High-throughput flow cytometry: Validation in ... - Wiley Online Library

© 2003 Wiley-Liss, Inc.

Cytometry Part A 53A:55– 65 (2003)

Technical Note

High-Throughput Flow Cytometry: Validation in Microvolume Bioassays Sergio Ramirez, Charity T. Aiken, Brett Andrzejewski, Larry A. Sklar, and Bruce S. Edwards* Cancer Research and Treatment Center, Department of Pathology, University of New Mexico Health Sciences Center, Albuquerque, New Mexico Received 20 June 2002; Revision Received 7 November 2002; Accepted 7 January 2002

Background: We recently reported an automated sample handling system, designated HyperCyt, by which samples are aspirated from microplate wells and delivered to the flow cytometer for analysis at rates approaching 100 samples per minute. In this approach, an autosampler and peristaltic pump introduce samples into a tubing line that directly connects to the flow cytometer. Air bubbles are inserted between samples to prevent sample dispersion. In the present work, we compare results of HyperCyt with those of conventional manual flow cytometric analysis in representative flow cytometric bioassays and describe a cell suspension method in which HyperCyt exploits the use of microvolume wells. Methods: Human eosinophils and neutrophils were treated with trypsin to generate a wide (⬎25-fold) range of membrane P-selectin glycoprotein ligand-1 (PSGL-1) expression and then stained with fluorescent anti–PSGL-1 antibodies. Human peripheral blood mononuclear cells were stained with fluorescein isothiocyanate- and phycoerythrin-conjugated monoclonal antibodies for multiparameter immunophenotype analysis. U937 cells labeled with PKH62GL were used to assess cell settling in microplate wells. Results: Differences in PSGL-1 expression levels were detected by HyperCyt autosampling of leukocytes from 96-well plates at an analysis rate of approximately 1.5 s/well. HyperCyt measurements linearly correlated with parallel manual

One of the most powerful aspects of flow cytometry is the ability to screen thousands of individual cells or particles per second on the basis of optical markers of cell phenotype or physiologic response. In contrast, flow cytometry has been slow to develop as an efficient means for rapid screening of multiple, discrete suspensions of cells or reagents. Such a capability promises to benefit a number of areas of biological investigation. For example, modern drug discovery involves testing of cellular targets against millions of potentially valuable compounds that may bind cellular receptors to effect clinically therapeutic

measurements (r2 ⫽ 0.98). Lymphocyte subpopulations were accurately distinguished and reproducibly quantified in multiparameter immunophenotyping assays performed over a range of HyperCyt analysis rates (1.4 –5.5 s/sample). When assay volumes were reduced to 10 ␮l/well in 60-well Terasaki plates, cells could be maintained in uniform suspension for up to 30 min by periodically inverting plates on a rotating carousel before HyperCyt analysis. HyperCyt analysis of five fluorescence-level Cyto-Plex beads sampled from Terasaki plate microwells at 2.5 s/well produced highly reproducible results over a wide range of input bead concentrations (from 7 ⫻ 105 to 20 ⫻ 106 beads/ml) that linearly correlated with manual analysis results. Conclusions: The HyperCyt autosampling system enabled a 10-fold or greater increase in sample thoughput compared with conventional manual flow cytometric sample analysis, with comparable analysis results. Assays were performed efficiently in 10-␮l volumes to enable significant reagent cost savings, use of quantity-limited reagents at otherwise prohibitive concentrations, and maintenance of uniform suspensions of cells for prolonged periods. Cytometry Part A 53A:55– 65, 2003. © 2003 Wiley-Liss, Inc.

Key terms: Flow cytometry; automation; drug discovery; biomolecular screening

Patent pending: A trademark has been filed for HyperCyt. Contract grant sponsor: National Institutes of Health; Contract grant numbers: GM60799 and RR14175; Contract grant sponsor: State of New Mexico Cigarette Tax. *Correspondence to: Bruce S. Edwards or Larry A. Sklar, Cytometry, CRF Room 217/219, UNM Health Sciences Center, 2325 Camino de Salud, Albuquerque, NM, 87131. E-mail: [email protected] or [email protected] Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/cyto.a.10035

56

RAMIREZ ET AL.

