Analyzing cell mechanics in hematologic diseases with microfluidic biophysical flow cytometry

Abstract

Pathological processes in hematologic diseases originate at the single-cell level, often making measurements on individual cells more clinically relevant than population averages from bulk analysis. For this reason, flow cytometry has been an effective tool for single-cell analysis of properties using light scattering and fluorescence labeling. However, conventional flow cytometry cannot measure cell mechanical properties, alterations of which contribute to the pathophysiology of hematologic diseases such as sepsis, diabetic retinopathy, and sickle cell anemia. Here we present a high-throughput microfluidics-based ‘biophysical’ flow cytometry technique that measures single-cell transit times of blood cell populations passing through in vitro capillary networks. To demonstrate clinical relevance, we use this technique to characterize biophysical changes in two model disease states in which mechanical properties of cells are thought to lead to microvascular obstruction: (i) sepsis, a process in which inflammatory mediators in the bloodstream activate neutrophils and (ii) leukostasis, an often fatal and poorly understood complication of acute leukemia. Using patient samples, we show that cell transit time through and occlusion of microfluidic channels is increased for both disease states compared to control samples, and we find that mechanical heterogeneity of blood cell populations is a better predictor of microvascular obstruction than average properties. Inflammatory mediators involved in sepsis were observed to significantly affect the shape and magnitude of the neutrophil transit time population distribution. Altered properties of leukemia cell subpopulations, rather than of the population as a whole, were found to correlate with symptoms of leukostasis in patients—a new result that may be useful for guiding leukemia therapy. By treating cells with drugs that affect the cytoskeleton, we also demonstrate that their transit times could be significantly reduced. Biophysical flow cytometry offers a low-cost and high-throughput diagnostic and drug discovery platform for hematologic diseases that affect microcirculatory flow.

Introduction

Flow cytometry has been an effective tool for high-throughput analysis of single cells and is widely used to quantify biological properties such as surface protein expression and signaling activity.¹ By fluorescently detecting rare cells and distinct sub-populations that lead to disease,^2,3 this tool has revolutionized medicine and has shown that measuring the properties of individual cells can be critical for effective health care.

Flow cytometry, however, cannot measure cellular mechanical properties, alterations of which contribute to the pathophysiology of various hematologic diseases, where occlusion of the microvasculature by blood cells can have profound effects on blood flow in the brain, lungs, and other vital organs.^4–6 Furthermore, as relatively few mechanically-altered blood cells are theoretically sufficient to induce microvascular pathology,^7,8 symptoms associated with hematologic diseases may be attributable to the minority subpopulation of pathologic blood cells that would not be detected with a bulk assay that measures average properties. Therefore, a need exists for high-throughput single-cell mechanical analysis to address basic questions in cell mechanics and provide diagnostic data for clinical hematology.

To address this need, several techniques have been developed to assess mechanical properties of blood cells, but none have successfully demonstrated high-throughput analysis of single-cell mechanical properties in clinically-relevant conditions. The existing techniques can be divided into bulk and single-cell approaches. Bulk microfiltration techniques have been used for several decades to study microcirculatory diseases by measuring changes in pressure or flow rate of blood cell solutions as milliliter-scale volumes of fluid are passed through micropore filters.^9,10 Advances in this technique involve the coupling of microfabricated sieves with optical microscopy for direct cell visualization^11–14 and improved geometric modeling of the microvasculature.^13,14 While these studies have revealed strong evidence of altered mechanical behavior of cells in various disease states, they are only able to measure population blood cell behavior or endpoint results, such as occlusion. Importantly, they cannot attribute aberrations in flow behavior to shifts in properties of the whole population or smaller sub-populations.

Single-cell approaches such as micropipette aspiration and atomic force microscopy are able to measure changes in the mechanical properties that are associated with disease at the single-cell level.^15–17 While precise, these approaches are limited by their inherent low throughput and high equipment cost, making them unlikely candidates for clinical use and unable to adequately describe cell populations.

Higher throughput techniques have been devised to measure single-cell transit time through micropores. Single-pore techniques, which quantify transit time across a single pore filter using either an opto-electronic^18,19 or resistive pulse detection technique,^20,21 can apply low physiologically relevant pressure across a capillary-sized pore, resulting in leukocyte transit times on the order of seconds.²¹ Single-pore techniques are limited in throughput however because a single cell can clog the system. A multiple pore technique (known as the Cell Transit Analyzer, or CTA)^22–27 has been developed that measures cell transit time across a filter with typically 30 pores and thus has a much higher throughput. However, the system discards transit time measurements when more than one cell is passing across the filter at once, because it is unable to separate the two transits from the electrical signal. As a result, this system may therefore be biased to ignore the contribution of slow transiting cells to the overall population. Furthermore, multi-pore systems work best with very fast transit times (1‒100 ms), which require high pressures and large pores, both of which make comparison to physiological flow through the microcirculation more difficult.

