assemblies (9 out of 9 deliveries), indicating a robust capillary-pressure-mediated barrier to leakage. This robust interfacial barrier was also not affected by the size difference between the delivery port (0.138 ± 0.009 mm) and the ports in the chamber array (0.081 ± 0.002 mm).

4. Discussion and Conclusions

This paper describes the requirements and fabrication strategy to achieve 3D printed SlipChips for the first time, and demonstrates 3D printing of a SlipChip device with a movable port for local stimulation of organ cultures as a case study. After optimization, DLP 3D printing produced smooth (RMS ≤ 0.3 µm), flat surfaces that were chemically modifiable by fluorination. The resin parts were biocompatible with the short-term (<15 min) exposures needed for local delivery to tissue slices. The delivery component was designed to be printed in a single step and contained an inlet, enclosed channel and small (0.138 ± 0.009 mm) terminating delivery port. The chamber component was produced via 3D printing followed by laser etching, which provided a monolithic culture chamber with an array of 0.081 ± 0.002 mm diameter ports at the bottom, while maintaining a smooth, flat interface for slipping. The device was able to perform multiple slipping and delivery steps without leakage in between the components. The spread of the delivery was dependent on the rate of pump-driven fluid flow, and the resolution was sufficient to target substructures in multiple locations in a tissue slice.

We anticipate that this DLP 3D printing fabrication method will enable the polymeric SlipChips, and in particular the movable port technology, to become accessible to other labs, by greatly simplifying the fabrication steps, materials, and time. Continued advances in biocompatibility of DLP resins [47,48,49] may eventually enable longer-term culture on the 3D printed chip, which was a limitation here. Furthermore, while the current device used binder clips, visual alignment, and manual slipping, 3D printing may enable rapid iteration in the future of other clamping methods and pre-programmed integration with manipulators. Thus, rapid prototyping with DLP 3D printing is expected to accelerate advances in movable port technology as well as other SlipChip device designs.

Selective fluorination of the surface of polymeric materials after stereolithography 3D printing

by

Selective fluorination of the surface of polymeric materials after stereolithography 3D printing

Megan A. Catterton, Alyssa N. Montalbine, and Rebecca R. Pompano

With the microfluidics community embracing 3D resin printing as a rapid fabrication method, controlling surface chemistry has emerged as a new challenge. Fluorination of 3D printed surfaces is highly desirable in many applications due to chemical inertness, low friction coefficients, anti-fouling properties and the potential for selective hydrophobic patterning. Despite sporadic reports, silanization methods have not been optimized for covalent bonding with polymeric resins. As a case study, we tested the silanization of a commercially available (meth)acrylate-based resin (BV-007A) with a fluoroalkyl trichlorosilane. Interestingly, plasma oxidation was unnecessary for silanization of this resin, and indeed was ineffective. Solvent-based deposition in a fluorinated oil (FC-40) generated significantly higher contact angles than deposition in ethanol or gas-phase deposition, yielding hydrophobic surfaces with contact angle > 110° under optimized conditions. Attenuated Total Reflectance-Fourier Transform Infrared (ATR-FTIR) spectroscopy indicated that the increase in contact angle correlated with consumption of a carbonyl moiety, suggesting covalent bonding of the silane without plasma oxidation. Consistent with a covalent bond, the silanization was resistant to mechanical damage and hydrolysis in methanol, and was stable over long-term storage. When tested on a suite of photocrosslinkable resins, this silanization protocol generated highly hydrophobic surfaces (contact angle > 110°) on three resins and moderate hydrophobicity (90 – 100°) on the remainder. Selective patterning of hydrophobic regions in an open 3D-printed microchannel was possible in combination with simple masking techniques. Thus, this facile fluorination strategy is expected to be applicable for resin-printed materials in a variety of contexts including micropatterning and multiphase microfluidics.

Keywords: Two-phase microfluidics, Droplet microfluidics, low surface energy, Digital Light processing (DLP), stereolithography printing (SLA)

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

Introduction

The microfluidics community has increasingly adopted 3D printing for device fabrication, including with fused deposition modeling1–3 and with resin-based methods such as stereolithography (SLA) and digital light processing (DLP) printing.4,5 As a result, methods to control the surface chemistry of 3D printed devices are emerging as a critical challenge, especially for microscale features produced by resin printing.6 In resin printing, UV/visible light is used to cross-link a photocurable, polymeric resin in a layer-by-layer fashion to produce a 3D structure.4,5 While methods for surface functionalization are well established for traditional materials such as glass and polydimethylsiloxane (PDMS), those methods do not necessarily translate directly to the polymeric materials used for 3D printing. A particular challenge is to generate a fluorinated surface on a 3D printed chip. Fluorinated surfaces offer many advantages for microfluidic device design, such as controlled surface wettability for passive fluidic control, chemical inertness, resistance to surface fouling, and low friction coefficient.7–10 These properties historically made fluorinated surfaces invaluable for multiphase microfluidic chips.11–16 By patterning fluorination amidst a non-fluorinated surface, patterned hydrophobicity has been used to generate droplets, create microarrays, and control microfluidic valving.17–19 Therefore, facile methods to selectively fluorinate the surface of polymeric SLA and DLP resins are required, particularly for the commercially available resins used by most laboratories.

Currently, there are few methods available to generate a fluorinated surface on 3D printed material, particularly a patterned surface. One option is to start directly with a fluorinated resin,20 but these are rare in practice due to limited commercial options. Additionally, fully fluorinated devices are not readily patterned at the surface due to their chemical inertness. Alternatively, selective surface patterning is possible by using printed pieces modified at the surface with fluorinated coatings.6,21,22 Polymeric liquid coatings provide a robust hydrophobic layer up to hundreds of micrometers thick,21 but may be inappropriate for microscale features that are easily blocked or filled in. A chemical vapor deposition method can be used to generate a thin, highly hydrophobic coating by polymerizing a fluorinated acrylate film on the surface, but has limited use in enclosed channels.23,24 Thin coatings can also be achieved by including a polymerization initiator in the resin, to provide covalent anchor points for fluorinated polymer brushes.22 However, polymer brushes may exhibit poor mechanical stability during abrasion.6

Silanization using fluorinated silanes is a reliable method for molecular-scale surface modification of glass and polydimethylsiloxane (PDMS),25,26 but silanization of polymeric materials can be challenging. Historically, polymers have been chemically modified primarily by strategies such as wet etching, plasma or corona treatment, or coatings, rather than direct silanization.27–31 Extensive surface oxidation is usually required to generate enough silane-reactive functional groups (e.g. hydroxyls) at the polymer surface, but not all polymers can withstand such treatment, as they may degrade after plasma exposure.7,27,28,32,33 So far there have been sporadic reports of silanization of resin 3D printed microfluidic devices, e.g. to fluorinate 3D printed molds for PDMS34 and to attach reactive functionalities for bonding of 3D printed pieces.35 In some cases, the printed polymer had to be coated with a layer of silica to enable silanization.36,37 To date, there has been little testing of the conditions required for direct fluoroalkyl silanization of resin printed pieces, nor characterization of the hydrophobicity and stability of the silanized surface.

Here, we aimed to develop a robust and straightforward silanization protocol using (tridecafluoro-1,1,2,2-tetrahydrooctyl) trichlorosilane, a fluoroalkyl silane, and a suite of commercially available SLA and DLP resins to generate a highly fluorinated surface for use in microfluidic devices. While optimizing the reaction conditions to generate the highest possible contact angle, we found, surprisingly, that surface oxidation using air plasma was unnecessary for silanization. To characterize the surface and investigate reactive groups involved in forming a covalent bond between the printed resin and the fluoroalkyl silane, we measured the air/water contact angle of the silanized surface and used infrared (IR) spectroscopy. We tested the ability of the method to selectively pattern hydrophobic regions in a 3D printed open microchannel, and further tested the applicability of the optimized method to four additional resins. The method is facile, versatile, and allows for dynamic patterning of a hydrophobic surface on a resin-printed piece.

Materials

A Bottle of the Master Mold for PDMS Device Resin

Master Mold Resin

Learn More

Clear Microfluidics Resin V7.0a

Learn More

M50

Experimental Section

3D Printing
Printed parts were designed using Autodesk Inventor 2018. The CAD files were sliced at 50 μM intervals using MII Utility Shortcut V 3.27 and printed using a CADworks3D M50–405 printer (MiiCraft, CADworks3D). The commercial resins included were BV-007A (Clear) (MiiCraft, CADworks 3D), Green Master Mold (MiiCraft, CADworks 3D), Dental LT Clear Resin (V2) (FormLabs), and Asiga PlasClear V2 (iMakr). A house-made photoresin consisting of 0.4 % w/v phenylbis(2,4,6-trimethylbenzoyl)phosphineoxide (Irgacure 819) (Therofisher) dissolved in poly(ethylene glycol) diacrylate (PEG-DA) (MW 250) (Sigma Aldrich) was also included in the suite of resins tested.38 The printer setting for each resin can be found in Table S1. Printed parts were rinsed with either 95% ethanol (Koptec), isopropanol (Fisher chemical), or methanol (Fisher chemical) as recommended for by the manufacturer for the resin. Printed pieces were post-cured in an UV-light box, then stored at room temperature on the bench top in polystyrene petri dishes (Fisher) prior to silanization.

Surface Treatment of 3D Printed Pieces
Where noted, some printed parts were plasma treated using a BD-20AC laboratory corona treater (Electro-Technic Products, Chicago IL, USA). Printed parts were placed 3 mm below the plasma source and treated for 5 – 60 s immediately prior to surface silanization. For gas-phase deposition, 200 μL of neat tridecafluoro-1,1,2,2-tetrahydrooctyl trichlorosilane (Gelest Inc., Morrisville PA, USA) was placed in a vacuum desiccator in a small polypropylene dish, followed immediately by the printed parts, and a vacuum was applied for 2 hours at room temperature. For solvent deposition, the surface of the printed part was submerged in a 10% v/v solution of tridecafluoro-1,1,2,2-tetrahydrooctyl trichlorosilane in solvent (Fluorinert FC-40 (Sigma Aldrich) or 200 proof ethanol (Koptec) for 30 min at room temperature, unless otherwise specified. After silanization, surfaces were rinsed with 95% ethanol and DI water and dried with a nitrogen gun.

Contact Angle Measurement
Surface air/water contact angles were measured using a ramé-hart goniometer (model 200–00, ramé-hart instrument co., Succasunna NJ, USA) and DROPimage Advanced software. Contact angle was measured for 3 separate printed pieces per condition, by pipetting one 5-μL droplet of DI water per print onto the silanized surface. 8×8×8 mm3 cubes were used for the printed piece, and oriented so the smooth flat face of the printed cube was tested.

Surface Chemistry Characterization with Infrared Spectroscopy
The surface chemistry of the printed parts was examined by using an iD7 ATR Nicolet IS5 FT-IR spectrophotometer (Thermo Fischer Scientific). The IR spectrum was measured on the flat smooth face of a 10×10×2 mm3 printed rectangular prism. The instrument was set to a constant gain of 4, and the background was collected prior to each session. Data was collected, visualized, and processed using the OMNIC software (Thermo Fischer Scientific).

Robustness testing
Printed pieces were silanized according to the optimized method. To test the resistance to mechanical damage, the parts were clamped with two binder clips against a clean petri dish to apply constant pressure and rubbed together for 30 s at a time. Air/water contact angles of the silanized surfaces were measured before and after the mechanical test. To test stability after storage, silanized printed parts were stored in a petri dish at room temperature under ambient light, and the air/water contact angles were repeatedly measured over time. Finally, contact angles were measured before and after soaking the printed parts for 2 hours in methanol.

Selective Patterning of 3D Printed Surfaces
Rectangular prisms (20×15×3 mm3) were printed using BV-007A resin. Each print contained an embossed cross-shaped open channel with a rectangular cross-section (1 mm deep, 2 mm wide). Scotch tape (3M) was cut and aligned manually to prevent the fluoroalkyl silane solution from coming into contact with portions of the printed surface inside the channel. Taped pieces were immersed in a solution of 10% v/v (tridecafluoro-1,1,2,2-tetrahydrooctyl) trichlorosilane in FC-40 for 30 min in a fume hood at room temperature. After treatment, pieces were rinsed with 95% ethanol and DI water and dried with nitrogen. To test the functionality of the patterned surface, solutions of food coloring in water were pipetted into the arms of the embossed features.

3D printed Droplet Generator
A simple T-junction was designed in AutoCAD, consisting of a 10 mm channel with a 0.5 × 0.5 mm cross-section, with a 3 mm channel length with a 0.5 × 0.5 mm cross-section channel that intersects the longer channel. The enclosed channel was fluorinated by filling the channel with a solution of 10% v/v (tridecafluoro-1,1,2,2-tetrahydrooctyl) trichlorosilane in FC-40 for 30 min, in a fume hood at room temperature. A syringe (1 mL, BD) with a 27 G needle (BD) was filled with FC-40 oil containing 0.5 mg/mL RfOEG (triethyleneglycol mono[1H,1H-perfluorooctyl]ether, a surfactant synthesized in house).29 Another syringe was filled with 1 M Fe(SCN)2+(aq) in water. Connections to the device were made with nonshrinkable PTFE TT-30 tubing (Weico Wire, Edgewood NY, USA). Pressure driven flow was achieved using a Chemyx syringe pump (Fusion 200, Houston TX, USA), using flow rates of 30 μL/min for the oil and 10 μL/min for the aqueous solution. Brightfield images were collected using an Zeiss AxioZoom macroscope (Carl Zeiss Microscopy, Germany) at 1.6 magnification with an Axiocam 506 Mono camera. Images were collected at 1 s intervals for 10 s. All images were analyzed in Zen 2 software.

Data Analysis
Statistical tests and curve fitting were performed using Graphpad Prism version 9. Half-lives and half-times of exponential fits were calculated according to half time = ln 2/k, where k is the rate constant from the fit.

Results and Discussion

Plasma oxidation was not necessary or effective for silanization of SLA printed pieces
While the precise composition of most commercial resins is proprietary, MSDS information states that many are based on acrylate and/or methacrylate polymers (Figure 1a). Silanization of related polymeric materials such as poly(methyl methacrylate) (PMMA) requires oxidation to generate hydroxyl groups that undergo condensation reactions with the silane reagent.7,14 Similarly, prior reports of silanization of an acrylate-based 3D printed material included activation of the surface with plasma treatment.34,35 Therefore, we first tested the efficacy of silanization of 3D printed pieces as a function of the duration of exposure to air plasma. As a case study, we selected a clear (meth)acrylate-based resin formulated specifically for printing microfluidic devices, BV-007A resin from MiiCraft, and sought to silanize it with (tridecafluoro-1,1,2,2-tetrahydrooctyl) trichlorosilane (Figure 1a). Two common methods of silanization were tested: gas-phase deposition29,34,39–41 and liquid-phase deposition.14,29 For the latter, we used a 10% v/v solution of silane in FC-40 fluorinated oil.t.

Figure 1: Effects of plasma treatment and silanization on the chemistry and hydrophobicity of DLP printed pieces. (a) Chemical structures of the fluoroalkyl silane and monomer acylate and methacrylate base used for many resin formulations. (b) Air/water contact angles of BV-007A after silanization by solution-phase (blue squares, FC-40 solvent) or gas-phase (pink dots) deposition after varied times of treatment with air plasma (n=3, mean ± std dev). The black triangle represents printed BV-007A pieces that received neither plasma treatment nor any silane treatment. Two-way ANOVA for solution vs gas-phase silanization (**** p<0.0001). (c) ATR-FT IR spectrum of the BV-007A surface with no exposure to air plasma (pink) and after 30 s plasma treatment (grey), without silanization. Spectra are offset to display spectral features. (d) Air/water contact angles of BV-007A surface after solution-phase silanization in FC-40 (pink dot) or ethanol (blue square). Two-way ANOVA with Sidak’s multiple comparisons to compare between solvents (****p <0.0001, *** p<0.001).

Surprisingly, we found that even in the absence of plasma treatment (0 s exposure), silanization significantly increased the air/water contact angle for both methods (gas phase, p<0.005; solvent, p<0.001) compared to the 60° contact angle of the unslianized printed piece angle (Figure 1b, Figure S1). While gas-phase deposition provided a contact angle near 90°, the lower boundary for hydrophobicity, the solution-phase method provided a significantly larger (p < 0.0001) contact angle close to 120°, the upper limit for a flat, fluorinated surface.26,42,43 Plasma treatment from 5 to 60 s did not further increase the contact angle. Wanting to further test the impact of plasma cleaning on the surface chemistry, we next used IR spectroscopy to investigate functional groups on the surface of printed BV-007A pieces.

We expected that sufficient exposure of BV-007A pieces to air plasma would oxidize the surface to form alcohol and/or carboxylic acid groups.44 To characterize the surface chemistry and investigate the extent of surface activation at short plasma treatment times, we collected surface ATR FI-IR spectra of the printed pieces (Figure 1c). As expected for (meth)acrylate-based BV-007A, the spectra closely resembled that of a commercial sheet of PMMA (Figure S2). The peaks at 2970, 2930, and 2870 cm−1 were assigned to alkane sp3 C-H stretching. A major C=O stretch peak at 1718 cm−1 was attributed to the carbonyl in the backbone of the (meth)acrylate-based polymer as well as other carbonyl-containing components of the resin, e.g. photoinitiators and photoabsorbers. The C-O-C stretching was assigned to the peaks ranging from 1000 – 1300 cm−1 in the fingerprint region.45 Treating BV-007A printed pieces with air plasma for 30 – 60 s did not alter the IR spectra substantially (Figure 1c and data not shown). In particular, no characteristically broad alcohol band (3550 – 3200 cm−1) was observed, and there was no change in the alkyl CH stretches or carbonyl peak. These data were consistent with plasma treatment neither affecting the contact angle of the material (Table S2) nor improving its silanization (Figure 1b). As a positive control, oxidation from the plasma treatment was verified using both glass and PDMS, whose contact angle decreased after 5 s of plasma treatment as expected (Table S2). Prior reports of plasma treatment of PMMA used longer treatment times (5 min and greater) to modulate the surface polarity,46,47 but we found that treatment of BV-007A pieces with air plasma for longer than 2 min generated cracks in the surface. Since plasma treatment was unnecessary for silanization and in fact was ineffective at oxidizing the BV-007A surface at short times, we proceeded to optimize and characterize the silanization of BV-007A pieces it its absence.

Solvent deposition was most effective when a fluorocarbon oil was used as a solvent.

Having established that solution-phase deposition was more effective than gas-phase deposition, we further optimized the choice of solvent and concentration of silane. Two solvents were tested: ethanol (200 proof), a common solvent for deposition of trichlorosilanes,29,30 and FC-40, a fluorinated oil.14 Whereas deposition from ethanol solution was largely ineffective (contact angles < 90°) regardless of silane concentration, deposition from FC-40 solution had a concentration-dependent effect, yielding an average contact angle of ~ 120° at 10 % v/v silane (Figure 1d). Exposure to FC-40 alone, without silane, did not increase the contact angle significantly (Figure S1). Therefore, 10% v/v of the fluorinated silane in FC-40 was used for all further experiments.

Time dependence of the reaction provides support for covalent bond formation

Next, we tested the time dependence of the silanization reaction. The contact angle increased in a time-dependent manner with a half-time of 3.4 min, reaching a plateau after 15 min that was unchanged for the rest of the testing period, up to 60 min (Figure 2a). To complement the contact angle data and assess the extent of bond formation between the fluoroalkyl silane and BV-007A, ATR-FT IR spectra were collected from these samples (Figure 2b). The spectra changed noticeably over this time period. In particular, the carbonyl stretch at 1716 cm−1 decreased in intensity over time (Figure 2b – c), and the peak area was well fit by exponential decay equation with a half-life of 3.5 min (Figure 2d). This observation suggested a molecular reaction between the resin and the fluoroalkyl silane that consumes a carbonyl. The data do not distinguish between the methacrylate carbonyl and any carbonyls that may be present in the resin’s photoinitiators or photoabsorbers, but these additives typically are a minor constituent of the resin. An immediate increase in fingerprint region intensity was consistent with the addition of fluoroalkyl silane to the surface of the print (Figure 2b, ,cc and ande).e). New peaks included those at 1023 cm−1, assigned to Si-O-R stretching,48 1232 and 1142 cm−1, consistent with asymmetric and symmetric C-F stretches, and 707 cm−1, assigned to the C-F wag.49 This increase had a half-time of only 2.0 min, shorter than the decay of the carbonyl, suggesting that physical adsorption of the silane may have preceded the covalent reaction (Figure S3). The -C-H stretch peaks at 2872, 2932, and 2971 cm−1 were still present after silanization (Figure 2b – c).50

Figure 2: Time dependence of the chemical reaction. (a) Contact angle of the fluorinated surface after various amount of silane treatment (n=3, mean ± std dev). The data were fit to an exponential curve, y = 111 – 46.9e−0.201x, R2 =0.844. Insets show images of droplets on BV-007A surface after 0 and 30 min of silanization. (b) ATR-FT IR spectrum of printed BV-007A pieces after various times of silane treatment. Two regions of interest are highlighted: the carbonyl peak at 1720 cm−1 and the finger print regions 650–1300 cm−1. (c) The chemical structures present in a methyl methacrylate-based resin and from the fluoroalkyl are labeled with the corresponding IR spectra peak. (d) The area under the carbonyl peak decreased in a time-dependent manner (n=3, mean ± std dev), fit to an exponential decay y = 55.9e−0.115x +43.3, R2 = 0.936. (e) The area under the curve of the finger print region increased in a time-dependent manner, fit to an exponential curve, y = 57.7 – 24.2e−0.348x, R2 =0.855.