cellular responses. High-throughput flow cytometry represents an important multifactorial approach for screening large combinatorial libraries of such compounds. Moreover, there are other routinely performed assays that might benefit from an automated approach that is fast and requires only microliter sample volumes. Although there have been several commercial efforts to automate flow cytometry, only one previous approach, which we designated Plug Flow Cytometry, permits analysis of as many as 9 to 10 samples per minute (1– 4). We recently reported a novel next-generation automated approach, designated HyperCyt, by which samples are aspirated from microplate wells and delivered to the flow cytometer for analysis at rates approaching 100 samples/min (5). The sampling probe of the autosampler is connected directly to the sample input port of a flow cytometer with commercial peristaltic tubing. A peristaltic pump is used to sequentially aspirate sample particle suspensions from multiwell microplates and insert air bubbles between individual samples. This results in the generation of a tandem series of bubble-separated samples that are delivered directly to the flow cytometer for serial analysis. The sample size and air bubble size are determined by the time that the autosampler probe is in a microplate well or above a well taking in air. The HyperCyt system differs from our previously described Plug Flow Cytometry system in its ability to use smaller sample volumes (⬍3 ␮l), its delivery to the flow cytometer of a “train” of tandem samples separated by air bubbles, and its direct peristaltic pumping of sample into the flow cytometer without an intervening coupling valve (4). Because of the unique features of the HyperCyt mechanism (in particular intersample air bubbles and direct peristaltic pumping), it is imperative to evaluate the use of HyperCyt in bioassays independently of the previous characterization of the Plug Flow system. The initial HyperCyt studies documented important features for optimizing optical measurement quality and sample carryover (5). In the present study, we show that this autosampling method produces results comparable to those of conventional manual sample analysis in representative flow cytometric bioassays. We also demonstrate that cells can be efficiently sampled and analyzed multiple times from microplate wells of 10-␮l volume capacity and that this autosampling format enables a novel solution to the problem of cell settling. MATERIALS AND METHODS Cells and Beads U937 and K562 cell lines were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum (both from Hyclone, Logan, UT) at 37°C, 5% CO2, in T75 Falcon tissue culture flasks (Becton Dickinson, Franklin Lakes, NJ). Cyto-Plex 4-␮m-diameter red fluorescent beads were obtained from Duke Scientific Corp. (Palo Alto, CA). The manufacturer’s designation for the five sets of beads, listed in order of increasing fluorescence intensity, were L2, L4, L6, L8, and L10. Eosinophils were prepared from peripheral blood of healthy volunteers, as previously described (6), by sepa-

ration of granulocytes on Percoll and hypotonic lysis of contaminating erythrocytes. Eosinophils were then purified from the granulocytes on a Vario-Max magnetic negative selection column (Miltenyi Biotec, Auburn, CA) using magnetic microbeads coated with monoclonal antibodies (mAbs) CD16 and CD3 (Miltenyi Biotec) to remove neutrophils and the small number of contaminating T lymphocytes, respectively. Such preparations consisted of 95% or more eosinophils. Unfractionated granulocyte preparations typically consisted of 95% or more neutrophils and only 1% to 3% eosinophils and thus were designated as neutrophils for analysis. Mononuclear cells were prepared from heparinized peripheral blood of a healthy volunteer that was separated on a discontinuous gradient of Ficoll-Paque (Pharmacia Biotech AB, Uppsala, Sweden). Cells were washed, resuspended in HHB/HSA buffer (110 mM NaCl, 10 mM KCl, 1 mM MgCl2, 1.5 mM CaCl2, 30 mM HEPES, 10 mM glucose, supplemented with 0.1% human serum albumin; Immuno-US, Rochester, MI) and stored on ice before use. Trypsin Cleavage and Labeling of P-Selectin Glycoprotein Ligand-1 (PSGL-1) Eosinophils and neutrophils were added to V-wells of 96-well microplates at 105 cells/well, washed once with 200 ␮l of phosphate buffered saline (GibcoBRL, Grand Island, NY), and resuspended in 50 ␮l of ice-cold phosphate buffered saline supplemented with trypsin and ethylene-diaminetetra-acetic acid (GibcoBRL) to final trypsin concentrations of 0 to 2.5 mg/ml. After 5 min of incubation on ice, 50 ␮l of HHB/HSA was added to neutralize the trypsin. Cells were then pelleted and stained as previously described (7) with phycoerythrin (PE)– conjugated anti– PSGL-1 (CD162) mAbs PL1 (Ancell, Bayport, MN) and KPL1 (Pharmingen, San Diego, CA) and with G1 mAb (anti-CD62P, Ancell) as an isotype-matched negative control. Cells were then resuspended in 200 ␮l of HHB/HSA for flow cytometric analysis. Immunophenotyping Mononuclear cells were stained in V-well microplates, as previously described (7), with fluorescein isothiocyanate (FITC)–and PE-conjugated mAbs directed against CD3, CD4 (BD Biosciences, Mountain View, CA), CD19 (DAKO, Carpenteria, CA), and CD16 (Caltag, Burlingame, CA). Cell Microplate Suspension Experiments After growth to a cell density of approximately 106 cells/ml, U937 cells were fluorescently labeled with PKH67GL (Sigma, St. Louis, MO) according to the manufacturer’s instructions. In some experiments, K562 cells were labeled for 15 min at 37°C with 5-,6-carboxyfluorescein diacetate, succinimidyl ester (CFSE; Molecular Probes, Eugene, OR) at indicated concentrations. Labeled cells were washed, resuspended in HHB/HSA buffer, and added to wells of Terasaki microplates at 20,000 to 25,000 cells in 10 ␮l/well. Microplates were then incubated in the microassay rotational suspension system (MARSS), in