Microfluidics offers the promise of high-throughput single-cell analysis of cell deformability for clinical applications. Using optical stretching,^28,29 shear flow,³⁰ or capillary-like microchannels,^31–34 microfluidic-based systems have enabled observation of differences between cell populations with single-cell resolution. Collectively, these devices enable what we generally refer to as ‘biophysical’ flow cytometry, and like standard flow cytometry, they are capable of high-throughput single-cell analysis. In addition, microfluidic devices can more accurately model physiological flow and branching in the microvasculature when compared to previous high-throughput single-cell devices. Although studies of these microfluidic techniques highlight device capabilities, the direct relevance of these systems for diagnostic or therapeutic purposes remains unproven.

Here we show that a simple microfluidic device can be used to rapidly generate clinically useful single-cell data that provides new insight into diseases states and new motivation for the measurement of mechanical properties with microdevices. Our biophysical flow cytometry system uses automated image analysis to track large numbers of individual cells as they traverse a microfluidic capillary network and measures the effect of cell deformability and cell size on cellular transit times. To demonstrate the value of quantifying changes in physical properties of blood cells at the single-cell level, we used two conditions as model disease states: (i) sepsis, a process in which the biophysical response of neutrophils to inflammatory mediators are well characterized,^5,27,35 and (ii) leukostasis, a poorly understood and often fatal complication of leukemia in which cell mechanical properties of cancer cells are thought to lead to microvascular obstruction. We show three primary results with the device. First, blood cell transit time into microchannels is not normally distributed. Second, inflammatory mediators and drugs that affect the cytoskeleton alter the shape and magnitude of the cell transit time population distribution. Third, the distribution of transit time correlates with disease pathology and symptomology in leukostasis.

Experiment

Microfluidic device

The main element of the device is the microfluidic capillary network (Fig. 1). Using standard microfabrication techniques, a SU-8 master mold of the capillary network was patterned onto a silicon wafer. PDMS was then molded onto the SU-8 master at 60 °C for at least 2.5 h, removed, and bonded after exposure to oxygen plasma to a PDMS spin-coated glass slide to form the capillary network. The glass slide was spin-coated with PDMS to ensure a consistent surface material on all sides of the device. Two wide channels bypass the capillary network to maintain a relatively constant pressure drop across the network even when cells plug the smallest microchannels (Fig. 1A).

Fig. 1.

Biophysical flow cytometer device. (A) Blood cells were loaded into a syringe and flowed into the device at a constant flow rate. The cytometry device trifurcates into two wide bypass channels and a network of bifurcating channels which split into 64 parallel capillary-like microchannels. Scale bar 1 mm. (B) 16 of these microchannels are shown here. Scale bar 100 µm. (C) A confocal image of fluorescein solution inside several of the microchannels.

The device was designed to be geometrically similar to capillary networks in vivo while creating a standard system for quantitative measurements. Capillaries in vivo range in diameter between 5–10 µm and length from 50–500 µm.³⁶ Linear flow rate through capillaries has been measured to be 0.42–1.3 mm s^{−1 36–38} with a pressure drop determined to be between 10–1000 Pa,^38,39 depending on capillary length. Comparatively, the microchannels are 5.89±0.08 µm wide (mean±SD) wide by 13.3 µm tall by 130 µm long (Fig. 1B, C). Flowing cell media at 1.0 µL min⁻¹, pressure dropped approximately 30 Pa across the microchannels, determined by Poiseuille flow equations,⁴⁰ which resulted in an average linear flow rate of 0.50 mm s⁻¹ in the smallest channels. Because of the low Reynolds and Womersly numbers of the in vivo microcirculation (<<1), which relate the inertial and pulsatile effects to viscous effects, forces due to transient and convective acceleration of peristaltic flow are negligible,³⁶ and hence, steady flow was used instead. Using constant flow with bypass channels instead of using pressure reservoirs makes the system less sensitive to air bubbles in the device and placement of the inlet and outlet holes to the device while also maintaining a high linear flow rate, which reduces the tendency of cells to settle onto the bottom of the device. Even with bypass channels, the constant flow approach can increase pressure drop across the microchannels as more of them become occluded by cells, though we show that this effect did not significantly alter cell transit time in our system (see data analysis subsection).

