Quantitative phase deformability cytometry for noninvasive high‐throughput characterization of cells

Currently available deformability cytometry, combining microfluidics with the bright‐field high‐speed imaging system, has allowed label‐free mechanical characterizations of cells at a rate of hundreds of cells per second. However, the poor contrast between the signal and background under bright‐field imaging, especially for abnormal cells or cells treated with drugs, may produce artifacts during the recognition of cellular contours and subsequently bias the analysis of cellular deformability. In this study, a quantitative phase deformability cytometry (QPDC), constructed from a quantitative phase imaging system, is demonstrated with an improved signal‐to‐noise ratio of the acquired images and mitigated image artifacts, allowing a truthful illustration of cellular contours, especially for cells subjected to actin depolymerization. We have also demonstrated that the phase value extracted from captured images may provide an additional biophysical characteristic of cells. The presented QPDC platform is therefore expected to complement the existing deformability cytometry and expand the capacity of label‐free characterization of cells for biological and clinical applications.

speaking, the Young's modulus of highly invasive ovarian cancer cells (HEY A8) has been measured at around 494 Pa, compared with 884 Pa from the less invasive counterparts (HEY). 4 The elasticity of mouse ovarian surface epithelial (MOSE) cells is also varied, comparing the earlystage MOSE cells (1.097 ± 0.632 kPa) and the late-stage MOSE cells (0.549 ± 0.281 kPa), indicating that ovarian cells become softer as they transform into malignant tumorigenic cells. 5 Furthermore, healthy red blood cells (RBCs) exhibit in discoid shapes with a deformable membrane to address their physiological needs. However, RBCs under pathological conditions such as diabetes mellitus are observed with altered shapes and reduced deformability, resulting in enhanced flow resistance within the blood vessels. 6 As a quantitative measure, the Young's moduli of RBCs are measured around 4.9 ± 0.5 kPa, softer than RBCs collected from diabetes mellitus patients (8.6 ± 0.5 kPa). 7 Additionally, the differentiation potential of mesenchymal stem cells (MSCs) is also observed closely related to their elastic and viscoelastic properties. [8][9][10] Recent studies have shown that MSCs of relatively softer mechanical properties are prone to differentiate into adipocytes, while osteogenesis is favored by stiffer cells. 10 To this end, various methods, including atomic force microscopy (AFM), 1 optical stretching, 11 and micropipette aspiration, 12 have been developed for the assessment of cellular mechanical properties (i.e., Young's modulus and cell deformability). Among them, AFM is the most prevalent method for the quantification of cellular elastic moduli featured in highly controlled applied force, high sensitivity and spatial resolution. AFM functions by scanning a microcantilever over the surface of a cell. When the sharp tip of the cantilever contacts the tracing surface, bending of the cantilever is then utilized for the estimation of cell stiffness based on the Hooke's law. As implied by the working principle, the results obtained by AFM may be affected by the tip configuration, such as the tip size and the spring constant of the microcantilever, as witnessed by the significantly different mechanical properties measured from identical cell lines across different studies. 13 Moreover, AFM earmarks high-resolution single-cell characterization, confining the processing throughput at around 1-20 cells per hour.
In recent years, microfluidics-based methods, such as deformability cytometry, have received increasing attention, especially for high-throughput mechanophenotyping of cells. [14][15][16] Cellular deformation in deformability cytometry is typically generated by flowing cells into a constriction region, where the cells are deformed by fluidic shear stress. The cellular deformation is then captured by a high-speed camera with a frame rate of around a few thousand frames per second (fps), followed by post-processing of the captured images to extract the cell size, change in shape, and subsequently deformability. The current processing speed of deformability cytometry ranges between 100 and 1000 cells per second. More importantly, the microfluidics-based deformability cytometry allows robust and tailorable assessment of cellular deformability under fluidic stress, collectively determined by the channel geometry and flow velocity. Combined with modules such as identification of targeted cells by machine learning or cell sorting by external force fields, 17,18 the microfluidics-based strategies may provide not only thorough and high-throughput characterization of cells, but also potential categorizations of cells based on their mechano-phenotypes. Given the ease of handling, the microfluidics-based deformability cytometry has shown significant progress in biological investigations and routine clinical testing, as witnessed by an array of applications, including distinguishing cells of different cycles, 14 classifications of blood cells, 19 and correlation of the physiological abnormalities with the deformability of immune cells. 20 The analysis in current deformability cytometry is mainly based on transmission-mode bright-field imaging, hereafter termed the BF-DC, where the amplitude or intensity of the illumination beam is reduced based upon the absorption of the specimen, that is, the cells. Given the transparency of living cells, the scale-down of amplitude is often subtle, intrinsically resulting in a poor contrast between the signal and background, or the signal-to-noise ratio (SNR). 21 The problem of poor contrast becomes aggravated for the imaging of dynamic objects, for example, imaging a deformed cell with a very short exposure time. On the other hand, the illuminating light retarded in wavefront or phase after passing through a specimen may be a more sensitive measure for label-free cellular imaging. 22 Imaging the quantitative phase shift (φ), associated with the thickness and refractive index of the specimen, by merging interferometry and microscopy constitutes the essence of quantitative phase imaging (QPI) for label-free imaging. Taking the advantages of full-field and single-shot image acquisition, off-axis QPI is capable of dissecting the phase changes at the millisecond time scales or less. 23,24 The rapid response of off-axis QPI has allowed the imaging of dynamic processes, for example, the membrane fluctuation of RBCs. 25 Besides, the combination of off-axis QPI and a microfluidic chip is developed for the measurement of cell dry mass and optical volume changes when cells pass through a narrowed microfluidic channel. 26,27 Herein, we combine the off-axis QPI with the microfluidics-based deformability cytometry, hereafter termed quantitative phase deformability cytometry (QPDC), for the recording of cellular deformation in a label-free and high-throughput manner. Using mammalian cells as a model, QPDC provides an improved SNR and moderate image artifacts, allowing a truthful illustration of cellular contours under deformation. The QPDC system is capable of measuring the altered deformability of mouse skeletal muscle (C2C12) cells subjected to actin depolymerization. The versatility of the QPDC has been demonstrated by assessing the different physical characteristics among three cell lines, namely, C2C12, human embryonic kidney (HEK293) cells, and pancreatic cancer (PANC-1) cells. We have also demonstrated that the phase value extracted from captured images may provide an additional measure of cellular biophysical properties.