MICROVOLUME HT FLOW CYTOMETRY

57

FIG. 1. HyperCyt configuration. System components used in the present studies include: (a) a PC laptop computer, Windows 95/98 operating system, CRFSoft software (in-house); (b) a Gilson autosampler (223 Sample Changer or 215 Liquid Handler); (c) a stainless steel sampling probe (0.01 in. inner diameter, 0.02 in. outer diameter; Small Parts, Inc., Miami Lakes, FL); (d) a 96 V-well (Costar, Cambridge, MA) or 60-well Terasaki microplate (Nalge Nunc International, Rochester, NY); (e) polyvinyl chloride tubing (PVC; 0.01 in. inner diameter; Spectra Hardware, Westmoreland City, PA); (f) a Gilson Minipuls 3 peristaltic pump; (g) junction between the PVC tubing and the flow cytometer sampling port; (h) FACScan flow cytometer (BD Biosciences); (i) Macintosh computer, OS9 operating system, Cell Quest software (BD Biosciences); and (j) microassay rotational suspension system.

which they were rotated between inverted and upright positions at approximately 4 rpm for the indicated time intervals at 24°C before HyperCyt flow cytometric analysis. The MARSS consisted of the revolving drum of a peristaltic pump, and the microplates were attached to the surface of the drum. In some experiments, stationary upright control microplates were incubated in parallel. HyperCyt Autosampling and Flow Cytometry The HyperCyt system for automated sampling of microplate wells and delivery to a flow cytometer was described elsewhere (5) (Fig. 1). Briefly, this system consists of a Gilson 215 Liquid Handler or 223 Sample Changer autosampler (Middleton, WI) equipped with a Gilson Minipuls 3 peristaltic pump and controlled by a laptop computer running a Microsoft Visual C⫹⫹ software program, CRFSoft, developed in-house. This approach uses air bubbles to separate samples introduced by the autosampler and peristaltic pump into a tubing line that directly connects to the flow cytometer. The sample size and air bubble size are determined by the time that the autosampler probe is in a microplate well or above a well taking in air. A FACScan flow cytometer (BD Biosciences, San Jose, CA) was used for all experiments. Fluorescence of all dyes was excited with the 488-nm line of an argon ion laser. Fluorescence emission was detected in the FL1 channel (530 ⫾ 15 nm) for cells labeled with FITC-conjugated mAbs, PKH62GL and CFSE, and in the FL2 channel (585 ⫾ 21 nm) for cells labeled with PE-conjugated mAbs. Sample transit time was defined as the time required for the first sample in a series to travel from its source well to

