Microfluidic Lysis of Human Blood for Leukocyte Analysis Using Single

Dec 9, 2011 - E-mail: [email protected]. Cite this:Anal. Chem .... Enabling the Development and Deployment of Next Generation Point-of-Care Diagnosti...
0 downloads 0 Views 384KB Size
Download PDF
Article pubs.acs.org/ac

Microfluidic Lysis of Human Blood for Leukocyte Analysis Using Single Cell Impedance Cytometry Xiaojun Han,† Cees van Berkel,‡ James Gwyer,‡ Lorenzo Capretto,† and Hywel Morgan*,† †

School of Electronics and Computer Sciences, University of Southampton, Southampton, SO17 1BJ United Kingdom Philips Research Laboratories, 101 Cambridge Science Park, Cambridge, United Kingdom

‡

S Supporting Information *

ABSTRACT: This paper demonstrates an integrated microfluidic system that performs a full blood count using impedance analysis. A microfluidic network design for red blood cell (RBC) lysis is presented, and the diffusive mixing processes are analyzed using experimental and simulated results. Healthy and clinical bloods analyzed with this system, and the data shows good correlation against data obtained from commercial hematology machines. The data from the microfluidic system was compared against hospital data for 18 clinical samples, giving R2 (coefficient of determination) values of 0.99 for lymphocytes, 0.89 for monocytes, and 0.99 for granulocytes in terms of relative counts and 0.94 for lymphocytes, 0.91 for monocytes, and 0.95 for granulocytes in terms of absolute counts. This demonstrates the potential clinical utility of this new system for a point-of-care purpose.

T

provide an optimal preparation of target cells. In particular, our previous publication3 demonstrated the use of a buffer that both lyses erythrocytes and also increases the discrimination of monocyte from neutrophils. This paper demonstrates such a multistep cell lysis method in a microfluidic system. It is used to process clinical blood samples prior to cell identification and enumeration using MIC. The properties of the system are analyzed, and the results from clinical blood samples are compared with data from central laboratory analysers.

he blood cell count is of prime interest for both medical and scientific applications and holds a central role in the diagnosis of many physiologic and pathologic conditions. It is performed with complex and expensive hematology analyzers located in central laboratories, based on flow cytometry techniques that were first developed in the 1960s.1 There is a need to develop low cost portable blood cell analyzers for point-of-care use in a similar fashion to a glucose test. Point-ofcare testing eliminates transport and waiting times associated with central lab processing. This could aid with diagnosis and may allow better matching between drugs and patient pathophysiology, reducing side effects and improving efficiency of therapy. Coulter developed the first high speed cell counting technology in the 1950s,2 based on measuring the electrical resistance across an aperture through which cells flow. Impedance spectroscopy is a development of this technology and provides a noninvasive method to analyze cell properties,3−6 measuring volume, membrane capacitance, and intracellular conductivity. Microfluidic impedance cytometry (MIC) offers high sensitivity and high throughput analysis using small sample volumes.4−6 MIC has been used to perform a differential white blood count using human blood.3 In human blood, the erythrocytes outnumber the white cells by 1000:1; therefore, in order to accurately count leukocytes, these red cells need to be removed, usually by chemical lysis. For a simple point-of-care instrument, an automated lysis and sample preprocessing approach is demanded. Ekberg et al.7 demonstrated a one step bulk sample preparation protocol on a microchip, using a small magnetic stirrer. A continuous flow single step lysis protocol was demonstrated by Sethu el al.8 More complex, multistep lysis recipes are required to © 2011 American Chemical Society

■

EXPERIMENTAL SECTION System Overview. Figure 1 shows a diagram of the integrated microfluidic analysis system used to perform whole blood lysis and enumeration. It contains three parts: the microfluidic lysis block, the impedance analysis chip, and the electronic detection and data collection processing section. Microfluidic Block. The microfluidic block is made from PMMA by micromilling and has dimensions of 50 mm × 120 mm. The microfluidic network and the cell analysis chip is shown in Figure 1. Blood and lysis solutions are mixed for a fixed time (defined by the sample flow rate) before the chemistry is quenched. The distance from junction 1 to junction 2 is 89 mm, and the distance between junction 2 and the inlet to the impedance chip is 500 mm. The depth of all channels is 100 μm. All channels are 200 μm wide, except the region from the quench inlet to junction 2 which is 100 μm wide.

