Multi-sample deformability cytometry of cancer cells

There is growing recognition that cell deformability can play an important role in cancer metastasis and diagnostics. Advancement of methods to characterize cell deformability in a high throughput manner and the capacity to process numerous samples can impact cancer-related applications ranging from analysis of patient samples to discovery of anti-cancer compounds to screening of oncogenes. In this study, we report a microfluidic technique called multi-sample deformability cytometry (MS-DC) that allows simultaneous measurement of flow-induced deformation of cells in multiple samples at single-cell resolution using a combination of on-chip reservoirs, distributed pressure control, and data analysis system. Cells are introduced at rates of O(100) cells per second with a data processing speed of 10 min per sample. To validate MS-DC, we tested more than 50 cell-samples that include cancer cell lines with different metastatic potential and cells treated with several cytoskeletal-intervention drugs. Results from MS-DC show that (i) the cell deformability correlates with metastatic potential for both breast and prostate cancer cells but not with their molecular histotype, (ii) the strongly metastatic breast cancer cells have higher deformability than the weakly metastatic ones; however, the strongly metastatic prostate cancer cells have lower deformability than the weakly metastatic counterparts, and (iii) drug-induced disruption of the actin network, microtubule network, and actomyosin contractility increased cancer cell deformability, but stabilization of the cytoskeletal proteins does not alter deformability significantly. Our study demonstrates the capacity of MS-DC to mechanically phenotype tumor cells simultaneously in many samples for cancer research.


I. INTRODUCTION
There is a growing interest in measuring and studying cell deformability, i.e., a cell's ability to deform or change shape under load. 6,7 Particularly in the context of cancer, investigations of cell deformability are gaining prominence since mechanical and biochemical cues can alter cancer cell deformability, which may in turn influence malignant transformation and tumor growth. 8,9 From a cancer diagnostics perspective as well, efforts are growing to develop cell deformability as a label-free marker to detect cancer cells in patient samples. [10][11][12] Deformability of cancer cells can also be potentially used as a functional readout during compound screening to identify cancer drug candidates. 13,14 To characterize cancer cell mechanical properties including deformability, numerous techniques have been developed. Popular among these methods are micropipette aspiration, 15,16 atomic force microscopy, 12,17 magnetic bead rheology, 18,19 and optical stretching. 20,21 These techniques provide reliable mechanical measurements of cells but suffer from low throughput, typically <O(1) Otto et al. 22 Linear channel $5 Â 10 3 -1.5 Â 10 4a HL60, whole blood O(100) This work Array of linear channels compounds in a pilot screen requires analysis of more than 100 samples, highlighting the need for multi-sample deformability cytometry (MS-DC) techniques. Advancement of methods that are not only high throughput, but also process numerous samples will bring cell mechanics into the realm of clinical cancer research. Literature reveals two reports of multi-sample evaluation of cell mechanical properties. Qi et al. 35 employed microfiltration using a custom-made well-plate where cells are forced through membrane filters in each well. In contrast, Cribb et al. 36 applied passive microbead rheology to cells cultured in multiwell plates and performed automated analysis to quantify cellular mechanical properties. In both cases, the multiwell format allowed analysis of multiple samples simultaneously.
Examining the available DC techniques for multi-sample deformability analysis, the approach of Otto et al. 22 seems favorable due to the simplicity of the channel geometry. In this work, we build on their approach but use an array of linear microchannels combined with a microfluidic manifold to perform multi-sample deformability cytometry (MS-DC). We use distributed pressure control to simultaneously deform cancer cells in ten independent channels enabling us to run ten samples per experiment, at a throughput of 100 cells/s in each sample. An automated image analysis routine was developed to quantify the deformation index (DI) for every cell in each of the ten channels.
Previously, Otto et al. 22 measured the deformability of HL-60 cells which is a blood-cancer cell line and it is not clear whether tissue cancer cells from solid tumors can be deformed due to flow in linear channels. In this study, we test a panel of cell lines derived from solid tumors, which are usually stiffer than HL-60. The library consists of six breast cancer cells and three prostate cancer cells having different molecular histotype, cellular origin and morphology, and metastatic potential. We also measured the deformability of cancer cells treated with six cytoskeletal intervening drugs. Broadly, our results validate MS-DC as a high throughput platform for mechanical phenotyping of individual cancer cells in multiple cell-samples.