the laser beam interrogation point. This was typically approximately 15 s under the standard volumetric flow conditions used in the present studies. Sample analysis time was defined as the average time between samples during passage through the laser beam interrogation point. This differed with respect to programmed autosampling parameters as described in individual experiments. The average sample analysis time was calculated as: (tlast ⫺ tfirst)/n, where tfirst and tlast are the time points for detection of the leading edge of the first sample and the trailing edge of the last sample, respectively, and n is the total number of samples in the series. All estimates of sample analysis rates were based on average sample analysis times. A computer program, FCSQuery, was developed in-house for efficiently analyzing the time-resolved data clusters generated by rapid multiwell sampling. Software algorithms automatically identified data clusters representing cells from individual sample wells and calculated mean and median fluorescence intensities and event number for each cluster. Data also were input automatically to Microsoft Excel spreadsheets for more detailed analysis. All analyses were done offline on flow cytometry list-mode data files stored in FCS 2.0 format. RESULTS Quantitative Analysis of Cellular Membrane Protein Expression We first evaluated the HyperCyt autosampling system for its ability to accurately discriminate different levels of cellular membrane molecule expression. The adhesion

58

RAMIREZ ET AL.

molecule PSGL-1 is a trypsin-sensitive molecule that is expressed at approximately twofold higher levels on eosinophils than on neutrophils (6,8). To produce a wide range of membrane PSGL-1 expression, neutrophils and eosinophils were briefly exposed to different concentrations of trypsin. The trypsin reactions were performed in V-bottom wells of a 96-well microplate, and the cells subsequently were stained immunofluorescently with mAbs that recognize two distinct PSGL-1 epitopes, PL1 and KPL1 (9,10). The cells in each well were then analyzed in the flow cytometer by three different methods. First, cells were sampled with the HyperCyt autosampling system with a 0.9-s sample aspiration time per well and a 0.9-s sampling probe wash step between wells. Second, the cells were sampled by the HyperCyt system with the wash step omitted. Third, after two rounds of sampling with each HyperCyt protocol, the remaining contents of each well were transferred to separate tubes for conventional manual sample analysis. In representative results from one round of HyperCyt no-wash autosampling, 29 wells were analyzed in approximately 44 s (Fig. 2, top). Each of the 29 discrete clusters of dots represents the log fluorescence distribution of individual cells from one well. Because of the relatively wide dynamic range of PSGL-1 fluorescence intensity, flow cytometric measurements were made on a log scale. The median log fluorescence intensity of each cluster was calculated and converted to a linear scale value for use in subsequent quantitative analyses (Fig. 2, bottom). When similar operations were performed on data from each of the three sampling methods, the PSGL-1 trypsin hydrolysis profiles obtained from both HyperCyt autosampling protocols were similar to each other and to the profiles obtained by conventional manual sampling (Fig. 3A, B). Response linearity was evaluated by plotting manual versus HyperCyt median fluorescence intensity results (Fig. 3C). The lines fitted by regression of the original log fluorescence intensity data had slopes of 0.97 ⫾ 0.01 and 0.95 ⫾ 0.01 (mean ⫾ standard deviation) for rinse and no-rinse HyperCyt protocols, respectively. No significant departure from linearity was detectable in a runs test of the data (P ⫽ 0.25 and 0.80 for rinse and no-rinse protocols, respectively). Response reproducibility was evaluated as the coefficient of variation (C.V.) for duplicate measurements made in each assay condition. This corresponded to a 2.2 ⫾ 3.2% C.V. for the rinse protocol and a 1.9 ⫾ 3.7% C.V. for the no-rinse protocol. Sample carryover from the HyperCyt rinse protocol results was estimated at 3.8 ⫾ 2.2%. Carryover was calculated as 100 ⫻ R/(S ⫹ R), where R is the number of cells detected during the rinse step and S is the number of cells detected in the sample just before the rinse. These results thus demonstrated a strong linear relation between HyperCyt analyses and data obtained by conventional manual analysis. Moreover, results of the HyperCyt no-wash protocol, in which effects of sample carryover should have been maximally manifested, were comparable to results of the wash protocol. This observation suggested that sample carryover between wells is sufficiently low for this type of quantitative fluorescence analysis so

FIG. 2. HyperCyt analysis of P-selectin glycoprotein ligand-1 (PSGL-1) adhesion molecule expression after graded cleavage with trypsin. In wells of a 96-well plate, eosinophils and neutrophils were incubated on ice for 5 min with the indicated trypsin concentrations. Cells were stained with fluorescent antibodies directed against PSGL-1 epitopes, PL1 and KPL1, and with G1 antibody as a negative control. Twenty-nine wells were analyzed in approximately 44 s by the HyperCyt at a 0.9-s sample aspiration time per well and no rinsing between wells (top). Because of the wide range of PSGL-1 expression, measurements were made on a log scale, with each discrete cluster of dots representing individual cells from one well. The median log fluorescence intensity of each cluster was calculated and converted to a linear scale value for use in subsequent quantitative analyses (bottom).