Received: October 11, 2011 Accepted: December 9, 2011 Published: December 9, 2011 1070

dx.doi.org/10.1021/ac202700x | Anal. Chem. 2012, 84, 1070−1075

Analytical Chemistry

Article

Figure 1. Schematic diagram of the integrated microfluidic leukocyte analysis system.

Signal Analysis. Data was processed and plotted as scatter plots. Digital signal processing was applied to the data, using analysis described in detail elsewhere.11 In a scatter plot, the experimental results of all cells are summarized by a distribution function N(Φ,O). This total distribution function is the sum of three separate distributions for lymphocytes, monocytes, and neutrophils. We assume that these are distributed normally around their means. For example, for the lymphocytes

All solutions were driven by syringe pumps (Harvard 11 plus, UK). Blood first meets the lysis solution at junction 1, as shown by the image (bottom right in Figure 1). The dark stream is the blood; lysis solution enters from the sides. The two samples flow and mix by diffusion until the quench solution is introduced at junction 2. The quench mixes with the lysate before the red blood cell (RBC) free suspension of white cells enters the impedance chip for analysis. An image showing the quench solution mixing with the lysate at junction 2 is also shown in the figure. The standard input flow rates of blood, lysis solution, and quench solution are 1.39 μL/min, 16.37 μL/min, and 7.24 μL/min, respectively, giving a 1:12:5.3 ratio and a total flow rate through the impedance chip of 25 μL/min. At these reference flow rates, blood is mixed by diffusion with the lysis solution for 6 s (between junctions 1 and 2), and the average residence quench time (between junction 2 and impedance chip inlet) is 24 s. Impedance Chip. The impedance chip is fabricated from glass and contains a small channel (40 μm × 40 μm) through which the cells flow. Two pairs of microelectrodes are used to measure cell impedance in a differential mode as single cells flow through the channel. The glass chips were fabricated using photolithography and full wafer thermal bonding; full details can be found elsewhere.9 Fluidic and electrical connections were made by clamping the chip within the microfluidic block. Spring loaded electrical connectors were used to make contact with the chip electrodes, and small rubber O-rings were used to seal the fluidic connections. Electronics. The impedance measurement system has been described previously.3,9,10 Two sinusoidal voltages at fixed frequencies were applied to both top electrodes, and the differential current between the electrodes were measured using custom built electronics and lock-in amplifiers. The output from the lock-in consists of the real component (in phase) and imaginary component (90° out of phase) of the electrical impedance at each of the two frequencies. As the cells pass through the chip, the low frequency (444 kHz) signal measures the electrical cell volume, while the high frequency (1776 kHz) signal measures the cell membrane capacitance. Data is plotted as opacity (O), the ratio of high frequency to low frequency impedance, against electrical volume (Φ), the magnitude of the low frequency impedance.

PL(Φ , O) =

1 2πσ1σ2 ⎡ ((Φ − Φ ) cos θ + (O − O ) sin θ)2 L L × exp⎢ − ⎢⎣ 2σ12 −

(O − OL) cos θ + (Φ − ΦL) sin θ)2 ⎤ ⎥ ⎥⎦ 2σ2 2

(1)

