1 Introduction

Biophysical properties of single cells mainly include biomechanical properties (e.g., membrane cortical tension, cytoplasmic viscosity, Young’s modulus) and bioelectrical properties (e.g., membrane capacitance and cytoplasmic conductivity) (Zheng et al. 2013). Variations in single-cell biophysical properties have been closely correlated with various biological processes, including (1) abnormalities of red and white blood cells (Yang et al. 2017; Liu et al. Feb 2018; Honrado et al. 2018; Guruprasad Feb 2019; Hassan, et al. 2017; Prieto 2016; Toepfner, et al. 2018; Bashant Jun 2019); (2) tumour cells in migration and evasion (Ahuja 2019; Ahmmed 2018; Che et al. 2017); and (3) stem cells in differentiation (Zhou et al. (2016); Xavier et al. 2017; Fan et al. 2019; Xavier 2016; Jacobi et al. 2019). Although these biophysical parameters are of importance, conventional characterization techniques (e.g., atomic force microscopy and electrorotation) suffer from limited throughputs and have difficulties in quantifying these biophysical properties simultaneously (e.g., ~ 10 cells per cell type based on atomic force microscopy (Cross et al. Dec 2007) and ~ 20 cells per cell type based on electrorotation (Becker et al. 1995).

Due to dimensional comparisons of microfluidics with biological cells, high-throughput quantification of single-cell biophysical properties have been realized by various microfluidic platforms (Xu et al. 2016; Liang et al. 2019; Darling and Carlo 2015; Wu 2018). For the characterization of single-cell mechanical properties, microfluidics-based deformability cytometry has been used for single-cell mechanical property characterization, where single cells travel rapidly in microfluidic channels and show deformations in response to geometrical constrictions and/or fluid stresses (Deng et al. 2017; Leblanc-Hotte 2019; Reichel et al. 2019). Based on equivalent biomechanical models, these raw data of cell deformations can be interpreted into intrinsic single-cell biomechanical markers including Young’s modulus and cytoplasmic viscosity (Rosendahl 2018; Kang Mar 2019; Armistead et al. 2019; Fregin, et al. 2019; Nyberg et al. 2017; Davidson 2019; Wang et al. 2019). Meanwhile, for the characterization of single-cell electrical properties, microfluidics-based impedance flow cytometry has also been used where single cells under measurements travel rapidly through microfluidic channels with impedance signals sampled (Jaffe and Voldman 2018; Ninno, et al. 2019; Feng et al. 2019; Yang and Ai 2019; Serhatlioglu et al. 2019; Reale et al. 2019; Wang et al. 2017). Based on equivalent bioelectrical models, these raw impedance data can be translated into intrinsic single-cell bioelectrical markers including specific membrane capacitance and cytoplasmic conductivity (Zhao 2018; Zhang 2019). Based on these approaches, although large populations of single-cell biophysical properties were reported, they can only characterize biomechanical or bioelectrical properties, not both and thus their performances in cell type classification and cell status evaluation were limited.

Compared with the quantification of biomechanical or bioelectrical parameters only, simultaneous quantifications of both types of biophysical parameters can definitely provide more insights in cell analysis (Zheng et al. 2013). Recently, high-throughput single-cell characterization of biomechanical and bioelectrical properties has been realized by combining deformability and impedance microfluidic flow cytometry where single cells are forced to deform through microfabricated constriction channels (with cross-sectional areas smaller than cells) continuously with corresponding images captured and impedance recorded, enabling high-rate classifications of wild and modified breast cancer cells of MCF-7 (Yang et al. 2019; Zhou et al. 2018), breast cancer cells of MDA-MB-231 and normal epithelial cells of MCF-10A (Ren et al. Jul (2019)), breast cancer cells of H460, H446, A549, 95D and 95C Wang (2017) and malaria-infected red blood cells (Yang et al. 2017). Although powerful, the majority of these previously reported approaches can only collect raw biophysical markers (e.g., transit times and impedance variations during cell travelling through constrictions) rather than intrinsic biophysical markers (e.g., cytoplasmic viscosity and membrane capacitance) due to the lack of sensing structures and equivalent models, leading to compromised performances in cell type classification.