A. MS-DC device design, set up and manifold characterization
The MS-DC device design consists of ten independent microchannels [ Fig. 1(a)], made using PDMS-based soft lithography techniques. 37 A PDMS manifold [Figs. 1(a) and 1(b)] chamber was fabricated to distribute the pressure across each on-chip reservoir. The manifold chamber was molded from a 3D-printed master containing a filament structure of height 3 mm, width 7 mm and length 44 mm. The test section of each channel has a width W ¼ 18 6 0.3 lm, height H ¼ 19.8 6 0.4 lm and length L ¼ 230 lm, respectively. This almost square cross-section of the test channels is also evident from the scanning electron micrographs shown in Fig. 1(c). To accommodate ten channels with reservoirs in a linear fashion we had to make the channels long. From inlet reservoir to outlet sample collector, the total length of each channel is 43 mm, which offers very high hydrodynamic resistance to the flow. To reduce the resistance due to the long channel, a section of the entry and exit (20 mm each side) were made with a larger crosssection (W ¼ 200 lm & H ¼ 100 lm) than the test section using two step lithography [inset of Fig. 1(a)]. The test section occupied %31% of the total channel resistance.
The experimental setup for MS-DC consists of the microfluidic device, a constant pressure source and an inverted microscope connected with a high-speed camera [ Fig. 1(b)]. Each of the microchannel accommodates a unique cell-sample. The deformed cells [inset of Fig. 1(b)] were analyzed using custom-written image analysis routines. Figs. 1(d) and 1(e) show the typical scatter density plot of deformation index (DI) versus the ratio of cell diameter to channel hydraulic diameter (D c /D h ), for breast and prostate cancer cells respectively.
To asses if the single pressure source and manifold were distributing the same pressure gradient in each of the channels, we measured the velocity of rigid polystyrene (PS) beads (Polyscience Inc., USA) in each of the channel at a driving pressure of 5 kPa. The manufacturer reported bead diameter is D ¼ 15.13 lm 6 6%. As shown in Fig. 2(a), the average velocity of PS beads (calculated from !20 beads/channel) varied less than 3% between the channels in the device and therefore the manifold was distributing pressure uniformly across each channel.
The slight variations in mean velocity between channels may be due to the variation in bead size as well as the off-centeredness of the particles inside the channel. 38 We note that since the channels are not interconnected and pressure controller maintains a constant pressure, in case one channel is clogged, the other channels remain unaffected.