Configurations of the QPDC and BF-DC, image acquisition, and image post-processing
The QPDC system introduces a Mach-Zehnder interferometer design, as illustrated in Figure 1A. A 532 nm laser (CNI Laser, China) is coupled into a 1 × 2 single-mode fiber-optic coupler (SMFC), diverting the beam into the sample beam and the reference beam. Upon illumination of a collimated sample beam through an objective lens, OL1, the scattered field from the specimen is collected by another objective lens, OL2 (63×/NA 1.25, oil immersion, Zeiss, Germany). The scattered field is further combined with the reference beam, producing an interferogram. At the final imaging plane, the interferograms are collected by a high-speed camera with a frame rate of 10,000 fps at full-field (1024 × 1024; Fastcam SA-X2, Photron, USA). The system has a field of view (FOV) of 56 × 56 μm 2 . Quantitative phase maps, Δφ(x, y) are retrieved through a Fourier transform-based method. 28 At each point, Δφ is linked to the local specimen information through Equation (1), where d is the thickness and Δn is the refractive index contrast between the specimen and the medium. Under the BF-DC system (setup not illustrated), the cells are illuminated by a light-emitting diode (peak wavelength, λ = 550 nm). The BF-DC images are captured at an identical frame rate, i.e., 10,000 fps, as the QPDC. To ensure sufficient signal intensity, the BF-DC images are acquired under an objective of 20× (20×/NA 0.45, Nikon, Japan) with a FOV of 410 × 250 μm 2 .
The signal-to-noise ratio (SNR) of acquired images is computed by Equation (2), 29 where μ sig is the average signal intensity with the region of interest (ROI) and σ sig is the standard deviation of the signal within the background region. A customized OpenCV image processing algorithm is applied for the image analysis. Briefly, the image is first subtracted by the background, followed by thresholding, smoothing, and contour finding to obtain the cellular contour. The projected cross-sectional area of a cell (A contour ) is computed by the number of pixels within the contour of an identified cell. The convex hull is then determined by fitting the cell surrounding with an enclosed envelope (A hull ), as detailed in Supporting Information. 30 With A hull as a descriptor of cell size, the cellular deformation (D) is computed based on Equation (3), where P is determined by the perimeter of the fitted cellular contour, that is, the convex hull. 14 The scatterplot of deformation and cell size is then plotted with the computed D and A hull , where the color of the scatter points indicates the density of events. Over 1000 cells were analyzed for each scatterplot. The reported phase value is the average phase value measured over the A hull .