To address this issue, this paper presents a microfluidic system with crossing constriction channels capable of characterizing intrinsic biophysical markers of single cells (e.g., cytoplasmic viscosity, cytoplasmic conductivity and specific membrane capacitance) continuously. A crossing constriction channel is composed of a major constriction channel for cell deformation and travelling while perpendicularly positioned side constriction channels were used for deformation and impedance measurements of single cells. Based on an equivalent biophysical model for cell travelling at the junction of the major and side constriction channels, these raw biophysical parameters were translated into intrinsic biophysical markers of cytoplasmic viscosity, cytoplasmic conductivity and specific membrane capacitance. In comparison to previous approaches of high-throughput quantifying single-cell biomechanical or bioelectrical markers only, the simultaneous quantifications of both biomechanical and bioelectrical properties of single cells by this approach can significantly improve the success rates of cell type classification.

2 Materials and methods

2.1 Materials

Unless otherwise indicated, all reagents for cell culture and treatments were purchased from Life Technologies Corporation (USA). The HL-60 cell line and the corresponding culture medium (RPMI 1640, w/o Hepes) were purchased from National Infrastructure of Cell Line Resources (China). Granulocytes were isolated from blood samples of healthy donors based on gradient density centrifugation enabled by Percoll (GE Healthcare, USA). Materials required for device fabrication included SU-8 photoresist (MicroChem Corporation, USA), AZ serial photoresist (AZ Electronic Materials Corporation, USA) and 184 silicone elastomer (Dow Corning Corporation, USA).

2.2 Working flowchart

Figure 1 shows the working flowchart for the microfluidic platform for simultaneous quantifications of single-cell cytoplasmic viscosity, cytoplasmic conductivity and specific membrane capacitance. There are mainly three steps of cell preparation, device operation, data processing and analysis. In cell preparation, HL-60 cells and granulocytes were prepared individually. For HL-60 cells, they were treated with CD to modulate cytoskeleton or ConA to modulate membrane properties. As to granulocytes, they were separated from whole blood samples of healthy donors based on gradient density centrifugation (see Fig. 1a).

Fig. 1
figure 1

Schematics for the microfluidic platform capable of characterizing single-cell cytoplasmic viscosity μcy, cytoplasmic conductivity σcy and specific membrane capacitance Csm, simultaneously, which mainly included key steps of a cell preparation, b device operation, c data processing and analysis. In operations, a travelling cell is forced to squeeze through the major constriction channel and aspirated into the side constriction channel with cell aspiration length captured by a high-speed camera and corresponding impedance variations monitored by a lock-in amplifier. Based on the equivalent mechanical and electrical model of a travelling cell at the junction of the major and the side constriction channels, these preliminary biophysical data were translated into intrinsic biophysical markers of cytoplasmic viscosity, cytoplasmic conductivity and specific membrane capacitance

Full size image

In device operation, suspended blood cells were forced to deform through the major constriction channel continuously and invade into the side constriction channel when travelling through the junction of the major and the side constriction channels. During this process, the cell images as a function of time were recorded by an inverted microscope in connection with a high-speed camera while impedance amplitudes and phases were sampled by a lock-in amplifier (see Fig. 1b).

In data processing and analysis, the cell aspiration length Lp, an indicator of cell invading into the side constriction channel was obtained based on image processing and fitted by an empirical model, enabling the extraction of the size-independent intrinsic biophysical marker of μcy. Experimental results of preliminary impedance data were translated into size-independent intrinsic bioelectrical markers of σcy and Csm, based on an equivalent theoretical model. Neural network-based pattern recognition was used to conduct cell-type classification based on μcy, σcy, Csm individually or in combination (see Fig. 1c).

2.3 Cell preperation

HL-60 cells were cultured in suspension with RPMI-1640 medium plus 20% fetal bovine serum. For cell treatments, native HL-60 cells were incubated with 1 μg/mL CD or 10 μg/mL ConA for 30 min, respectively. Before experiments, the concentration of HL-60 cells was adjusted to ~ 5 × 106 cells per mL based on centrifugation and resuspension.

As for normal granulocytes, they were separated from peripheral blood samples of two healthy donors based on density centrifugation. More specifically, solutions of 78% Percoll, 60% Percoll and the whole blood samples were centrifuged at 400 g for 30 min. Then, the layer of granulocytes was collected and resuspended in RPMI-1640 medium at a concentration of 107 cells per mL for the following experiments.