B. Influence of driving pressure on deformation behavior of cells
Our approach of measuring deformability of cancer cells relies on flowing them through the microchannel using a constant pressure drop across the channel. In the microchannel, cells are expected to change their initial spherical shape to bullet-like due to the combination of shear and pressure forces. Since these forces depend on the magnitude of applied pressure drop, we investigated the influence of driving pressure to identify the driving pressures that can induce cell deformation; and how this deformation varies as the cell is moving from the entrance to the exit of the microchannel. Figure 3 shows the deformation behavior of breast cancer cells (MCF-7) at driving pressures of 3, 5, 7, 10, 15, 25, and 30 kPa. We observe that before entering the channel, cells had a mean DI less than our maximum error limit of 0.024 for all the driving pressures indicating that they are nearly spherical [ Fig. 3(a)]. As the cells pass through the entrance of the channel, for all the driving pressures, the DI jumps to a value higher than the error to comply with the increased fluid stress due to the reduced channel cross-section. For all driving pressures, DI gradually increases after the initial jump to a maximum before cells exit the channel. After exiting the channel, cells tend to relax as their DI decreases.
We find that for all the driving pressures, cells experienced the maximum deformation before exiting the channel. We use this maximum DI before exiting the channel as the measure of their deformability and plot in Fig. 3(b) by binning the cells having confinement 0.85 < D c /D h < 0.95, where D c is the undeformed cell diameter and D h is the hydraulic diameter of the channel. The lower limit of the bin was chosen to be 0.85 because below this confinement we did not observe much deformation and the upper limit of the bin was chosen as 0.95 to avoid possible contact between the cell surface and channel wall. Figure 3(b) shows that deformation of MCF7 cells increases linearly with increasing driving pressure.
The results of Fig. 3 also enable us to select a working driving pressure for cancer-related applications. The selection depends on the trade-off between the amount of cell deformation and the volume of images generated. At higher driving, pressure cells deform more but the volume of images generated due to the higher recording frame rate is also high. In this study, we choose 15 kPa as the working driving pressure since this pressure generates enough deformation as well as requires a relatively lower recording frame rate. Also at 15 kPa driving pressure, the cell reaches a steady shape inside the test section of the channel [ Fig. 3(a)].
C. Label-free mechanical phenotyping of metastatic capacity of cancer cells As a first demonstration of the utility of MS-DC, in this section, we present the mechanical phenotyping of six breast cell lines and three prostate cancer cell lines having different metastatic potential, histotype, morphology, and origin [summarized in Figs. 4(a) and 4(b)], using deformability as a label-free marker. This study was conducted by injecting 2-3 different cell lines and their replicates simultaneously in the MS-DC device.
Among the breast cell lines tested, two are non-tumorigenic (MCF10A & HMS50), one is weakly metastatic (MCF7), one has an intermediate metastatic potential (MDA-MB468), and two are strongly metastatic (MDA-MB231 and HCC1419). 39,40 To compare the deformability of these cancer cell lines, we binned the cells with 0.85 < D c /D h < 0.95, and performed the Mann-Whitney U test since the measured deformability has a non-normal (log normal) distribution [ Fig. 4(c)]. We find that the strongly metastatic cells (MDA-MB-231 and HCC1419) are much more deformable than both the less-metastatic cells (MCF-7 and MDA-MB468) and normal cells (MCF10A and HMS50). Interestingly, we find that the enhanced deformability of the strongly metastatic cells does not correlate with the histotype, but correlates well with the myoepithelial morphology of these cell types.
In addition to breast cancer cell lines, we tested prostate cancer cell lines: LNCaP, CL2, and PC3 with different metastatic potentials. [41][42][43] Here, CL2 is androgen-dependent derivative of androgen-independent LNCaP, hence these two cell lines are isogenic but vary in their metastatic potential. 44 In contrast to breast cancer cell lines, in the case of prostate cancer cells, we find that the weakly metastatic cells LNCaP are more deformable than the strongly metastatic cells CL2 and PC3. In these cell lines as well, we observed that the deformability does not correlate strongly with the histotype. In sum, using two tissue-specific cancer cell types, we have shown that MS-DC can certainly distinguish highly metastatic cell lines from weakly metastatic and healthy cells, but the correlation between metastatic potential and deformability might be dependent on the cancer type. Thus, the deformability cytometry technique originally developed by Otto et al. 22 can be expanded to cancer cells of solid tumors.