where ΦL and OL are mean lymphocyte volume and opacity and σ1, σ2, and θ are parameters that describe the spread and orientation of the ellipsoidal multivariate distribution. Similar expressions are written for PN (Φ, O) and PM (Φ, O), the neutrophil and monocyte populations, respectively. We can obtain a goodness of fit, provided by a correlation coefficient, of the experimental results to this compound theoretical distribution. For experimental results that show good separation between the different cell types, typical correlation coefficients are up to 98%. Experiments with poorer cell differentiation are characterized by lower correlation coefficients. In the Results and Discussions section, we use this to determine the optimal microfluidic lysis conditions. Although microfluidic lysis produces a substantially RBC free sample, it is not 100% efficient and leaves some debris. These fragments are small and have low opacity (low membrane capacitance). Data processing is used to exclude particles of this type, e.g., below a minimum size or above a maximum opacity. The process of 2D Gaussian fitting, gating, and counting is fully automated. Blood Samples. Whole blood was drawn into Vacutainer tubes (EDTA K3E15%). The first set of experiments was performed with blood from healthy donors. For each of these, a second sample was sent to a hospital central laboratory (Addenbrookes) for analysis on a standard hematology analyzer. A second set of experiments was performed with blood collected from patients at University College London Hospital providing a full range of 1071

dx.doi.org/10.1021/ac202700x | Anal. Chem. 2012, 84, 1070−1075

Analytical Chemistry

Article

pathological samples. The blood was analyzed at the hospital lab and sent to our laboratory for microfluidic analysis. Residual samples were selected from the hematology laboratory after all testing was complete; medical and personal information was removed prior to acceptance of the samples. Blood tubes were placed on a roller and continually mixed at room temperature before use. All samples were analyzed within 12 h. Microfluidic Lysis. The microfluidic lysis is based on the bulk lysis protocol described elsewhere.3 In this process, whole blood is mixed at the bench with lysis solution (0.12% v/v formic acid, 0.05% w/v saponin) in a 1:12 ratio for exactly 6 s. Subsequently, a quench solution (0.6% w/v sodium carbonate, 3% w/v sodium chloride) is added in 5.3:13 ratio. The microfluidic network reproduces this protocol in continuous flow. Reagents and blood were loaded into three Hamilton syringes (250 μL for lysis solution, 250 μL for quench solution, and 10 μL for blood, respectively). The syringes were connected to inlets using PTFE tubes. After each experiment, the system was flushed with DI water, followed by 4 M NaOH for 30 min. The system was then rinsed with 2 mL of DI water and then 1 mL of Hanks’ balanced salt solution prior to reuse. The time taken for cleaning is necessary because in this set of experiments cartridges were reused multiple times. In clinical practice, cartridges would be single use only.

■

HEMOGLOBIN DIFFUSION In order to study lysis in the continuous flow microfluidic network, we compared experimental observations of the hemoglobin (Hb) diffusion profile with numerical simulations under different theoretical assumptions. The bloodstream sheathed by the two adjacent lysis streams was numerically simulated with a two-dimensional model using Ansys Fluent 12.1.4 (ANSYS Inc., Canonsburg, PA). For further details, see Supporting Information. A concentration-dependent viscosity (i.e., mass-weighted-mixing-law) was assumed for the fluid, considering an initial dynamic viscosity of 0.89 and 5.3 mPa·s for lysis solution and blood.12,13 For the simulation, it is assumed that the lysis solution (predominantly, formic acid) diffuses into the bloodstream and immediately destroys the RBC membranes. Two types of conditions were imposed at the channel walls. The “no-slip” condition imposed zero velocity at the wall, giving the familiar parabolic flow profile for pressure driven flow. In contrast, the “free-slip” condition imposes no resistance at the channel walls and results in a uniform flow profile. The comparison between the two is of interest for comparing sample residence and reaction times. The literature value for the diffusion constant of Hb is 6 × 10−11 m2·s−1. Hb is a tetramer at neutral pH but undergoes a reversible dissociation to dimers at low pH.14 The Hb diffusion coefficient was experimentally determined both in PBS and in the lysis solution. This was performed by flowing a solution of Hb through the lysis device (Figure 1) and using either the standard lysis solution (pH = 2.95) or PBS (pH = 7.4) as the sheath flow. The Hb stream width was measured as follows: Optical microscope images of the profiles were taken at different distances from junction 1. The images were processed and fitted to a Gaussian profile using Origin. The width of the Hb distribution was determined from the standard deviation of this profile. This data was used to calculate the Hb diffusion constant. The width of the Hb stream during lysis of whole blood was measured in the same way. Figure 2a shows the flow profile of blood and lysis solution at the start of the reaction when both samples enter the network

