The metastasis of cancerous tumours relies on the release of circulating cells that migrate to distant sites and form secondary tumours1,2. The factors that determine the invasiveness of these circulating tumour cells (CTCs) remain poorly deﬁned, and it is not yet possible to distinguish CTCs that have high versus low metastatic potential. Studying CTCs that are directly collected from unprocessed blood samples is a challenge given their rarity (parts per billion) in the bloodstream3,4. Moreover, these cells are highly heterogeneous: multiple cell phenotypes can exist within a given tumour, and their properties evolve dynamically once they leave a tumour and enter the bloodstream1.
Fluorescence-activated cell sorting (FACS) is a powerful presentday method to characterize and sort heterogeneous cell subpopulations. Unfortunately, FACS does not possess the sensitivity required to enable the routine characterization of CTCs at the levels at which they are present in the bloodstream, and is therefore not broadly applicable to the analysis of rare cells in clinical specimens. Microﬂuidics-based approaches have provided a new avenue to study CTCs5–17; however, existing techniques are generally limited to the capture and enumeration of CTCs and do not report on the phenotypic properties of CTCs.
New methods are urgently needed to characterize and sort CTCs according to their detailed phenotypic proﬁles so that the properties of invasive versus noninvasive cells can be identiﬁed. High levels of sensitivity and high resolution are required to generate proﬁles that will provide biological and clinical insights. We recently reported a method that allowed us to sort CTC subpopulations coarsely according to their phenotypic properties18. The resolution that was achieved, however, enabled discrimination among surface expression levels only when very large differences were at play. We hypothesized that much greater resolution would be required to accurately proﬁle the phenotypes of CTCs to connect their molecular-level properties with invasiveness.
Here, we report a novel approach that exploits nanoparticlemediated cell sorting, and relies on a unique chip architecture that achieves excellent control over an applied magnetic ﬁeld along a channel. In this way, this new system accomplishes high-resolution phenotypic ranking of CTCs. We term the new approach, which is based on the longitudinal proﬁle of magnetic ﬁeld gradients, magnetic ranking cytometry (MagRC). MagRC generates a phenotypic proﬁle of CTCs using information collected at the single-cell level. We show that it allows sorting of CTCs into one hundred different capture zones. We ﬁnd that MagRC has a very high level of sensitivity and is able to proﬁle CTCs accurately even when they are present at low levels (10 cells per ml) in unprocessed blood. The strategy allows the dynamic properties of CTCs to be tracked as a function of tumour growth and aggressiveness. Using blood samples both from xenografted mice and from human cancer patients, we show that the increased resolving power of MagRC provides distinct new information that is not accessible using existing methods.
Overview of MagRC
The MagRC approach leverages immunomagnetic separation19 for proﬁling CTCs as a function of their surface marker expression. A whole blood sample is incubated with antibody-functionalizedmagnetic nanoparticles that bind speciﬁcally to a correspondingsurface marker, and microengineered structures inside the deviceenable the rare cell proﬁling capability of MagRC. X-shaped struc-tures within the microﬂuidic channel generate regions with slowﬂow and favourable capture dynamics18, a requirement for thecapture of cells that are tagged with magnetic nanoparticles;whereas highly discretized sorting of subpopulations is achievedvia the introduction of differently sized nickel micromagnets20.Thelocal magnetic force is engineered to vary systematically withinthe device via the footprint of the micromagnets (Fig. 1b,c). Themicromagnets are positioned concentrically within the X-shapedmicrostructures, creating regions with low ﬂow and high magneticﬁeld gradients that are ideal for capturing CTCs with even low levelsof magnetic loading (Fig. 1a). This device, coupled with immunostain-ing of captured cells, is intended to generate high-resolution proﬁles ofcells captured from whole blood (Fig. 1d).
A quantitative physical model of the device (see SupplementaryFigs 1–4) was developed to explore how cells exhibiting variedexpression levels would generate different MagRC proﬁles that man-ifested their distinct phenotypes. A capture volume was deﬁned as aregion in which the magnitudes of the magnetic and drag forces arecomparable. As a result, those cells that pass through a capture zonewill be deﬂected and captured. For a cell covered with an abundanceof bound magnetic nanoparticles, the capture zones generated byeven the smallest micromagnets are sufﬁcient to ensure completecapture in the earliest zones of the MagRC Chip (Fig. 1c, top).Cells with low surface marker expression are deﬂected only if theyare close to edges of the micromagnets, where the magnetic forceacting on the nanoparticles is highest (Fig. 1c, bottom). As eachmicromagnet is positioned concentrically with an X-structure, theregions in the MagRC chip that exhibit the highest magneticforces and ﬁeld gradients also correspond to the regions thatexhibit the slowest ﬂows. This has the beneﬁt of creating localizedregions with favourable capture dynamics (low drag and highmagnetic forces), while also contributing to the high-resolutionsorting capability of the chip.