D. Deformability based drug dose-response analysis
In this section, we present the capability of MS-DC in performing drug dose-response analysis using deformability as the label-free marker. Current drug dose-response analysis approaches rely heavily on viability and other molecular markers. Deformability measures represent whole-cell physiological readouts and can complement molecular marker analysis. Existing deformability methods 1,2,4,22 do not allow convenient testing of different doses of drugs in the same run. In fact, it is common in cell mechanics literature to pursue investigations with cells that are treated at a single dosage rather than a range of concentrations. Here, we show that due to access to ten independent channels, in the MS-DC approach, drug-treated cells at different doses can be simultaneously deformed in a single run.
To demonstrate that drug dose-response analysis can be achieved using the ten channel format, duplicates of four different drug concentrations and the control were run simultaneously.  Fig. S1 for these cell lines. The statistical significance was determined using the one-tailed Mann-Whitney U test with a significance level of 0.01: ns (p > 0.01); *** (p < 0.0001); and **** (p < 0.00001). The horizontal dashed line in (c) and (d) shows the maximum error limit in DI measurements.
A library of six drugs that target the function of cytoskeletal proteins was chosen. Actinintervention drugs included Latrunculin A, Jasplakinolide, Blebbistatin, and Y-27632. Microtubule-affecting drugs were nocodozole and paclitaxel. The mechanism of action of each of these drugs is listed in Table II. To compare the data for different drug doses for each drug quantitatively, the cells with 0.85 < D c /D h < 0.95 were binned and the results are shown in Figs. 5(a)-5(f) (a representative scatter density plot for the Lat A-treated MCF7 cell is shown in supplementary material Fig. S2). Overall, it was observed that disruption of the actin network, microtubule network, and actomyosin contractility increased cancer cell deformability, but stabilization of the cytoskeletal proteins did not alter deformability significantly.
To quantitate the dose-response behavior, the DI for each drug concentration with control (no drug) was compared and the minimum concentration (C min ) of the drug was determined as the point at which a change in deformability was observed, i.e., either increase or decrease in DI, as listed in Table II. Blebbistatin, Y-27632, and Nocodazole were observed to be more potent than Latrunculin A since they altered cell deformability at lower dosages, while paclitaxel did not affect cell deformability across the tested range. Thus, MS-DC is capable of identifying drugs and their doses that can alter cancer cell deformability.
To check whether the applied stress in MS-DC had any deleterious effect on the viability of the cells, viability assays were performed. The viability of the to-be-injected cells as well as cells collected from the channel outlets for the drug Latrunculin A was compared [ Fig. S3(a)]. The viability was measured for the remaining drugs, after the specified treatment time [ Fig.  S3(b)]. There was no statistical difference in the viability of injected and collected cells from all five samples [see supplementary material Fig. S3(a)] of Latrunculin A-treated MCF7 cells, indicating that the fluid stresses in the channels did not induce any undesirable effect on the cancer cells. For the remaining drugs, the viability was also not affected. Thus, MS-DC is capable of drug dose-response analysis using deformability as a label-free indicator, and it can do so without affecting cell viability.

A. Technical advances of MS-DC compared to other deformability cytometers
There have been two attempts to perform multi-sample characterization of cell mechanics in multiwell plates. In one approach, a microfiltration strategy was used in which cells were forced through membrane filters integrated into a custom multiwell plate. 35 The fraction of cells passing through the filter in each well are recorded making it a scalable approach with quick population-level readouts. Potential drawbacks of this approach include fouling, lack of readouts on individual cells, and inability to resolve cell-size effects. In another approach, passive bead-based microrheology was applied to cells cultured in multiwell plates. 36 The benefit of this approach is that cancer cells in the adherent state can be interrogated mimicking the tumor microenvironment in vivo. The disadvantage of this approach is the lack of single-cell resolution and inadequate control of the attached bead position which makes it difficult to interpret cell rheology from passive bead motion. Compared to these two prior approaches, this study builds on recent advances in microfluidics-based cell deformability techniques to achieve multi-sample capability. We built our work on the approach of Otto et al. since the constriction devices 2,4,28 are prone to clogging and the extensional deformation of cells requires high camera frame rates 1,45 which poses an issue for continuous recording and storing images from multiple cell-samples.
In this study, we tested 6 breast cell lines, 3 prostate cell lines, and 6 drugs (Â4 concentrations) targeting function. Thus, we have used our multi-sample deformability cytometer to analyze 77 cell samples (including replicates) at rates of 100 cells per second per sample, which could only be possible because of the major improvements we have made. Specifically, (i) Otto et al. 22 use a constant flow rate to deform cells, which is difficult to multiplex, whereas our distributed pressure control allows operation of ten independent channels. (ii) The on-chip reservoirs introduced in our device greatly facilitate multiplexed sample loading, and reduce dead volumes. Sample requirement is at most $50 ll, which could be essential for scarcely available rare and patient-derived cells. (iii) Otto et al. 22 report that about 10% of their devices clog, however, MS-DC is inherently fail-safe due to an array of independent channels and our protocol involving two channels per sample. (iv) A data analysis software that tracks cells in each sample and outputs deformed shapes. About 50 000 images were collected from our multichannel device and processed on a standard desktop computer at a speed of 10 min per sample.
In the future, our MS-DC approach can be reconfigured to deliver new functionalities. For example, rather than using channels of the same width, channels of different width or geometry can be incorporated so that cellular confinement or deformation response is manipulated. In addition, rather than using the same driving pressure, different driving pressures could be integrated into the microfluidic manifold. Such flexibility could be used to obtain multidimensional readouts on the same cell sample, which could particularly benefit cancer-related applications given the heterogeneity of tumor cells.