Figure 2. (a) Simulated and experimental results of flow profile of blood and lysis solution at junction 1; (b) experimental and simulated data of the Hb sample stream width.

at junction 1. Because the viscosity of the blood (5.3 mPa·s) is larger than that of the lysis solution (0.89 mPa·s), the result is a double peaked flow profile. Initially, the Hb is confined in a 13 μm wide sample stream at the junction and then diffuses into a distribution with a width that increases along the channel length. Figure 2b shows both experimental and simulated data of the Hb sample stream width. The Hb in PBS data (blue line star marker) demonstrates slow diffusion, which is consistent with the normal value of the diffusion constant for Hb. For comparison, the simulated data (no slip) indicated by a dashed blue line matches the experimental for D = 8 × 10−11 m2/s which is close to the literature value for Hb. In the lysis solution, the Hb denatures and diffuses much more quickly, as shown by the orange line, circle marker, data set. The blood lysis sample stream width is also shown by the green line (diamond marker) and is close to the width measured for Hb solution. Again, simulation (no slip) indicated by a dashed green line shows that the diffusion constant is now 27 × 10−11 m2/s. Hb is a roughly spherical tetrameric molecule of 5.5 nm diameter. It is known to dissociate reversible into dimers at low pH (>pH 4), but under extended exposure to lower pH (3.6), the molecule denatures irreversibly.13 The observed diffusion constant for the Hb in lysis is equivalent to a molecule of diameter 1.6 nm, implying that the Hb has probably denatured into monomers. For interest, also shown are simulations for free-slip (dashed purple) and a simple analytical calculation 1072

dx.doi.org/10.1021/ac202700x | Anal. Chem. 2012, 84, 1070−1075

Analytical Chemistry

Article

Figure 3. Scatter graph of differentiated leukocytes from a healthy donor using (a) bulk lysis and (b) microfluidic lysis: 1, debris; 2, lymphocytes; 3, granulocytes; 4, monocytes.

(dashed red) based on 1-D diffusion, where x = (2Dt)1/2. Though there is a noticeable difference between these calculations, the overall result shows that diffusion of denatured Hb is the dominant factor in the observed experimental results rather than the effects of residence time distribution. This data confirms the assumption made above that the lysis solution quickly diffuses into the bloodstream and that RBCs lyse almost instantly. Moreover, this analysis shows that simple analytical methods can be used to design these microfluidic sample processing networks.

Table 1. Relative Cell Count Data for the Scatter Graph Shown in Figure 3 hospital data bulk lysis MF lysis

LYM

MON

GRN

28% 30% 26%

8% 9% 9%

64% 61% 65%

flow rates were varied to examine the efficiency of lysis against the bulk process. Keeping all ratios constant, the total flow rate was changed to 0.8, 0.9, 1.1, 1.2, and 1.3 times the reference flow. It was observed over several experiments that the scatter diagrams (cf Figure 3) showed much poorer separation of the different cell populations at the lower (0.8) or higher (1.2) flow rates than at the standard flow rates. Figure 4 quantifies this

■

RESULTS AND DISCUSSIONS Microfluidic Lysis on Blood from Healthy Donors. Figure 3 shows a typical scatter graph for leukocytes from a blood sample from a healthy donor. Figure 3a shows data obtained using conventional bulk lysis, while Figure 3b shows the same blood processed using microfluidic lysis. In each case, the data is plotted as opacity against electrical volume.5 The data was analyzed as described in Signal Analysis. The fitted populations of the 2D Gaussian are illustrated by the three ovals. The correlation coefficients of the fits to the Gaussians reflect how well the three populations are visible in the scatter plot. The correlation coefficient for this data is over 95%, indicating that the Gaussian distributions provide a good description of the measured data. The green lines mark positions of equal probability deviation between the different populations. They provide the gating boundaries from which an accurate count for lymphocytes, granulocytes, and monocytes is obtained. The microfluidic lysis scatter plot shows similar data to that obtained from the bulk lysis. The debris, as shown in region 1 in each graph, has a small electrical volume and high opacity (small membrane capacitance) and is most likely RBC ghosts and other liposomes. Higher debris incidence is typically observed in microfluidic lysis compared to bulk lysis. For this particular data set, the relative cell counts for the two lysis methods are summarized in Table 1. Also shown is the data obtained from the Hospital laboratory demonstrating that the data from both bulk and microfluidic lysis agrees well with the standard technique. Optimization of Flow Rates. The timing for the bulk lysis method has been optimized for bench preparation. There is no active mixing in the microfluidic channel, and consequently, the timing of the lysis chemistry needs to be verified. Therefore, the