For each cell in each zone, the probability of that cell encounteringa capture region was calculated and reported as the captureparameter. Because the nickel micromagnets generate ampliﬁed mag-netic ﬁelds near the bottom of the microﬂuidic channel, the captureparameter of a given cell within the chip depends strongly on its ver-tical position. Additionally, the extended length of the chip relative toits height leads to long residence times and the potential for cells tosettle towards the bottom of the chip. The vertical dependence ofthe capture parameter for cells having different levels of magneticloading is illustrated in Fig. 2a. Thousands of model cells were simu-lated, each having a randomly assigned initial height ranging from 5to 45 µm at the inlet of the microﬂuidic chip. The overall modellingresults presented in Fig. 2b show the predicted capture locations forthree types of cells with high, medium and low levels of magneticloading. (See Supplementary Information for a detailed explanationof the parametric model).
Resolution, sensitivity and versatility of the MagRC approach
In a ﬁrst suite of experiments, we used four cell lines with knownlevels of expression of the epithelial cell adhesion molecule(EpCAM) to challenge the capture and sorting capabilities of theMagRC chip. EpCAM is a surface marker that is commonly usedto target CTCs. It is known that CTCs lose EpCAM when theyundergo the epithelial to mesenchymal transition (EMT) duringcancer progression21,22, and therefore tracking this marker shouldallow EMT to be monitored. Four different target cell lines—MCF-7, SKBR3 (breast adenocarcinoma cells), PC-3 (humanprostate cancer cell line) and MDA-MB-231 (a breast cancer cellline with mesenchymal characteristics that mimics triple negativebreast cancer cells)—were incubated with 50 nm nanoparticlescoated with anti-EpCAM in buffered solution. After capture, anuclear stain was introduced into the chip to identify capturedcells, and the capture efﬁciency was assessed by counting the cap-tured cells using ﬂuorescent microscopy. Experiments for each cellline were repeated three times. Figure 3a shows the ﬂuorescentmicroscope images of an SKBR3 cell captured at the edge of anickel micromagnet (where the magnetic ﬁeld and ﬁeld gradientsare at a maximum).
The four different cell lines tested exhibited markedly differentdistributions within the device (Fig. 3b and Supplementary Fig. 5).High recoveries of the cells injected into the device are achieved(MCF-7 95 ± 5%, SKBR3 93 ± 4%, PC-3 91 ± 6%, MDA-MB-23194 ± 5%; Fig. 3c), indicating that this approach has a high level ofsensitivity. MCF-7 cells, which have the highest level of EpCAMexpression, were found primarily in the earlier zones where themicromagnets are the smallest. However, PC-3 and MDA-MB-231 cells (which had the lowest level of EpCAM expression) wereonly captured after they encountered the large micromagnetscloser to the outlet of the chip. The relative levels of EpCAMexpression of the cell lines were con ﬁrmed via ﬂow cytometry(FCM, inset to Fig. 3b). T-test analysis was used to assess the statisti-cal signiﬁcance of the MagRC proﬁles obtained from different celllines (Supplementary Tables 1–3). The calculated P values(<0.0001) conﬁrm the statistical signiﬁcance of the uniqueness ofthe MagRC proﬁles and that the resolution of this technique ishigh. On the basis of these results we can conclude that theMagRC chip is able to sort cells according to the expression levelof a targeted surface marker. Moreover, it efﬁciently captures cellsthat exhibit even low levels of a target surface marker, and can beapplied widely to target surface markers for which a correspondingantibody is available.
This MagRC approach is amenable to the use of a wide range ofsurface antigens as the basis for proﬁling. We proﬁled the SKBR3cell line using three different surface markers that are oftenexpressed in epithelial cancer cells: human epidermal growthfactor receptor 2 (HER2)/neu, EpCAM and N-cadherin (Fig. 3d).The inset in Fig. 3d shows the level of these three surface markersin SKBR3 cells measured by FCM. HER2 is known to be highly over-expressed in this cell line, and experiments with magnetic nanopar-ticles coated with anti-HER2 led to cell capture within the veryearliest zones of the chip. In contrast, capture with anti-N-cadherincoated nanoparticles showed most cells being captured in the laterzones of the chip. EpCAM levels are intermediate for these cells, afact also reﬂected in the MagRC proﬁles.
The data presented indicate that MagRC produces proﬁles thatare comparable to those reported by FCM. FCM is a powerful androbust approach useful in analysing protein expression and hetero-geneity in living cells. It is limited in its sensitivity, however, andrequires cell numbers of 104or higher for accurate results23.Asshown here, MagRC reports on protein expression with a similarresolution, but using much smaller collections of cells. It is alsonoteworthy that the MagRC approach is a gentle analysis methodthat allows high recoveries of viable cells (Fig. 3e). 92% of capturedcells can be recovered, and 98% of the recovered cells are viable(Supplementary Fig. 7).
We then proceeded to challenge the system using unprocessedwhole blood samples, and found that MagRC retains its sensitivityand pr oﬁling capability . When whole blood samples (1 ml) containingbetw een 10 and 40 cells were proﬁled using EpC AM as a targetmark er, reproducible pr oﬁles w ere obtained (Fig. 4). We comparedthe performance of the MagRC appr oa ch with the CTC gold s tandardFDA-clear ed CellSear ch as say (Fig. 4b). Spiked blood sa mples contain-ing 100 SKBR3, PC-3 and MDA-MB-231 cells per millilitr e w er eprepared and analysed using both the MagRC chip and CellSearch.High re co veries of the spik ed samples injected into the MagRC chip