that a sampling-probe wash step is not required. Moreover, the small sample volumes (⬃2.5 ␮l), the rapid sampling rates, and the use of air bubbles to separate samples in the autosampling protocols did not appear to significantly compromise the quality of the data. Multiparameter Immunophenotype Analysis A second set of studies evaluated the performance of HyperCyt in a multiparameter fluorescence analysis format. Prototypic of such a format is the conventional immunofluorescence analysis of leukocyte phenotype, in which multiple leukocyte membrane determinants are probed in parallel by using antibodies conjugated with distinct fluorophores. Mononuclear cells were stained with fluorescent mAbs in 96 V-well microplates and resuspended in 200 ␮l of HHB per well for subsequent HyperCyt analysis. The mAbs were used in FITC- and PE-conjugated pairs to distinguish T lymphocytes from natural killer (NK) lymphocytes (CD3-PE, CD16-FITC), T lympho-


FIG. 3. Comparison of HyperCyt with manual flow cytometric analysis. Eosinophils and neutrophils were treated with trypsin and stained as in Figure 2 with antibodies PL1 (A) and KPL1 (B). Cells were suspended in 200 ␮l per well and sampled first by HyperCyt at a 0.9-s sample aspiration time per well with a 0.9-s rinse of the sampling probe between wells (circles). The cells were then sampled again by HyperCyt with the same aspiration time per well but no intervening rinse step (squares). Remaining cells in each well were then sampled manually (triangles). Results were plotted after transformation of log fluorescence data to a linear scale. C: A plot of manual versus HyperCyt measurements for both monoclonal antibodies in the original log fluorescence intensity units indicated a significant linear correlation for the rinse (circles) and no-rinse (triangles) protocols (r2 ⫽ 0.98, P ⬍ 0.0001 for each). Results are representative of two separate experiments.

cytes from B lymphocytes (CD3-PE, CD19-FITC), and CD4 T lymphocytes from CD8 T lymphocytes (CD4-PE, CD3FITC). Light scatter distributions characteristic of lymphocytes and monocytes were distinguishable under HyperCyt anal-

59

ysis conditions, but were shifted to the left on the forward light scatter intensity axis as compared with mononuclear cells analyzed under conventional manual conditions (Fig. 4, top left panel). This finding was consistent with previously observed effects of HyperCyt analysis on light scatter distributions of cells (5). To evaluate lymphocyte subset measurement reproducibility, the HyperCyt system was programmed to aspirate triplicate samples from each well, with an air bubble inserted between each sample. In a representative experiment (Fig. 4, top right and bottom panels), triplicate samples were taken from each of three wells with an aspiration time of 2 s/sample. The sampling probe was not washed between samples. The nine total samples were resolved as temporally discrete data clusters when FITC fluorescence intensity was plotted as a function of time (Fig. 4, top right panel). The two-color fluorescence intensity dot plots for each of the nine data clusters are illustrated in the bottom panel of Figure 4. Lymphocyte subset percentage measurements were generally reproducible within each triplicate group. Cells from the same donor also were seeded in V-wells at 1.5 ⫻ 106 and 3 ⫻ 106 cells/ml and immunostained as before, and the aspiration time for each sample was changed. At sample aspiration times of 0.9, 2, and 5 s, the average times required to analyze cells were 1.4 ⫾ 0.3, 2.5 ⫾ 0.3, and 5.5 ⫾ 0.3 s/sample (mean ⫾ standard deviation), respectively. The number of cells available for analysis in each sample ranged from 460 to 4430, depending on the sample aspiration time and number of cells initially seeded in each well (Table 1). The lymphocyte subset percentage measurements were remarkably similar over the entire range of experimental variation. These results illustrate how sample size can be significantly increased by relatively simple means to enable, e.g., more accurate measurements of low-frequency cell subpopulations. It is noteworthy in the above set of experiments that a 10-fold increase in sample size was achieved while sustaining analysis rates in excess of 10 samples/min. In a previous study (11), the mononuclear cells of this donor were rigorously characterized in 30 separate determinations over a 19-month period. Edwards et al. detected 58 ⫾ 4% CD3, 9 ⫾ 1% CD20 (which detects most but not all B cells detected by CD19), and 39 ⫾ 3% CD3/CD4 double-positive lymphocytes. These results corresponded quite well to the 58 ⫾ 2% CD3, 13 ⫾ 1% CD19, and 37 ⫾ 1% CD3/CD4-positive lymphocytes detected by HyperCyt in the present study (mean ⫾ standard deviation of pooled data from Table 1). Accounting for the relatively high percentage of CD3/CD19 double-negative lymphocytes was the presence of a high frequency of NK cells in this donor. This subset corresponded to the 19% CD16-positive NK cells (Fig. 4, Table 1) and an additional 10% or so of NK cells detected with CD56 mAbs (data not shown). Another point of interest is that the two CD3 mAb conjugates detected similar frequencies of T cells despite the relatively dim staining with the CD3-FITC mAb (57 ⫾ 1% and 58 ⫾ 2% for the FITC and PE conjugate results, respectively, in Table 1). The dim CD3-FITC staining likely reflected the use of a suboptimal mAb preparation be-