Figure 4. Goodness of fit to a Gaussian function for the three cell populations. Data is plotted as a function of normalized flow rates. Five samples were measured for each flow rate; error bar is one standard deviation.

observation by showing the goodness of fit to the sum of three Gaussian cell population distributions for different flow rate conditions. Each point was measured for 5 different blood samples. The correlation peaks at 95% for a normalized flow rate of 1.0 and 1.1. For comparison, bulk prepared lysate shows slightly higher results at 96%. For the concordance studies below, the two optimum flow rates (total flow rates of 25 μL/min and 27.5 μL/min) were used. 1073

dx.doi.org/10.1021/ac202700x | Anal. Chem. 2012, 84, 1070−1075

Analytical Chemistry

Article

Figure 5. Comparison between the impedance data and hospital data in terms of relative counts at 100% flow rate (a) and at 110% flow rate (b) on blood samples from healthy donors. Linear regression was performed to each group individually with an intercept at zero.

Figure 6. (a) Typical scattering graph of clinical samples at 100% flow rate. The comparison between the impedance data and hospital data over pathological range in terms of relative counts (b), absolute counts of each population (c), and absolute counts of total leukocytes (d). Linear regression was performed to each group individually with an intercept at zero.

Concordance Study. Healthy Volunteers. Figure 5 compares the relative counts for the three leukocyte subpopulations obtained from the microfluidic impedance system against data acquired using the central lab blood analysis system. Blood samples from 5 different healthy donors were processed on both machines. The correlation coefficient (R2) measures how well the two data sets (percentage count) correlate with each other. There is a good correlation between the impedance cytometer data and central lab data at the reference flow rate, with R2 values of 0.99 for lymphocytes, 0.97 for monocytes, and 0.99 for granulocytes, respectively. Increasing the flow rate by 10% reduces R2 for the monocytes (Figure 5b). Clinical Blood Samples. While the concordance from healthy donors is excellent, any new Point-of-Care technology must be able to process clinically relevant samples. To this end, 18 separate clinical samples were measured using both the

microfluidic cytometer and a hospital hematology analyzer. The clinical samples reflect a diverse range of blood cell parameters, with the absolute total white cell count (WBC) ranging from 0.77 to 50.45 × 109/L and relative counts ranging from 1.4% to 92.9% for lymphocytes, from 0% to 28.5% for monocytes, and from 4.7% to 98.6% for granulocytes, respectively. Figure 6 shows data and a concordance plot for these samples using the standard (25 μL/min) flow rate. Figure 6a shows a representative scatter plot for leukocytes, and the concordance is shown in Figure 6b. The R2 value of all results is summarized in Table 2. For relative counts, the R2 values are 0.99 for lymphocytes, 0.89 for monocytes, and 0.99 for granulocytes; the concordance with hospital analysis is good. The absolute counts for each subpopulation are shown in Figure 6c, and the total WBC count is shown in Figure 6d. The table also shows similar data for the 110% flow rate. 1074