Fabrication of the microfluidic chips
The microfluidic chip employed in this study was designed with an overall thickness of 200 μm, as well as a constriction area of 30 μm × 30 μm × 300 μm (height × width × length) ( Figure 1B). The patterns of microchannels were illustrated by a computer-aided design (CAD) software program and printed on a Cr/Au mask (MicroCAD Photo-Mask Ltd., China). The microfluidic chips were fabricated following the standard soft lithography process. 31 Briefly, a negative photoresist of SU8-3050 (Kayaku Advanced Materials) was spun on a 4″ silicon wafer at a designated spinning speed of 6000 rpm for 30 s to obtain a height of 30 μm. The photoresist-coated wafer was then exposed to ultraviolet light through a Cr/Au mask under a mask aligner (ABM-USA). The developing and baking procedures of SU8-3050 were conducted following the manufacturer's instructions. The fabricated molds were then hard-baked at 180 • C for 3 h.
To fit the microfluidic chip between two objective lenses in the QPDC system, the microfluidic chips ( Figure S1) were tailored made by a specialized protocol as described in Supporting Information and briefly described in the following. The polydimethylsiloxane (PDMS) prepolymer and curing agent (SYLGARD™ 184 Silicone Elastomer Kit) were well mixed at a ratio of 10:1 and spin coated on the master mold at a speed of 500 rpm for 30 s to obtain a PDMS film thickness of around 200 μm. The PDMS film layer was cured at 110 • C for 10 min. Two pre-made PDMS blocks (20 mm × 20 mm × 3 mm) were then aligned to the inlet and outlet regions, followed by bonding with the PDMS film layer upon oxygen plasma activation (Harrick Plasma, USA). The whole assembly was then baked at 110 • C for 10 min. The produced assembly was then peeled, hole punched, and bonded with a clean cover slide after oxygen plasma treatment, followed by baking at 110 • C for 20 mins. The microfluidic chips and tubing accessories were sterilized before every set of experiments.