B. Relationship between cell deformability and metastatic potential
Our study expands the capability of deformability cytometry to simultaneously analyze multiple cell-samples, enabling us to explore the relationship between cell deformability and metastatic potential. With the breast panel, we have chosen cell lines that differ in molecular histotype, morphology, and metastatic capacity. In breast carcinomas, molecular characterization based on ER, PR, and HER2 receptor expression is used to guide decision-making on patient prognosis and treatment. 46 Our panel of cell lines spanned different molecular subtypes, and we find no strong correlation between cell deformability and molecular sub-type. Fascinatingly, what seems to correlate the best between deformability and metastatic capacity is the cellular origin and morphology of the cells. We find that myoepthelial/basal cells (MDA-MB231 and HCC1419) are more deformable than epithelial-like cells (MCF-7 and MDA-MB468), which is consistent with the basal-like phenotypes having mesenchymal signatures. 47 Given that breast cancers characterized with HER2 and basal-like phenotypes correlate with poor prognosis and/or resistance to chemotherapy, 48 suggests that highly deformable breast tumor cells might lead to aggressive cancers. In striking contrast with our study on breast cancer cells where we found that highly deformable cells are more metastatic, in prostate cancer cells, we found the opposite behavior. The reason for this observation is not fully clear although studies have shown that in brain and pancreatic cancer cells, 5,49 less deformable cells are more metastatic. Thus, the relationship between cancer cell deformability and metastatic capacity needs to be further explored in a broader range of cancers and MS-DC offers a means to pursue such investigations.

IV. CONCLUSIONS
In this study, we have presented a novel microfluidic technique called multi-sample deformability cytometry (MS-DC) which can measure the deformability of multiple samples simultaneously using a distributed pressure control system and on-chip reservoirs. We have characterized the manifold and standardized the deformability measurements using rigid spherical PS beads. Optimal driving pressure conditions have been identified to adequately deform tissue cancer cells. The MS-DC system was validated by testing breast and prostate cancer cell lines of different metastatic potential and mapping the deformability changes due to different doses of drug treatment. We believe the capacity to process several samples with MS-DC at a throughput of 10 min per sample will be useful in cancer research where there is a need to handle multiple samples for diagnostics, drug discovery, and genetic screens.
In the drug experiments, MCF7 cells were treated with different concentrations of Latrunculin A (Lat A) for 2 h, Jasplakinolide (Jasp) for 1 h, Blebbistatin (Bleb) for 1 h, Y-27632 for 30 min, Nocodazole (Noco) for 1 h, and Paclitaxel (Pac) for 1 hr. All the drugs were dissolved in dimethyl sulfoxide (DMSO) to prepare the stock solution. The required concentrations of the drug were made by diluting the stock solution using PBS. In all cases, the percent of DMSO in the final cell solution was <0.1%, thus avoiding the toxicity from DMSO. Drug solution in PBS was freshly prepared before each experiment and used on the same day. Prepared drug solutions were added to a suspension of 2.25 Â 10 5 cells/ml in cell media in a 12 well plate and incubated at 37 C and 5% CO 2 environment. After the specified hours of incubation, cells were trypsinized and cell media with drug was removed as a supernatant by centrifugation. The final cell solution for the experiment was prepared using PBS. A cell concentration of 1.5 Â 10 6 cells/ mL was used in all cell experiments. All the experiments were completed within 30 min of cell harvesting.

B. Microfluidic device preparation
The MS-DC master mold was made using SU-8 photolithography. Polydimethylsiloxane (PDMS) monomer and the curing agent were mixed at a 10:1 ratio, degassed, poured on the master mold to get an approximate 6 mm height, and cured in an oven for at least 2 h at 70 C.
Cured PDMS was cut, peeled off from the mold, and 3 mm diameter inlet and outlet holes were punched using a biopsy puncher (Miltex, Japan). The Inlet holes were used as on-chip sample reservoirs and the outlet holes facilitated the collection of the samples from each channel. Ten holes were punched into the PDMS manifold to introduce samples into the on-chip reservoirs and subsequently all but one were blocked by pins to connect the pressure source.
We clean the PDMS replica, PDMS manifold, and cover glass (No. 2, Fisher Scientific, 45 Â 50 mm) using isopropanol for bonding. Bonding is done in two steps. First, we bond the PDMS replica with the cover glass by activating the surfaces with air plasma for 1.5 min (Plasma Cleaner PDC-32G, Harrick Plasma) then we bond the PDMS manifold on top of the inlet reservoir holes using air plasma again. In the second step of bonding, we ensure that the chamber of the manifold is aligned well so that all the on-chip reservoirs are open to the chamber which facilitates the distribution of driving pressure to each reservoir. The entire assembly was baked again at 70 C for 4 min to have a permanent seal. The sealed device is then filled with PBS to maintain the hydrophilicity of the microchannels.