From both the contact angle measurements and the IR spectra, we concluded that the silanization reaction likely resulted in a covalent bond, and that 30 min was sufficient for reaction completion and generation of a highly hydrophobic surface. We note that the mechanism for such a reaction does not match that of typical silanizations, which occur through a condensation reaction with hydroxyl groups on the surface of the material. In this case, there were no detectable hydroxyl groups, yet surface modification still occurred. We were unable to find a precedent for the reaction of a tri-substituted silane with (meth)acrylate; the closest reaction we found in literature was that of silyl radicals attacking alkenes and acrylates,51 but we would not expect formation of a silyl radical in this system.

Robustness and stability of fluorination procedure

To establish the practical utility of the method, we considered the sensitivity of the procedure to the state of the printed piece and characterized the stability of the hydrophobic surface. First, we considered that the surface chemistry of the printed piece may change over time and potentially alter the reactivity with the trichlorosilane, e.g. due to slow cross-linking of residual monomer under ambient light.52,53 To test the efficacy of silanization as a function of light-induced aging, printed parts were treated with either the manufacturer-recommended 20 s or an extended 360-s UV exposure during the post-curing process. We estimate that continuous 360-s exposure was an equivalent dose of light as being on a bench top under ambient light for 32 days (Table S3). The extended UV cure created discoloration and warped some of the pieces, so only pieces with a flat top surface were used for subsequent silanization. No significant difference was observed in the water contact angles of the control pieces (20 s) compared to the pieces with extended UV exposure (360 s), either before or after silanization (Figure 3a). This result was consistent with our informal observations that month-old BV-007A pieces yielded similar contact angles after silanization as recently printed (1–3 days old) pieces. Therefore, the silanization method appears insensitive to the age of the piece, at least in this timescale, which enables robust fabrication procedures.

Figure 3: Robustness of the method to the age of the printed piece, abrasion, and storage time after silanization. (a) Contact angle of DLP printed pieces (BV-007A) that were silanized with or without extended UV curing (n=3 printed parts for each condition, mean ± std dev). Two-way ANOVA with Tukey’s multiple comparisons (ns, p>0.05, ** p<0.005). (b,c) Contact angle of silanized BV007 after (b) deliberate mechanical abrasion under constant pressure or (c) long term storage. n=3 printed parts for each condition, mean ± std dev. Some error bars too small to see. One-way ANOVA (ns, p>0.05).

Next, we assessed the robustness of the silanized surface when subjected to mechanical damage and extended storage, a property that affects the range of potential uses, handling, and storage. Microfluidic chips must be able to withstand mild abrasion during the movement of the device, and in particular we anticipated using this method to generate fluorinated SlipChips, which rely on sliding parts past one another.11 Therefore, silanized printed pieces were subjected to gentle mechanical damage by manually rubbing the piece against a clean polystyrene surface, mimicking normal wear and tear during use. The water contact angle of the fluorinated pieces of BV-007A was not significant altered by this process (Figure 3b), indicating that the surface is stable under mild abrasion conditions. Similarly, when silanized pieces of BV-007A were stored on the bench, the contact angles remained unchanged for at least 154 days, the longest time point measured (Figure 3c). We did observe that the initial contact angle in these experiments was slightly lower than in previous experiments, which we attribute to hydrolysis of the trichlorosilane during storage because replacement of the silane stock improved the hydrophobicity (data not shown). We concluded that the silanized surface was quite stable and the method was robust to the age of the resin though sensitive to the quality of the silane stock, all of which are consistent with the formation of a covalent bond during the silanization reaction.

Patterning of surface hydrophobicity on 3D printed parts

Compared to printing with a fully fluorinated resin, site-specific patterning is an advantage of post-print modifications, offering the potential for passive fluidic control. Therefore, we tested the ability of the silanization protocol to selectively pattern hydrophobic patches on the surface of BV-007A resin, using a pair of intersecting open channels in a simple, recessed cross design. The arms of the cross were protected from the silanization using adhesive tape, while the center square was silanized to generate a pattern of four separate fluid compartments, separated by a surface tension barrier. In the non-silanized control, colored solutions pipetted into the arms of channel mixed readily in the center of the cross (Figure 4a, Not Treated), whereas a micropatterned hydrophobic patch in the center of the cross successfully constrained the solutions to the arms (Figure 4a, Pattern). These data demonstrate that because the silanization method requires contact of the liquid silanization solution with the printed surface, it is easily patterned by physical masking strategies to define the silanized area.

Figure 4: Application of the optimized silanization procedure for surface patterning and droplet generation. (a) Selective surface patterning of an open channel. Parts printed in BV-007A were patterned so that the center of the cross was hydrophobic. In a non-silanized piece (top), the blue and yellow food dyes mixed in the center; in the piece patterned by local silanization (bottom), the droplets remained distinct from each other. The width of the channels in these photos was 2 mm. (b) Fluorination of 3D printed fluidic channels in a T-junction droplet generator. (Top row) Photos of empty 3D printed chip at low and high magnification, with inlets and outlet marked. (Bottom row) Images of two-phase fluid flow with and without silanization of the interior channels. The dark liquid is the aqueous solution; the fluorinated oil is colorless. Droplets formed only in the silanized system. Scale bar 1 mm.

Silanization of an enclosed channel for droplet formation

In addition to surface features, many microfluidic devices feature enclosed channels whose surface chemistry must be controlled, e.g. to present a fluorophilic interior surface for droplet microfluidics.54 We reasoned that the liquid-based silanization described here to could be used to fluorinate enclosed channels by simply filling the channel with silanization solution for 30 min and rinsing afterwards; no prior surface oxidization is needed. To test this prediction, a simple T-junction droplet generator was printed (Figure 4b) and used to create water-in-oil droplets using an aqueous solution of Fe(SCN)32+ and fluorinated oil. As expected, when the 3D printed device was not silanized, the aqueous solution wetted the channel and droplets did not form (Figure 4b, Not Treated), whereas fluorosilanization by this protocol prevented wetting and enabled the formation of droplets (Figure 4b, Silanized).

Silanization of a suite of SLA resins demonstrates broad applicability

This silanization protocol would be most useful if applicable across a variety of SLA and DLP resins. Therefore, in addition to BV-007A, we tested three commercially available resins: Dental (FormsLab), Green Master Mold resin (CADworks3D), and Plasclear (iMakr), plus a polyethylene glycol diacrylate (PEG-DA)-based resin developed by the Folch laboratory.38 The resins chosen come from 3 different companies and span a breadth of applications (PDMS mold, dentistry, small part/figurine prints) and properties (low viscosity/high resolution, biocompatible, heat stable) of interest to microfluidic device fabricators. Based on our prior data that the extent of reaction correlated with diminished absorbance from the carbonyl in the IR spectrum, we hypothesized that any resin with an acrylate (BV007A and PEG-DA) or methacrylate (Dental, Green Master Mold, and Plasclear) backbone, or possibly with carbonyl-containing photoinitiators or photoabsorbers, would react with the fluoroalkyl silane. Following the optimized protocol, all printed pieces were submerged in a 10% (v/v) solution of fluorinated silane in FC-40 oil for 30 min, without plasma treatment. This procedure successfully increased the contact angle for each material compared to its non-treated control (Figure 5a). The Green Master Mold, Dental, and BV007 resins were highly hydrophobic after silanization, with contact angles of ~ 115–118°. In contrast, the PEG-DA and Plasclear resins had a mildly hydrophobic contact angle, near 100°. This trend was reproduced in two independent experiments.

Figure 5: Testing the optimized silanization method with a variety of SLA/DLP resins. (a) Air/water contact angles for resins prior to silanization, after silanization with the optimized procedure, and after soaking in methanol (n=3 printed parts for each condition, mean ± std dev). Two-way ANOVA with Tukey’s multiple comparisons tests (ns, p>0.05, ** p<0.005). (b) ATR FT-IR spectra of pieces printed using the various resins, before and after silanization (n=3 printed parts for each condition). Spectra are offset to display spectral features.

To test the extent to which the high contact angles may have been due to physically adsorbed silane, the silanized pieces were soaked for 2 hours in methanol.55 First, the efficacy of methanol to remove a significant fraction of physically adsorbed silane was confirmed on a piece of acrylic (Figure S4). Next, silanized 3D printed parts were tested; in all cases, the hydrophobic surface persisted, again suggesting that the silane was covalently bound (Figure 5a). Surprisingly, the contact angle increased for both the Plasclear and PEGDA resins, an observation that remains to be explored. While here we tested stability to methanol treatment, we recommend that the stability of the fluorinated surface be tested under the experimental conditions relevant to the intended use of the printed piece, e.g. under varied pH or solvents.

We examined the surface chemistry of the printed pieces by ATR FT-IR to potentially explain the difference in susceptibility to silanization between resins (Figure 5b). The pair of peaks at 1453 and 1510 cm−1 are useful to distinguish PMMA from poly(methyl acrylate) (PMA).56 The 1453 cm−1 peak, which was present in all samples, is attributable to a methylene vibration -CH2- found in both PMMA and PMA. The peak at 1510 cm−1, attributable to the methyl vibration C-CH3, is indicative of PMMA. This peak was present in all four commercial resins tested, suggesting the presence of methacrylates in these materials; as expected, it was not seen in the PEGDA sample. A small peak at 1630 cm−1, assigned to C=C bonds from residual (meth)acrylate monomers,57 was present in all samples.

Next, changes in the surface IR spectra after silanization were examined. The three resins that exhibited a larger change in contact angle after silanization (BV-007A, Green Master Mold, and Dental) also showed a larger decrease in the intensity of the carbonyl peak at 1719 cm−1 (Figure 4c). Furthermore, the decrease in the carbonyl peak correlated with appearance of peaks consistent with deposition of the fluoroalkyl silane. The peaks at 1232, 1142 and 707 cm−1 were again assigned to stretches and wagging of the fluoroalkyl chain,23,49 and they increased after silanization for the four commercial resins. Similarly, the peak at 1010 cm−1 that increased after silanization in the commercial resins may be a part of the Si-O-R stretch (usually a strong and broad stretch, 1000–1100 cm−1).58 In contrast, Plasclear and PEGDA, which had smaller changes in contact angle, showed less consumption of the carbonyl, and PEGDA showed no increase in the finger print region. From these data, we concluded that while all five resins showed an increase in contact angle that was resistant to removal by methanol, only a fraction of them formed a covalent bond that consumed a carbonyl. It may be significant that PEGDA, which has no added photoabsorbers, was the least reactive of the materials towards the silane; this possibility was not tested further here. Go to:

Conclusions

We have demonstrated a robust and versatile strategy to control the surface chemistry and hydrophobicity of DLP 3D printed parts by reacting the printed surface with an alkyl-fluorinated silane. The optimized method consisted of simply placing a DLP-printed part directly into a solution of 10% v/v (tridecafluoro-1,1,2,2-tetrahydrooctyl) trichlorosilane in FC-40 for 30 min, then thoroughly rinsing the part with ethanol and water and drying under nitrogen. This method did not require any pre-treatment of the printed piece. The reaction between the silane and the resin appeared to consume a carbonyl present in the resin material, and was consistent with covalent bond formation by an unknown mechanism. This method created a hydrophobic surface with air/water contact angles close to 120 deg. Additional work would be needed if superhydrophobicity (> 150°) was required, e.g. by adding multi-scale surface roughness to the printed surface prior to silanization.59 The fluorinated surface was resistant to mechanical damage, methanol soaking, and 154 days of storage, and the method was compatible with printed parts even after significant light exposure. Selective patterning of a hydrophobic surface was demonstrated in 3D printed open channels by a simple masking method. Furthermore, the method was effective with a suite of (meth)acylate based resins, with higher contact angles correlating with greater consumption of the carbonyl. We anticipate that simple approach to controlling the surface chemistry of resin 3D printed microfluidic parts, including for selective fluorination of specific regions, will advance the fabrications of complex two-phase devices and enable greater control of the wettability of 3D printed parts.

Design and characterization of a 3D-printed staggered herringbone mixer

by

Design and characterization of a 3D-printed staggered herringbone mixer

Vedika J Shenoy , Chelsea ER Edwards , Matthew E Helgeson  & Megan T Valentine

3D printing holds potential as a faster, cheaper alternative compared with traditional photolithography for the fabrication of microfluidic devices by replica molding. However, the influence of printing resolution and quality on device design and performance has yet to receive detailed study. Here, we investigate the use of 3D-printed molds to create staggered herringbone mixers (SHMs) with feature sizes ranging from ∼100 to 500 μm. We provide guidelines for printer calibration to ensure accurate printing at these length scales and quantify the impacts of print variability on SHM performance. We show that SHMs produced by 3D printing generate well-mixed output streams across devices with variable heights and defects, demonstrating that 3D printing is suitable and advantageous for low-cost, high-throughput SHM manufacturing.

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

Method Summary

We investigate the use of 3D printing to create staggered herringbone mixers (SHMs) and show that such devices generate well-mixed output streams across devices with variable heights and defects. This demonstrates that 3D printing is suitable and advantageous for low-cost, low-effort, high-throughput micromixer manufacturing.

Keywords: 3D printer,calibration 3D printing, microfluidics,micromixers, staggered herringbone mixer

Microfluidic mixers such as the staggered herringbone mixer (SHM) [1] promote eddy-like mixing of laminar flow streams, avoiding prohibitively long channel lengths and enabling applications in drug delivery and discovery [2], chemical synthesis [3], sample concentration [4] and biological analysis [5–7]. In SHMs, asymmetric herringbone grooves embedded in the floor or ceiling of the rectangular channel cause transverse flow within fluid streams, promoting mixing by increasing local vorticity (Figure 1) [1,8]. Alternating the grooves' offset between cycles increases flow irregularity, further mixing the two streams [9]. The SHM's mixing proficiency and mechanism has been characterized extensively [8,10] due to its efficiency and ease of design compared with other grooved micromixers [11]. Their low shear flow properties [6] and ability to circulate flow within the channel [4] make SHMs particularly advantageous for biomedical applications,

h/H ∼ 0.36, w/d ∼ 0.45, and a ∼ 95°. See Supplementary Materials for additional information.

Despite these advantages, the widespread use of SHM devices for on-chip diagnostics [4,6,12] is hindered by the reliance on time-intensive photolithography-based fabrication to generate molds for polydimethylsiloxane (PDMS)-based microfluidic devices. Photolithography requires cleanroom training and costly equipment and reagents [13]. Consequently, 3D printing is emerging as an attractive substitute for photolithography due to its comparative affordability, simplicity [14], and the ease of fabricating multilevel designs [13]. 3D printing also enables rapid prototyping by reducing typical fabrication times from days to hours [14].

To establish the use of 3D printing to generate molds for PDMS-based microfluidic devices such as the SHM micromixer, experimental validation of printer and device performance is required. For current 3D printers [15], the minimum feature size is significantly larger than that provided by photolithography, print-to-print variability is expected and random disfigurations can occur, all potentially influencing device performance. Moreover, there are often discrepancies between the targeted and actual dimensions of the 3D printed parts [16]. Unfortunately, not only is information concerning such limitations generally unavailable for any given 3D printer model, but the lack of reporting on chip-to-chip variability is a known barrier for large-scale manufacturing of microfluidic devices [17]. Thus, understanding how print quality influences pattern transfer and mixing performance is especially important for low-cost, low-effort manufacturing that maintains efficacy, as required for microfluidic diagnostic platforms.

To experimentally investigate print quality, we printed a series of raised channels, with heights and widths ranging from ∼25 to 700 μm, using a Miicraft Ultra 50 3D digital light processing (DLP) printer, with a horizontal and vertical resolution of ∼30 μm and ∼5 μm, respectively, and using Resin Works 3D Master Mold Resin for PDMS. We sized the 3D-printed parts by imaging with a Keyence VHX-5000 microscope (Supplementary Figure 1), and found them to be consistently smaller than the programmed design dimensions, with greater discrepancies in height than in the planar dimensions. These differences were consistent across all printed molds, so the programmed design dimensions could be calibrated to achieve printed parts with the correct size (Supplementary Figure 2).

This calibration critically informed mold design, which was performed in Solid Works. Features with design dimensions ≥120 μm (∼100 μm actual dimension) were reproducibly printed, whereas features with smaller design dimensions were often deformed or occasionally missing. We then printed a series of herringbone grooves with a fixed width w of 100 μm and herringbone spacing d that varied from 50 to 500 μm (Figure 1). Printed grooves with designed d < 300 μm (∼220 μm real spacing) frequently fused with the adjacent herringbones, with increasing severity for decreasing d (Supplementary Figure 1A & B). We note that these disfigurations existed within the molds themselves, not only the PDMS replicates. Thus, we established a minimum feature size of 100 μm and fixed d = 300 μm in the design of SHM molds for further study.

These constraints required SHM features in 3D-printed molds to differ somewhat from the geometries recommended by prior work [18], which find mixing to be most effective near w/d ∼1 and w = 50 μm. By contrast, our mixer has w/d ∼ 0.45 and w = 100 μm. However, we achieved a ratio of herringbone height h to channel height H, h/H ∼ 0.36, and herringbone angle a ∼ 95°, the latter of which was informed by computational work that suggested that angles in the range of ∼90°–105° produce optimal transverse flow [1,19]. Ten herringbones per each half cycle were included in accordance with prior experimental studies [10]. To compensate for the larger feature sizes and spacing requirements of 3D-printed molds and prevent the microchannel length from being excessively long, we limited the number of mixing cycle elements to five on a single chip (total channel length ∼41 mm) (Supplementary Figure 3). We then investigated a range of flow rates to produce well-mixed streams with these larger dimensions.

To demonstrate the mixing capabilities of PDMS-based micromixer devices generated using the 3D-printed SHM molds, we examined the mixing performance of two exemplar devices generated with two identically designed, but individually printed, molds. In detail, PDMS, obtained as commercially available Sylgard 184 (Dow Chemical, MI, USA), was mixed at a 10:1 ratio of PDMS to cross-linker, poured in the treated molds, degassed for approximately 2 h, then cured at 60–80°C for 4 h and removed from the mold. We did not observe qualitative deviations in the PDMS replicates formed from either mold, consistent with the well-established high-fidelity pattern transfer from 3D-printed molds to PDMS [20]. Each device was tested at flow rates from 0.5 to 20 μL min-1, corresponding to Peclet numbers Pe = 200–8000, a useful range for low shear-stress mixing of biological materials [21]. The Peclet number describes the ratio of advective to diffusive transport and is given by Pe = uL/D where u is the average velocity defined by u = q/WH and q is the volumetric flow rate, L is a characteristic length scale given here by W/2 and D is the Stokes–Einstein diffusivity.

For each test, a 0.30 μM aqueous solution of 70 kDa fluorescein isothiocyanate (FITC)-dextran was delivered into Input 1, while neat water was delivered into Input 2 at an equivalent flow rate. The mixing of the two streams under steady-state flow conditions was observed via fluorescence microscopy (see Supplementary Materials for details). Representative intensity maps of the flow fields along a cross-section of the channel after one, three and five mixing cycles are shown in Figure 2. For quantitative analysis, the intensity was normalized by dividing by the maximum value (Supplementary Figure 4), and its coefficient of variation (CV), defined as the standard deviation divided by the mean, was calculated (Figure 3 & Supplementary Figure 5). We consider any profile with a CV < 0.1 to be well mixed [10].

The color scale indicates the presence of fluorescein isothiocyanate (FITC)-dextran, introduced through Inlet 1 and mixed with neat water introduced through Inlet 2. The dark blue regions indicate low fluorescence signal (i.e., mostly water), whereas red regions indicate high FITC signal; uniform light blue/green regions indicate homogeneous mixing. At Pe = 200 streams appeared homogeneous and well-mixed after three cycles, while at Pe = 8000 striations remained even after five cycles.

Devices generated using 3D-printed SHM molds demonstrate good mixing performance. The number of cycles required to achieve good mixing depends on Pe, as expected. At Pe = 200, where diffusion is prominent, only three mixing cycles are required to achieve CV < 0.1, and only four to five mixing cycles are needed at Pe values of 800–6000 (Figure 3). We could not achieve good mixing for Pe > 6000 with this design due to the limited channel length. We observed a modest decline in mixer performance compared with that of devices with smaller feature sizes achieved via photolithography, where it is possible to achieve CV < 0.1 with two mixing cycles at Pe = 625 and with five cycles at Pe = 6250 [10].

In both cases well mixed streams with CV < 0.1 black dashed line) are achieved for Pe < 6000. Error bars were calculated via propagation analysis as described in the Mixing Analysis section in the Supplemental Methods.

Finally, we examined the effects of print-to-print variation on mixing by comparing the results obtained using two devices containing chipped herringbones, with the second device containing twice as many defects as the first, and modest differences in herringbone properties (Supplementary Tables 1–4). Device 2 exhibits larger CV values across almost all Pe values after one and two cycles of mixing; however, these differences vanish after three to five cycles of mixing (Supplementary Figure 5). That print-to-print variability does not compromise performance suggests that rigorous measurements of each mold are not necessary beyond the initial characterization of 3D printer capabilities and 3D printing can deliver SHM devices that yield reproducible results over multiple prints, even in the presence of defects.

In summary, we introduced methods to characterize and calibrate 3D printer outputs and adapt the SHM geometry to enable 3D printing of replica molds. We demonstrated good mixing performance despite the modified dimensions and in the presence of print-to-print variation and defects. This method provides significant advantage for applications that benefit from rapid, low-effort manufacturing.

Apparatus

Master Mold for PDMS Resin

Master Mold for Resin

Learn More

M50

Learn More

3D Printing for Microfluidics: PDMS Devices

by
3D Printing has several microfluidic applications, a key example being PDMS Device fabrication. 3D Print Master Molds for PDMS with design features that are as small as 50µm

3D Printing for Microfluidics: PDMS Devices

< 5 minute read

Polydimethylsiloxane (PDMS) is the ideal material for soft lithography devices because it has high optical transparency, it is permeable to oxygen and is biocompatibility, which makes it suitable for biological research.   

Master molds are essential for scientific research in the field of PDMS devices. The CADworks3D printer for Microfluidics  with the Master Mold Resin for PDMS Devices is the ideal tool needed to make molds easier, faster and cost effective. Confidently create a variety of iterations of designs and concepts without disclosing IP, not meeting project timelines or budget shortfalls. 