2.4 Device design, fabrication, and operation

The crossing constriction channel developed in this study consists mainly of a major constriction channel and a side constriction channel with rectangular cross-sections. A single cell to be measured needs to be squeezed into the major constriction channel and tightly sealed with the channel walls. Thus, a slightly smaller cross-sectional area should be used for the major constriction channel, comparing to that of the cell under measurement. Based on diameter ranges of the adopted HL-60 cells (i.e., 10–12 μm) and granulocytes from human samples (i.e., 8–10 μm), the geometries of the major constriction channels were chosen as 6 μm in width × 4 μm in depth for HL-60 and 4 μm in width × 4 μm in depth for granulocytes, respectively.

As the deformed cell flows through the crossing point of the major and the side constriction channels, the cytoplasm is partially aspirated into the side channel due to the negative pressure. As to the cross-section dimensions of the side channel, they should be as small as possible to avoid fluid leakage at the corners between the cell and channel walls. However, to decrease the basal impedance without travelling cells, the dimensions should be as large as possible. After a trade-off process, the cross-section dimensions of the side constriction channel were selected as 3 μm in width × 4 μm in depth for HL-60 cells and 2 μm in width × 4 μm in depth for granulocytes, respectively.

The device fabrication was conducted based on conventional microfabrication techniques, which mainly focused on the replication of PDMS-based two-layer crossing constriction channels from a mould master made of silicon at bottom and SU-8 on top, respectively. The key fabrication steps included the formation of the silicon-based layer of crossing constriction channels using deep reactive ion etching, the formation of the SU-8 based second layer using photolithography, the formation of PDMS-based crossing constriction channels using soft lithography, the formation of the glass slide layer with patterned electrodes based on lift-off technologies and bonding between the PDMS layer and the metal patterned glass layer.

In device operation, the microfluidic device was first fully filled with culture medium, followed by the addition of cell suspensions. Negative pressures of 5 kPa and 3 kPa generated from two pressure calibrators (DIP-610 and PACE 5000 pressure calibrators, Druck, UK) were applied on the exit end of the major constriction channel and the two ends of the side constriction channel, respectively, so that cells were forced to travel through the major constriction channel and deform into the side constriction channel. Images of the deformed cell were captured by a high-speed camera (M320S, Phantom, USA) which was connected with an inverted microscope (IX71, Olympus, Japan) at a speed of 1000 frames per second. A two-frequency sinusoidal voltage signal (100 kHz + 250 kHz) was generated by a lock-in amplifier (model 7270 DSP lock-in amplifier, Signal Recovery, USA) and applied on the two ends of the side channel to monitor impedance profiles of the travelling single cell.

2.5 Data processing

To measure Lp as a function of time for each cell tested, a sequence of image processing steps, including frame differencing, thresholding, particle removal using erosion, and edge detection were applied to the images acquired during experiments. Then, Lp was translated into intrinsic biomechanical parameter of μcy using a numerical model modified from the previous study (Wang et al. 2019) (see Fig. 1c). More specifically, a numerical relation between cytoplasmic viscosity μcy and aspiration pressure at the two ends of the side constriction channel Ps, equivalent side channel width Ws, cell radius Rc, aspiration length Lp as a function of time t was obtained based on finite element simulations. This relationship was further represented in a dimensionless format as Lp/Ws = f(Rc/Ws, t × Ps/μcy) where f is a nonlinear function obtained from simulations. This format was further used to fit experimental data of Lp as a function of time through nonlinear curve fitting with μcy as the fitting parameter, from which the cytoplasmic viscosity of single cells cell can be obtained.

The measured impedance data with amplitude and phase at 100 kHz (amp@100 kHz and phase@100 kHz) and 250 kHz (amp@250 kHz and phase@250 kHz) were translated into intrinsic bioelectrical parameters of σcy and Csm using a theoretical model modified from the previous study (Zhang 2019) (see Fig. 1c). Briefly, impedance values for individual cells at 100 and 250 kHz were processed to Rleak (an indicator of sealing properties of deformed cells with the walls of the side constriction channel), membrane capacitances Cm and cytoplasm resistances Rcy. Then, Rcy was translated into σcy based on the format of σcy = Lc/(SSC× Mf × Rcy) where Lc represents cell elongation length at the crossing point, SSC represents cross-sectional area of the side constriction channel and Mf represents a modification factor due to divergence of electrical lines at the crossing point. Meanwhile, Cm was further translated into Csm based on the format of Csm = Cm/Ssc.