Our device has several advantages when compared to previous filter-based devices which can measure single-cell transit times. Optical imaging of cells deforming into multiple channels at the same time offers higher throughput than the single-pore devices and more physiological flow conditions than the multiple-pore devices. Optical imaging also allows for measurement of multiple parameters including cell size and transit time, as shown in this study, as well as protein expression and cell surface markers through integration of fluorescence. Lastly, the successive bifurcations of one large channel into multiple microchannels is a significant improvement over existing techniques and more closely mimics physiological conditions. In particular, leukocytes have been noted to collect preferentially at bifurcations into small capillaries,⁴¹ Also, total cross-sectional area of the vasculature increases as vessels decrease in size, typically increasing by a factor of 1.1–1.4 at each branch.³⁷ This increase in cross-sectional area results in a reduction of the linear flow rate as blood transits to the small vessels. Our device captures this deceleration into the smallest microchannels, as compared to filter-based techniques which produce the opposite effect of a large increase in linear flow rate as the fluid crosses the small cross section of the pore(s).

Clearly differences exist between this microfabricated device and in vivo capillaries including rectangular instead of a circular cross section, which results in flow around cells stuck within channels, microchannels in parallel without collateral vessels bridging them, stiffer vessel walls, the lack of a living endothelium, and the lack of adhesion molecules. Based on these simplifications, our device serves as a first-order model to specifically study blood cell deformation under physiologic flow conditions in the absence of other factors.

Cells

Fresh patient leukemia cells used for experiments were obtained from newly diagnosed acute leukemia patients with detectable leukemia cells in their peripheral blood. Each sample was then immediately isolated via centrifugation with Histopaque 1077 (Sigma–Aldrich). Leukemic cells, which comprised >90% of the mononuclear fraction, were then either used immediately or cryopreserved for future use. Cell viability (trypan blue exclusion) was >95%. We have shown previously with atomic force microscopy (AFM) that the freeze–thaw cycle does not significantly alter cell stiffness.¹⁶ To confirm this, we flowed a fresh acute myeloid leukemia sample through the device and later flowed cells from this same sample after they had been frozen and thawed and found no significant difference in cell transit time (Mann–Whitney, p = 0.66).

Red blood cells (RBCs) and neutrophils were obtained from the peripheral blood of healthy adult volunteers using a dextran/Histopaque protocol published previously.⁴² Briefly, peripheral blood was collected into phlebotomy tubes with EDTA as an anticoagulant, which has been reported to minimize neutrophil activation during the isolation process.^43,44 The blood was then transferred to a dextran–EDTA solution, and RBCs were purified from the leukocyte/plasma suspension via sedimentation at gravity. Neutrophils were further purified with Histopaque 1077 and 1119 (Sigma-Aldrich). Institutional review boards approved all experiments, and informed consent was obtained for each sample.

The HL60 myeloid leukemia cell line (ATCC) was used for drug experiments. Because HL60 cells are larger in diameter than patient samples, a device of the same geometry but with slightly larger microchannels (7.9 × 12.9 µm) was used, and the flow rate was decreased from 1.0 µL min⁻¹ to 0.6 µL min⁻¹. In several experiments, cells were exposed to different drugs. Neutrophils were exposed to 10 ng mL⁻¹ fMLP (N-formyl-methionyl-leucyl-phenylalanine) for 5 min, and HL60 cells were exposed to 2 µm cytochalasin D for 1 h or 1 mM pentoxifylline for 3 h.

Cytometry experiment

After bonding, the channels were passivated for 30 min with either 5% autologous human plasma or 20 mg mL⁻¹ bovine serum albumin (Sigma) when plasma was not available to block non-specific adhesion. Similar protocols have been used previously with micropipette aspiration experiments to reduce non-specific adhesion.^45–48 Cells were diluted in RPMI cell culture media (Gibco) to a concentration of 3000 cells µL⁻¹. Cells were loaded into a 250 µL heated glass syringe (Hamilton) (37 °C) and pumped with a syringe pump (Harvard Apparatus) into the heated device (37 °C) at 1.0 µL min⁻¹ (Fig. 1). After passing through the device, the cells drained to a waste container. Experiments were imaged with a 20× objective (Edmund Optics) and live images were streamed at 30 frames per second with a CCD camera (Watec) via a frame grabber (National Instruments) into Labview 8 (National Instruments) and saved as an AVI for future analysis. As an example, neutrophils can be observed flowed into the device in Movie 1 (see ESI†, scalebar 50 µm). Thirty-two channels were tracked at a time in all experiments. Between experiments, the device was flushed with media from an additional syringe to clear residual cells from the microchannels.