There are several key benefits when producing everything in-house. First of all, your intellectual property is kept within the walls of your company. In addition, once a mold is created in-house, you can make any changes immediately. While outsourcing any projects, there will be a much longer waiting period. For instance, once you receive your mold, you might be required to make additional changes and once again outsource a new file. 

The Master Mold Resin for PDMS Applications is extremely versatile and can print raised channels as fine as 50μm (results based on the M50 printer). The printed mold can be used to cast up to a 2,000 PDMS devices (Statistics provided by the Kelley Lab from University of Toronto).

The Master Mold Resin for PDMS devices has been formulated specifically for microfluidics PDMS devices. The unique features of this material is its high print resolution and being able to cast PDMS directly onto the master mold without the need of a release agent.

The Cadworks3D line of 3D printer and Master Mold Resin for PDMS application are ideal for soft lithography research. The Cadworks3D 3D printer specifications allow it to print fine details required for Microfluidics. At CADworks3D we further calibrate the machine further to meet these requirements. However, a 3D printer is only one part of the equation. The other half would be the resin: 

Master Mold Resin goes hand in hand with the 3D Printer and allows for details as fine as 50μm (model dependent). Meanwhile, other resins on the market cannot be used for fabrications. The Master Mold Resin has been specifically formulated for PDMS applications.

3D Printing of Inertial Microfluidic Devices

by

3D Printing of Inertial Microfuidic Devices

Sajad Razavi Bazaz, Omid Rouhi, MohammadAmin Raouf, Fatemeh Ejeian, MohsenAsadnia, Dayong Jin and Majid Ebrahimi Warkiani

Inertial microfluidics has been broadly investigated, resulting in the development of various applications, mainly for particle or cell separation. Lateral migrations of these particles within a microchannel strictly depend on the channel design and its cross-section. Nonetheless, the fabrication of these microchannels is a continuous challenging issue for the microfluidic community, where the most studied channel cross-sections are limited to only rectangular and more recently trapezoidal microchannels. As a result, a huge amount of potential remains intact for other geometries with cross-sections difficult to fabricate with standard microfabrication techniques. In this study, by leveraging on benefits of additive manufacturing, we have proposed a new method for the fabrication of inertial microfluidic devices. In our proposed workflow, parts are first printed via a high-resolution DLP/SLA 3D printer and then bonded to a transparent PMMA sheet using a double-coated pressure-sensitive adhesive tape. Using this method, we have fabricated and tested a plethora of existing inertial microfluidic devices, whether in a single or multiplexed manner, such as straight, spiral, serpentine, curvilinear, and contraction-expansion arrays. Our characterizations using both particles and cells revealed that the produced chips could withstand a pressure up to 150 psi with minimum interference of the tape to the total functionality of the device and viability of cells. As a showcase of the versatility of our method, we have proposed a new spiral microchannel with right-angled triangular cross-section which is technically impossible to fabricate using the standard lithography. We are of the opinion that the method proposed in this study will open the door for more complex geometries with the bespoke passive internal flow. Furthermore, the proposed fabrication workflow can be adopted at the production level, enabling large-scale manufacturing of inertial microfluidic devices.

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

Introduction

Inertial microfuidics has been broadly investigated, resulting in the development of various applications, mainly for particle or cell separation. Lateral migrations of these particles within a microchannel strictly depend on the channel design and its cross-section. Nonetheless, the fabrication of these microchannels is a continuous challenging issue for the microfuidic community, where the most studied channel cross-sections are limited to only rectangular and more recently trapezoidal microchannels. As a result, a huge amount of potential remains intact for other geometries with crosssections difcult to fabricate with standard microfabrication techniques. In this study, by leveraging on benefts of additive manufacturing, we have proposed a new method for the fabrication of inertial microfuidic devices. In our proposed workfow, parts are frst printed via a high-resolution DLP/SLA 3D printer and then bonded to a transparent PMMA sheet using a double-coated pressure-sensitive adhesive tape. Using this method, we have fabricated and tested a plethora of existing inertial microfuidic devices, whether in a single or multiplexed manner, such as straight, spiral, serpentine, curvilinear, and contraction-expansion arrays. Our characterizations using both particles and cells revealed that the produced chips could withstand a pressure up to 150psi with minimum interference of the tape to the total functionality of the device and viability of cells. As a showcase of the versatility of our method, we have proposed a new spiral microchannel with right-angled triangular cross-section which is technically impossible to fabricate using the standard lithography. We are of the opinion that the method proposed in this study will open the door for more complex geometries with the bespoke passive internal fow. Furthermore, the proposed fabrication workfow can be adopted at the production level, enabling large-scale manufacturing of inertial microfuidic devices.

Continuous separation of particles and cells is required for a wide variety of applications that include mineral processing, chemical syntheses, environmental assessments, and biological assays1 . A number of conventional methods exist for this purpose; however, they have several drawbacks. Membrane fltration-based techniques, while effici t and simple, are limited by flter fouling and clogging. Centrifugation methods are also plagued by problems of particle adhesion and clogging, along with their high cost and inability for continuous processing. Likewise, techniques based on sedimentation are prone to particle adhesion and slower processing time, which increases the non-viability of cells in biological applications. Also, methods based on magnetic-activated cell sorting (MACS) and fuorescence-activated cell sorting (FACS) are proven to be low throughput and expensive2–4 . With the evolution of microfabrication and rapid prototyping techniques, microfluidic technology has emerged as an alternative to improve upon conventional separation techniques5,6 . Tese microfuidic techniques are grounded on the unique characteristics of microscale fl w phenomena and have recently gained prominence as effici t tools for the control and focusing of microbeads. Amongst existing microfuidic systems, inertial microfuidics has experienced massive growth in many applications ranging from cell separation7,8 , cytometry9,10, multiplexed bio-assays11,12, and also fuid mixing13. Despite great advantages of inertial microfuidics, the commercial impact and scalability of this technology have been restricted due to fabrication issues. As a passive technique, inertial microfuidic systems manipulate cells and particles by taking the advantage of hydrodynamic forces in microchannels with a variety of cross-sections. To date, several microchannels (i.e.,straight, spiral, and serpentine) with diferent cross-sections (i.e., square, rectangular, triangular, trapezoidal, and circular) have been proposed to enhance particle sorting by optimizing the synergetic efects of inertial and Dean drag forces14–18. Tese devices are mainly fabricated by casting PDMS on a master mold, which is made by either standard microfabrication techniques (i.e., silicon etching or SU8 lithography) or using conventional micromilling on an aluminum or polymethylmethacrylate (PMMA) sheets19–21. While this approach has been the workhorse behind the development of majority of these devices, the inability to build non-orthogonal and non-planar structures, cost, and labor intensiveness of the process have hampered its widespread applications and commercialization22,23.

Besides the aforementioned approach, other groups attempted to develop alternative strategies for the fabrication of inertial microfuidic devices. For instance, several groups reported the usage of femtosecond laser irradiation and CO2 laser ablation techniques to produce straight and spiral shape microchannels inside a glass or PMMA24–28. Despite the simple fabrication process, the complexity of building non-rectangular cross-sections, poor surface fin sh, and lengthy etching steps are making them less user-friendly. Some groups also proposed the utilization of metal micro-wires or a sacrific al template in conjunction with soflithography to produce inertial microfuidic devices29–31. In spite of the simplicity of this method in the fabrication of circular channels, PDMS rupture or distortion and the presence of residuals in microchannels during the template removal restrict its utility.

Recently, the fabrication of PMMA microchannels using hot embossing technique has also been reported. While this method is attractive for rapid prototyping and high volume production of microfuidic systems with microscale features, the necessity of using sophisticated equipment limits its widespread usage32,33. Recently, additive manufacturing has emerged as a powerful platform to fabricate 3D functional microfuidic systems from a variety of polymeric materials34. Th s outstanding technology enables investigators to build microstructures with complex shapes and geometries in a short time35,36. Benefting from the stereolithography apparatus (SLA) technique37, Lee et al. directly fabricated a 3D helical trapezoidal microchannel to separate E. coli bacteria using magnetic nanoparticle clusters38. However, due to the poor transparency of the fabricated channel, imaging (whether fuorescent or bright fi ld) was not feasible through the channel. Besides, to remove residuals from channels, the channel width is in the order of millimeter-sized dimensions, which is not suitable for most of the inertial microfuidic applications where small cells or particles are of interest. More lately, 3D printing of sacrific al molds combined with soflithography has gained signifi ant attention due to its simplicity and cost-efectiveness39. Gaining the effici cy of the fused deposition modeling (FDM) printer, Tang and colleagues40 fabricated various microchannels with unconventional cross-sections to study the efect of geometry on elasto-inertial focusing. While this approach is suitable to fabricate microchannels with diferent cross-sections, the resolution of printed parts is not high enough due to inherent limitations of FDM printing. Although direct fabrication of microchannels using SLA and digital light processing (DLP) method is a suitable candidate, inertial microfuidic devices ofen operate in channels in the order of micrometer (e.g., rectangular with 200 µm width and 40µm height) where removing resin residuals from the channel is a challenging issue41.

To address these inadequacies, we have developed a robust protocol for large-scale manufacturing of inertial microfuidic systems. Tanks to the capabilities of DLP and SLA 3D printing42,43, we have printed a wide range of microchannels with diferent geometries, capable of performing particle and cell focusing for various Reynolds numbers (Re). Te approach makes the use of a double-coated pressure-sensitive adhesive tape that perfectly binds open 3D-printed microchannels with optically transparent acrylic sheets, producing a leakage-free interface for inertial microfuidic applications. Te bonding strength is quantifi d, and the compatibility of the concept for the fabrication of new generation of inertial microfuidic devices is evaluated using cells and particles.

Materials

Clear Microfluidics Resin V7.0a

Learn More

M50

Learn More

Results and discussion

Fabrication and characterization of 3D-printed channels.
Soflithography using PDMS and a master mold is a frequently used method for the fabrication of microfuidic systems. Th s strategy has several advantages; for instance, PDMS is biocompatible, optically transparent, and gas permeable, which makes it suitable for a myriad of biological applications44. Also, 3D printing of PDMS has been reported using stereolithography approach45. However, certain drawbacks such as lack of chemical stability, deformation under pressure, and adsorption of small hydrophobic molecules have hindered its industrial-scale utilization. Moreover, the manual molding, cleaning, and bonding process complicate the mass production.

Although theoretically straightforward, scaling up of PDMS-made microchannels for commercialization application in inertial fl ws is challenging since these devices are flex ble and prone to rupture or collapse at high fl w rates. In addition, infation and hysteresis feature of PDMS create a big question mark regarding the exact focusing position of particles. Tis becomes more serious in CFD modeling where the “fxed wall boundary condition” is not truly correct in inertial regimes within PDMS-made microchannels. It is not surprising that the results of numerical simulations must be validated with hard chips rather than soft (PDMS-made) microchannels. Apart from these issues, the inherent limitations in the standard microfabrication and soflithography techniquesmake researchers unable to explore particle migration in unconventional cross-sections (e.g., right-angled triangular or hexagonal). For instance, particle focusing within a triangular curved microchannel has never been explored due to the fabrication difculty. As such, there is a great need to develop standardized protocols for the fabrication of inertial microfuidic devices to facilitate ground-breaking research, while enabling quick translation into commercial products.

In this study, we have proposed a novel approach for the fabrication of inertial microfuidic devices based on the 3D printing method. Figure 1 demonstrates an overview of the fabrication process. Gaining the effici cy of a high-resolution 3D printer, the desired microchannel is printed while its face (where the design pattern exists) is outer, and the base is attached to the build plate (Fig. 1AI). Th s method is particularly signifi ant since the change of cross-sections in inertial microfuidics is of great interest. However, the printing parameters need to be optimized for the fabrication of a channel with proper and accurate dimensions. Te slice thickness

Figure 1. (A) Schematic illustration of the proposed workfl w for the fabrication of inertial microfuidic devices. I. Te desired channel geometry was printed by a high-resolution SLA/DLP 3D printer II. Afer cleaning the part by isopropanol, it was bonded to a PMMA sheet by means of a double-coated pressuresensitive adhesive tape. Te entire process just takes less than two hours. III. Beneftting from the PMMA transparency, high-speed, fuorescent, bright fi ld, or phase contrast microscopy can be performed from the bottom side of the channel (B) An actual complicated inertial microfuidic device containing a spiral and serpentine microchannel. (C) Fluorescent microscopy from the bottom side of the channel.

in Z direction, curing time of each layer, and total thickness of the part are the most critical factors to have a high-quality channel with a great surface fin sh. Various cross-sections, ranging from right-angled or isosceles triangular to hexagonal, were fabricated and the best optimized parameters were identifi d (Fig. S1). To complete the fuidic network, 3D-printed inertial microchannels need to be bonded to a substrate with enough optical transparency and rigidity for subsequent testing. In this work, a variety of scenarios has been evaluated, and upon extensive evaluations and characterizations, permanent bonding of 3D-printed channels to a PMMA sheet via a double-coated adhesive tape was selected as the most promising and reproducible method. A transparent double-coated pressure-sensitive adhesive tape (ARcare, Adhesive Research) having 25.4 µm clear polyester flm coated with AS-110 acrylic medical grade adhesive was cut with a similar size of PMMA sheet (Fig. 1AII). Afer the attachment of one side of the tape to the PMMA sheet, the 3D-printed inertial part was manually placed over the other side of the tape and pressed with a tweezer until no bubble was observed at the interface (Fig. S2).

An important feature of PDMS is its optical transparency, which makes it suitable for a broad range of microscopic applications. Given the fact that commercial DLP/SLA resins are not typically transparent, the attachment of 3D-printed microchannels to PMMA sheets provides enough transparency for the optical and fuorescent microscopy (Fig. 1AIII). What makes this approach attractive for a wide range of communities (e.g., biologists and chemists) is its user-friendliness for people without prior knowledge about microfabrication and soflithography. Te entire process from CAD drawing to printing and then testing takes less than 2 hours, portraying the versatility of this method for inertial microfuidic research. More importantly, devices made using this technique are not prone to the deformation and leakage compared to the PDMS-made devices, making them suitable to study new physics, especially at high Re. Furthermore, by considering the fabrication cost, time, and efforts of a complicated inertial microfuidic device, our suggested method is rapid and utilizes a low-cost raw material which are valuable features, especially in areas where resources are limited. Figure 1B,C depict a fi al device fabricated using this technique. Te internal channels are flled with red food color for the sake of illustration.

In order to investigate the bonding quality, a straight microchannel with dimensions of 50 µm height, 200 µm width, and 4 cm length was fabricated and tested accordingly. We have monitored the device performance for the appearance and growth of Safman-Taylor fi gers until it becomes stable, called “infation stability” (Fig. 2A). Te results are presented in a 2D diagram to identify the channel behavior at a given pressure, as shown in Fig. 2C. Our results revealed that the holding strength of double-coated adhesive tape was able to achieve a leak-proof interface between the 3D-printed part and PMMA sheet, not only at typical operating pressure reported in literature46, but also more than the capability of PDMS-made channels in withstanding high fl w rate conditions. Shear rate distribution across a line parallel to the channel width was also evaluated, and as Fig. 2B revealed, increasing the fl w rate leads to imposing more shear forces at the edges of the channels. Te more the fl w rate, the larger the appearance of Safman-Taylor fi gers (insets of Fig. 2B). Te green area in Fig. 2C shows the safe zone for performing inertial microfuidic experiments where no Safman-Taylor fi gers appear during the operation. We have found that at pressures more than 82.6 psi, Safman-Taylor fi - gers begin to appear; however, this does not impose any detrimental efect on the device performance (i.e., no leakage or bonding collapse). Also, we did not observe any delamination or deformation in channels afer consecutive runs at high pressures (i.e., 120 psi), all of which are common in PDMS-based inertial microfuidic devices (see Figs. S3 and S4 for the pressure drop, velocity profle inside the microchannels for a wide range of operating fl w rates).

The surface characteristic of the double-coated adhesive tape was also investigated using a proflometer. As Fig. 3 illustrates, the roughness of the tape is homogenous and is in the submicron range. Te values of Ra and Sa were about 250 and 240nm, respectively. Also, the roughness of the 3D printed parts was evaluated and value of Sa was less than 300 nm. Tese nanometric rugosities indicate that the roughness of tape does not have any efect on the fow profle and particle focusing. Although optically transparent, the optical characteristics of the PMMA sheet (2-mm-thick) and adhesive tape were evaluated to identify the possibility of accurate fuorescence imaging47. Hence, the UV-visible absorbance spectra for a wide range of wavelengths (i.e., from 200 to 1100nm) were recorded via a spectrophotometer (Cary 60 VU-Vis spectrophotometer, Agilent Technologies). Figure 3B,C reveal that the light loss is negligible for both PMMA and adhesive tape within the visible spectrum, resulting in no trace of autofuorescence residual.

Straight microchannel.
Straight microchannels with rectangular or square cross-sections are arguably the most widely used inertial microfuidic systems. Tanks to their ease of fabrication and the ability for parallelization, a myriad of applications have been developed using these platforms over the past decade48. Te required channel length for inertial particle migration to the equilibrium positions is L H = πμ /ρ α U f f m L 2 2 where fL is estimated in the range of 0.02 to 0.05 for (H/W) from 2 to 0.5, and the corresponding fl w rate for inertial migration is calculated as Q ≈ 2 / πμWH 3ρ αL fL 3 2 48. Channel Re (Re = ρUD/μ) and particle Reynolds number ( ) Re Re p H 2 = 2 α are two dimensionless numbers for the characterization of particle migration in a straight microchannel. When particle Re is much smaller than 1, viscous drag becomes dominant, and particles follow the streamline. Increasing particle Re augments inertial forces, causing inertial particle migration become obvious in the microchannel49,50.

Particle migration within a straight channel strictly depends on its cross-section. In square straight microchannels (with an aspect ratio (AR) (width/height) of 1), particles migrate to four equilibrium positions located at the center of each wall. Changing the cross-section to rectangular disturbs this focusing pattern where in a rectangular straight microchannel with AR of 0.5, focusing positions reduce to two near the center of long walls51. Th s behavior was explained by Zhou and Papautsky where they identifi d two-stage particle migration in rectangular straight microchannels21. Further increase in the AR results in the more unpredicted focusing behavior of particles. Generally, in channels with high AR, stable focusing positions are reduced. However, by exceeding Re from a critical value, the number of stable equilibrium positions increases which is a function of particle size, channel dimensions, and Re. Based on reported experimental results, = . κ κ ≤ ≤ ≤ ≤ − . Re A 697( R A / ) (4 5 / R R 60, 5 e 660) c 0 79 was identifed52. Te abovementioned results elucidate that particle focusing is strongly afected by channel cross-section. However, due to the fabrication limitations, dependency of various cross-sections to channel geometries was not systematically investigated. Recently, triangular and semi-circular cross-sections were fabricated using Si anisotropic etching with potassium hydroxide53, a brass for mold fabrication54, FDM for creation of sacrific al mold40, or unconventional micromilling14. However, critical fabrication limitations do not allow for further investigation on the dependency of triangular angle or type (e.g., right-angled triangular) on focusing patterns of the particles. Here, as a showcase, a straight microchannel with rectangular cross-section and AR of 4 (all channel dimensions are provided in Section S3 and Tables S1–5) was fabricated, and the results are illustrated in Fig. 4A. As the results indicate, for low Re (Fig. 4AI), 20µm particles focus at the center of long walls of the channel cross-sections, shown previously in PDMS-made microchannels. Nonetheless, the focusing pattern for particles at higher Re does not obey a specifc role. As clearly can be seen, increasing fl w rates leads to generation of additional focusing positions within the microchannels where side walls are also added to the equilibrium positions of particles (Fig. 4AII–IV). Furthermore, lateral migration of MDA-MB-231 and DU-145 cells at low fl w rates (10~20ml/hr) (Fig. 4BI–III) illustrates their single-line focusing within the rectangular straight microchannel, which is promising for fl w cytometry applications. Moreover, to showcase the versatility of the proposed method, a triangular straight microchannel was fabricated and the results are shown in Fig. 4D. Te results are completely in line with those reported in the literature where 10µm particles and cells occupy one lateral position in the channel for low fow rates (Fig. 4DI,II). For high

Figure 2. (A) Analyzing the Safman-Taylor fi ger criteria for the bonding quality in a microchannel versus various fl w rates. For fl w rates lower than 1.5ml/min, the Safman-Taylor fi gers do not appear, while for fl w rates more than 1.5ml/min, Safman-Taylor fi gers become discernible. (B) Shear rate distribution across a line parallel to the channel width. (C) Te more the pressure, the faster the creation of Safman-Taylor fi gers. In the green area, Safman-Taylor fi gers do not appear during the experiments. Also, the results show that there is not any evidence of channel burst or delamination during the bonding quality test.

Figure 3. (A) Surface topography of the double-coated adhesive tape. Results in a vertical line (green line), horizontal line (blue line), or across an area (red rectangular) show that the tape has homogenous roughness with a nanometric value, which does not impose any interference on the channel performance. Te absorbance amount of (B) double-coated adhesive tape and (C) PMMA sheet, implying that these two materials are transparent for visible spectra range, and there is not any signifi ant absorption.