C. Experimental protocol
The experimental setup for MS-DC consists of the microfluidic device, a constant pressure source, and an inverted microscope connected with a high-speed camera. Samples were loaded using a 25-gauge hypodermic needle (Excel International disposable) into each of the reservoirs ($50 ll/reservoir) through the holes punched on the roof of the manifold chamber and then all but one hole on the roof of the manifold were closed using 1 mm diameter pins (Instec, USA). The reservoirs were loaded to 85% of the maximum capacity to avoid the overflow of the cell samples and thus the possibility of any cross-contamination. A constant driving pressure was applied through the open hole across the device using a pressure controller (MFC8-FLEX4C, Fluigent Inc.) and 0.02 in. inner diameter Tygon tubing (Cole Parmer) connected with 20-gauge hollow blunt pin (Instech, USA). A constant driving pressure of 15 kPa was applied for all experiments unless noted otherwise.
High-speed video imaging was conducted to track the cell shape at a single cell level as it passes through the test section of the channel using a combination of an inverted microscope (Nikon eclipse TiU) and a high-speed CMOS camera (Phantom v710 12-bit, Vision Research). The region of interest (ROI) was recorded with a reduced resolution of 464 Â 504 pixels at a frame rate of 15 000 fps using 20Â magnification. The ROI included the area of test section and five channel-widths before and after the test section. The effective pixel size for this optical setup is 0.97 lm and the depth of focus is $6 lm. An exposure time of 1 ls was used to record blur-free motion [inset of Fig. 1(b)] of cells traveling at velocity $0.3 m/s. We measured the deformability of MCF7 cells using 20Â and 30Â magnifications in the same experiment to check whether the higher magnification improves statistical accuracy. The effective pixel sizes for these two magnifications are 0.97 lm and 0.64 lm, respectively. The results (supplementary material, Fig. S4) show that the lower magnification used in this study does not significantly affect the accuracy of the deformability metric.

D. Image processing
The recorded images were analyzed off-line using a custom-written MATLAB routine capable of automated image processing and data analysis. The images were segmented to have a binary image using different filters to enhance contrast, subtract background, and identify the presence of a cell. After segmentation, each cell was tagged with an identification (ID) number when it appears in the ROI for the first time. Using the MATLAB built-in function "regionprops", the cell's projected area, centroid location, perimeter, and frame number were recorded against that ID number for all the frames that cell took to pass through the ROI. The undeformed diameter of each cell was calculated from the projected area before it enters into the constricted channel.
To capture the change in cell's shape, the deformation index (DI) was calculated for each cell which was defined as 22 where A is the cross-sectional area and p is the perimeter. The DI of a perfect sphere is zero. As the cell deviates from the spherical shape, the value of DI increases from zero accordingly. Image frames with multiple cells in the test section of the channel were discounted due to potential hydrodynamic interactions between cells which can interfere with the deformation behavior of cells. To determine the minimum DI that can be reliably measured with our set up, we used rigid spherical polystyrene beads and the resulting DI distribution is shown in Fig.  5(b). We find that the beads show a maximum DI of 0.024 as shown by the horizontal dashed line in Fig. 5(b). Therefore, DI ¼ 0.024 was used as the maximum error in the measurement of DI of cells. In addition, the measured bead diameter was found to be 15.83 lm 6 4%, which is close to the manufacturer reported size distribution of 15.13 lm 6 6%.

SUPPLEMENTARY MATERIAL
See supplementary material for effects of magnification on the measured deformability, scatter density plots, and viability assays.