Throughput for the device is approximately 50–100 cells min⁻¹ and is dependent on flow rate, number of microchannels in the device and in the field of view, and cell concentration. It also depends on the mechanical properties of the cells themselves, which determines how often channels become occluded and thus not available for measurement of other cells transiting through the device. Throughput could be raised by increasing the flow rate and therefore pressure drop, but then the cells would be subjected to non-physiologically relevant conditions. Increasing cell concentration could also increase throughput but would also increase cell–cell interactions, making it difficult to isolate the effects of single-cell properties on flow through the microchannels.

Automated transit time analysis

Essential to a high-throughput technique for quantifying single-cell biophysical properties is a method to quantify the transit time of large numbers of individual cells through the microchannels. All analysis was performed with custom-written scripts in Matlab 7. To detect cell transit time, regions of interest (ROIs) were selected for each of the microchannels and the channels leading into them (Fig. 2A, B). The standard deviation (SD) of pixel intensity was tracked for each ROI for each frame (Fig. 2C, movie frames correspond to those of ESI Movie 2‡). A cell’s presence increased the SD of the pixel intensity of the ROI due to light scattering. The SD of each ROI was thresholded to determine the presence of a cell. Pertinent data such as transit time and blockage initiation was subsequently determined. Cells were included for analysis only if they did not come in contact with another cell while transiting through the microchannels. This technique effectively detects differences in transit time of cells passing through the cytometer.

Fig. 2.

Image analysis of cell transit time through microchannels. (A) Using video microscopy, cells are tracked as they pass into the smallest of the capillary channels. 16 of the 64 parallel microchannels are shown here, though 32 were tracked at a time for the data reported in this paper. Scale bar 50 µm. An example of a cytometry experiment can be seen in ESI Movie 1‡ (scale bar 50 µm). Regions of interest (ROIs) are selected to measure cell transit time into the channels, which is defined as the amount of time it takes for a cell to enter ROI 1 and leave either ROI 4a or 4b (Bi). This neutrophil is visually tracked as it transits through ROI 1 (Bi), ROI 2 (Bii), ROI 3 (Biii), and ROI 4 (Biv). Scalebar 10 µm. Cells are tracked in each ROI by measuring the standard deviation of the pixel intensity in that region. If there is no cell present, the standard deviation is low. When a cell is present, the light scattering creates contrast in the ROI and standard deviation goes up. (C) Standard deviation of pixel intensity for each ROI is displayed versus time, with blue being low standard deviation and red being the highest standard deviation. Frames (Bi–iv) are highlighted by arrows. Two additional cells are shown passing through the ROIs at later times. These corresponding cells can be seen transiting through the microchannels in ESI Movie 2‡ (scale bar 20 µm). A typical histogram of the distribution of neutrophil transit time is shown in (D). Cells that did not transit in less than 8 s are pooled into the “more” bin.

Data analysis

To compare transit times of cell populations from different samples, a non-parametric ranked Mann–Whitney test was used to quantify differences with SPSS software. Cells that became stuck were given equal ranks in each population and included in analysis. All box plots represent the 25th, 50th, and 75th percentiles of the data. Chi-square analyses were used to determine significant differences in percent of cells that occluded vessels. Lastly, to determine differences in occlusion rate of capillaries a moderated linear regression was used.

No statistically significant differences in transit time between different microchannels were found, indicating consistency of flow through the parallel channels (p = 0.17–0.82, n = 5 experiments). Cell transit time was also consistent over the course of an experiment, which normally ran between 4–8 minutes, and there was no statistical correlation between cell transit time and the order that the cells entered the channels (which is proportional to elapsed experiment time, range of r = 0.008 to r = 0.16, n = 6 experiments), indicating that the pressure changes that did occur due to clogged channels did not have a significant effect on transit time. No significant differences in transit time were observed between different experiments of the same cell type (p = 0.41–0.75, n = 6 experiments).