Sinusoidal and serpentine microchannel.
Inertial microfuidics in sinusoidal (curvilinear) microchannel has gained traction due to its improved focusing performance, the ability of massive parallelization, and small footprint. In the sinusoidal microchannel, the curvature direction changes in each loop, resulting in an intricate phenomenon that help in particle focusing. Indeed, by alternating the curvatures, the direction of Dean fl ws changes, and secondary fl ws may not reach a steady-state condition. Th s design was fi stly developed by Di Carlo in 2007 and its capability in wide ranges of Dean number ( / De = Re D R2 h , where Re is channel Re, Dh is characteristic length of channel, and R is the radius of channel curvature) was evaluated56. By assuming that Dean drag forces were balanced with shear gradient lift forces, his team proposed the ratio of inertial lift forces to Dean drag forces as F F L D / 2 = ra /Dh 2 3 (where r is radius of channel and a is particle diameter). Generally, if FL/FD ≫ 1, secondary fl ws do not afect particles, and if FL/FD ≪ 1, particles are entirely afected by secondary fl ws57. Th application of this microchannel was even expanded where it was used for high-throughput separation of micron and sub-micron bioparticles (cyanobacteria)58 and a microfuidic concentrator for harvesting of cyanobacteria59. In addition, in a comprehensive study, the design principle of curvilinear microchannels was investigated, and a map for various focusing phenomena was provided, based on F F/ ( ~ Re / ) De ( / a D ) f L D h L 2 2 3 where fL was approximated by Zhou and Papautsky21 as fL ~ 1/Re(Dh/a)2 60. Te dependency of curvature angle61 and various cell lines62 on the focusing positions was also evaluated. In order to showcase the adaptability of our method for fabrication of various inertial microfuidic devices, a curvilinear microchannel with rectangular cross-section (Fig. 5A) was designed, fabricated, and evaluated. Figure 5B reveals that 15 µm particles fi st occupied two focusing positions and by increasing the fl w rate, it reduced to a single focusing line across the channel (e.g., at 5th loop for fl w rate of 900µl/min (Fig. 5C)) which is consistent with the previously reported results60.

By altering the design of curvilinear to straight, serpentine microchannel with a square-wave pattern is created. Th s channel proves to have unique features for size-based particle focusing. Generally, particle focusing achieves when a/Dh > 0.07 and Rp~1. However, additional secondary fl ws in serpentine microchannel lead to focusing of particles with smaller diameters compared to those calculated by the above formula. Tese devices can beneft from parallelization along their vertical direction, thereby increasing their throughput. Tree focusing

Figure 4. (A) Inertial microfuidics in a rectangular straight microchannel with height and width of 50 µm and 200 µm, respectively. I. At fi st, 20 µm particles occupy the center of the channel as their focusing position. Te intensity profle also illustrates that particles are focused at the center of the channel. II–IV. Later, looking at lateral position and intensity profles reveal that by increasing the fl w rate, side walls are added to the focusing position of the particles, and the focusing band of particles at center becomes wider. To extract these images, we have used “max intensity” feature from Fiji Sofware (https://fji.sc). (B) Te equilibrium position of MDA-MB-231 cells at fl w rates of I. 10ml/hr and II. 20ml/hr and III. DU-145 cells at fl w rate of 20ml/hr confi ms the single-line focusing of cells within the rectangular straight microchannel (from top view). (C) A surface proflometry of the rectangular cross-section which shows the rectangular profle of the microchannel. Results show that the channel has perfect shape and quality which is suitable for inertial microfuidics. (D) In triangular microchannel, particles fi st migrate to I. and II. one focusing position and then this increases to III. three separate points. Th s trend is similar to those reported in the literature14,53,54. Te results for MDA-MB-231 cells at fl w rate of 200 µl/min illustrate a single-line focusing position, and at fl w rate of 500 µl/min depict three focusing positions. (E) Surface proflometry of the triangular straight microchannel with height and width of 40 µm and 300 µm, respectively

Figure 5. (A) Inertial microfuidics in a curvilinear microchannel with height and width of 50 and 200µm, respectively. (B) Te results show that the equilibrium position of particles depends on the fl w rate and has diferent focusing modes. Particles fi st focus at two equilibrium positions and then occupy just one focusing line. Eventually, by further increasing the fl w rate, Dean drag forces become dominant, resulting in defocusing of particles. Te trend is similar to that reported in the literature60. (C) 15µm particle migration throughout the channel for fl w rate of 900µl/min, demonstrating that particles are focused at the 5th loop. (D) Inertial microfuidics in a serpentine microchannel with height and width of 40 and 200µm, respectively. Te number of lateral positions depends on the applied fl w rate at the entrance of the channel. (E) Focusing behavior of 10µm particles at 0.7ml/min. (F) As intensity profle elucidates, at the 10th loop, particles reach the stable equilibrium position. Lateral migration of MDA-MB-231 cells at fl w rate of (G) 0.7ml/min and (H) 0.8ml/min shows a single-line focusing of these cells at the center of the channel. (I) Surface proflometry of the channel with rectangular cross-section with width and height of 200 and 50µm, which shows the accuracy and high-quality of the channels appropriate for inertial microfuidics.

patterns can be identifi d by increasing the input fl w rate, i.e., two-sided focusing, transition focusing, and central single-line focusing. If inertial efects dominate the secondary fl ws, particles occupy two lines near the walls. In contrast, dominance of secondary fl ws results in a single-line focusing at channel center. If these two efects have the same order, particles focus as a wide streak. Gaining the efciency of two-sided focusing for small particles and central focusing for big ones give us the opportunity of size-based particle separation. Based on the literature, a serpentine microchannel with cross-section of 40×200µm (H×W) and 15 loops was fabricated and used to showcase the focusing of 10 µm particles (Fig. 5D). In a straight channel with 40×200 µm (H×W) cross-section, a/Dh for 10 µm is 0.056, which is less than the focusing criteria (0.07); theoreically, these particles cannot focus in a straight microchannel. However, in the serpentine microchannel with the aid of secondary fl ws, 10 µm particles can effici tly be focused. Figure 5E elucidates that 10 µm particles at fow rate of 0.7ml/min can be focused at the center of channel at the 10th loop and occupy central equilibrium position at the outlet, which is consistent with the insets as normalized intensity profle of the particles (Fig. 5F). More importantly, the performance of device was tested with MDA-MB-231 cells (Fig. 5G,H), and the results are consistent with those reported in the literature63. Te surface proflometry of the channel cross-section is also provided in Fig. 5I, indicating the high-accuracy of the proposed method for fabrication of inertial microfuidic devices.

Spiral microchannel.
Spiral defi es as a curve winding around a center point with continuous decreasing or increasing manner. When fl w passes the curvature, velocity mismatch occurs in the curve section of the channel, resulting in the generation of secondary fl ws. In inertial microfuidics, spiral microchannel has progressed signifcantly, and nowadays, most of the particle/cell separations are performed using these microchannels64. De is used for the characterization of secondary fows within the channel. Intuitively, smaller channel curvature or larger channel size or Re leads to higher De, thereby imposing stronger secondary fl ws within the channel. For a given De, average transverse Dean velocity (UDe=1.84×10−4 De1.63) and Dean drag force = = πµ α π . × µ α − . ( 3 F U 5 4 10 De ) D De 4 1 63 can be identifi d. However, the exact behavior of particle migration at the downstream of the fuid was not thoroughly investigated, and all results are based on experimental data. Te most appealing feature of spiral inertial microfuidics is its high-throughput where 2100 particles per second can be processed9 . Particle sorting is one of the most signifcant applications of spiral microfuidics. Previously, the potential of a rectangular spiral microchannel for continuous and simultaneous isolation of 10, 15, and 20µm based on soflithography was investigated (Fig. 6AI) 65. Dean fl w dynamics for a low-aspect-ratio rectangular spiral microchannel was also thoroughly explored66. Beyond a simple rectangular spiral microchannel, various geometry modifcations for regulation of Dean forces and performance enhancement of the device have been proposed. Beneftting from micromilling (Fig. 6AII), trapezoidal spiral microchannels illustrate promising results in redistribution of lateral focusing positions of particles appropriate for size-based particle separation. In these channels, smaller particles focus along the outer wall, whereas larger ones migrate toward the inner wall67. Th s superior advantage has been widely investigated by our group, among other groups, for circulating tumor cell (CTC) and circulating fetal trophoblasts (CFT) isolation19,68, blood plasma separation69, isolation of microcarriers from mesenchymal stem cells70,71, microalgae separation72, and synchronizing C. elegans73. Also, multiplexing using stack of attached PDMS layers to boost the throughput is illustrated previously69,74. However, most of the aforementioned applications are just doable by utilizing cleanroom facilities or employing conventional micromachining (e.g., metal machining or laser cutting) for the fabrication of microchannel. Besides, micromachining has its own limitations such as inability to make sharp corners or difculty in making spiral loops close to each other. Tese challenges highlight an unmet need for the fabrication of spiral microchannels using a versatile method which is robust and can surmount aforementioned issues.

As a showcase of the versatility of our proposed method, we have fabricated a spiral microchannel with trapezoidal cross-section with a width of 600 µm and heights of 80 and 130 µm. Tese results are then put aside a PDMS chip with similar dimensions, and the data is provided in ESI (Fig. S5). Despite all progress in spiral inertial microfuidics, there is not any report of a spiral with cross-sections rather than rectangular or trapezoidal. In other words, a huge amount of potential remains intact to study spiral microchannels with diferent cross-sections such as triangular (Fig. 6AIII). For this aim, for the fi st time, we have fabricated a spiral microchannel with right-angled triangular cross-section (as schematically shown in Fig. 6B) where the width and height are 600 and 210µm, respectively. As the results are illustrated in Fig. 6C, there is a tight band focusing for particles larger than 10 µm, which is suitable for high throughput fl w cytometry applications where single line focusing is desired. Also, we observed double-band focusing behavior for 20µm particles at fl w rate of ≥4ml/min. Te dimensions (Fig. 6D) and channel cross-section (Fig. 6E) show the accuracy of the proposed method the for fabrication of right-angled triangular spiral microchannel (check Fig. S6 for contraction-expansion array microchannel results). Tese results illustrate the flex bility of this method where a complex cross-section can be fabricated in less than two hours with high robustness and stability. Our results hold promise for leveraging the potential of additive manufacturing for the fabrication of inertial microfuidic devices, which is more challenging using conventional microfabrication methods (see Section S6 for multiplexing of 3D printed inertial microfuidic devices).

Cellular studies.
PDMS-made inertial microfuidic devices have been widely used for the cell separation using biological samples such as blood and urine. While PDMS is proven to be a biocompatible material with minimum side efects on cells, we have tested the 3D printed devices using DU145 cells, assessing their viability and functionality post-separation. Te collected cells from the device outlet were cultured back into a petri dish for 5 days, showing similar morphological features to the control group as shown in Fig. 7A,B. Te fl w cytometry tests (Fig. 7C) showed that the viability of the cells was not compromised during the operation using 3D printed devices. Te real-time PCR analysis was utilised to assess the expression of genes related to the general activities and stress responses in both treated and untreated cells (Fig. 7D). Te similar expression level of GAPDH and CDKN2A confi med that neither cellular metabolism nor cell cycle progression were afected afer processing

Figure 6. (A) Illustration of a spiral microchannel where the fuid direction is from outside to inside. I. Firstly, several groups (e.g., Bhagat et al. 9 , Papautsky et al. 66,77, etc.) have shown the capability of rectangular spiral microfuidics for such applications as fl w cytometry or microparticle/cell separation. Te fabrication of these devices was based on photolithography. II. Gaining the effici cy of micromilling, many groups (e.g., Guan et al. 67, Warkiani et al. 20, etc.) made an attempt to get the advantage of trapezoidal spiral microchannel for particle/cell fltration and fractionation. III. In this study, for the fi st time, we have shown the fabrication of a right-angled triangular spiral microchannel with the aid of additive manufacturing. (B) Schematic illustration of the microchannel where the inset shows the cross-section of the microchannel. (C) Results reveal that for particles larger than 10 µm, a tight focusing band appears at the outlet of the channel. Also, for larger particles at high fl w rates (i.e., 4 ml/min) double-band focusing appears. (D) dimensions of the right-angled triangular spiral microchannel where the inner wall is 210 µm and the width is 600 µm. Te hydraulic diameter of this channel is similar to a spiral with a trapezoidal cross-section and the dimension of 80×130×600 µm. (E) Illustration of a right-angled triangular cross-section, which shows the accuracy of the fabrication process.

through the microchannels. Also, there are no signifi ant changes in the expression of TXNIP and MAPK14, which are two well-known regulators of cellular stress. In all, the 3D-printed inertial microfuidic device does not alter the cell activity and is safe to be used in biological assays.

Conclusion
In this study, we have showcased a robust and versatile workfl w for the manufacturing of inertial microfuidic devices for both laboratory experiments and industrial applications. Te proposed method involves 3D-direct printing of a channel with the desired structure and geometry via a high-resolution DLP/SLA 3D printer. Our approach relies on the bonding of 3D-printed devices (i.e., with high-quality surface fin sh) to a transparent PMMA sheet via a double-coated pressure-sensitive adhesive tape. Given the great transparency of PMMA layer, these devices can be utilised in bright fi ld, phase contrast, fuorescent, or high-speed microscopy. Te bonding quality was evaluated by Safman-Taylor fi ger criterion, and the results showed that the device is capable of withstanding pressure as high as 150psi (nearly triple of the value reported for PDMS). Te versatility of this method allowed us to fabricate and evaluate a plethora of existing inertial microfuidic devices as exemplifi d for fabrication of straight, spiral, serpentine, curvilinear, and contraction-expansion arrays microchannels. Te total time frame, from designing a part to starting an experiment, takes less than two hours, allowing multiple experiments in a single day. Also, for the fi st time, we have fabricated and examined a new inertial microfuidic device, i.e., spiral microchannel with right-angled triangular cross-section which is theoretically impossible to fabricate using photolithography. We believe that the proposed workfl w will provide inertial microfuidic devices within the reach of any research groups involving in particle/cell manipulation without strong microfabrication background.

Figure 7. (A) Monitoring the morphological feature of cells during fve days post-experiment, compared to the control group. (B) Fluorescent staining of F-actin flaments in expanded cells on day 5 (green = phalloidinFITC, blue = nucleus) (C) Representative plot and mean value ± SEM of fl w cytometric analysis of live/dead population for control and test group. (D) Cycle of threshold (Ct) value (expression level) for GAPDH (cellular metabolism related gene), CDKN2A (cell cycle regulatory gene), TXNIP, and MAPK14 (genes involved in cellular stress response). Results are expressed as mean value ± SEM from three independent experiments. Scale bars are equal to 100 µm in large images and 20 µm in inset ones

Materials and methods

Fabrication method
In this study, inertial microfuidic devices were fabricated using a high-resolution DLP/SLA 3D printer (Ultra 50, Miicraft, Hsinchu, Taiwan) featuring 30 µm XY resolution and 32×57×120 mm3 printing area. Te desired inertial microfuidic device is fi st drafed using a commercial CAD drawing sofware (SolidWorks 2016) and then translated into STL format, a suitable fle for 3D printer language. Te fle is then sliced in Z direction using the Miicraft sofware (Version 4.01, Miicraft). Te slicing in Z direction (slice thickness) can be adjusted from 5 to 200µm with an increment of 5µm. Te slicing option is related to the complexity of the geometry. In geometries with ramp or step, the slice thickness should be small, whereas for planar or orthogonal structures, higher slice thickness is suitable. Te detailed dependencies of printing parameters are provided in electronic supplementary information (ESI) (Section S1 and Fig. S1). Te sliced fle is then sent to the 3D printer with UV wavelength of 358–405nm. Te UV light is projected from the bottom of the resin bath (flled with BV-007 resin) and passes through a transparent Tefl n flm. BV-007 is an acrylate-based resin containing 80–95% acrylate components and 10–15% photoinitiator and additives. Once the UV light cures one layer, the Z-stepper motor moves one slice upward, and the next layer starts to be polymerized. Th s process continues until the part is printed, successfully. When the part is removed from the picker, it should be rinsed and washed with isopropanol, thoroughly and air-dried by an air nozzle. Aferward, the microchannel is exposed to a UV light with 405 ± 5nm wavelength within a curing chamber for post curing process. One superior advantage of this method compared to the soflithography is that it does not require punching the holes for inlets and outlets since those are printed as a one-body part by the 3D printer. Eventually, tubes (Tygon tubing, inner diameter: 0.020″, outer diameter: 0.060″) were connected to the microchannel by a tweezer.

Preparation of bead suspension
Fluorescent microbeads (Fluoresbrite Microspheres, Polysciences Inc, Singapore) with 0.01% volume fraction and various diameters were added to the MACS bufer. Te primary usage of MACS bufer is to prevent nonspecific adhesion of microbeads to the tubing of the microchannel. Te distribution of particles is illustrated using standard deviation, minimum, or maximum light intensity plots, as reported previously54. Preparation of bead suspension Fluorescent microbeads (Fluoresbrite Microspheres, Polysciences Inc, Singapore) with 0.01% volume fraction and various diameters were added to the MACS bufer. Te primary usage of MACS bufer is to prevent nonspecific adhesion of microbeads to the tubing of the microchannel. Te distribution of particles is illustrated using standard deviation, minimum, or maximum light intensity plots, as reported previously54.

Bonding quality test
Inertial devices are operated at high fl w rates; hence, the bonding technique must provide enough strength to prevent leakage from the interface. To evaluate the bonding quality of our proposed technique, a simple 3D-printed straight channel featuring 50 µm height, 200 µm width, and 4 cm length was bonded to a 2-mm-thick PMMA layer. A high-pressure syringe pump (Chemyx Fusion 4000, Chemyx, TX, USA) was used to inject fuids inside the channel from a small syringe (6ml). Increasing the fl w rate leads to the generation of Safman-Taylor fi gers around the inlet, in which the most pressure in the channel present (Section S2). Safman-Taylor fi gers are generated by the movement of a viscous fuid within a porous material75,76. As the bonded adhesive tape forms a porous zone between the connecting parts, this theory is applicable for the bonding evaluation. An increase in the applied pressure leads to developments of the Safman-Taylor fi gers until the bonding fails. A CCD camera (DP80, Olympus, Tokyo, Japan) mounted on an inverted microscope (IX73, Olympus, Tokyo, Japan) was used for monitoring the bonding integrity. All recorded data were obtained immediately afer the bonding of the 3D-printed channels to a PMMA sheet.

Surface characterization
For the surface characterization of double-coated adhesive tape, a 3D laser microscope (Olympus LEXT OLS5000) was used, and an LMPLFLN 20x LEXT objective lens (Olympus) was selected. Arithmetic mean deviation (Ra), the arithmetic mean of absolute ordinate Z (x, y) documented across a line, and arithmetical mean height (Sa), the arithmetic mean of the absolute ordinate Z (x, y) recorded across a region were chosen to evaluate the surface characterization of the tape.

Cell culture, harvesting, and device operation
DU145 cells (human prostate cancer cell line) were cultured and expanded under standard culture condition (37 °C and 5% CO2) using Roswell Park Memorial Institute medium (RPMI, TermoFisher) supplemented with 10% fetal bovine serum (FBS, Gibco) and 1% Penicillin Streptomycin (Pen/Strp, Gibco). Cells were harvested when the fask was 80% confuence. To obtain a homogenous cell suspension without cell clumps, a suffici t volume of TryplE (Gibco) was added to cover the whole fask, and the fask was incubated at 37 °C for 5 min. Te cells were then collected in a 15 ml falcon tube and counted with a hemocytometer. TryplE was then replaced with phosphate bufer saline (PBS, Gibco), and cells were diluted to 106 cells/ml concentration by PBS. Aferward, cells were introduced to 3D-printed microchannels (straight rectangular microchannel with a length of 4 cm, width of 200µm, and height of 50µm) at the fl w rate of 0.3ml/min. A group of untested cells was kept as control.

Morphological analysis and cell viability assay
In order to evaluate the viability of cells afer passing through the channels, a live-dead assay was performed using Live and Dead Cell Assay kit (Abcam, Cambridge, UK) afer the test. One group of collected DU145 cell suspension was diluted to 5×105 cell/ml and incubated with the staining solution for 10min under room temperature, based on the kit manufacturer’s instruction. Ten, the stained cells were processed through fl w cytometry (Olympus CKX53, Tokyo, Japan) and analyzed by CytExpert sofware (Beckman Coulter, Inc.). Live and dead cells were detected by green (λ excite/emit = 488/515nm) and red (λ excite/emit = 488/617nm) fuorescent, respectively. Another group of DU145 cells was cultured in a culture fask and stained by live and dead staining afer 24 hours without detaching by TryplE. Fluorescent microscopic imaging was taken for adhered cells to identify live cells (with green cytoplasm) from dead ones (with red nucleus) (Fig. S8). Te morphology of the attached cells was monitored under an inverted microscope for up to fve days. Furthermore, the morphological features of both test and control groups were visualized by fuorescence staining of cytoskeleton afer three days. Te F-actin flaments were fi ed and permeabilized by 4% paraformaldehyde (PFA, Sigma) and 0.2% Triton X-100 (Sigma) and then labeled by Phalloidin- FITC (Sigma).

Real-time PCR analysis
The effect of shear stress on the expression of genes related to proliferation and survival was measured by real-time PCR (BioRad CFX 96 thermocycler). Briefy, the processed cells were reseeded on a culture dish and kept under standard culture conditions for one day. Next, total RNA of cells were extracted by using PureLink RNA Mini Kit (TermoFisher), and cDNA was synthesized by applying Revert Aid First Strand cDNA Synthesis Kit (TermoFisher). Real-time PCR was performed using specific TaqMan primer sets and TaqMan PCR Master Mix (TermoFisher) with following cyclic conditions: 95 °C for 10min, followed by 40 cycles of 95 °C for 10 s, 60 °C for 1min, and 72 °C for 10 s.