2.3
Cell culture and sample preparation C2C12 (ATCC catalog number CRL-1772), HEK293 cells (ATCC catalog number CRL-1573), and PANC-1 cells (ATCC catalog number CRL-1469) were maintained in Dulbecco's modified Eagle medium (Gibco, USA) supplemented with 10% of fetal bovine serum and 1% penicillinstreptomycin (Gibco). All cells were grown at 37 • C in a humidified atmosphere with 5% CO 2 , and the culture medium was replaced every 2 days. To artificially alter the cellular deformability, Cytochalasin D (Cyto D, Sigma-Aldrich, Germany) was used to disaggregate the microfilaments and destroy the intracellular microfilament cytoskeleton as shown in previous studies. 32,33 Briefly, 10 mM of Cyto D stock solution was prepared by dissolving the intended amount of Cyto D powder in dimethyl sulfoxide (Sigma-Aldrich, Germany). Cyto D stock solution was diluted to the working concentration, that is, 1 and 10 μM, by adding the intended amount of Cyto D stock solution into the cell culture medium. The culture medium was replaced by 1 mL of 1 or 10 μM Cyto D into the six-well culture plate containing C2C12 cells, followed by incubation at 37 • C for 20 mins. The treated C2C12 cells were then washed with phosphate-buffered saline (PBS) twice and trypsinized. Prior to the measurements, cells were resuspended in the MC-PBS buffer (0.5% [w/v] methylcellulose [MC] in PBS) to a final cell concentration of 1 × 10 6 cells/mL. Noted that Cyto D was retained in the cell suspension throughout the experiments to maintain the treatment effect. 34 The as-prepared cell suspension was introduced into the microfluidic chip together with the sheath buffer (MC-PBS buffer) at a volumetric flow rate of 10 μL/min controlled by syringe pumps (PHD 2000, Harvard Apparatus, USA). The employed volumetric flow rate was optimized based on the retained cell viability (∼98% viable cells, Figure S3). The maximum shear stress acting on the cell surface was estimated at ∼2.2 kPa as computed by a simulation model described in Supporting Information and illustrated in Figure S2.

Fluorescence staining and confocal imaging
Cells were washed twice with PBS and fixed with 4% formaldehyde (Sigma-Aldrich, USA) for 40 mins. Following permeabilization with PBS containing 0.1% Triton X-100 (Beyotime, China) for 15 mins, cells were stained with 4′,6-diamidino-2-phenylindole (DAPI, Sigma-Aldrich, USA) and Phalloidin-iFluor 555 Reagent (ab176756, Abcam, UK) for 40 mins in the dark at room temperature. Fluorescent images of Cyto D-treated and -untreated cells were acquired by an inverted confocal laser-scanning microscope (C2 Plus, Nikon, Japan) under a 60× objective (Nikon, 60×/NA 1.4, oil immersion). Signals from DAPI and phalloidin were acquired by excitation with 405 and 561 nm lasers, whereas the emissions were filtered through the filters of 438/24 nm (bandpass) and 561 nm (longpass), respectively. Confocal images were analyzed using Fiji software. Quantification of the mean fluorescence intensity per cell was conducted after image thresholding and a selection of the ROI, that is, every cell. 35

AFM characterization
AFM indentation tests were performed by a JPK NanoWizard II (JPK Instruments, Germany) mounted on an inverted optical microscope (Axioobserver D1, Zeiss). A cylindrical tip with a radius of 2 μm fabricated by a focused ion beam (DualBeam system FEI Quanta 200 3D, Thermo Fisher Scientific, USA) was housed on the cantilever (MLCT, Bruker, spring constant of 0.04 N/m, USA). Prior to the indentation tests, the sensitivity was set by measuring the slope of force-distance curves acquired on glass, whereas the spring constant was calibrated by the thermal noise method. The tip was then aligned to the center of a target cell, followed by an indentation of 1.5 μm in depth and 0.3 μm/s in the extending velocity. The Young's modulus was extracted from the force-distance curves by the Hertz model built in the JPK Data Processing (JPK Instrument, Germany) software. At least 10 cells were analyzed for each condition.

Statistical analysis
Statistical analyses were performed using GraphPad Prism v.5.0 (GraphPad Software, Inc., USA). The statistical significance (p-value) was determined by two-tailed unpaired Student's t-tests or one-way analysis of variance, with the computed p-values denoted on the presented results. All analyses were conducted with at least three biological replicates.