In this study, the measurements of multiple samples were conducted with results summarized in scatter plots, box plots and distribution plots. More specifically, scatter plots use the method of drawing scatter points to represent the data distribution range and aggregation. Box plots use the minimum, first quartile, median, third quartile, and maximum values in a set of data to reflect the center position and spread of the data distribution. Distribution plots show the probability distribution of the data in different value intervals.

In addition, neural network based on pattern recognition was conducted to classify cell types where a two-layer (hidden and output layer) feed forward neural network with sigmoid hidden and softmax output neurons as well as a backpropagation algorithm was used (MATLAB 2016, MathWorks Corporation, USA). In this study, for the classification of two cell types (HL-60 vs. HL-60 + CD or HL-60 vs. HL-60 + ConA), the values of cytoplasmic viscosity, cytoplasmic conductivity and/or specific membrane capacitance were used as inputting datasets and the corresponding 0/1 matrices (0 represents HL-60 and 1 represents HL-60 + CD or HL-60 + ConA) were used as output matrices. The complete dataset made of a specific inputting dataset and a corresponding output matrix was divided into a 70% training data used by the network for its adjustment based on generated errors, a 15% validation data used to evaluate the generalization of the network, and a 15% testing data used to independently quantify the performance of the network after training and validation. As to the results, the neural network-based pattern recognition generated a confusion matrix where the orange, green and blue squares represent incorrect responses, correct responses, and the overall accuracies. More specifically, the successful classification rate of classifying two types of cells can be found in the lower right blue square with green color.

3 Results and discussion

With the development of microfluidics, a variety of microfluidic structures have been proposed for single-cell biophysical property characterization (Xu et al. 2016; Liang et al. 2019; Darling and Carlo 2015; Wu 2018). Among these sensing structures, crossing constriction channels have been recently used for single-cell biomechanical or bioelectrical property characterization.

More specifically, when the exit end of the major constriction channel and two ends of the side constriction channel were applied with negative pressures, travelling single cells were aspirated through the major constriction channel from inlet and when they reached the junction of the major and side constriction channels, cells deformed into side constriction channels partially. This deformation process was captured and translated into cytoplasmic viscosity based on the corresponding biomechanical model (Wang et al. 2019).

Meanwhile, when the exit end of the major constriction channel was applied with a negative pressure and two ends of the side constriction channel were sealed from environmental pressures but connected with an impedance analyzer, single cells were aspirated through the major constriction channel from inlet and when they reached the junction of the major and side constriction channels, cells sealed rather than invaded into the side constriction channel, leading to blocked electric lines and varied impedance data. These impedance variations were sampled and translated into cytoplasmic conductivity and specific membrane capacitance based on the corresponding bioelectrical model (Zhang 2019).

This study was aimed to characterize both biomechanical and bioelectrical properties of single cells leveraging the crossing constriction channel by addressing the conflicting point that previously cells were forced to aspirate into the side constriction for mechanical property characterization and seal rather than invade into the side constriction for electrical property characterization. To deal with this issue, in this study, negative pressures were applied on the exit end of the major constriction channel and the two ends of the side constriction channel, respectively, so that cells deformed into the side constriction channel with image captured and translated into cytoplasmic viscosity. Meanwhile, a lock-in amplifier was applied on the two ends of the side channel to monitor impedance profiles of the travelling single cell invading into the side constriction channel while both images and impedance profiles were used for the interpretation of cytoplasmic conductivity and specific membrane capacitance.

Figure 2 shows microscopic pictures of a HL-60 cell (a, b, c), a CD-treated HL-60 cell (e, f, g), a ConA-treated HL-60 cell (I, j, k) and a granulocyte from a healthy donor (m, n, o), travelling in the major constriction channel and deforming into the side constriction channel. Experimentally measured aspiration lengths, two-frequency amplitude and phase profiles as a function of time for these four representative cells are shown in d, h, l and p. The red lines and the crossing labels show the results of curve fitting and the location of the maximal impedance amplitudes, which were further used for single-cell mechanical and electrical property characterization, respectively.