60

RAMIREZ ET AL.

FIG. 4. Immunophenotype analysis with HyperCyt. Peripheral mononuclear cells from a healthy donor were stained with fluorescent monoclonal antibodies to distinguish B, T, and natural killer (NK) lymphocyte subpopulations. Top, left: Light scatter analysis of cells under manual (top) and HyperCyt (bottom) analysis conditions. Enclosed within circles are the putative lymphocyte populations on which fluorescence analyses were gated. Top, right: Three wells of a 96well microplate were sampled in triplicate by HyperCyt, with an aspiration time of 2 s/sample. In post-data acquisition analysis with FCSQuery software, the resulting nine data clusters were resolved automatically on the horizontal time axis into separate groups for fluorescence analysis. Each rectangle encloses a distinct cluster of events representing cells labeled with CD16-FITC plus CD3-PE mAbs (A1–3), CD19-FITC plus CD3-PE mAbs (B1–3), and CD3-FITC plus CD4-PE mAbs (C1–3). Only the left and right boundaries of each rectangle, the time domain boundaries, were used to gate the fluorescence analysis; the top and bottom rectangular boundaries are visual aids to help distinguish how the data were partitioned on the time axis. The solid bar beneath the horizontal time axis represents a 20-s span. Bottom: Dot plots of FITC and PE fluorescence intensity data for each of the time domain-gated data clusters A1–3 (top row), B1–3 (middle row), and C1–3 (bottom row), distinguished as described above. Numbers in dot plot quadrants represent the percentage of total events detected in the quadrant. Approximately 1,530 ⫾ 160 (mean ⫾ standard deviation) events were analyzed in each dot plot. FITC, fluorescein isothiocyanate; mAbs, monoclonal antibodies; PE, phycoerythrin.

cause the other FITC-conjugated mAbs produced wellresolved staining. HyperCyt Analysis in Small-Volume Microplate Wells Because the autosampling method worked reliably with aspirated sample volumes of 2.5 ␮l or smaller, we investigated the use of microplates with smaller wells. Terasaki microplates have truncated conical wells of approxi-

mately 10-␮l volume capacity arranged in a 6-row, 10column, 60-well format. A 10-␮l fluid volume was considered large enough that evaporative fluid loss was unlikely to be a significant problem under most assay conditions but small enough to substantially reduce total amounts of cells and reagents required for assays. An additional feature of fluids in microwells of this geometric scale is the strong influence of liquid surface tension forces. Terasaki microplates can be completely

61


Table 1 Effect of Cell Concentration and Sample Aspiration Time

a

Cell phenotype

CD16-FITC/CD3-PE

Cells/ml 1.5 ⫻ 106 3.0 ⫻ 106

CD19-FITC/CD3-PE

1.5 ⫻ 106 3.0 ⫻ 106

CD3-FITC/CD4-PE

1.5 ⫻ 106 3.0 ⫻ 106

Time of aspirations

No. cells analyzed in sample

Upper left

Upper right

Lower left

Lower right

0.9 2 5 0.9 2 5 0.9 2 5 0.9 2 5 0.9 2 5 0.9 2 5

590 980 2,740 940 1,940 4,290 460 1,230 2,780 800 1,860 3,950 590 1,310 3,170 750 1,730 4,430