Ultrasensitive and rapid quantification of rare tumorigenic stem cells in hPSC-derived cardiomyocyte populations

by

Ultrasensitive and rapid quantification of rare tumorigenic stem cells in hPSC-derived cardiomyocyte populations

Zongjie Wang 1 2, Mark Gagliardi 3 4, Reza M Mohamadi 5, Sharif U Ahmed 5, Mahmoud Labib 5, Libing Zhang 5, Sandra Popescu 5, Yuxiao Zhou 5, Edward H Sargent 1, Gordon M Keller 3 4 6, Shana O Kelley 2 5 7

The ability to detect rare human pluripotent stem cells (hPSCs) in differentiated populations is critical for safeguarding the clinical translation of cell therapy, as these undifferentiated cells have the capacity to form teratomas in vivo. The detection of hPSCs must be performed using an approach compatible with traceable manufacturing of therapeutic cell products. Here, we report a novel microfluidic approach, stem cell quantitative cytometry (SCQC), for the quantification of rare hPSCs in hPSC-derived cardiomyocyte (CM) populations. This approach enables the ultrasensitive capture, profiling, and enumeration of trace levels of hPSCs labeled with magnetic nanoparticles in a low-cost, manufacturable microfluidic chip. We deploy SCQC to assess the tumorigenic risk of hPSC-derived CM populations in vivo. In addition, we isolate rare hPSCs from the differentiated populations using SCQC and characterize their pluripotency.

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

Introduction

Human pluripotent stem cells (hPSCs) hold great promise for cell therapy given their ability to differentiate into many different cell types (1). Numerous studies have demonstrated the considerable potential of hPSC derivatives in treating chronic diseases, including neuron degeneration (2) and chronic heart failure (3). Typically, a dose of cell therapy for treating heart failure requires 0.1 to 1 billion of de novo hPSC-derived functional cells (4); therefore, the translation of cell therapy from bench to bedside heavily depends on reliable manufacturing of high-quality cell products (4–6).

The self-renewal and pluripotent properties of hPSCs are also associated with a high level of tumorigenicity in vivo (7). Undifferentiated cells can persist in differentiated populations following long periods of time in culture (8, 9), and rare contaminating hPSCs, even at concentrations less than 0.025%, can lead to teratoma formation in animal models (10–12). As a result, quantitation of the percentage of rare hPSCs is a key quality control parameter that needs to be monitored in manufactured populations to be used for cell therapy applications (4, 13).

Flow cytometry (FCM) and the polymerase chain reaction (PCR) are powerful methods for the analysis of rare cells. Unfortunately, neither of these methods is sensitive enough to rapidly and accurately identify rare hPSCs in relevant samples. FCM has intrinsic limitations including sampling losses and dead volumes (14) that reduce accuracy at exceptionally low levels of hPSC contamination that may still represent a potential risk of tumor formation (15). PCR-based methods are often problematic due to the high background from differentiated cells relative to the rare undifferentiated controls (16) and reverse transcription–induced artifacts, including primer-independent complementary DNA (cDNA) synthesis and template switching (17).

Here, we report a new approach to the quantitation of rare undifferentiated cells: stem cell quantitative cytometry (SCQC). This method takes advantage of a rare cell profiling approach based on a strategy for monitoring cancer cells (18, 19). SCQC uses a microfluidic chip that is scalable, cost-effective, and compatible with the requirements of manufacturing and quality control. SCQC has excellent sensitivity and is able to profile rare hPSCs robustly even when present at concentrations as low as 0.0005% in populations of hPSC-derived cardiomyocytes (CMs). Through the analysis of CM samples containing defined numbers of spiked hPSCs, we demonstrate that SCQC detects rare contaminants with unprecedented performance. Comparative studies show that SCQC can accurately quantify hPSCs at levels that are not reliably detected by either FCM or droplet digital PCR (ddPCR). Last, we use SCQC to isolate live rare hPSCs from differentiated CM populations and characterize their pluripotency.

Materials

A Bottle of the Master Mold for PDMS Device Resin

Master Mold Resin

Learn More

H Series

Learn More

Results

Development of microfluidic SCQC
The device used for SCQC relies on immunomagnetic labeling to profile cells based on their surface marker expression (Fig. 1A) (20). Cells are labeled with magnetic nanoparticles (MNPs) that bind to a surface marker expressed by hPSCs, but not CMs. Labeled cells are subsequently introduced into a microfluidic device with a flow velocity gradient and a constant magnetic field (Fig. 1, A to C, and fig. S1, A and B). Cells with more MNPs experience a higher magnetic force to compensate the downstream drag force generated by the flow velocity. As a result, the hPSCs that bind more MNPs than the differentiated CMs can withstand high flow velocities and remain in the high-velocity regions of the device. The CMs with few or no MNPs are captured in the low-velocity regions or flushed away. At the completion of the run, the cells trapped in the microfluidic device are quantified by immunofluorescence and microscopy to generate a cytometric profile that includes number, phenotype, and distribution of trapped cells. The information from the cytometric profile provides an assessment of the total number of hPSCs.

Fig. 1 SCQC for rare stem cell analysis.
(A) Overview of the SCQC approach. Cells are labeled with MNPs functionalized with an antibody against specific stem cell surface markers. Labeled cells are magnetically captured in a 3D-printed microfluidic device with a flow velocity gradient. Stem cells labeled by a high number of MNPs can withstand higher flow velocities and are captured in earlier zones. The number of stem cells in each zone is quantified by immunostaining and microscopy to generate a cytometric profile that can be used for further analysis. (B) Illustration of the microfluidic device for SCQC. Eight sequential capture zones with a decreasing flow velocity gradient are generated by linearly increasing the channel height. (C) Picture of the fabricated device. A dye-containing solution was introduced into the channel to visualize the changes in channel height. Error bar indicates the SD of the mean from three experiments (B).

Sensitivity and specificity of SCQC

We used the HES2 hPSC line and derivative CMs to characterize and optimize the performance of SCQC. The CMs were differentiated for at least 20 days following an established protocol (fig. S2) (21, 22). The hPSC-derived CMs contain more than 85% cardiac troponin T (cTNT)–positive cells. In the first suite of experiments, we used FCM to benchmark eight surface markers (SSEA-1, SSEA-4, TRA-1-60, TRA-1-81, EpCAM, CD90, CD9, and E-cadherin) and three intracellular markers (Oct4, SOX2, and Nanog) for distinguishing hPSCs and hPSC-derived CMs (fig. S3). FCM results revealed that TRA-1-60 and EpCAM offer the highest separation (equivalently the best contrast) among all markers. Hence, we chose TRA-1-60 and EpCAM MNPs for on-chip optimization.

HES2 hPSCs or derivative CMs were labeled with antibody-conjugated MNPs, profiled using SCQC and stained by a (4′,6-diamidino-2-phenylindole) DAPI/NucDead 488 cocktail for visualization (fig. S4A). TRA-1-60 MNPs produced higher levels of hPSC capture and hPSC-derived CM depletion (fig. S4B) than EpCAM MNPs. We also investigated the specificity of the TRA-1-60 MNPs for labeling hPSCs by transmission electron microscopy (fig. S4C). More than 10 major clusters of MNPs were observed on the surface of hPSCs, while 0 or 1 major cluster of MNPs was detected on the CMs. On the basis of these findings, we concluded that the TRA-1-60 has enhanced performance for capturing rare hPSCs in hPSC-derived CMs compared with other surface markers. Next, we tested the direct isolation of hPSCs using TRA-1-60 MNPs via magnetic-activated cell sorting (MACS). However, MACS had poor capture and recovery efficiency for both live and fixed cells (fig. S5). Notably, around 20% of cells, regardless of cell type, remained trapped in the columns and could not be recovered. This is consistent with previous studies (23, 24). Therefore, it is necessary to use the microfluidic chip to enable more efficient recovery and in situ immunostaining of rare cells for quantification.

We next assessed the specificity of SCQC to evaluate its utility for the quantitation of undifferentiated hPSCs. In the absence of an external magnetic field or the absence of MNPs, less than 0.05% hPSCs were captured at the flow rate of 2 ml/hour. Subsequently, the flow rate, which dominates the flow velocity on-chip, was optimized to balance the capture of hPSCs and the depletion of hPSC-derived CMs (fig. S4D). At the optimized flow rate (10 ml/hour), we captured 85.7% of hPSCs while depleting 99.7% of CMs (Fig. 2, A and B). Over 90% of captured hPSCs were detected in zones 1 to 5, which generates a reproducible cytometric profile. In addition, we characterized the dynamic range of the SCQC using the optimized flow rate. The chip had a consistent capture efficiency and cytometric profile in the range of 0 to 5000 hPSCs (Fig. 2C and fig. S4E). The chip offered sufficient depletion (>99%) up to 500,000 CMs (fig. S4E). Also, the chip maintained a consistent capture performance when using other hPSC cell lines, including an induced pluripotent stem cell line (fig. S4F) and other hPSC-derived populations (i.e., definitive endodermal cells) (fig. S4G). On the basis of these results, we conclude that SCQC is a highly sensitive and selective method with an excellent dynamic range for the capture and analysis of hPSCs.

Fig. 2 Characterization of sensitivity and selectivity of the SCQC approach. (A and B) Representative cytometric profiles of captured (A) hPSCs and (B) hPSC-derived CMs at the flow rate of 10 ml/hour (n = 3). (C) Capture performance of hPSCs in CM-free buffer (n = 3). (D) Representative microscope images of captured hPSCs in CMs. HPSCs are quantified as DAPI+, Oct4+, and Nanog+. (E) Capture performance of hPSCs spiked in 1 million hPSC-derived CMs (n = 4). Error bar indicates the SD of the mean from all experiments (A to E). Cell capture experiments (A, C, and E) were performed at the flow rate of 10 ml/hour and the volume of 1 ml. The number of hPSCs (A) or CMs (B) was 500.

We determined the limit of detection (LOD) of SCQC using samples of hPSC-derived CMs spiked with defined numbers of hPSCs. As the SCQC device nonspecifically captures a small fraction of hPSC-derived CMs, immunostaining was used to quantify the number of hPSCs on the fluidic chip. The hPSCs were defined using a cocktail of DAPI, Oct4, and Nanog (Fig. 2D). We found that SCQC can clearly identify the difference between the negative control (zero hPSC in 1,000,000 hPSC-derived CMs) and the 0.0005% sample (five hPSCs in 1,000,000 hPSC-derived CMs as shown in Fig. 2E). Hence SCQC achieves a LOD of 0.0005% for quantifying rare hPSCs.

Quantitative comparison between SCQC, FCM, and ddPCR

We conducted a comparative study to systematically evaluate the performance of SCQC, FCM, and ddPCR for rare hPSC detection. We generated populations of hPSC-derived CMs containing 0.01 to 5% of spiked HES2 hPSCs. For FCM, we used TRA-1-60 and EpCAM as the hPSC markers with a two-laser six-color flow cytometer. For ddPCR, we monitored the expression of three hPSC genes: POU5F1, SOX2, and CD326. TBP or B2M was included as a housekeeping control. For SCQC, we applied the TRA-1-60 MNPs and the flow rate of 10 ml/hour, as detailed above.

The representative profiles obtained by FCM are shown in Fig. 3A and fig. S6A. The hPSC signal (TRA-1-60+ EpCAM+) decreased rapidly with the decreasing number of hPSCs. FCM was unable to detect hPSCs in the 0.1% sample, as the signal for these samples was the same as the signal for the samples that did not contain hPSCs. The combination of the two most sensitive markers (TRA-1-60/EpCAM) yielded the optimal LOD around 0.2% to 0.3% (see fig. S6, A and B). Previous literature suggests that the number of total events that must be collected to detect a population present at a frequency of 0.1% is at least 10 million (25). Routine collection of 10 million events by FCM is neither cost- or time-effective.

Fig. 3 Benchmarking the performance of SCQC, FCM, and ddPCR for rare hPSC quantification. (A) Representative cytometric profile of FCM for samples with spiked hPSCs. (B and C) Detection performance of ddPCR using EpCAM and (B) TBP or (C) B2M as the housekeeping control for normalization. (D to F) Detection performance of SCQC and FCM in the range of (D) 0 to 5%, (E) 0 to 2%, and (F) 0 to 0.1% (n = 3 for SCQC, FCM, and ddPCR; 50,000 cells were analyzed for each replicate). Error bar indicates the SD of the mean from three experiments (B to F). Cell capture experiments (D to F) were performed at the flow rate of 10 ml/hour using a total volume of 1 ml. Each cell suspension contained 50,000 hPSC-derived CMs spiked with various amounts of undifferentiated hPSCs in the desired final concentration, as indicated on the x axis.

The representative ddPCR results are shown in Fig. 3, (B and C) and fig. S6 (C and D). From the three primer sets tested (POU5F1, SOX2, and EpCAM), we found that the combination of EpCAM as the target and TBP as a housekeeping gene yielded the lowest LOD, calculated at 0.2 to 0.4% (Fig. 3B). The LOD obtained using POU5F1 and SOX2, which are the definitive genes for pluripotency, was higher than 1% (fig. S6D). This is not an unexpected result as previous literature has indicated that pluripotent genes (i.e., POU5F1, SOX2, Nanog, Klf-4, and Lin28) are weakly expressed at the RNA level in hPSC-derived cells, typically in the range of 0.01 to 1% relative to undifferentiated hPSCs (16, 26). As a result, the signal from rare hPSCs is concealed by the large background generated by a substantial excess of hPSC-derived cells. Therefore, given the lack of a specific hPSC marker, ddPCR is not a practical method for quantifying rare hPSCs in differentiated population.

We summarized the comparative results in Fig. 3 (D to F). The LOD of SCQC, FCM, and ddPCR is <0.01, 0.2 to 0.3, and 0.2 to 0.4%, respectively (table S1A). In addition to having a superior LOD, SCQC also offered the best linearity of detection (R2 > 0.99) among the three technologies (table S1B). On the basis of this comparative study, it is clear that SCQC outperforms the existing cell detection methods when quantifying rare hPSCs in batches of differentiated CMs. In addition, SCQC also has other advantages for cell manufacturing including cost-effectiveness, scalability, and compatibility with the U.S. Food and Drug Administration Current Good Manufacturing Practice (cGMP) regulations (27) (table S2). SCQC facilitates accurate detection and quantitation of undifferentiated cells within an hour at a cost of $30 per chip per run. Given its compact design and scaled fabrication method, this technology can be easily set up to support parallelized and large-scale operations.

Analysis of the tumorigenicity of rare hPSCs

To benchmark whether the performance of SCQC is relevant for the analysis of batches of therapeutic cells, we used SCQC to assess the tumorigenicity of samples containing different levels of hPSCs (Fig. 4). We prepared samples of hPSC-derived CMs (1 million) spiked with different numbers of hPSCs to yield contamination frequencies of 0.03 and 0.3%. CM populations with no additional hPSCs (0%) were used as controls. The three samples were injected into the testis of male NOD/SCID/Gamma (NSG) mice (Fig. 4A) to measure teratoma potential. A small portion (50,000 cells) of each sample was used for hPSC detection by SCQC and FCM analyses (Fig. 4B). As shown in Fig. 3B, SCQC was the only technology that correctly identified the percentage of hPSCs in differentiated CMs before injection. FCM analysis failed to distinguish the differences among three samples [P > 0.05 when performing the analysis of variance (ANOVA) between any of two samples].

Fig. 4 Rare hPSCs form teratomas in vivo. (A) Workflow of the teratoma-forming assay. Exogenous rare hPSCs were spiked into hPSC-derived CMs to form cell mixtures for testicular injection. After 10 weeks, the mice were euthanized to examine teratoma formation. (B) Quantification of hPSC concentration in the samples used for injection (n = 3 for SCQC and n = 5 for FCM). (C) Representative pictures of fixed teratoma from 0% hPSCs, 0.03% hPSCs, and 0.3% hPSCs and to hPSC-derived CMs. (D) Percentage of teratoma formation in mouse models. (E) Weight of teratoma in mouse models. (F) The 0.03% and 0.30% hPSCs added to hPSC-derived CMs can form a mature teratoma that contains three germ layers, as visualized by histology. Error bar indicates the SD of the mean from all experiments (B). Whisker, box, cross, and horizontal line indicate the minimum/maximum, first/third quartile, mean, and median from each group, respectively (E). Dots represent data points (E). Cell capture experiments (B) were performed at the flow rate of 10 ml/hour using a total volume of 1 ml. Each cell suspension contained 50,000 hPSC-derived CMs spiked with various amounts of undifferentiated hPSCs in the desired final concentration, as indicated on the legend.

All the mice in both experimental groups developed teratomas after 10 weeks (Fig. 4, C and D). The averaged testis weight in the 0.03 and 0.3% hPSC group underwent a marked increase from 0.1 g to over 2 g (Fig. 4E). Conversely, mice in the control (0%) group were teratoma free, and no significant change in testis was found. This result matched with the previous studies that showed that populations consisting of 0.025% hPSCs diluted in feeder fibroblasts could initiate teratoma formation within 12 weeks (28).

We further characterized the teratomas by histopathology (Fig. 4F) and detected multiple cell types including pancreatic, respiratory, and intestinal epithelium (endoderm); cartilage, bone, fibrous, and adipose connective tissue (mesoderm); and melanocytes and glial cells (ectoderm). The histopathological finding indicates that the hPSCs retain strong pluripotency and are capable of developing mature tumors in vivo. Together, these findings demonstrate that quantification of rare hPSCs with an LOD of 0.03% or lower is required to avoid the formation of teratoma in animal models. However, the minimal number of hPSCs that is sufficient to form teratoma remains unclear and will depend on several variables including the hPSC cell line, the number of injected cells, the format of injected hPSCs (clumps or single cell), and the site of injection (29, 30). While determining this number was beyond the scope of this study, we envision being able to take advantage of the sensitivity of SCQC to determine the number of hPSCs required to form teratoma under a clinically relevant dosage.

Isolation and characterization of live rare hPSCs
Isolating live rare hPSCs in hPSC-derived CMs may provide insights into the origins of heterogeneity for in vitro differentiation processes. We next applied SCQC to the isolation and characterization of live rare hPSCs in CM populations. For these studies, we generated CM populations from HES2 and HES3-NKX-2.5GFP hPSCs using both monolayer- and embryoid body (EB)–based protocols and profiled the samples using SCQC. After capture, the external magnetic field was removed, and the cells in the chips were isolated and expanded in culture (Fig. 5A).

Fig. 5 Isolation and characterization of live rare hPSCs from manufactured batches of CMs using SCQC.
(A) Workflow of the live cell isolation. Batches of hPSCs-derived CMs were profiled using SCQC. Captured rare TRA-1-60+ cells were recovered and cultured up to 15 days for analysis. (B) Representative microscope images of colony-forming rare TRA-1-60+ cells from hPSC-derived CMs cultured in a monolayer (day 8) and as EBs (day10). The colony-forming cells maintained a high level expression of Oct4 and Nanog (n = 3 to 8). (C and D) Assessment of the pluripotency of rare hPSCs. Rare hPSCs were successfully differentiated into endoderm [FOXA2+ and SOX17+], mesoderm [SMA+ or CD144+ cells], and ectoderm [PAX6+ and Nestin+] as quantified by (C) IF and (D) FCM. (E and F) Analysis of the pluripotency-related gene expression of rare hPSCs (normal hPSCs as control). (E) Microarrayed mRNA profile of rare hPSCs (n = 4). (F) Global analysis of the state of rare hPSCs. Rare hPSCs hold a higher expression of pluripotency-related mRNA (*P < 0.05). Error bar indicates the SD of the mean from four experiments (E). Whisker, box, cross, and horizontal line indicate the minimum/maximum, first/third quartile, mean, and median from each category of genes, respectively (F).

We initiated live cell experiments by optimizing the capture and culture conditions using CM populations containing spiked pluripotent HES2 hPSCs at the frequencies of 0.01 and 0.03%. We slowed the flow rate to 4 ml/hour to secure a capture efficiency of 90 to 95%. After capture, cells were released from the SCQC chips and cultured in StemFlex medium. This medium contains bovine serum albumin (BSA) and heat-stable fibroblast growth factor (FGF), which better support the survival of rare hPSCs. After 15 days of culture, no hPSCs were detected in the negative groups that contain the cells released from the chips (fig. S7A). In contrast, we observed the formation of multiple colonies in positive groups from 0.01 and 0.03% samples (fig. S7A). It typically took 6 to 10 days to allow the rare hPSCs to recover and grow from a single cell to a colony. No noticeable internalization of TRA-1-60 MNPs was observed, and 98.5% of MNPs on the cell membrane detached in 2 days (fig. S7B). The floating MNPs were removed during regular medium change. This allows the rare hPSCs to grow in an MNP-free environment to avoid unwanted cell-MNP interaction that could hamper cell function over long time periods (31). In addition, the CMs in the negative groups were found to adhere and form a network of cells within 3 days and formed beating monolayers at day 6. This demonstrated that SCQC is a gentle cell sorting method that poses minimal stress on profiled cells.

We next proceeded to profile the differentiated batches of cardiac cells generated from monolayer-based [D4 cardiac progenitor cells (CPCs) and D8, D12, and D16 contracting CMs] and EB-based differentiation protocols (D3 CPCs and D10 and D20 contracting CMs). The phenotypes of the manufactured batches of cells were characterized as shown in fig. S7C (percentage of cTNT+ cells) and fig. S7D (representative images and video clips showing the contractility of the samples). We captured rare hPSCs in all CPC samples (three of three) (fig. S7E), most of the D8 samples (two of three) (Fig. 5B), and some of the D10 samples (two of eight) (Fig. 5B). We did not find any rare hPSCs in the D12, D16, and D20 samples. These results indicate that the rare hPSCs are mostly present in early CPCs and CMs undergoing the maturation process (D8 to D20). They also show that expression of mature cardiac markers is not an indication of a lack of rare hPSCs, as undifferentiated cells were detected in day 10 EB populations that contained greater than 75% cTNT+ cells.

Next, we characterized the pluripotency of the rare hPSCs isolated from D8 HES2 hPSC–derived CMs. At the phenotypic level, a trilineage differentiation was performed to verify the pluripotency. The rare hPSCs retained the capacity to differentiate into FOXA2+ SOX17+ definitive endoderm, SMA+ smooth muscle cells or CD144+ endothelial mesoderm-derived cells, and PAX6+ Nestin+ neural stem cells, as verified by immunofluorescence (Fig. 5C and fig. S7F) and FCM (Fig. 5D). To characterize gene expression, a quantitative polymerase chain reaction (qPCR) microarray was used to analyze the expression of key pluripotent, naïve, primed, and differentiated genes. Compared with the standard unsorted HES2, the rare hPSCs had little alteration in the expression of key genes as all fold changes remained in the range of 0.3 to 9. The highest up-regulated and down-regulated genes were EGLN1 (8.6-fold) and KHDC1L (0.37-fold), respectively (Fig. 5E). However, global analysis revealed that the rare hPSCs had statistically higher expression of pluripotent markers (P = 0.02). Together, the characterization here demonstrates the feasibility of using SCQC for identifying and isolating rare cells in hPSC-derived differentiated populations.