Fig. 2
figure 2

Microscopic pictures of a HL-60 cell (a, b, c), a CD-treated HL-60 cell (e, f, g), a ConA-treated HL-60 cell (i, j, k) and a granulocyte from healthy donor I (m, n, o) travelling in the major constriction channel and deforming into the side constriction channel. Experimentally measured aspiration lengths, two-frequency amplitude and phase profiles as a function of time for these four representative cells are shown in (d, h, l and p). The red lines and the crossing labels show the results of curve fitting and the position of the maximal impedance amplitudes, which were further used for single-cell mechanical and electrical property characterization. Note that R2 represents the degree of fitting of the regression curve to the experimental data where the closer the value of R2 is to 1, the better the degree of fitting

Full size image

Based on raw image and impedance data shown in Fig. 2, the processing durations for single cells travelling through the crossing constriction channel were in the ranges of ~ 0.1 s and thus the throughput of this microfluidic platform was roughly ~ 1 cell per second. In this study, when HL-60 cells with diameters of 10–12 μm and granulocytes from human samples with diameters of 8–10 μm were characterized by this microfluidic platform, the blockage of the major constriction channel was not often noticed. As to the blockage of the side constriction channel, this was not a concern since assymmetries of aspiration length in the two side constriction channels were seldom located. Currently, the bottleneck of this microfluidic platform is the high-speed camera which was used to record images of cell aspiration in the crossing constriction channel. In operations, the high-speed camera had to record images for several minutes with its internal memory fully occupied and then a few more minutes were required to transfer data from the high-speed camera to a computer. Due to this limitation, the high-speed camera cannot work continuously and thus it took roughly half an hour to collect both impedance and image data from hundreds of single cells.

Figure 3(a, b, c) shows scatter plots of μcy vs. Rc, box and distribution plots of μcy for HL-60 cells, CD-treated HL-60 cells and ConA-treated HL-60 cells. Specifically, three quartile values and peaking ranges of μcy of HL-60 cells under different conditions were quantified as 21.5 Pa•s, 40.4 Pa•s, 63.8 Pa•s and 42.9–68.7 Pa•s (ncell = 329, HL-60), 14.8 Pa•s, 27.2 Pa•s, 58.8 Pa•s and 16.8–26.8 Pa•s (ncell = 320, HL-60 + CD), 15.8 Pa•s, 32.0 Pa•s, 70.1 Pa•s and 26.8–42.9 Pa•s (ncell = 320, HL-60 + ConA).

Fig. 3
figure 3

a, b, c: Scatter plots of μcy vs. Rc (a), box (b) and distribution plots (c) of μcy for HL-60 cells, CD-treated HL-60 cells, and ConA-treated HL-60 cells. d, e, f: Scatter plots of σcy vs. Rc (d), box (e) and distribution plots (f) of σcy for HL-60 cells, CD-treated HL-60 cells, and ConA-treated HL-60 cells. g, h, i: Scatter plots of Csm vs. Rc (g), box (h) and distribution plots (i) of Csm for HL-60 cells, CD-treated HL-60 cells, and ConA-treated HL-60 cells. Confusion matrix for the classifications of HL-60 cells vs. CD-treated HL-60 cells (j) and HL-60 cells vs. ConA-treated HL-60 cells (k) based on single-cell parameters of μcy, σcy, Csm, σcy and Csm and μcy and σcy and Csm

Full size image

The quantified values of μcy for HL-60 cells fell into the previously reported ranges (~ tens of Pa•s) (Wang et al. 2019) where only biomechanical parameter of μcy was measured, suggesting that electrical phenotyping of cells has no significant effects on cellular mechanical properties. HL-60 cells with the CD treatment indicated lower values of μcy in comparison to HL-60 cells (27.2 Pa•s Vs. 40.4 Pa•s), which agreed with the assumption that the treatment of cells with CD can compromise cytoskeletons. Neural network-based pattern recognition was conducted to classify HL-60 cells vs. HL-60 + CD cells based on μcy, producing a success rate of 59.6% (see Fig. 3(j)). These results indicate that although μcy can be used to differentiate HL-60 cells with and without the treatment of CD, the only use of μcy can only provide limited insights in cell type classification.

Meanwhile, although HL-60 cells with the ConA treatment indicated marginally lower values of μcy in comparison to HL-60 cells (32.0 Pa•s Vs. 40.4 Pa•s), since a low success rate of 54.2% was obtained to classify HL-60 cells vs. HL-60 + ConA cells (see Fig. 3k), it is concluded that μcy is insensitive to the ConA treatment of cells, since ConA only binds with glycoproteins at the membrane domain and has limited effects on the cytoplasmic domains of cells.