Improved signal-to-noise ratio by QPDC
As shown in Figure 2A, the representative raw and contour-fitted images of C2C12 cells were acquired by BF-DC and QPDC, respectively. Image contrast, collectively determined by the signal collected from imaged objects and background, was quantitively assessed by the SNR, estimated at around 9.53 and 18.92 dB for the BF-DC and QPDC images, respectively, indicating improved image contrast acquired by the QPDC ( Figure 2B). As shown in Figure S4, the BF-DC acquired images were typically processed by the steps of background subtraction, thresholding, contour identification, and convex hull. On the other hand, the QPDC-acquired images demonstrated an improved SNR, thereby allowing effective thresholding without the demand of background subtraction.
Noted from Figure S4c,d, the fitted A hull was larger than the A contour for images acquired from both the BF-DC and QPDC, given that the convex hull fitting identifies the concavity artifacts and corrects them. Moreover, the raw A contour was also observed varied by the optical settings. Therefore, the area ratio (R), defined as Equation (4), was used to quantify the level of concavity artifacts or the quality of contour fitting: While the area ratio of a perfectly fitted contour for a smooth boundary, as anticipated for C2C12 cells in the suspension format, was observed to be close to 1, an area ratio over 1.2 was typically considered a shape containing significant amount of concave artifacts. 17,30 Figure 2C showed the analysis of R ratio of images acquired by BF-DC and QPDC after filtering those with R > 1.2. The area ratio R was observed consistently lower for QPDC (R ∼ 1.046) images compared with BF-DC (R ∼ 1.054). Taken together, the presented QPDC system has eased image processing in deformability cytometry by eliminating the need for background subtraction with the improved SNR.

Ameliorated illustration of irregular cellular contours by QPDC
Previous studies have reported that abnormal cells or cells treated with drugs may present an irregular shape composed of concave sections, liable to produce artifacts for the recognition of cellular contours and subsequently biased analysis. 36 To validate whether the improved performance of QPDC may genuinely illustrate the contours of abnormal cells, Cyto D was introduced to artificially alter the cell shape. The effect of cytoskeleton destruction was firstly validated by bright-field investigation and fluorescent staining of the actin fibers (phalloidin staining, red) as shown in Figure 3A. Similar to previous studies, progressive loss of F-actin filaments ( Figure 3B) and formation of F-actin aggregation were observed in both the 1 and 10 μM Cyto D-treated cells compared to the untreated counterparts. 37 Subsequently, both BF-DC and QPDC were employed to measure the deformation of C2C12 cells treated with 10 μM of Cyto D. As noted from Figure 4A, Cyto D treatment not only altered the cell shape but also reduced the image contrast (SNR ∼ 7.63 dB) significantly under the BF-DC imaging, presumably due to significant alterations in cytoskeleton arrangement resulting in a relatively flat intensity profile. 38,39 On the other hand, the contrast of images acquired by QPDC remained minimally altered (SNR ∼ 17.42 dB). Quantification of the area ratio was sub-sequently conducted without filtering. The improved SNR offered by QPDC was observed to be competent in maintaining the accuracy of cell contour recognition ( Figure 4B) and the quality of contour fitting ( Figure 4C, R = 1.08 ± 0.09 for untreated cells; R = 1.11 ± 0.1 for Cyto D-treated cells). In contrast, images acquired by BF-DC appeared particularly poor after the Cyto D treatment ( Figure 4C, R = 1.44 ± 0.49 for Cyto D-treated cells; R = 1.1 ± 0.24 for untreated cells). The results from this set of experiments suggest that the QPDC system may provide an authentic illustration of cellular morphologies, particularly for those suffering from abnormalities or intrinsically in irregular shapes.
While the dosage dependency was not clearly seen from the bright-field images, the deformation of cells under QPDC appeared visually different ( Figure 5A). Cellular deformation was quantitatively analyzed by QPDC as shown in the density scatterplots ( Figure 5B). The mean cell size of C2C12 cells was observed to increase with the Cyto D concentration ( Figure S5a 0.063). These observations demonstrated that the deformation may serve as a label-free biomarker to quantitatively assess the organization of cytoskeletons. While the fluorescence staining may provide a visual inspection of the structural arrangement of actin filaments, QPDC is able to quantify the level of disruption and morphological changes by the collective reporting of cell size and deformability.