DISCUSSION

The SCQC method described here provides an ultrasensitive, rapid, inexpensive, and scalable means of quantifying and isolating rare hPSCs in hPSC-derived CM populations. This approach is more sensitive and cost-effective than conventional methods including FCM and ddPCR. In a manufacturing environment, SCQC provides an effective way to monitor the quality of the manufactured population with respect to the presence of contaminating hPSCs.

In addition, we found that the rare hPSCs can be detected in populations of CPCs and immature CMs using SCQC. This highlights and validates the safety concerns surrounding stem cell–based cell therapy, especially for the therapies involving progenitors and differentiated cells at the early stage. As these cells have been used in small-scale clinical trials (32, 33), the quality assessment enabled by the SCQC is critical to fulfilling the demand of safeguarding cardiovascular cell therapies (34, 35).

In general, the concept underlying SCQC is broadly applicable to all surface markers and even intracellular mRNAs via the sequence-specific MNPs clustering (36). Hence, the implementation of the SCQC can be easily extended to the quantification of other rare cells in therapeutic products or patient samples, such as circulating tumor cells and chimeric antigen receptor therapy (CAR-T). Recent work has highlighted the importance to improve manufacturing technologies to quantify rare misprogrammed leukemic B cell for safeguarding CAR-T therapy (37).

MATERIALS AND METHODS

Device design and simulation
The SCQC device implements a fluidic channel with increasing heights to generate a flow velocity gradient with eight discretized flow velocities, which correspond to eight capture zones (Fig. 1B). The height of the first zone is 50 μm, and the stepwise increment is 50 μm per zone. X-shaped structures within the microfluidic device generate capture pockets that significantly improve trapping efficiency (20). Numerical simulations of the flow velocity profiles were carried out by COMSOL Multiphysics (version 5.3; COMSOL Inc., USA) using 3D creeping flow module. The key parameters were set as below: wall condition, no slip; boundary condition, pressure of 0 Pa; suppression of backflow, yes; mesh size, physics-controlled, normal; vector field shape, normal inflow velocity; and inlet velocity rate, 3.5 mm/s. The simulated flow velocity field was processed by MATLAB R2017b (MathWorks, USA) to extract the normalized linear velocity per zone. Simulated results suggest that the normalized flow velocities range from 100 (1×) to 14% (0.14×) (fig. S1, A and B). The multidepth design has two major advantages over the previously reported planar design (19, 20, 36). First, the device remains compact when adding more zones, which reduces the fabrication cost and accelerates the microscope scan. Second, manipulating heights offers easy and fine control over the flow velocity gradient.

Design of fabrication workflow
The fabrication of multidepth microfluidic devices usually involves multiple photolithography and mask-alignment processes that markedly reduce the cost-effectiveness and scalability (38). Although three-dimensional (3D) printing has shown the potential to provide a rapid solution for fabricating multidepth microfluidic devices, existing techniques could not achieve high resolution (dot feature sizes <200 μm) in a cost-effective and robust manner (39–42). To overcome these challenges, we carefully optimized the printing conditions of a desktop stereolithographic 3D printer with a pixel size of 30 by 30 μm (fig. S1C). The 3D printer supports the formation of positive structures (i.e., microposts) compared with negative structures (i.e., microwells). The minimal printable dot and line feature is 100 and 30 μm, respectively. This optimized condition allows the successful fabrication of various positive multidepth structures with a maximal aspect ratio up to 5 (fig. S1D) within an hour at the material cost of $50. To further improve the throughput and reduce the cost, multiple molding processes have been introduced (fig. S1E). Negative molds are first generated by casting polydimethylsiloxane (PDMS) on 3D-printed positive molds. The negative molds are subsequently treated by detergent and used as a new mold to generate the microfluidic devices. In this way, one 3D-printed mold can create multiple PDMS molds for mass production. We have achieved a throughput of 40 devices per day per operator at the laboratory scale and reduced the cost to $4 per chip. The details of the X-shaped structures with high aspect ratios can be transferred properly, granting the high quality of fabricated chips (fig. S1F). The measured thickness of each zone is within ±4% of the designed thickness (fig. S1G).

Device fabrication
Positive molds were fabricated by a stereolithographic 3D printer (μMicrofluidics Edition 3D Printer, Creative CADworks, Canada) using the “CCW master mold for PDMS” resin (Resinworks 3D, Canada). The layer thickness is set to 50 μm. Negative molds were fabricated by casting PDMS (Dow Chemical, USA) on positive molds and baked at 70°C for 2 hours. Negative molds were then treated by saturated detergent solution (Sparkleen, Thermo Fisher Scientific, USA) in 70% ethanol at room temperature (RT) for at least an hour. PDMS-positive replicas were generated by casting PDMS on negative molds and baked at 70°C for 2 hours. The cured replicas were then peeled off, punched, and plasma bonded to thickness no. 1 glass coverslips (Ted Pella, USA). The bonded chips were left in a 100°C oven for 30 min to secure a robust bonding. Afterward, the silicon tubing was attached to the inlet and outlet of the device. Before use, the devices were conditioned with 1% Pluronic F68 (Sigma-Aldrich, USA) in phosphate-buffered saline (PBS) for at least 1 hour to reduce the nonspecific adsorption. Each device was sandwiched between two arrays of N52 NdFeB magnets (K&J Magnetics, USA; 1.5 mm by 8 mm) with alternating polarity. A syringe pump (Chemyx, USA) was used for the duration of the cell capture process.

Device characterization
For the characterization of microstructures, printed positive molds, PDMS negative molds, and PDMS positive replicas were sputter coated with 20-nm Au (Denton Desk II, Leica, Germany) and observed under field emission scanning electron microscopes (Hitachi SU-5000 or FEI Quanta FEG 250) using 5-kV accelerating voltage. PDMS-positive replicas were also measured by a thickness gage (Mitutoyo, Japan) to determine the thickness of each zone.

Culture of hPSC lines
 HES2 (karyotype: 46, XX) was purchased from WiCell (USA). The HES3-NKX-2.5GFP reporter cell line (karyotype: 46, XX) was provided by E. Stanley and A. Elefanty (Monash University, Australia). BYS-0113 (karyotype: 46, XY) was purchased from the American Type Culture Collection (USA). hPSCs were maintained on Matrigel (Corning, USA)– or vitronectin (Thermo Fisher Scientific)–coated well plates in feeder-free hPSC culture medium consisting of DMEM/F12 (Cellgro, Corning) supplemented with 1% penicillin/streptomycin (Thermo Fisher Scientific), 2 mM l-glutamine (Thermo Fisher Scientific), 1× nonessential amino acids (Thermo Fisher Scientific), 55 μM β-mercaptoethanol (Thermo Fisher Scientific), 20% KnockOut serum (Thermo Fisher Scientific), and rhbFGF (50 ng/ml ) (Thermo Fisher Scientific).

CM differentiation of hPSC lines
Both HES2 and HES3-NKX-2.5GFP cell lines were differentiated into CMs using a modified version of previously published cardiac differentiation protocols (21, 22). Briefly, hPSCs were grown to 80 to 90% confluence and dissociated into single cells and reaggregated to form EBs in StemPro-34 medium (Thermo Fisher Scientific) containing 1% penicillin/streptomycin (Thermo Fisher Scientific), 2 mM l-glutamine (Thermo Fisher Scientific), transferrin (150 mg/ml; Roche, Switzerland), ascorbic acid (50 mg/ml; Sigma-Aldrich), and monothioglycerol (50 mg/ml; Sigma-Aldrich), 10 mM Y-27632 (ROCK inhibitor, Tocris, UK), and rhBMP4 (1 ng/ml; R&D Systems, USA) for 18 hours on an orbital shaker. At day 1, the EBs were transferred to mesoderm induction media consisting of StemPro-34 medium with above supplements (-Y-27632) and rhBMP4, rhActivinA (R&D Systems), and rhbFGF (R&D Systems) at the optimal cardiac differentiations for each line given in fig. S6. At day 3, the EBs were harvested, washed with Iscove's modified Dulbecco's medium, and transferred to cardiac mesoderm specification medium consisting of StemPro-34 medium, 2 mM IWP2 (Wnt inhibitor, Tocris), and rhVEGF (10 ng/ml; R&D Systems). At day 6, the EBs were transferred to StemPro-34 with rhVEGF (5 ng/ml) for an additional 7 days under hypoxic conditions (5% O2). The cultures were further matured for another 8 to 10 days in StemPro-34 medium without additional cytokines under ambient oxygen conditions. At day 20, the hPSC-derived CMs were analyzed on the basis of the expression of cTNT via FCM. The EBs were cultured in ultralow attachment six-well dishes (Corning) throughout the differentiation, which routinely generated cultures with greater than 85% CMs, as determined by cTNT expression.

Definitive endoderm differentiation of hPSC lines
HES3-NKX-2.5GFP cell lines were differentiated into definitive endodermal cells using a commercially available kit (PSC Definitive Endoderm Induction Kit, A306260, Thermo Fisher Scientific). Briefly, hPSCs were seeded and grown to 10 to 20% confluence. At day 0, the medium was changed to PSC definitive endoderm induction medium A for 24 hours; after which, the medium was changed to PSC definitive endoderm induction medium B for 24 hours. The cells were then recovered for analysis and SCQC capture experiments. The differentiation routinely generates greater than 95% definitive endodermal cells based on the FCM analysis of SOX17 expression.

Generation of samples containing diluted or spiked hPSCs
Confluent hPSCs (50 to 70%) were dissociated by TrypLE (Thermo Fisher Scientific) for 3 min at RT. Dissociated cells were centrifuged, and the cell number was quantified by an automated cell counter (Countess II, Thermo Fisher Scientific) by taking the average of three to five individual counts. Low concentration solutions were achieved by serial dilution (maximal 9:1 ratio per dilution). Day 20 hPSC-derived EBs were dissociated to single cells by collagenase type 2 (300 U/mg; Worthington Biochemical Corp., USA) in Hanks’ buffer (Thermo Fisher Scientific) at 37°C for 90 min, followed by 3 min TrypLE treatment. Confluent hPSCs (50 to 70%) were dissociated by TrypLE for 3 min at RT, quantified by the cell counter, and serially diluted to achieve low concentrations of hPSCs. Populations of hPSCs and hPSC-derived CMs were combined together in the end to generate spiked samples containing 0.0005 to 5% HES2 cells in CMs. Total number of cells for each experiment is indicated in the figure captions.

Flow cytometry
For surface marker analyses, diluted or spiked samples were fixed by 4% methanol-free paraformaldehyde (PFA; Thermo Fisher Scientific) at RT for 10 min, blocked by 1% BSA (Sigma-Aldrich) in PBS (Wisent Bioproducts, Canada) on ice for 30 min, and stained by antibodies of SSEA-1, SSEA-4, TRA-1-60, TRA-1-81, CD324 (E-cadherin), CD326 (EpCAM), CD9, or CD90 (all from Miltenyi Biotec, Germany) for 10 min at 4°C in a flow buffer containing 1% BSA in PBS. For intracellular marker analyses, samples were fixed by 4% PFA at RT for 10 min, permeabilized by 0.5% Triton X-100 (Sigma-Aldrich) in PBS at RT for 10 min, blocked by 1% BSA in PBS on ice for 30 min, and stained by antibodies of SOX2, Oct3/4, Nanog (all from Miltenyi Biotec), or cTNT (BD Biosciences) for 30 min at RT in flow buffer. Detailed information regarding conjugations and dilutions is given in table S3. Stained samples were analyzed using the FACSCanto flow cytometer (BD Biosciences, USA) or the fluorescence-activated cell sorting (FACS) LSR Fortessa flow cytometer (BD Biosciences). Data were analyzed using FlowJo software (FlowJo LLC., USA). To characterize the LOD of FCM, three individual tubes were prepared for each concentration. The LOD was defined as means + 3 SD.

Droplet digital PCR
Total RNA was isolated from the spiked samples by using a single-cell RNA purification kit (51800, Norgen Biotek Corp., Canada) and used for ddPCR. The isolated RNA was used for cDNA synthesis using the First-Strand DNA Synthesis Kit (Invitrogen, USA), which contained random hexamer primers and Superscript III Reverse Transcriptase, according to the manufacturer’s protocol. The cDNA was submitted to the Centre for Applied Genomics (The Hospital for Sick Children, Toronto, Canada) for a standard ddPCR performed by a QX200 ddPCR system (Bio-Rad, USA). The TaqMan primers for target genes were purchased from Thermo Fisher Scientific: POU5F1 (OCT3/4, Hs00999634_gH), SOX2 (Hs04234836_s1), and CD326 (EpCAM, Hs00901885_m1). The TBP or B2M gene was used as the housekeeping control. The results were analyzed by the Centre for Applied Genomics using QuantaSoft Analysis Pro Software (Bio-Rad).

Characterization of magnetic labeling
Diluted samples were fixed by 4% PFA at RT for 10 min and labeled by anti–TRA-1-60 (dilution: 1:50; Miltenyi Biotec) in 1 ml of 1% BSA for 30 min at RT. Labeled samples were washed with 1% BSA in PBS twice and centrifuged at 2000 rpm for 4 min to form pellets. Pellets were then dehydrated with increasing concentrations of ethanol at 10-min intervals and embedded with Quetol-Spurr resin (Sigma-Aldrich) overnight. Samples were sliced to 70- to 80-nm-thick layers by an ultramicrotome (Ultracut RMC MT6000, Leica Microsystems, Germany) and deposited on electron microscopy grids (Ted Pella Inc.). Samples were observed under a transmission electron microscope (FEI Tecnai 20, Thermo Fisher Scientific) using 100-kV accelerating voltage.

Stem cell quantitative cytometry
Diluted or spiked samples were fixed by 4% PFA at RT for 10 min and labeled by anti–TRA-1-60 or anti-CD326 microbeads (dilution: 1:50; Miltenyi Biotec) in 1 ml of flow buffer for 30 min at RT. Labeled samples were loaded into the chips and profiled at flow rates ranging from 2 to 10 ml/hour. For the quantification of capture and depletion efficiency, captured cells were stained by DAPI and NucDead 488 (Thermo Fisher Scientific) for 10 min at the flow rate of 1 ml/hour. For the quantification of spiked hPSCs in hPSC-derived CMs, captured cells were permeabilized by 0.5% Triton X-100 in PBS at RT for 10 min at the flow rate of 1 ml/hour and stained by cocktails of antibodies [DAPI, NucDead 488, Oct3/4–PE (phycoerythrin), and Nanog–APC (allophycocyamin)] for 30 min at RT in a buffer containing 1% BSA and 0.1% Tween 20 (Bio-Rad, USA) at the flow rate of 400 μl/hour. Detailed information regarding antibody dilutions is given in table S3. After staining, the cells were washed with flow buffer for 10 min at the flow rate of 1 ml/hour. Washed chips were stored at 4°C and scanned within a week of profiling. To quantify the number of captured hPSCs, the chips were tile scanned using a Nikon Ti-E microscope with automated stages. The exposure time is 20 ms for DAPI, 10 ms for NucDead 488, 200 ms for Oct3/4-PE, and 400 ms for Nanog-APC. Scanned images were combined into a large image using Nikon NIS-Elements software (high content analysis version) and quantified using IMARIS software (Bitplane, Oxford Instrument, UK) via colocalization analysis. Cells (hPSCs) were defined as DAPI+, NucDead+, Oct3/4+, and Nanog+. To characterize the LOD of SCQC, three to five individual runs were performed for each concentration. The captured cell numbers were divided by 0.85 to normalize the effect of capture efficiency. The LOD was defined as means + 3 std.

Magnetic-activated cell sorting
For live cell separation, 1 million of HES2 hPSCs or derived CMs were labeled by anti–TRA-1-60 microbeads (dilution: 1:50) in 1 ml of flow buffer for 10 min at RT, as instructed by the manufacturer. Labeled samples were applied to MS columns (Miltenyi Biotec) and washed twice using the flow buffer. TRA-1-60–positive cells were recovered from the column by firmly pushing the plunger into the column twice. Recovered cells were centrifuged and immediately processed for cell counting using an automated cell counter. For stained cell separations, 1 million of HES2 PSCs or derived CMs were fixed, permeabilized, and stained by DAPI, Oct3/4-PE, and Nanog-APC. These cells were subsequently labeled with anti–TRA-1-60 microbeads (dilution: 1:50) in 1 ml of flow buffer for 10 min at RT. Labeled samples were then sorted using MS columns. Recovered cells were centrifuged and immediately processed for cell counting.

Teratoma formation and analysis
All animal experiments were carried out in accordance with the protocol approved by the University of Toronto Animal Care Committee. Male NOD/SCID/interleukin 2 receptor Gamma chain null (NSG) strains of mice at 6 to 8 weeks of age were purchased from the Jackson laboratory (USA) and maintained at the University of Toronto animal facility. Spiked sample with 1 × 106 cells in 15 μl of Matrigel (Corning) was injected into the pericardium of testis. Ten weeks after injection, mice were euthanized, and the formation of teratomas was examined. Extracted teratomas were weighed and fixed in 10% formalin (Sigma-Aldrich). Formalin-fixed, paraffin-embedded teratomas were sectioned (5-μm thickness) and stained with hematoxylin and eosin. Histological examination was performed by a licensed veterinary pathologist blinded to the difference of samples to identify the germ layers in the teratomas.

In vitro colony-forming assay for spiked samples
Spiked samples were labeled by anti–TRA-1-60 microbeads in 1 ml of 1% BSA in PBS for 30 min at RT. Labeled samples were loaded into the chips and profiled at flow rates of 4 ml/hour. After profiling, magnets were removed from the chip. The negative groups were obtained from the syringe, and the positive groups were obtained by withdrawing cells in the chips with a new syringe. The profiled groups were centrifuged and resuspended in the hPSC culture medium for reculturing on vitronectin-coated well plates. At days 3, 6, 10, and 15 after profiling, the recultured cells were fixed by 4% PFA at RT for 10 min, permeabilized by 0.5% Triton X-100 in PBS at RT for 10 min, and stained by antibodies of DAPI, TRA-1-60-Vio488 (Miltenyi Biotec), Oct3/4-PE, and Nanog-APC for 60 min at RT in flow buffer. Detailed information regarding conjugations and dilutions is given in table S3. The plates were tile scanned using the Nikon Ti-E microscope. The exposure time is 20 ms for DAPI, 100 ms for TRA-1-60-Vio488, 200 ms for Oct3/4-PE, and 400 ms for Nanog-APC. hPSCs were defined as DAPI+, NucDead+, Oct3/4+, and Nanog+. Number of colonies (>4 hPSCs per colony) per well was quantified manually.

Isolation and characterization of rare hPSCs isolated from manufactured batches
In addition to the abovementioned EB-based protocol, a commercially available cardiac differentiation kit (A2921201, Thermo Fisher Scientific) was also used to generate batches of CMs from a monolayer. Briefly, HES2 and HES3-NKX-2.5GFP cells were maintained in vitronectin-coated well plates in Essential 8 medium (Thermo Fisher Scientific) for 2 days (days −2 to 0). The confluency of cells at day 0 is between 50 and 70%, as suggested by the manufacturer. Then, the medium was replaced with CM differentiation medium A and B at days 0 and 2, respectively. At day 4, the medium was changed to Cardiomyocyte Maintenance Medium (A2920801, Thermo Fisher Scientific) and changed every 2 days until day 16. Contracting CMs appeared at day 8, and the beating conditions of the CMs were monitored at days 8, 12, and 16 using the Nikon Ti-E microscope. The percentages of cTNT-positive cells at days 8, 12, and 16 were quantified by FCM using the protocol described in the “Flow cytometry” section. At days 4, 8, 12, and 16, monolayers of hPSC-derived CMs were dissociated by TrypLE for 4 min. At days 10 and 20, hPSC-derived CMs grown as EBs were dissociated to single cells by collagenase type 2 (300 U/ml) in Hanks’ buffer at 37°C (30 min for day 10 and 90 min for day 20), followed by 3-min TrypLE treatment. Dissociated cells were labeled by anti–TRA-1-60 microbeads and profiled using the protocol same to the spiked samples. The positive groups were recultured in StemFlexTM medium (A3349401, Thermo Fisher Scientific). At days 2, 6, and 10 after profiling, the positive cells were fixed, stained, and quantified using the same protocol used for the spiked samples. If rare hPSCs (DAPI+, TRA-1-60+, Oct3/4+, and Nanog+) were found at day 2, the same groups were passaged when it reached 80 to 90% confluency up to 14 days to allow rare hPSCs to proliferate. To examine the pluripotency of isolated rare hPSCs, proliferated hPSCs were differentiated into three germ lineages using a trilineage differentiation kit (130-115-660, Miltenyi Biotec), which typically takes 7 days. At day 7, the cells were fixed, permeabilized, and stained with DAPI, FOXA2, and SOX17 (for endoderm); DAPI, smooth muscle actin (SMA), and CD144 (for mesoderm); and DAPI, PAX6, and Nestin (for ectoderm). Detailed information regarding conjugations and dilutions is given in table S3. The stained plates were observed using the Nikon Ti-E microscope. The exposure time is 20 ms for DAPI, 100 ms for FOXA2-PE, 400 ms for SOX17-AF647, 20 ms for SMA-PE, 400 ms for CD144-AF647, 200 ms for PAX6-PE, and 600 ms for Nestin-AF647. To examine the naïveness of isolated rare hPSCs, total RNA was isolated from the proliferated hPSCs following the same protocol used for ddPCR. A comparative CT experiment was performed on an Applied Biosystems 7500 Real-Time PCR System (Thermo Fisher Scientific) using hPSC naïve-state qPCR array (07521, Stemcell Technologies, Canada). The assay was carried out using 5 μl of TaqMan Universal Mix, 4 μl of nuclease-free water, 1 μl of cDNA (10 ng/μl) for each sample in a 96-well plate. Cycling conditions for the qPCR were 95°C for 10 min, followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. The post-PCR analysis was performed by an online tool provided by the manufacturer (https://stemcell.shinyapps.io/qpcr_tool/).

