In initial experiments using the original ETC system [21
], microspheres were coated with monoclonal antibodies to the lymphocyte surface markers CD3, CD4, or CD8, followed by microfluidic delivery of fluorescently labeled lymphocytes from whole blood obtained from non-HIV-infected participants. Although lymphocytes were readily captured, precise quantification of cell numbers and CD4 cell counts were not possible using the microsphere as a surface for lymphocyte capture (data not shown). We next modified the flow cells with a disposable, microporous membrane filter for lymphocyte capture. A single polycarbonate, track-etch membrane with 3-μm pores was immobilized and secured within the flow cell, creating a lymphocyte capture surface with a surface area of 80 mm2
. Whole blood samples were delivered to the flow cell from a sample reservoir tube, and the membrane within the flow cell was washed with PBS from a second reservoir. As in the original ETC system, cells were imaged under fluorescence optics using a mercury arc lamp light source and a CCD camera ().
To confirm that cells could be adequately captured, 33 μl of unprocessed whole blood from non-HIV-infected participants was incubated for 8 min with fluorophore-conjugated anti-CD4 antibodies, and delivered by a peristaltic pump to the modified microfluidics chip. Red blood cells passed readily through the pores under appropriate fluid flow conditions. In contrast, the majority of white blood cells were captured onto a single imaging focal plane (). This mechanical separation of autofluorescent red blood cells allows for the imaging and counting of white blood cells from unprocessed whole blood without additional sample processing, such as centrifugation or red blood cell lysis. Using the digital imaging system originally developed for microsphere-based capture in the ETC system, fluorescently labeled white blood cells can then be imaged directly on the chip and counted.
The Membrane Flow Cell Selectively Captures Lymphocytes and Provides for the Removal of Red Blood Cells without Sample Processing
To assess the analytical validity of the membrane-based microchip system, we first performed a dilution control study to evaluate the correlation between total fluorescence intensity and the absolute number of purified CD4 cells from non-HIV-infected participants (labeled with fluorophore-conjugated anti-CD4 antibody) captured in the microchamber. The results show a linear correlation between the number of cells in the sample and the intensity of light emitted from the membrane filter (R2 = 0.999) for a range of CD4 cell counts relevant to advanced HIV disease (0–200 CD4 cells/μl blood) (). This dose–response study established proof of the concept that a modified microfluidic flow cell and a digital image analysis system can accurately detect and measure populations of whole blood lymphocytes labeled with fluorescent markers.
We next quantified the percentages of CD3, CD4, and CD8 cells in whole blood samples from healthy control participants using this system. Prior to delivery to the flow cell, we labeled a 33-μl whole blood sample with 3 μl of fluorophore-conjugated anti-CD3 and anti-CD4 antibodies for 8 min off chip, then diluted the sample with 961 μl of PBS, and delivered 500 μl of the resulting sample (containing 16.5 μl of blood) to the flow cell using a fluidics controller. Digital images from one region of the lymphocyte capture membrane were obtained with two different emission filters, one specific for the AlexaFluor-488-conjugated antibody used to stain CD4+ T lymphocytes green (A), and the other specific for the AlexaFluor-647-conjugated antibody used to stain CD3+ T lymphocytes red (B). Automated digital merging of the two images and image processing allowed the system to distinguish the CD3+CD4+ T lymphocytes of interest (i.e., “CD4 cells”), which appear yellow, from the CD4+CD3− monocytes (green) and the CD3+CD4− T lymphocytes (red) (C).
Data Collection and Processing for Digital Images Obtained from a Single Diluted Whole Blood Specimen from an HIV-Infected Participant
We next developed a custom algorithm for translating these digital images into accurate CD4 and CD8 T cell counts using pixel analysis with the aid of a commercial image processing package. Automated counting of the three subsets of cells was based on object size, aspect ratio, and uniformity, iterated across the range of color intensity levels. As shown in D, a binary mask first removes the unwanted cell types, and residual objects representing CD4 T cells are counted. A similar protocol was applied to a second aliquot of blood stained with AlexaFluor-647-conjugated CD3-specific antibody and AlexaFluor-488-conjugated CD8-specific antibody to visualize and count CD3+CD8+ T lymphocytes.
In order to calculate an absolute CD4 count with standard flow cytometry, one of two measures must be undertaken to calculate a concentration in cells per microliter. Either a standardized reference reagent, such as calibration beads at a known concentration, can be added to the assay (“single-platform” flow cytometry), or an absolute total lymphocyte count in cells per microliter can be obtained on a hematology analyzer (“dual-platform” flow cytometry). The microchip assay we describe here uses a direct volumetric method and functions as a single-platform approach. By delivering a consistent volume of blood to the flow chamber (16.5 μl of stained whole blood, diluted to a total volume of 500 μl of PBS), and calculating the unit volume of blood per digital image (0.18 μl), we were able to count the total number of CD4+CD3+ cells in 0.9 μl of blood, and determine the absolute CD4 count per microliter.