Figure 3(d, e, f) shows scatter plots of σcy vs. Rc, box and distribution plots of σcy for HL-60 cells, CD-treated HL-60 cells, and ConA-treated HL-60 cells. Specifically, three quartile values and peaking ranges of σcy of HL-60 cells under different conditions were quantified as 0.54 S/m, 0.60 S/m, 0.65 S/m and 0.55–0.60 S/m (ncell = 329, HL-60), 0.67 S/m, 0.74 S/m, 0.78 S/m and 0.70–0.75 S/m (ncell = 320, HL-60 + CD), 0.44 S/m, 0.50 S/m, 0.62 S/m and 0.45–0.50 S/m (ncell = 320, HL-60 + ConA).

The quantified values of σcy for HL-60 cells in this study fell into the previously reported ranges (~ 1 S/m (Wang 2017)) for single-cell electrical property characterization, suggesting that the simultaneous mechanical phenotyping of cells has no significant effects on electrical properties of cytoplasm. HL-60 cells with the CD treatment indicated higher values of σcy in comparison to HL-60 cells (0.74 S/m Vs. 0.60 S/m), which may again result from cytoskeleton compromise. Neural network-based pattern recognition was conducted to classify HL-60 cells vs. HL-60 + CD cells based on σcy, producing a success rate of 81.1% (see Fig. 3j). These results indicate that σcy can function as an effective parameter in classifying HL-60 cells and HL-60 + CD cells.

Meanwhile, HL-60 cells with ConA treatment indicated marginally lower values of σcy in comparison to HL-60 cells (0.50 S/m vs. 0.60 S/m) and neural network-based pattern recognition produced a low success rate of 52.5% when these two cell types were classified. These results again indicate that σcy is insensitive to the ConA treatment of cells, since ConA only binds with glycoproteins at the membrane domain and has limited effects on the cytoplasmic domains of cells.

Figure 3(g, h, i) shows scatter plots of Csm vs. Rc, box and distribution plots of Csm for HL-60 cells, CD-treated HL-60 cells, and ConA-treated HL-60 cells. Specifically, three quartile values and peaking ranges of Csm of HL-60 cells under different conditions were quantified as 2.19 μF/cm2, 2.46 μF/cm2, 2.94 μF/cm2 and 2.20–2.50 μF/cm2 (ncell = 329, HL-60), 3.03 μF/cm2, 3.24 μF/cm2, 3.42 μF/cm2 and 3.10–3.40 μF/cm2 (ncell = 320, HL-60 + CD), 2.88 μF/cm2, 3.09 μF/cm2, 3.34 μF/cm2 and 2.80–3.10 μF/cm2 (ncell = 320, HL-60 + ConA).

The quantified values of Csm for HL-60 cells in this study fell into the previously reported ranges (~ 1 μF/cm2 (Wang 2017) for single-cell electrical property characterization, suggesting that the simultaneous mechanical phenotyping of cells has no significant effects on electrical properties of cell membranes. HL-60 cells with both CD and ConA treatments indicated higher values of Csm in comparison to HL-60 cells (3.24 μF/cm2 vs. 3.09 μF/cm2 vs. 2.46 μF/cm2), which may result from the fact that both the treatments of CD and ConA can affect membrane properties of single cells. Neural network-based pattern recognition produced success rates of 80.1% for HL-60 cells vs. HL-60 + CD cells and 72.5% for HL-60 cells vs. HL-60 + ConA cells based on Csm. These results indicate that Csm is sensitive to both CD and ConA treatments of cells and can be effectively used for cell type classification.

In addition, in comparison to the cell type classification based on μcy only, when two or three intrinsic parameters were used in combination, significantly higher success rates were obtained. For the classification of HL-60 cells with and without CD treatments, the classification rate based on σcy and Csm (83.4%) was higher than the classification rates based on σcy (81.1%) or Csm (80.1%) only, which further validated the observation that CD treatment can produce differences in both membrane and cytosolic portions. As to the classification of HL-60 cells with and without ConA treatments, the classification rate based on σcy and Csm (75.6%) was marginally higher than the classification rate based on Csm (72.5%) only, which further validated the observation that ConA treatment can only produce difference on cell membranes without significantly modulating cytosolic portions. Furthermore, the classification rates based on μcy and three parameters in combination were quantified as 59.6% vs. 88.0% for HL-60 cells vs. HL-60 + CD cells and 54.2% vs. 75.6% for HL-60 cells vs. HL-60 + ConA cells, which confirmed the importance of collecting multiple biophysical markers in cell type classification and cell status evaluation.