Results and discussion

Blood cell transit time through microchannels is not normally distributed

Statistical descriptions of cellular properties, such as deformability, in terms of a mean and standard deviation implicitly assume the property is normally distributed. Live tracking of individual cells passing through the microchannels can directly examine blood cell mechanical heterogeneity in microcirculatory flow conditions and identify different populations of cells within the same cell type. We found that typical histograms of cellular transit times through the microchannels show distinct non-Gaussian distributions, indicating that blood contains biophysically heterogeneous cell populations within the same cell type (Fig. 2D). These results confirm the previously proposed existence of multiple subpopulations of blood cell deformability^44,49 based on mathematical models of bulk microfiltration data. Transit time distributions of all blood cell populations analyzed in this study were found to be significantly positively skewed, with long right-sided tails (skewness range: 1.4 to 7.8, range of standard error of skewness: 0.012 to 0.21), agreeing with histogram data obtained from single-cell transit time measurements through micropores.^21,22,50 Measurements of dispersion that ignore skewness, such as standard deviation or interquartile range, were large and consistent with previously published measurements of cell mechanical properties using comparable techniques,^15,35,51,52 suggesting that information about subpopulation properties is masked by a simple mean and dispersion about the mean. Thus bulk methods, which cannot assess the impact of outliers, or low throughput techniques, which do not yield large enough sample sizes, cannot accurately describe cell biophysical properties. Biophysical flow cytometry, however, allows for robust comparisons of different blood cell types and biophysical effects of drugs and biological modifiers.

Inflammatory mediators change the shape and magnitude of cell transit time population distribution

The ability of neutrophils to successfully deform into capillaries much smaller than their diameter is essential for their circulation through the vascular system.³⁶ Increased rigidity of neutrophils has been associated with their increased capillary retention leading to tissue ischemia seen in sepsis and acute respiratory distress syndrome (ARDS).^12,15,27,53 Studies have shown that inflammatory mediators, such as N-formyl-methionyl-leucyl-phenylalanine (fMLP) or Tumor Necrosis Factor-α (TNF-α), are responsible for this stiffening behavior and can alone induce clogging of microfilters and increase neutrophil retention in the pulmonary capillaries.^5,27,54 As a control experiment to test the sensitivity of the microchannels in our device to changes in cell stiffness and also to address the question of how a population distribution of transit times will be affected by inflammatory mediators, we evaluated neutrophils exposed to inflammatory mediators with our device.

Populations of neutrophils with and without exposure to fMLP were flowed into the device (Fig. 3). A typical control neutrophil (without fMLP exposure, Fig. 3A) took less time to enter a microchannel than an fMLP exposed neutrophil (Fig. 3B). As a population, the transit time of fMLP exposed neutrophils was significantly longer than control neutrophils (control n = 204, fMLP n = 182, p<0.001). The histogram of transit times of the control neutrophils shows a single peak at 0.24 s with a long tail having relatively few cells (Fig. 3C). A small fraction of the cells (5%) took longer than 8 s to enter into the channels. After addition of fMLP, the single peak in the histogram at short timescales dropped considerably and nearly doubled to 0.40 s (Fig. 3C). Interestingly, the most marked difference was in the increase of cells that took longer than 8 s to transit (increased to 21%), an indication that the tail of the population distribution had been extended. The population behavior is summarized in a boxplot (Fig. 3C inset) that shows the 25th, 50th, and 75th percentiles of the data shifting significantly upwards.

Fig. 3.

Exposure to the inflammatory mediator fMLP increases neutrophil transit time through microchannels. To determine sensitivity of the device to physiologically relevant conditions, neutrophils were passed through the device both with and without exposure to fMLP. (Ai–vi) A neutrophil not exposed to fMLP transits into and across the channel in less than 0.60 s. (Bi–vi) After exposure to fMLP, a neutrophil takes longer than 1.50 s to compress into the channel and travel across it. Cells appear as double cells in Ai, Aiv, and Bi due to interlacing. Scale bar 10 µm. (C) The histograms of the fMLP-exposed (red) and non-fMLP exposed (blue) neutrophil populations are distinctly different. After exposure to fMLP (red), the distribution of transit times increased significantly, with over 20% of neutrophils passing in greater than 8 s or clogging within the device. Median transit time of the population (inset) increased from 0.4 s to 3.1 s (middle bar), and the 25th (bottom of box) and 75th (top of box) percentiles shifted up after FMLP exposure. n = 204 and 184 for control and fMLP exposed cells, respectively.