We next tested this rapid, whole blood microchip assay in a series of samples acquired in an HIV reference laboratory in Botswana. Seventy consecutive HIV-infected participants presenting to the HIV reference laboratory for standard CD4 counting as part of a vertical transmission study were enrolled, of whom 64 were adult women and six were infants. Parallel samples were processed by standard four-color flow cytometry on a Becton Dickinson FACSCalibur. The time from blood collection to complete analysis and results reporting using the chip-based assay was approximately 15 min per sample. Three adult participants did not have valid flow cytometry results available, leaving 61 adults and six infants for analysis.
Representative processed data images from three participants, two adult women and one infant, are shown in . A shows a 31-y-old woman with an absolute CD4 count by flow cytometry of 83 cells/μl. While numerous CD3+ T cells (red) are present as well as scattered monocytes (green), her low CD4 count is reflected in the few double-labeled CD3+CD4+ T cells (yellow) seen in the image. Similar representative data images from a young woman with a CD4 count of 271 cells/μl by flow cytometry and a 5-mo-old infant with a CD4 percentage of T lymphocytes of 0.39 by flow cytometry are also shown in B and C, respectively. These images illustrate the dynamic range of the membrane capture and digital image analysis system, including the ability to quantify both absolute CD4 counts and CD4 percentages.
Representative Processed Data Images from Three Participants in Botswana
We compared results from our microchip assay with results available from flow cytometry, the latter obtained on a FACSCalibur through standard clinical laboratory operating procedures. The data for adult absolute CD4 counts are plotted in the Bland–Altman methods comparison plot shown in . For 61 adult participants with CD4 counts ranging from 35 to 1,087 cells/μl (mean, 372 cells/μl) by flow cytometry, results show a good correlation between absolute CD4 counts measured by our microchip assay and those measured by flow cytometry. Bland–Altman methods comparison analysis shows a bias of −50 cells/μl (95% confidence interval, −81 to −20 cells/μl), and good 95% limits of agreement (). Several of the results from participants at the higher end of absolute CD4 counts fall outside the 95% limits. For these participants, individual lymphocytes may overlap in the digital images (as seen in C), which can interfere with the accuracy of the lymphocyte counting algorithm. In resource-limited settings, the primary use of CD4 counts is as a trigger to initiate antiretroviral therapy, which typically occurs at a CD4 count of 200 cells/μl. Higher CD4 count thresholds of 350 and 500 cells/μl are also used to increase the intensity of monitoring. For these values, the sensitivity and specificity of our method are: CD4 < 250, sensitivity = 0.86, specificity = 0.81; CD4 < 350, sensitivity = 0.97, specificity = 0.83; and CD4 < 500, sensitivity = 0.96, specificity = 0.85.
Methods Comparison and Correlation Studies for Absolute CD4 Counts in 61 Adults in Botswana
One important application of our method is in pediatric HIV monitoring. The wide range of normal absolute CD4 counts in infants and children requires the use of CD4:CD8 ratios or CD4 percentages in pediatric infection. Results for CD4:CD8 ratios and CD4 percentages of T lymphocytes for all 67 participants (61 adults and six infants) are shown in . Agreement, bias, and correlations between the microchip method and flow cytometry are excellent for both CD4 percentages of T lymphocytes (A and B) and CD4:CD8 ratios (C and D). Bland–Altman plots for both CD4 percentages of T lymphocytes and CD4:CD8 ratios show low proportional bias, with tight 95% limits of agreement. Correlations are excellent for both CD4 percentages of T lymphocytes (r = 0.98, p < 0.0001) and CD4:CD8 ratios (r = 0.98, p < 0.0001). Overall, the data show that all three approaches to measuring CD4 cell counts can be accurately quantified using the microchip method, and that both adult and pediatric CD4 results can be obtained.
Methods Comparison and Correlation Studies for CD4 Percentages of Total T Cells and CD4:CD8 Ratios in 67 Human Subjects
To determine assay variability, we examined 20 replicate samples of blood from a single participant over the course of one day, using the established basic protocol. We determined that the coefficient of variance was 12% (data not shown), which is similar to other methods of CD4 counting [27
]. Although the assay described here introduced 16.5 μl of blood into the system, the actual volume of blood analyzed by digital image analysis is only 0.90 μl. We have conducted preliminary studies that suggest that we can accurately measure CD4 counts from less than 5 μl of blood obtained via fingerstick (data not shown); additional studies will be required to assess the correlation between CD4 counts obtained by fingerstick and by venipuncture.