Acknowledgments

We would like to thank members of the Kelley and Keller laboratory, especially B. Green, X. Fan, A. Garcia, S. Ogawa, and S. Protze, for experimental advice and critical comments on the manuscript, A. Elefanty and E. Stanley (Monash University) for providing the HES3-NKX-2.5GFP reporter cell line, M. Ly, K. Patel, and H. Patel (Creative CADworks) for establishing the protocol for 3D printing, T. Paton (The Hospital for Sick Children) for assistance in ddPCR, M. Ganguly (University Health Network) for assistance in histology, and M. Larsen (Mbed Pathology) for assistance in pathology. Funding: Research reported in this publication was supported in part by the Canadian Institutes of Health Research (grant FDN-148415 to S.O.K. and grant FDN-159937 to G.M.K.). This research is part of the University of Toronto’s Medicine by Design initiative, which receives funding from the Canada First Research Excellence Fund. Z.W. was supported by a Connaught International Scholarship. Author contributions: Z.W., M.G., E.H.S., G.M.K., and S.O.K. conceived and designed the experiments. Z.W., M.G., R.M.M., S.U.A., M.L., L.Z., S.P., and Y.Z. performed the experiments and analyzed the data. All authors discussed the results and contributed to the preparation and editing of the manuscript. Competing interests: G.M.K is a founding investigator, equity holder, and a paid consultant for BlueRock Therapeutics LP and a paid consultant for VistaGen Therapeutics. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from S.O.K.

A rapidly prototyped lung-on-a-chip model using 3D-printed molds

by

A rapidly prototyped lung-on-a-chip model using 3D-printed molds

Jesus Shrestha, Maliheh Ghadiri, Melane Shanmugavel, Sajad Razavi Bazaz, Steven Vasilescu, Lin Ding and Majid Ebrahimi Warkiani

Organ-on-a-chip is a microfluidic cell culture model that replicates key organ-specific microarchitecture and pathophysiology in vitro. The current methods to fabricate these devices rely on softlithography, which is usually tedious, laborious, and requires adroit users as well as cleanroom facilities. Recently, the use of 3D-printing technologies for the rapid fabrication of molds for polydimethylsiloxane (PDMS) casting is on the rise. However, most of the 3D-printed materials are unsuitable for PDMS casting. To address this issue, we have improved the existing techniques and introduced a modified protocol for the surface treatment of 3D-printed molds, making them ideal for repeated long-term PDMS casting. Using this protocol, we have fabricated a simple open well lungon-a-chip model to simulate the in vivo environment of airway at air-liquid interface under dynamic condition. To validate the functionality of the developed chip, Calu-3 cells were cultured in the chip and maintained at an airliquid interface. The model demonstrated that the cultured cells replicated the 3D culture-specific-morphology, maintained excellent barrier integrity, secreted mucus, and expressed cell surface functional P-glycoprotein; all indicative of a promising in vitro model for permeability assays, toxicological tests, and pulmonary drug delivery studies. To validate the suitability of this lung-on-a-chip in vitro model, the effects of cigarette smoke extract (CSE) on Interleukin-6 (IL-6) and Interleukin-8 (IL-8) release from cultured Calu-3 cells were examined. CSE treated cells showed significantly higher secretion of IL-6 and IL-8 over 24 h compared to the cells treated with both CSE and Budesonide, an anti-inflammatory drug. Moreover, our results illustrated that CSE reduced the expression of Ecadherin as an adherent junctional protein. In conclusion, the proposed protocol demonstrated an easy and lowcost fabrication technique which will allow a biologist with minimal technical skills to rapidly prototype molds for different/versatile organ-on-a-chip models

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

Introduction

The burden of respiratory diseases throughout the world is on the rise, with almost 4 million premature deaths from chronic respiratory diseases every year, indicating a serious health issue (Forum of International Respiratory Societies, 2017). Therefore, there is an urgent need to develop new respiratory drugs, which require developing better and physiologically relevant respiratory disease and drug testing models (Shrestha et al., 2020). The conventional methods used to reproduce the microenvironment, functions, and physiology of the human lung are two dimensional (2D) monolayer cell culture and transwell-based immersion cultures (Huh et al., 2011). However, these models do not accurately replicate in vivo three-dimensional (3D) cellular structure, cell-cell interactions, the air-exposed environment of the alveolar cells, and the physiological functions at the organ level.

The inability of existing cell culture models to accurately and reliably model respiratory disease has led to the emergence of microfluidic cell culture systems. Microfluidic systems precisely replicate the physiological conditions required for both basic research and drug development, enabling drug discoveries through systematic testing (Neuzil et al., 2012; Azadi et al., 2019). Several microfluidic organ-on-a-chip models replicating the 3D microarchitecture and mechanical and physiological features of different organ tissues have been developed (Esch et al., 2015). However, conventional softlithography techniques used to fabricate organ-on-a-chip devices impose severe limitations on time, geometric complexity, and cost, often requiring a separate cleanroom facilities and adroit users, as a result of which, impedes the pace of development and innovation in microfluidic applications. Therefore, a method for rapid fabricating microfluidic devices with little technical expertise would be ideal for scientists with little knowledge of microfabrication.

Recently, additive manufacturing has emerged as an alternative for the fabrication of microfluidic devices (Vaezi et al., 2013; Erkal et al., 2014; Bhattacharjee et al., 2016). With further developments in 3D-printing, there has been a significant increase in the utilization of 3D-printed part as molds to generate PDMS replicas. 3D-printing has superiority over other techniques of mold fabrication such as micro-milling or laser cutting, enabling higher resolution features and a control over height parameters (Guckenberger et al., 2015; Condina et al., 2019). Among all 3D printing methods, stereolithography apparatus (SLA) and digital light processing (DLP) offer great advantages and are therefore preferred for microfluidics and biomedical applications (Macdonald et al., 2017). This allows researchers to rapidly design and alter complex microstructures without expending large amounts of time or resources. In addition, 3D-printed molds are more adaptable as they can include multiple design features at different channel heights not possible using conventional photolithography techniques. However, most of the 3D-printed molds printed via SLA/DLP techniques are not immediately suitable for PDMS casting since residual monomers and oligomers on the surface of the 3D-printed parts impede PDMS polymerization. Several groups have proposed surface treatment methods for 3D-printed molds to make it suitable for casting PDMS, including ink or lubricant infused coatings, plasma treatment, salinization, and heat cycling (Waheed et al., 2017; Chan et al., 2015; Comina et al., 2014; Villegas et al., 2018). Reported protocols are time-consuming, labor-intensive, and lack reproducibility (Razavi Bazaz et al., 2019). Although many protocols contain similar basic parameters such as treatment time, curing temperature, and UV exposure, these parameters are subject to change according to feature dimensions. Some groups use the services provided by commercial 3D-printing companies to manufacture their molds (Park et al., 2019; Novak et al., 2018). However, these molds can be expensive and have waiting periods associated with manufacturing and delivery (Ellison et al., 2016). This type of arrangement does not allow rapid prototyping to be performed, where quick changes in design parameters are required. Curing temperature of PDMS on the 3D-printed molds is a vital step that requires careful optimization to prevent material strain and microstructure deformation. Therefore, it is crucial to develop an optimized surface treatment process for 3D-printed molds ensuring long-term cell viability in the organ-on-a-chip devices.

Herein, we present an enhanced protocol for surface treatment of 3Dprinted molds to fabricate a simple open access lung-on-a-chip design. The protocol presented here enables the quick fabrication of molds for long-term use without the development of any cracks or channel deterioration through carefully optimized steps and parameters. This treatment process allows high-resolution repetitive PDMS casting using 3Dprinted molds. The fabricated chip was further optimized by testing different membranes and ECM coatings for cell growth and extended viability. The chip allowed the lung epithelial cells to be cultured at an air-liquid interface under dynamic conditions; the transparency of PDMS enabled real-time cell visualization and chip monitoring (Fig. 1). Calu3 cells are known to highly express the tight junction proteins Occludin and E-cadherin, which makes it a suitable cell type for analyzing tightjunction formation and cell barrier functions (Kreft et al., 2015; Haghi et al., 2010). Mucus production, differentiation, and the expression of transport proteins are other features of Calu-3 cells, making them useful for modelling the airway epithelium. Different study groups have already shown the suitability of this cell line to use it as a respiratory in vitro model (Zhu et al., 2010; Foster et al., 2000; Florea et al., 2003). Thus, the Calu-3 cell line was chosen for our lung-on-a-chip device. Using Calu-3 cells, we demonstrate the versatility of our lung-on-a-chip model through the assessment of CSE effects and Budesonide treatment on the secretion of inflammatory markers and cellular expression of the junction protein E-cadherin. Furthermore, we provide a functional analysis of the epithelium cell layer generated. The flexibility of direct 3D-printing utilized in this study will aid the fabrication of novel organ-on-a-chip designs within a short time frame. The printing process we report here is versatile enough to be adapted for multiple organ-on-a-chip models beyond what is studied here.

Fig. 1. Microfluidic model of human Lung-on-achip design and fabrication: A) A conceptual schematic of the experimental setup showing the human respiratory system B) A cross-section of human airway tissue. C) 3D printer was used to fabricate the open well design of the chip model with upper and lower layers to recapitulate human lung. D) The top layer contains a central open well for cell seeding, and an inlet and an outlet for media in the lower channel. The lower layer includes a channel for media flow. The porous PC membrane is carefully placed and aligned between the two layers, where the cells attach and grow. E) Once the cells were confluent, the effects of CSE on the cells were analyzed.


2. Materials and methods
2.1. Mold fabrication and surface treatment
The mold was fabricated using a digital light processing (DLP) 3Dprinter, (MiiCraft Ultra 50, MiiCraft, Hsinchu, Taiwan) with a printing area of 57 X 32 X 120 mm and XY resolution of 30 μm. The printer projects a 385–405 nm UV wavelength through the resin (BV-007) on the resin bath. The design process began with Computer-Aided Design (CAD) modelling of the required geometries in SolidWorks (2016), after which designs were exported in an STL file format to the Miicraft printer software (MiiCraft 125, Version 4.01, MiiCraft Inc). To enable highresolution printing of mold features, the print options were carefully tailored to each design. Smaller design features required a slice thickness of 10 μm and a curing time of 1 s per slice, where features were less fine, slice thicknesses of 30 and 50 μm were used. Considering the size of the printed molds, a base layer was used to ensure the part adhered to the picker for the duration of the print. The curing time for the base layer was set to 24 s. A buffer layer was used to facilitate the transition from the base layer into the printed part.

To prevent the PDMS from sticking to the 3D-printed mold, surface treatment of the resin mold is mandatory. We have optimized the surface treatment method proposed before (Waheed et al., 2017) to shorten the overall duration, making it suitable for channels with smaller dimensions and with the ability of repeated casting without affecting the mold structures. First, the 3D-printed mold was washed with isopropanol (IPA) followed by high-pressure air drying after the print. After that, the post-curing of the mold is required for 3 min (steps of 10 s). To make sure that uncured monomers and oligomers on the surface of the mold are eliminated, the prepared mold was then soaked inside 100% ethanol for 2 h. The surface of the mold must be prepared for the step of an easy detachment of the PDMS; hence, Oxygen plasma treatment (Basic Plasma cleaner PDC-002, Harrick Plasma) was carried out for 2 min. In the end, surface of the mold was silanized using trichloro (1H, 1H, 2H, 2H-perfluoro-octyl) silane (Sigma-Aldrich, Australia) in a desiccator under vacuum for 1.5 h. The workflow for mold design and fabrication is illustrated in Fig. 2.

2.2. Flow simulation and diffusion in the channels
To characterize how the microchannel impacts fluid flow and its effect on the membrane, a 2D model of the proposed design was analyzed using Comsol multiphysics 5.3a, a commercial CFD package (Bazaz et al., 2018). To study the effects of fluid flow on the membrane, a coupled equation in free and porous media flow is mandatory. The flow in the microchannel is described using continuity and Navier-Stokes equations.

Here, u represents the vector of fluid velocity, ρ denotes fluid density, μ refers to dynamic viscosity, and p is the pressure. In this study, Brinkman equations are used for the momentum transport over the membrane.

where εp is porosity, κ is permeability, Qbr denotes mass source or mass sink, and F equivalents to the external forces. Using an oversimplified Darcy's law results in the neglection of the viscous effect which arises from the fluid media. As an alternative, the Brinkman equation was employed to solve the fluid within the porous media. A hydrophilic Isopore polycarbonate membrane (Sigma-Aldrich, Australia) is used in this study. The pore size of the membrane (dpÞ is 0.4 μm while the porosity ðεp Þ is 15%, and membrane thickness is 10 μm. The membrane permeability can be calculated via a packed bed model, and its value

Fig. 2. Protocol for the surface treatment of the 3D printed resin molds: A) CAD design of the desired mold using Solidworks. B) The finalized designs are printed using the 3D printer. C). The mold is then washed with IPA, followed by UV curing. D) and E). The mold is then dipped in ethanol before plasma treating. F) and G) Before casting PDMS, the mold is silanized. Once the PDMS is cured, it is carefully cut out from the molds to form two slabs. The PDMS slabs are then cleaned with alternative IPA and Ethanol washing before being plasma treated, aligned, and bonded carefully to make a complete lung-on-a-chip adjusted to 4.98e-18 m2 and 4.98e-19 m2 in the absence and presence of the cells on the membrane, respectively. The physics-controlled mesh was used as the domain grid in this study. The normal inflow velocity was applied at the inlet, while zero static pressure was considered at the outlet.

2.3. Device fabrication
Following surface treatment of the resin mold, a mixture of PDMS base and curing agent (ratio of 10:1) was prepared (Sylgard 184 from Dow Corning, MI, USA) (Fig. 2). The mixture was then degassed in a desiccator for 20 min and poured into the treated molds. The PDMS was left to cure in the hot air oven for 4–5 h at 45 C. The PDMS layer was gently lifted off from the mold, trimmed to desired shapes, and holes are punched for inlets and outlets. The PDMS layers were thoroughly cleaned with IPA and ethanol, at least three times, followed by air drying between each wash. Polycarbonate membrane was cut into a square shape, large enough to cover the central circular well and carefully placed on the lower PDMS layer. The contact surface of both the upper and lower PDMS layers was plasma treated for 1 min and aligned precisely to ensure perfect bonding. The bonded chip was then kept in the oven again for 4 h at 45 C to bond, followed by testing for leakage using food dyes.

2.4. Cell culture
Human airway epithelial cell line, Calu-3 from American Type Cell Culture Collection (ATCC, Rockville, IN, USA) were cultured in Dulbecco's Modified Eagle's medium: F-12 (DMEM: F12) supplemented with 10% fetal bovine serum (FBS), 1% L-glutamine, 1% penicillin ( Gibco, Life Technologies, Australia) and 1% non-essential amino acids (SigmaAldrich, Australia). The chip was sterilized by flushing 70% ethanol through the channels, followed by hot air drying in the oven. It was then exposed to UV light in a biosafety cabinet for 30 min before the extracellular matrix (ECM) coating. Once the membrane was coated with ECM, the chip was kept in a humidified incubator at 37 C with 5% CO2 for 2 h before seeding cells. Residual ECM coating in the channels was removed by passing fresh culture media. After seeding with cells, the chips were maintained in 37 C and 5% CO2. The media in both the upper and lower channels were replaced daily until the cells were confluent (day 5). Once the confluency was achieved, the media from the upper channel was aspirated to allow cells to grow in the air-liquid interface. The lower channel was then attached to a syringe pump (Fusion 200, Chemyx Inc.) with a flow rate of 30 μl/h to maintain a dynamic condition with fresh media flow.

2.5. Cell adhesion to the membrane and ECM coating
Different combinations of membranes and ECM coatings were tested to guarantee optimal cell growth and the formation of a confluent monolayer. The different membranes tested were Isopore membrane Polycarbonate filter (PC) (Merck, Australia), Nuclepore track-etch PC membrane (Whatman, Australia), Polyester (PE) from transwell cell culture inserts (Corning Incorporated, Australia), and Millipore from Millicell culture plate inserts (Sigma- Aldrich, Australia), all with 0.4 μm pore size. The membranes were first tested without any ECM coating. Calu-3 cells were seeded at 8000 cells/mm2 and incubated at 37 C, 5% CO2. Media in both channels were changed every day. The cells were stained with a working concentration of 1 μg/ml of Hoechst 33342 (Abcam, Australia) and incubated for 15 min on day 5. After washing with phosphate-buffered saline (PBS), it was observed under the microscope to identify the most suitable membrane with the highest level of cellular attachment. To further enhance the cell attachment, selected membrane was coated with different ECMs: (i) 5 μg/ml of Fibronectin (Corning, 356008), (ii) 3 μg/μl of Collagen-I (Col-I) (Corning, 354236), (iii) 6 μg/μl of Matrigel (Corning, 354234) and (iv) a mixture of 6 μg/μl Matrigel and 3 μg/μl Col-I. This was followed by incubation for 2 h at 37 C, 5% CO2, and washing with PBS (Sigma- Aldrich, Australia) to get rid of excess ECM. Cells were seeded and incubated. The media from both channels was changed the next day. The cells were then stained with Hoechst 33342 on day 5 and observed with Olympus Ix73 Inverted Microscope for cellular attachment to determine the optimal ECM coating for Calu-3 cell culture.

2.6. Viability and functionality testing of the device
After the selection of the membrane and ECM coating, the following experiments were conducted in the final chip design.

2.6.1. Cell viability and proliferation
Live/dead cell double staining kit (Sigma-Aldrich, Australia) was used to distinguish viable cells from non-viable cells. The assay solution of the stain was prepared by adding 10 μl of Calcein-AM (Solution A) and 5 μl of Propidium Iodide (Solution B) in 5 ml of PBS. The cells were washed with PBS before adding the assay solution. 30 μl of the assay solution was added to the cell layer in the upper channel and incubated at 37 C for 15 min. Olympus Ix73 Inverted Microscope was used to simultaneously observe live and dead cells.

2.6.2. Mucus staining
The mucus production from the Calu-3 cells in the chip was characterized on day of 3, 5, 7, 9, and 11 of culture by staining the glycoprotein in the mucus using Alcian blue (1%. (w/v) in 3% (v/v) acetic acid/water at pH 2.5) (Sigma- Aldrich, Australia). The monolayer of the cells was washed twice with PBS. After fixing the cells with 4% (v/v) Paraformaldehyde for 20 min, PBS wash was repeated. Finally, 50 μl of Alcian blue stain was added. The chips were incubated for 15 min and washed multiple times with PBS until the rinsate was clear. Images were obtained using an Olympus Ix73 Inverted microscope. Depending on the ratio of red, green, and blue (RGB) from the microscopic images of mucus staining, data were analyzed to generate a semi-quantitative estimate of mucus concentration (Haghi et al., 2010). The mean RGB values were obtained with Image J (v1.52p, NIH) with Color Profile (Dimiter Prodanov; Leiden University Medical Centre, Leiden, Netherlands). The mean RGBB was divided by the total sum of RGB values for each image (RGBR þ RGBG þ RGBB) to calculate the ratio of blue (RGBB ratio). The mean RGBB of eight images was used to quantify the secretion of mucus by the Calu-3 cells for days 3, 5, 7, 9 and 11.

2.6.3. Paracellular permeability of sodium fluorescein (Flu-Na)
 The barrier integrity of the cells grown on the membrane was assessed by using Flu-Na (MW 0.367 kDa, Sigma-Aldrich, Australia) on day 7. The cell layer in the upper channel was washed with warm PBS (37 C) after removing the media from both the channels. The lower channel was filled with prewarmed PBS, and the cells were incubated at 37 C for 2 h. The upper channel was filled with 50 μl of flu-Na solution (2.5 mg/ml), while PBS was flushed through the lower channel at 0.5 μl/min for both blank devices and seeded chips. Samples were collected from the lower channel every 30 min for a total of 120 min. The fluorescence of flu-Na present in each sample was measured in Corning full black clear bottomed 96-well plates using a fluorescence plate reader (Infinite 200 PRO; TECAN), using excitation and emission wavelengths of 485 and 520 nm, respectively.

2.6.4. Flow cytometry detection of cell surface P-Glycoprotein
On day 7, the Calu-3 cells were harvested from the membrane of the chip by trypsinization (TrypLE Express; Gibco). After washing twice with PBS, the cells were labelled using 20 μl of FITC-anti P-gp (clone17F9, BD Pharmingen, USA). It was then incubated in the dark at room temperature for 30 min. The cells were washed twice again with PBS before resuspending them in 200 μl of PBS, spiked with (1 μg/ml) propidium iodide (PI) (Sigma- Aldrich, Australia). Samples were analyzed by flow cytometry (FCM) using the CytoFLEX LX (Beckman Coulter, Life Sciences, USA) and CytExpert Software. 