This data confirms previous findings that fMLP significantly increases neutrophil stiffness and retention in capillary-like structures,^{5,24,27,54,55} and it shows the ability to detect physiologically relevant changes in blood cell deformability with our cytometer. It also adds new information about the population response of white blood cells to inflammatory mediators, specifically that the response was not a linear shift from the neutrophil control distribution but rather an increased skew of it. This suggests that obstruction of capillaries seen in the normal inflammatory response and in sepsis and ARDS may be due to a smaller number of outlying cells rather than due to an entire population of moderately stiffer cells.

Distribution of transit times correlates with hematologic disease pathology in leukostasis

While the effects of altered deformability of neutrophils in sepsis have been well studied,^5,27,56,57 cell deformability in some other disease pathologies have not been as thoroughly investigated. Leukostasis, a poorly understood and often fatal condition of acute leukemia in which leukemia cells aggregate in the vasculature, causes respiratory failure and brain hemorrhage,⁵⁸ and we have shown previously that increased leukemia cell stiffness correlates with symptoms of leukostasis.⁵⁹ Although leukostasis is known to occur more often in patients with acute myeloid leukemia (AML) than those with acute lymphoid leukemia (ALL),⁵⁸ no effective methods currently exist to diagnose or predict leukostasis. Furthermore, chemotherapeutic treatments of leukemia cells have been shown to increase their stiffness by more than an order of magnitude,¹⁶ making monitoring of blood cell stiffness during treatment clinically important. Biophysical flow cytometry can determine if there are differences in transit time of the leukemia cells from leukostasis-symptomatic patients and whether those differences are due to a shift in the entire population or due to an increase in the number of outliers. Better knowledge of the biophysical differences among leukemia cells in leukostasis-symptomatic patients would be useful in developing assays to identify patients at risk for leukostasis and could significantly improve treatment options and reduce mortality.

To answer these questions, we analyzed samples from an AML patient with leukostasis symptoms (AML2), an AML patient without leukostasis symptoms (AML1), and two ALL patients without leukostasis symptoms, as well as neutrophils and RBCs from healthy volunteers. The leukostasis-symptomatic AML patient sample cells took significantly longer to deform into the microchannels than both the leukostasis-asymptomatic AML patient and the ALL patients, as well as the normal neutrophils and RBCs (p < 0.001, Mann-Whitney). The histograms of the leukemia samples show that the majority of the cells of all leukemia samples deformed into the microchannels in less than one second, with the majority of the populations forming bell curve distributions (Fig. 4, Ai–iv). Median transit times of all blood cells were remarkably similar, ranging between 0.13–0.50 s (Fig. 4B, Table 1). The leukostasis-symptomatic sample, however, had an increased distribution of slow transit time outliers. This behavior becomes more apparent when looking at both the number of cells that transited in greater than four seconds (Fig. 4, Aiv) and in the 75th percentiles of transit time (Fig. 4B, Table 1), which is over nine times higher in the leukostasis-symptomatic sample than the leukostasis-asymptomatic samples.

Fig. 4.

The distribution of transit time of cells from a leukostasis-symptomatic patient is distinct from cells of leukostasis-asymptomatic patient cells. (A) Four patient leukemia samples were flowed through the device to determine transit time differences—two leukostasis-asymptomatic ALL samples (ALL1 and ALL2, Ai–ii, n = 239, 128 respectively), one leukostasis-asymptomatic AML sample (AML1, Aiii, n = 418), and one leukostasis-symptomatic AML sample (AML2, Aiv, n = 239). Distributions of the lower 50th percentile of the data are largely similar, but the upper 50th percentile substantially deviate. The leukostasis-symptomatic AML2 sample (Aiv) has a secondary subpopulation of cells with transit times ranging from 1–3 s. In addition, the fraction of cells that transit in greater than 4 s is substantially larger in the leukostasis-symptomatic AML2 (29%) than in the ALL and leukostasis-asymptomatic AML1 samples (9%, 8%, and 17%, respectively). (B) When looking at boxplots of the 25th, 50th, and 75th percentiles of the distributions, the differences in the population appear markedly different. While median transit times are substantially similar, the 75th percentile of the leukostasis-symptomatic AML2 sample is significantly higher than that of the leukostasis-asymptomatic samples and the RBCs and neutrophils.

Table 1.