dx.doi.org/10.1021/ac202700x | Anal. Chem. 2012, 84, 1070−1075

Analytical Chemistry

Article

(4) Ayliffe, H. E.; Frazier, A. B.; Rabbitt, R. D. J. Microelectromech. Syst. 1999, 8, 50. (5) Cheung, K.; Gawad, S.; Renaud, P. Cytometry, Part A 2005, 65A, 124. (6) Gawad, S.; Schild, L.; Renaud, P. Lab Chip 2001, 1, 76. (7) Ekberg, B. A.; Larsen, U. D.; Fogh-Anderson, B. Point of Care 2005, 4, 64. (8) Sethu, P.; Anahtar, M.; Moldawer, L. L.; Tompkins, R. G.; Toner, M. Anal. Chem. 2004, 76, 6247. (9) Holmes, D.; She, J. K.; Roach, P. L.; Morgan, H. Lab Chip 2007, 7, 1048. (10) Morgan, H.; Sun, T.; Holmes, D.; Gawad, S.; Green, N. G. J. Phys. D: Appl. Phys. 2007, 40, 61. (11) Sun, T.; van Berkel, C.; Green, N. G.; Morgan, H. Microfluid. Nanofluid. 2009, 6, 179. (12) Sullivan, S. P.; Akpa, B. S.; Matthews, S. M.; Fisher, A. C.; Gladden, L. F.; Johns, M. L. Sens. Actuators, B: Chem. 2007, 123, 1142. (13) Sahu, K. C.; Ding, H.; Valluri, P.; Matar, O. K. Phys. Fluids 2009, 21. (14) Field, E. O.; O’Brien, J. R. Biochem. J. 1955, 60, 656.

Table 2. Concordance between Hospital Cell Counts and the Impedance Cytometer for Patient Samples

Lym Mon Grn total WBC

100% flow rate relative cell counts

110% flow rate (relative)

100% flow rate absolute cell counts

110% flow rate (absolute)

0.99 0.89 0.99

0.98 0.79 0.99

0.94 0.91 0.95 0.95

0.82 0.92 0.97 0.96

Although the relative WBC counts demonstrated excellent concordance with hospital laboratories, the absolute WBC count was lower than expected. This could be due to sedimentation of cells in the syringes and tubing during the experiment. In order to test this, the system was run with all pumps and tubing mounted vertically, and in this case, the total WBC increased to between 96 and 97% of the value from the hospital lab. This suggests that sedimentation in the external syringe pumps is responsible for the systematic undercounting, rather than the microfluidic lysis and impedance cytometry system. We anticipate that full integration of pumps and sample loading into a single disposable cartridge will eliminate this issue in the future.

■

CONCLUSIONS Microfluidic lysis combined with impedance cytometry can be used to analyze a wide range of clinical blood samples. Some of these samples had extremely low subpopulation counts (indicative of neutropaenia and lymphocytopaenia), and the automated fitting algorithms performed well in all these situations. For this system, the microfluidic lysis of blood is well described by a simple model of fast diffusion of the acid into the bloodstream and the diffusion of denatured Hb out into the lysis stream. Cell sedimentation in the external syringe is the cause of low cell recovery but does not influence the relative counts. An integrated point-of-care system will include integrated sample metering and pumping, eliminating the problems associated with external microfluidic drive.

■

ASSOCIATED CONTENT

S Supporting Information *

Additional information as noted in text. This material is available free of charge via the Internet at http://pubs.acs.org.

■

AUTHOR INFORMATION

Corresponding Author

*Tel: 44 (0)23 8059 3330. Fax: 44 (0)2380 593029. E-mail: [email protected].

■

ACKNOWLEDGMENTS The authors acknowledge the Technology Strategy Board and EPSRC for funding this work. We also thank the team at Philips for their support and Carol Briggs of University College London Hospital (60 Whitfield Street, London W1T4 EU) for the access to clinical blood samples.

■

REFERENCES

(1) Toner, M.; Irimia, D. Annu. Rev. Biomed. Eng. 2005, 7, 77. (2) Coulter, W. H. Proc. Natl. Electron. Conf. 12 1956, 1034−1040. (3) Holmes, D.; Pettigrew, D.; Reccius, C. H.; Gwyer, J. D.; van Berkel, C.; Holloway, J.; Davies, D. E.; Morgan, H. Lab Chip 2009, 9, 2881. 1075

dx.doi.org/10.1021/ac202700x | Anal. Chem. 2012, 84, 1070−1075