2.7. Preparation of CSE and treatment of the cells
Cigarette smoke extract (CSE) was prepared by a method modified from a publication (Laurent et al., 1983). One Marlboro Red cigarette (Philip Morris, Victoria, Australia) was bubbled through 25 ml of DMEM in a T-75 flask at a constant rate, which was considered as 100% concentration CSE. The collected CSE was filtered and diluted to the required CSE concentrations in the media. Prepared CSE was immediately used and diluted within 30 min. To stimulate Calu-3 cells with a non-toxic concentration of CSE, first cytotoxicity assay was performed. Briefly, Calu-3 cells were cultured in 96-well plates were stimulated with serial dilutions of CSE from 0.0675% to 100%. The cytotoxic concentration of CSE on Calu3 cells over 48 h was tested using MTS assay kit (Promega, CellTiter 96® AQueous One Solution Cell Proliferation Assay- Australia). Then, cells grown on the chip were stimulated with appropriate concentration of CSE (the concentration below IC50 (Inhibitory concentration at 50%). The experiment was categorized into four groups of chips, and the cells were grown as mentioned previously for 7 days. One group was first treated with 100 nM Budesonide for 24 h from the lower chamber and then treated with CSE from the top (Bud-CSE). The second group was treated with CSE and 100 nM Budesonide at the same time (CSE-Bud). The third group was only treated with CSE and the fourth group was the control (no treatment). The lower channel was connected to a syringe pump and media was collected in a tube for 24 h; this media was used to measure IL-6 and IL-8 with ELISA technique. The media collected from the chips were stored at -80 C.

2.8. Enzyme-linked immunosorbent assay (ELISA) for IL-6 and IL-8
The media collected from the chips were stored at 80 C until performing the experiment. After thawing them, the levels of secreted IL-6 and IL-8 were analyzed according to the manufacturer's instructions, using commercial human IL-6 and IL-8 ELISA kits (BD Pharmingen, San Diego, CA, USA). The absorbance was read at 450 nm/570 nm using plate reader (Infinite 200 PRO; TECAN).

2.9. Immunofluorescence staining
To visualize the effects of CSE and Budesonide treatment on the tight junctions, the Calu-3 cells were assessed by imaging of immunolabelled stains of tight junction protein, E-cadherin. Images were taken using a Nikon A1 confocal microscopy (Japan). All immunostaining steps were conducted at room temperature. The cells were fixed with 4% volume/ volume (v/v) paraformaldehyde (Sigma- Aldrich, Australia) in PBS after washing three times with PBS. It was then incubated for 15 min and washed with PBS twice. This was followed by permeabilization of the cell membranes with 0.1% (v/v) Triton X 100 (Sigma- Aldrich, Australia) for 10 min and blocking with 1% (w/v) bovine serum albumin (BSA) (SigmaAldrich, Australia) in PBS for 1 h. After further washing with PBS, cells incubated with 50 mM ammonium acetate in PBS for 10 min. Then cells were washed with PBS and incubated at 37 C with CD324 (E-Cadherin) monoclonal antibody (10 μg/ml in PBS) (Invitrogen) for 1 h. Rewashing with PBS was done before adding AlexaFlour 594 goat anti-mouse IgG1 (10 μg/mL in PBS) (Invitrogen, Australia) to incubate for 1 h. After washing with PBS, the cells were counterstained with 1 μg/mL 40 , 6-diamidino-2- phenylindole (DAPI) in water for 10 min before washing with PBS. The membrane was carefully cut from the open well of the chip and mounted on a microscopic glass slide and covered by a coverslip. Care was taken to prevent curling and tearing of the membrane, and the slides were stored at 4 C. The slides were viewed the next day using the Nikon A1 Confocal Laser Microscope with NIS-Elements C Software.

2.10. Statistical analysis
Data were analyzed using IBM SPSS Statistics 25 software (USA). The ANOVA one-way analysis was used to determine significance (P < 0.05). All results are expressed as the mean standard deviation (SD) of at least three independent determinants.

3. Results and discussion

3.1. Fabrication of the mold and the device
Lung-on-a-chip models typically have closed system designs with straight channels (Huh et al, 2010, 2012, 2013; Jain et al., 2018; Benam et al., 2016). Although this allows uniform flow, manipulating cell suspensions and micro-volumes of fluids through microchannel is challenging. Functional tests such as immunostaining, mucus secretion, and permeability assays become difficult to perform within a closed design chip. The open well design of our device allows easy access to the membrane for uniform coating, cell seeding, fluid manipulation, and sample collection. Moreover, the open access design allows the cells to be directly exposed to the CSE, drugs or nanoparticles, making transport or migration studies more controllable. Multiple open well chambers can be interconnected to each other to conduct independent as well as parallel studies. This was first suggested by Blume et al. who developed a similar open well design compatible with commercially available Transwells to be interconnected. (Blume et al., 2015). However, this is a complicated process compounded by the number of components comprising their chip design. We have effectively simplified the lung-on-a-chip model to facilitate the on chip cell culture with the potential for multichip interconnections. The simple maintenance and usability of our chip will allow a person familiar with conventional cell culture methods to conduct their experiments in a more relevant microenvironment, even with minimum microfluidic knowledge.

The open well lung-on-a-chip model presented here consisted of an upper PDMS layer with a large circular well, an inlet, and an outlet on either side (Fig. 3A). The lower layer comprised of a central chamber connecting to two straight channels from either side. The thin, porous PC membrane separated the two PDMS layers. The 3D-printed mold consisted of raised channels to imprint the PDMS surface. Surface treatment of the mold before casting PDMS is vital for PDMS casting (Chan et al., 2015). This treatment cures the surface areas of the mold left uncured from the 3D-printing process. To achieve this, the mold must be free of any debris, residual monomers or oligomers. The process of silanization provides a hydrophobic fluorinated monolayer on the 3D-printed mold that prevents the sticking of the PDMS to the resin mold. This methodology allows easy peeling off the PDMS from the mold. Curing PDMS at 45 C for 4–5 h was found to yield the best surface finish for the 3D-printed resin molds. PDMS incubation with the mold at higher temperatures induced the formation of surface fractures within the mold, negatively impacting the surface finish of the PDMS piece, resulting in leaking of the bonded device. Hence, it can be reasoned that the temperature has a significant influence on the molds.

This approach of fabricating and treating the 3D-printed resin molds is a simple, cost-effective, and time-efficient method for producing 3D microfluidic lung-on-a-chip models compared to the conventional softlithography techniques using silicon wafers. The 3D-printing of the molds, surface treatment, and chip fabrication can be completed within a day. This helps to avoid the typically lengthy processing and delivery time of the commercially manufactured molds. This time reduction is a result of optimized fabrication steps minimizing the risk of human error. Thus, the approach presented here enables researchers to rapidly prototype multiple designs of different geometries within a short period. The molds created can be used repeatedly without concern over the reproducibility between chips. However, despite these advantages, 3D-printing has certain limitations regarding the resolution of printable features and quality of the surface. For instance, ink-jet type 3D-printing is an option available for 3D-printing molds to cast PDMS and fabricate chips (Kamei et al., 2015). Nevetheless, ink-jet 3D-printing can result in increased surface roughness and may cause altercations in channel profiles.

Fig. 3. Simulations of fluid flow in the lung-on-a-chip model: A) Geometry and boundary conditions used for computational modelling. B) Velocity profile along the length of the microchannel. It is evident that the velocity distribution across the upper channel is small enough to not negatively impact attached cells. C) Velocity distribution along the length of the channel. It is shown that the velocity profile in lower channel has a parabolic profile. D) Velocity profile at the upper channel. The order of velocity is small enough so that cells are not influenced by shear rate.

Materials

M50

Learn More

3.2. Fluid behaviour and diffusion simulation in the microchannel
To better quantify the fluid behaviour within the lung-on-a-chip device, the flow velocity was simulated, as shown in Fig. 3B. The numerical results reveal that the flow exists at a steady-state condition, and the relative velocity is higher in the lower channel compared to the upper channel. Streamlines in the lower channel run along the length of the channel from inlet to outlet and shows that the velocity profile in the lower channel is parabolic (Fig. 3B) while, in the upper channel, fluid is driven from the lower channel across the porous membrane before exiting from the membrane. These results indicate that fluid can traverse the membrane, passing between the upper and lower channels continuously. Thus, the velocity distribution along the length of membrane is calculated and illustrated in Fig. 3D. According to the value of velocity in the upper channel, the effect of media flow on the cells growing on the membrane is negligible. This allows the secreted molecules from the cells cultured on the membrane to remain undisturbed in the interstitial fluid since the flow of media is limited to the lower surface of the cell layer (Nalayanda et al., 2009; Walker et al., 2004). This was further validated by mucus staining performed over several days.

3.3. Cell adhesion to the membrane and ECM coating
To determine the optimal membrane and ECM coating combination to support the adhesion and growth of the Calu-3 cells in the device, several types of membranes and ECMs were tested. We evaluated the results based on the cell adhesion and area coverage by cells. Chips were seeded with 8000 cells/mm2 which ensured a viable cell population covering a large area. Based on these findings, 8000 cells/mm2 seeding density was used for all subsequent experiments in this study. On the fifth day of incubation, the chip was observed for cell attachment; to help confirm cellular attachment, the cells were stained with Hoechst (Fig. 4AI). The PE membrane consistently showed large empty areas with minimal cellular attachment. In comparison the Millipore membrane provided greater cell attachment, but the membrane with the most consistent cell attachment was PC (Merck, Australia). Therefore, for further experiments on ECM coatings, PC (Merck, Australia) membranes were used.

To identify which ECM coating achieved the highest level of cellular attachment, cell viability and the formation of a monolayer, multiple PC membranes were coated with Fibronectin, Collagen-I, Matrigel, and a mixture of Matrigel and Collagen-I. A mix of 6 μg/μl Matrigel and 3 μg/μl Col-I was selected based on findings suggested by Humayun et al. (2018). Seeded membranes were then incubated and washed with PBS to remove all detached and dead cells. The cells were then stained with Hoechst to compare the results of different ECM coatings on day 5 (Fig. 4AII). Fibronectin coated chips had few attached cells, with the majority of the membrane left unoccupied. The membrane coated with Collagen-I had comparatively more cells attached but still had large sections void of any cells. Matrigel coating had a significantly larger proportion of cells attached to the membrane with few empty spots. The coating with the highest cell coverage was the mixture of Matrigel and Col-I, similar to the findings of Humayun M., et al. Fig. 4AII illustrates the distribution of area coverage by cells in each membrane and ECM scenario tested. The most confluent monolayer can be observed in the combination of collagen and Matrigel on a PC membrane.

3.4. Testing the functionality of the lung-on-a-chip model
3.4.1. Live and dead staining
To confirm the cell viability, cells grown in the chips for 3, 5, and 7 days were stained with live and dead staining kit. The aim was to note the day for the cells to get fully confluent with minimal dead cells in the chip. The chip was then connected to a syringe pump to provide a constant stream of media in the lower channel, while the cells were cultured at the air interface in the upper channel by removing the media. The number of live cells increased gradually from day 3 and were confluent by day 5. On day 5, the media on the upper channel was removed to create an airliquid interface. The cells were viable and growing well on day 7 i.e., after 48 h at the air-liquid interface (Fig. 4C). Thus, we decided to perform further characterization of the cell layer lung-on-a-chip model

Fig. 4. Calu-3 adhesion in the chip: A) I. Cell adhesion to the membrane A-II Selection of coating ECM on the PC membrane: Once the PC membrane was identified as the suitable membrane, different ECM coating was used to identify the optimal ECM for cell attachment and proliferation. All cells were stained with Hoechst stain. (Scale: 100 μm) B) Cell viability: Live and dead staining of the cells in the chip on day 7 (Scale: 100 μm). 

on day 7. With optimized seeding density and proper ECM coating, the results showed that the chip was able to grow and maintain a viable cell population up to 2 weeks.

3.4.2. Mucus secretion
Airway mucus is an extracellular gel, which traps inspired toxins and carries them out of the lung through ciliary beating and coughing, making the lungs highly defensive to environmental harm (Fahy and Dickey, 2010; Knowles and Boucher, 2002). Calu-3 cells contain submucosal glands, which are a source of airway surface liquids such as mucus, making it an optimal cell line for assessing cellular functionality (Knowles and Boucher, 2002). Alcian blue staining enabled the detection of secreted mucus in Calu-3 cells (Haghi et al., 2010; Inglis et al., 1998). A film of mucus was seen on the surface of the Calu-3 cell layer after day 3 (Fig. 5A). The representative microscope images shown indicate an increase in mucus secretion over the 7 days of growth, as the blue staining became progressively darker, covering more areas of the membrane. The amount of mucus produced over several days, measured as an RGBB ratio (Fig. 5B), increased with time.

Here, mucus secretion is used as an indicator of a differentiated cell

Fig. 5. Mucus staining of Calu-3 in the chip A) Microscopic images of mucus staining of Calu-3 cells at I. Day 3 and II. Day 7 B) A plot of RGBB ratio across several days of Calu-3 grown in the chip (Mean SD, n ¼ 8) (Scale: 100 μm). J. Shrestha et al. Organs-on-a-Chip 1 (2019) 100001 7

layer to goblet cells (mucus-producing cells) and, therefore, by extension, the physiological functionality of cells (Haghi et al., 2010). The RGBB value is an indirect measure of mucus production and shows variations in mucosal secretion, which enables a comparison of mucus secretion across several days. Since the secretion increased with time, it can be concluded that the cell layer grown in the chip represented physiological functionality.

3.4.3. Sodium fluorescein permeation
The permeability of the cell monolayer was tested by studying the permeation of sodium fluorescein (flu-Na) across the cell monolayer. There was a significant decrease in the permeability values in the cultured devices as compared to the blank devices (Fig. 6A). The significant difference in flu-Na concentration indicates a barrier formation by the cell junctions in the monolayer of cells growing on the membrane. This ability of the model to culture cells at an air-liquid interface and develop a tight barrier against flu-Na permeation, is another physiological function of respiratory epithelial cells.

3.4.4. Cell surface P-gp expression
Transport proteins are integral transmembrane proteins involved in the pharmacokinetics of many drugs (Haghi et al., 2010). They play a crucial role in the absorption of the drugs, its distribution, and elimination of waste metabolites; as such they are vital in maintaining the pharmacological barrier integrity throughout the body. P-gp is a plasma membrane glycoprotein, magnesium (Mg2þ)-dependent ATPase, and is a predominant drug transporter protein (Bebawy et al., 2001). P-gp is physiologically expressed on the surface of respiratory epithelial cells (Madlova et al., 2009). Commonly used drugs for treating respiratory diseases have shown interaction with P-gp, making it relevant for respiratory drug transport studies (Forbes and Ehrhardt, 2005). The P-gp expression of the cultured Calu-3 cells in the chips was analyzed using flow cytometry by direct immunolabelling. In comparison with the isotype control, chip grown cells displayed P-gp expression, which increased gradually over time (Fig. 6B). Nevertheless, this result further validates the functionality of Calu-3 cells in our lung-on-a-chip model and favours the suitability of the model to be used for in vitro pulmonary drug transport studies.

3.5. Effects of CSE on IL-6 and IL-8 production from cultured Calu-3 cells
Cigarette smoke contains over 4000 individual chemicals, which are noxious and carcinogenic (Burns, 1991). These chemicals, once deposited on the surface of the airways, absorb across the alveolar-capillary membrane to be circulated into the blood. The inflammatory response contributes to the structural and functional changes in the lung, leading to their destruction and subsequently the development of lung diseases. Cigarette smokers are associated with altered levels of inflammatory cytokines secretion in their bronchoalveolar lavage (BAL) (Maestrelli et al., 2001; Mikuniya et al., 1999). IL-8 is a potent chemoattractant for neutrophils and eosinophils; released by phagocytes and various tissue cells when exposed to inflammatory stimuli (Baggiolini and Clark-Lewis, 1992). IL-6 is a pro-inflammatory cytokine secreted by epithelial cells and macrophages in the airways (Mikuniya et al., 1999). Several studies have reported increased levels of IL-6 and IL- 8 in the BAL and induced sputum obtained from smokers (Mikuniya et al., 1999; Mio et al., 1997). Budesonide, a glucocorticoid, has been successfully used in asthma and inflammatory disorders like COPD (Lung and Institute, 2002; Keatings et al., 1997). The cytotoxicity assay showed that 10% CSE was not toxic to the cells (Supplementary Materials). Thus, 10% CSE was used for the

Fig. 6. A) Flu-Na permeation: Transport of flu-Na from apical to basolateral in blank and seeded chips on day 7 of culture (n ¼ 3, * ¼ p < 0.01) B) Comparison of the P-gp expression in Calu-3 cells cultured in the chip on different days: The expression of P-gp increased with days. C) Flow cytometry results of measuring P-gp expression by Calu-3 cells on day 7: I. Control group with no staining; II. Control group stained with FITC-anti P-gp III. Control group stained with PI IV. Samples stained with both FITC-anti P-gp and PI.

experiments. Significantly increased secretion of IL-6 (p < 0.01) and IL-8 (p < 0.05) was observed with CSE treatment. The cells treated with 10% CSE only released significantly higher amounts of IL-6 (p < 0.01) and IL-8 (p < 0.05) over 24 h compared to cells treated with both CSE and Budesonide (Fig. 7A). The CSE-Bud and Bud-CSE group had lower amounts of IL-6 and IL-8 secretion, indicative of the anti-inflammatory effects of Budesonide (Keatings et al., 1997; Confalonieri et al., 1998). The increased secretion of IL-6 and IL-8 on exposure to CSE are in substantial agreement with previous studies (Mikuniya et al., 1999; Mio et al., 1997; Wang et al., 2000). The capability of our model to successfully replicate the effects of CSE and Budesonide makes it a suitable in vitro model for toxicological and inflammatory studies.

3.6. Effects of CSE on cellular expression of E-cadherin
The junctional complexes, tight junctions, and adherens junctions maintain the epithelial cell integrity (Jang et al., 2002). The tight junctions are located apically and form a barrier controlling paracellular permeability of ions and solutes, while the adherens junctions hold the cells together through calcium-dependent adhesion molecules (Tsukita et al., 2001; Zihni et al., 2016). E-cadherin, an adherent junction protein, is observed in the cell membrane or cell-cell junctions. It is required for proper localization of essential tight junctional proteins like claudin-1 and 4, and ZO-1, into junctions to maintain a correct and intact tight junction formation (Tunggal et al., 2005). The loss of this cell transmembrane protein destabilizes the epithelial cell integrity, leading to dissociation between neighbouring cells and reduced its polarity (Willis and Borok, 2007). Cigarette smoke exposure has been associated with the disruption of the airway epithelium barrier integrity and increased permeability (Rusznak et al., 2000). Also, the disruption of the intercellular junctional integrity in the absence of tight junction proteins when exposed to CSE has been previously noted (Schamberger et al., 2014). To confirm the effects of CSE treatment on the expression of the surface protein, E-cadherin, the Calu-3 cells were stained with monoclonal antibodies against E-cadherin and observed with a confocal laser scanning microscope. Single confocal slices are shown in Fig. 7 B illustrate the differentiated tight junction and nucleus stained regions. The cells grown in absence of CSE maintained their morphology and strongly expressed E-cadherin protein. However, treatment with 10% CSE for 24 h, showed reduced expression of the epithelial marker. Similarly, the cells treated with both CSE and Budesonide also had a slightly reduced expression of the E-cadherin. The results agree with earlier findings of the effects of CSE on the barrier integrity and E-cadherin expression in cultured airway epithelial cells (Schamberger et al., 2014; Oldenburger et al., 2014; Xi et al., 2000). This re-affirms the validity of our microfluidic lung-on-a-chip model as a functional platform to conduct pulmonary studies.

4. Conclusion
A simple fabrication technique of lung-on-a-chip devices using surfacetreated 3D-printed molds has been proposed in this paper. The fabrication technique allows the chip to be fabricated in less than a day, and the molds can be used for repeated PDMS casting. Thus, the technique is simple, robust, and cost-effective. Validation of our design was conducted by investigating the effects of CSE on the production of inflammatory markers, IL-6, and IL-8, along with effects on cellular expression of adherent junction protein, E-cadherin. CSE augmented the release of IL-6 and IL-8 from the epithelial cells cultured within a lung-on-a-chip model. The budesonide treatment helped to reduce the effects of CSE on the release of both IL-6 and IL-8. Moreover, our results show that the CSE disrupted the airway epithelial barrier by affecting the junctional proteins. This result is supportive of the findings of numerous studies suggesting that cigarette smoke is the precursor of inflammatory response and

Fig. 7. Effects of CSE on cultured Calu-3 cells in the chip: A) IL-6 and IL-8 release from Calu-3 cells in response to CSE: The cells were cultured until confluency in the upper channel and were treated with 10% CSE. Normal media alone or media containing 100 nM Budesonide was flown at 30 μl/h and collected in a tube. After 24 h of incubation, the collected media was later assessed for IL-6 and IL-8 by ELISA. B) Changes in the E-cadherin expression: Imaging of immunofluorescence-stained Calu3 cells cultured in the absence of CSE showed strong surface expression of E-cadherin. The Bud-CSE group and CSE-Bud group both showed a slight reduction in expression of the epithelial marker, E-cadherin. However, with treatment with 10% CSE for 24 h, the expression of E-cadherin was reduced significantly. E-cadherin staining is shown in green, and DAPI staining is shown in blue. Scale bar: 50 μm. White arrows indicate disrupted tight junctions after treated with 10% CSE. Data are expressed as mean SD (n ¼ 3). * indicates p < 0.05, ** indicates p < 0.01. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.) contributes to the development of smoke-related respiratory diseases. Therefore, these results confirm that our design produces similar results to the in vitro static experiments but in a dynamic condition which has far more comparability to the real physiological conditions. The model developed using 3D-printed molds was able to maintain excellent barrier integrity, expressed cell surface functional P-gp, and secretion of mucus layer, providing a platform for permeability assays, transport mechanisms, and pulmonary drug delivery studies. Also, the ability to rapidly prototype these molds with little technical skills makes organ-on-a-chip modelling accessible to a broad group of researchers.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments
We would like to acknowledge the Australian Research Council for supporting through Discovery Project Grants (Grant Nos. DP170103704 and DP180103003) and the National Health and Medical Research Council through the Career Development Fellowship (Grant No.) APP1143377. Dr. Ghadiri is the recipient of Ann Woolcock Fellowship from Woolcock Institute of Medical Research.