Summary of transit time and microchannel occlusion data in Figs. 4 and 5. Leukostasis symptomatic AML2 cells had significantly higher transit times than leukostasis-asymptomatic AML1 and ALL cells as well as RBCs and neutrophils (p<0.001, Mann–Whitney)

		Transit time (s)
Cell type	n	25%	50%	75%	Pass through %
RBCs	196	0.13	0.13	0.23	100.0
Neutrophils	254	0.23	0.33	0.53	98.4
ALL1	239	0.20	0.30	0.57	95.4
ALL2	128	0.13	0.23	0.40	94.5
AML1	418	0.10	0.20	0.83	91.5
AML2	239	0.23	0.50	7.75	81.2

The fraction of cells from the leukostasis-symptomatic AML patient that were able to pass through the microchannels was also significantly lower than both the leukostasis-asymptomatic AML patient and the ALL patients, as well as the normal neutrophils and RBCs (Fig. 5A) (Chi-square, p<0.001 for all comparisons with the leukostasis-symptomatic AML patient sample). A higher proportion of leukemia cells as compared to normal neutrophils and RBCs were not able to compress into the microchannels, with the leukostasis-asymptomatic AML sample having the second lowest fraction of cells passing through. This is consistent with clinical data showing myeloid leukemias having a higher propensity to cause leukostasis than lymphoid leukemias.⁶⁰

Fig. 5.

Leukostasis-symptomatic patient cells have distinctively aberrant behavior when compared to leukostasis-asymptomatic patient cells and normal blood samples. (A) The fraction of cells that did not obstruct the channels is significantly lower in the leukostasis symptomatic AML2 sample versus the other samples. (B) The change in number of open channels is plotted versus the cells that entered the channels (a normalization of time for the density of cells entering the channels). Change in open channels was normalized to account for channels that were already occluded at the beginning of the experiment. A substantially reduced number of channels stayed open over time in the leukostasis-symptomatic AML2 sample (red) than the leukostasis-asymptomatic AML1 sample (light blue) or leukostasis-asymptomatic ALL samples (dark blue, grey).

Each microchannel that becomes occluded prevents upstream cells from flowing through. We tracked change in number of unblocked channels versus the order of cells that entered into the device (Fig. 5B). While all leukemia cells created at least a small reduction in the number of unblocked microchannels during the experiments, the leukostasis-symptomatic AML patient sample occluded the microchannels at a significantly higher rate (moderated linear regression, p<0.001 for all comparisons with the leukostasis-symptomatic AML patient sample), with only 10% of the original channels still open (3 of 31) at the end of the experiment (Fig. 5B). In acute leukemia, where leukemia cell concentrations can reach over 10 times normal white blood cell concentrations (5000–10 000 cells µL⁻¹), this rate of occlusion could have profound effects on microvascular flow in the brain, lungs, and other vital organs. Taken together, this data shows that biophysical flow cytometry enables the detection of leukostasis in acute leukemia, even among patients with the same leukemia type, and offers a possible platform to predict and diagnose this potentially fatal complication.

Transit times of cells passing through small channels might be expected to be very sensitive to cell size, and it is not initially obvious whether variations in cell deformability or cell size are playing the most significant role in the measured transit time distributions. Similar to the multiple parameter measurements of standard flow cytometry, biophysical flow cytometry can simultaneously measure image-based parameters such as cell size with cell transit time. To investigate the impact of cell diameter on transit time, cell diameters of both the leukostasis-symptomatic and leukostasis-asymptomatic AML patient samples were measured before the cells deformed into the capillaries. Cells from the leukostasis-symptomatic AML patient sample were found to be larger than the leukostasis-asymptomatic AML patient sample (mean diameter = 9.6 µm vs. 9.1 µm, respectively, p < 0.001), which would be expected to increase their transit time through microchannels. Interestingly, even when controlling for cell diameter using analysis of covariance, significant differences in transit time were still observed (p < 0.001). For both types, there was a significant but weak correlation between size and transit time (R²_AML1 = 0.17, R²_AML2 = 0.26). Taken together, these results indicate that although larger cells tend to transit in longer times, other factors such as deformability play more significant roles.

Drug treatment shifts the transit time distribution of leukemia cell populations

To test the use of biophysical flow cytometry as a platform to determine the effect of drugs on aberrant blood cell transit, HL60 cells exposed to drugs were flowed through the device. HL60 cells are an acute myeloid leukemia cell line that show a transit time distribution similar to the leukostasis-symptomatic AML sample, with a large number of long transit time outliers (Fig. 6A, B) and a significant fraction of cells unable to deform into the channels (Fig. 6C).

Fig. 6.