Cell size and cellular deformation measurement on different cell lines by QPDC
Subsequently, QPDC was applied for the assessment of different cell lines, including C2C12, HEK293, and PANC-1 cells, as shown in Figure 6. Noted from Figures 6A and S5c,d, the C2C12 cells were measured with smaller cell area (A hull = 274.1 ± 77.5 μm 2 ) and lower deformation (D = 0.026 ± 0.03) compared with HEK293 cells (A hull = 298 ± 88.8 μm 2 , D = 0.042 ± 0.02). The measured deformation was inversely correlated with their elastic moduli as characterized by AFM (E C2C12 = 1.884 ± 0.467 kPa and E HEK293 = 0.8 ± 0.242 kPa). These findings are unsurprising for the functional and morphological differences between C2C12 and HEK293, essentially the skeletal myoblasts as the fundamental units in muscle development, 40,41 and the epithelial cells characterized by their high proliferation capability in culture, 42 respectively. On the other hand, PANC-1, derived from primary pancreatic ductal adenocarcinoma (PDAC) malignancies, exhibited a significantly larger mean cell size ( Figure  S5c) with a moderate deformation degree ( Figure 6A). In agreement with the AFM characterization ( Figure 6B) and other previous studies, 43 the deformability of PANC-1 cells implies their invasive potential of metastasis, as PDAC is considered an aggressive cancer. These results suggested that QPDC may be used to categorize different cell types based on the collective measurements of cell size and deformability.

Phase information extracted by QPDC
An additional physical parameter embedded in the QPDC images, the phase, was extracted as shown in Figure 7. The quantitative phase map was reconstructed at the single-cell level as shown in the representative intensity plot of Figure 7A. The average phase value ( Figure 7B) of untreated C2C12 was quantified as 4.001 ± 0.558, while the phase value was observed to decrease for C2C12 treated with 10 μM of Cyto D (φ = 3.489 ± 0.534), presumably because the cytoskeleton was destructed. Besides, the quantitative phase analysis of three cell types ( Figure 7C) also showed recognizable differences, implying that the averaged phase value may distinguish among the three cell types (C2C12: φ = 3.92 ± 0.45, HEK293: φ = 2.344 ± 0.277, and PANC-1: φ = 2.584 ± 0.422). Moreover, the trend of changes in phase intensity was observed in line with the stiffness of cells characterized by AFM ( Figure 7D); that is, softer cells exhibited relatively low phase values, corroborated by previous studies, 44 and plausibly rationalized by the phase being proportional to the sample thickness and inversely proportional to the refractive index. Furthermore, the refractive index may reflect how cells deformed along the axial direction, 45,46 and the biophysical parameters of cells, such as dry mass, wet mass, and protein concentration. Previous studies have observed a correlation between the cell refractive index and the protein concentration, 47,48 as well as the DNA content, for differentiating the G2/M phases from the G1/S phases. 49 Although beyond the scope of this study, the presented QPDC is expected to fuel the field of label-free phenotyping of cells by providing multiple parameters through the measurements of phase.

CONCLUSION
We have demonstrated that QPDC is able to acquire images with an improved SNR compared to the BF-DC. The comparison shown in this study suggests that the enhanced contrast enabled by phase measurements not only eases the image post-processing, but also allows a faithful illustration of cellular morphologies, which is considered particularly important for the assessment of abnormal cells or cells in irregular shapes. In addition, the phase values extracted from QPDC may provide additional biophysical characteristics of cells. The QPDC platform is therefore expected to complement the existing deformability cytometry and expand the capacity of label-free characterization of cells for biological and clinical applications.

C O N F L I C T O F I N T E R E S T S TAT E M E N T
The authors declare no conflict of interest.