Furthermore, the microfluidic platform was used to quantify cytoplasmic viscosity, cytoplasmic conductivity and specific membrane capacitance of single granulocytes from healthy donors. Figure 4(a, b, c) shows scatter plots of μcy vs. Rc, box and distribution plots of μcy for donor I and II. Specifically, three quartile values and peaking ranges of μcy of these granulocytes were quantified as 9.5 vs. 10.4 Pa•s, 15.6 vs. 17.2 Pa•s, 24.6 vs. 27.8 Pa•s and 10.5–16.8 vs. 16.8–26.8 Pa•s (ncell = 406 of donor I vs. ncell = 458 of donor II). In addition, the success rate of cell type classification based on μcy was quantified as 53.8% for donor I vs. donor II. Comparable values of μcy for donor I and II and the low success rate of cell-type classification between donor I and donor II suggest that there may exist a range of μcy for normal granulocytes (e.g., 10–20 Pa•s).

Fig. 4
figure 4

a, b, c Scatter plots of μcy vs. Rc (a), box (b) and distribution plots (c) of μcy for granulocytes of Donor I and II; d, e, f: Scatter plots of σcy vs. Rc (d), box (e) and distribution plots (f) of σcy for granulocytes of Donor I and II. g, h, i: Scatter plots of Csm vs. Rc (g), box (h) and distribution plots (i) of Csm for granulocytes of Donor I and II

Full size image

Figure 4(d, e, f) shows scatter plots of σcy vs. Rc, box and distribution plots of σcy for donor I and II. Specifically, three quartile values and peaking ranges of σcy of these granulocytes were quantified as 0.42 vs. 0.38 S/m, 0.49 vs. 0.43 S/m, 0.58 vs. 0.53 S/m and 0.45–0.50 vs. 0.40–0.45 S/m (ncell = 406 of donor I vs. ncell = 458 of donor II). In addition, the success rate of cell type classification based on σcy was quantified as 55.3% for donor I vs. donor II. Comparable values of σcy for donor I and II and the low success rate of cell type classification between donor I and donor II suggest that there may exist a range of σcy for normal granulocytes (e.g., 0.4–0.6 S/m).

Figure 4(g, f, i) shows scatter plots of Csm vs. Rc, box and distribution plots of Csm for donor I and II. Specifically, three quartile values and peaking ranges of Csm of these granulocytes were quantified as 2.06 vs. 2.01 μF/cm2, 2.25 vs. 2.25 μF/cm2, 2.55 vs. 2.52 μF/cm2 and 2.20–2.50 vs. 1.90–2.20 μF/cm2 (ncell = 406 of donor I vs. ncell = 458 of donor II). In addition, the success rate of cell type classification based on Csm was quantified as 53.8% for donor I vs. donor II. Comparable values of Csm for donor I and II and the low success rate of cell type classification between donor I and donor II suggest that there may exist a range of Csm for normal granulocytes (e.g., 2.0–3.0 μF/cm2).

In summary, the microfluidic platform developed in this study was demonstrated to characterize cytoplasmic viscosity, cytoplasmic conductivity and specific membrane capacitance from hundreds of single cells simultaneously. Based on these intrinsic biophysical markers, (1) successful classifications of HL-60 vs. HL-60 + CD cells with compromised cytoskeletons and HL-60 vs. HL-60 + ConA with membrane glycoproteins bound were realized; (2) estimated ranges including 10–20 Pa•s for μcy, 0.4–0.6 S/m for σcy and 2.0–3.0 μF/cm2 for Csm of normal granulocytes were obtained. From the perspective of technical developments, future studies may focus on the measurements of more biophysical parameters (e.g., cortical tension and transmembrane potential) at the single-cell level to further improve the success rates of cell type classification. As to applications, this microfluidic platform can be further used to test more samples of white blood cells from both patients and benign counterparts to evaluate the potential in disease diagnosis at a label-free manner.