Drug treatment improved flow through microchannels. (A) HL60 cells, a model AML leukemia line with a histogram profile similar to the leukostasis-symptomatic AML2 sample, were flowed through the device without exposure to any drug (red, n = 146), after exposure to pentoxifylline (grey, n = 157), and after exposure to cytochalasin D (blue, n = 117). Substantial shifts in the histograms can be seen after exposure to drugs, with a large fraction shifting from > 4 s towards the median transit time. (B) The 25th, 50th, and 75th percentiles of the population all shifted downward after exposure to drugs, with cytochalasin D exposed cells experiencing more reduced transit time than pentoxifylline exposed cells. (C) The fraction of cells passing through the microchannels also significantly improved after exposure to cytochalasin D and pentoxifylline. (D) This resulted in a higher number of microchannels remaining open over time when drug-treated cells versus non-treated cells were flowed into the device.

Two drugs were used. The first, cytochalasin D, is a common cell biology tool for disrupting the actin cytoskeleton.⁶¹ Though it cannot be used as a clinical treatment due to its systemic toxicity, cytochalasin D is known to significantly decrease cell stiffness and provides a positive control. The second drug, pentoxifylline, is a phosphodiesterase inhibitor that has been shown to reduce stiffness of neutrophils exposed to TNF-α or fMLP while also reducing lung injury associated with hemorrhagic shock.^56,62,63 We hypothesized that this drug could also attenuate stiffness of AML cells to decrease microvascular obstruction in patients with leukostasis.

We found that both drugs significantly reduced transit time of HL60s through the microchannels (p<0.001 for both drugs), and both drugs also created significant shifts in the distribution of transit times of HL60 cells. The non-treated cell population transit time was well spread with a low peak and a large number of slow-transiting cells (>4 s) (Fig. 6A). Both cytochalasin D and pentoxifylline shifted the distributions towards shorter transit times and reduced the number of slow-transiting cells (Fig. 6A). Median transit time was slightly reduced from 0.63 s to 0.23 s for cytochalasin D and 0.37 s for pentoxifylline, but most remarkably, the 75th percentile of transit time reduced from > 100 s (over 25% of the cells could not deform into the microchannels) to 0.70 s and 3.47 s for cytochalasin D and pentoxifylline, respectively. The fraction of cells that were able to deform into the microchannels significantly increased from 73% to 97% and 89% for cytochalasin D and pentoxifylline, respectively, (Chi-square p < 0.001 for both drugs). Improved flow through the device was seen with both drugs due to reduced numbers of blocked channels (Fig. 6D), indicating that reducing the outlying occluding cells significantly increases flow (p < 0.001). These results suggest that biophysical flow cytometry can be useful in identifying drugs that may be able to modulate the pathology of leukostasis.

Conclusions and future outlook

Many hematologic diseases are associated with reduced deformability of blood cells. We have shown that a simple microfluidic device and analysis system is able to quantify differences in blood cell deformability and that these differences are consistent with clinical outcomes. Specifically, we demonstrate here that we can detect distinct differences between normal and aberrant blood cells that cause hematologic complications, such as microvascular occlusion. This occlusion is caused by an increased number of outliers rather than an overall shift in the cell population. In addition, we show that we can alter the distribution of transit time towards longer or shorter transit times by exposure to inflammatory mediators or drugs that affect the cytoskeleton, respectively.

Biophysical flow cytometry offers the potential to be a low-cost and straightforward tool to explore the mechanical properties of cell populations in a single-cell manner, highlighting that biophysical phenotype of individual cells may be more predictive of pathology than average population measurements. With our cytometer, we have shown here that blood cell subpopulations under microcirculatory flow conditions can have a disproportionate effect on disease pathology, necessitating the clinical use of these types of high-throughput single-cell devices. In addition to the study of sepsis and leukostasis, this system is relevant for the investigation of other hematologic diseases involving microvascular occlusion including diabetic retinopathy, reperfusion injury in stroke and myocardial infarctions, and sickle cell anemia, among others.^{4,15,27,64–69}

Currently, there are no effective methods to predict or diagnose leukostasis in acute leukemia, and clinical deterioration is often rapid and irreversible, with a large proportion of cases being fatal. A clear clinical need exists for improved techniques to identify patients at high-risk for leukostasis as well as to measure therapeutic efficacy. As measurements obtained with our cytometer were able to clearly show differences between cells taken from a leukostasis-symptomatic patient versus leukostasis-asymptomatic patients, biophysical flow cytometry may serve as a diagnostic and drug discovery tool to meet these needs.