60 58 59 56 58 61 58 60 61 59 60 60 3 2 4 4 5 5

2 6 2 2 2 2 1 1 1 1 1 1 38 39 36 37 38 37

18 19 20 22 22 19 28 26 25 26 26 26 40 39 40 38 39 39

20 17 19 21 19 18 14 13 13 14 13 13 19 20 19 21 19 20

Cells in quadrant (%)

a

Fluorescein isothiocyanate (FITC) is the x-axis parameter (left to right), and phycoerythrin (PE) is the y-axis parameter (lower to upper).

inverted without spilling of well contents. We expected that this latter feature could be exploited as a novel means to avoid the problem of cell settling, a confounding factor in flow cytometric automation. To test this concept, Terasaki microplates containing 10-␮l suspensions of U937 cells were attached to the surface of a vertically revolving wheel on which they continuously orbited the axis of rotation, reciprocating between inverted and upright orientations, at a rate of approximately 4 rpm. When microplates were incubated up to 20 min on this MARSS, the numbers of U937 cells recovered from 10 replicate well suspensions by HyperCyt were consistently comparable to initial samples aspirated at 0 min (Fig. 5). In contrast, cell recoveries progressively decreased over 20 min in stationary control microplates (Fig. 5). Thus, the use of small microvolume wells in combination with periodic plate inversion was a simple and practical means for prolonged maintenance of uniform cell suspensions. To assess HyperCyt analysis of cell fluorescence in Terasaki microplates, K562 cells were labeled with CFSE at two concentrations (200 nM and 2 ␮M). The resulting cell preparations were strongly fluorescent in comparison with unlabeled cells, falling in the third and fourth decades of the log intensity scale, respectively (Fig. 6A). Labeled cells from each preparation were dispensed into alternating wells in the first five wells of each six-well microplate row. Unlabeled cells (autofluorescence control) were dispensed into the wells at the end of each row. When all 60 wells were sampled by HyperCyt and analyzed within a 90-s span, 60 clearly resolved data clusters were detected (Fig. 6B). The three levels of fluorescence intensity were reproducibly detected over the entire span of the microplate when the cells were sampled immedi-

ately after dispensing into wells (Fig. 6C) and after a subsequent 30-min interval, during which the microplate was incubated on the MARSS (Fig. 6D).

FIG. 5. Use of small-volume microplate wells to enable prolonged suspension of cells. Terasaki plates containing 10 replicate 10-␮l U937 cell suspensions were attached to the microassay rotational suspension system so that the microplates periodically alternated between inverted and upright positions at a rate of approximately 4 rpm. Surface tension prevented sample loss from inverted wells. For up to 20 min on the rotating carousel, cell recoveries were consistently comparable to the initial samples (circles). In contrast, cell recoveries significantly declined in control plates that were kept upright and stationary (triangles).

62

RAMIREZ ET AL.

FIG. 6. Quantitative fluorescence analysis of cells in Terasaki microplates. A: Manual fluorescence analysis of K562 cells that were unlabeled (light gray histogram) or labeled with 200 nM (dark gray histogram) and 2 ␮M (black histogram) of 5-,6-carboxyfluorescein diacetate, succinimidyl ester (CFSE). B: Labeled and unlabeled cells were dispensed in 10-␮l aliquots into a 60-well Terasaki plate, with unlabeled cells in every sixth well and the two levels of CFSE-labeled cells alternating in the remaining wells. All 60 wells were sampled by HyperCyt and analyzed within a 90-s span (solid bar, horizontal time axis). Each discrete cluster of events represents cells from a separate well. C: The median green fluorescence intensity of cells from each well when the microplate was sampled immediately after dispensing cells into wells. D: The median green fluorescence intensity of cells sampled from the same microplate after it was incubated for 30 min in the microassay rotational suspension system (MARSS; 4 rpm, 24°C). The means ⫾ standard deviations of the number of cells sampled from each well were 844 ⫾ 210 in immediately sampled wells and 938 ⫾ 242 in wells sampled after 30 min in the MARSS.

HyperCyt Analysis Over a Broad Range of Particle Concentrations To more rigorously assess the effect of input particle concentration on HyperCyt sampling performance, wells of Terasaki plates were seeded with mixtures of fluorescent Cyto-Plex beads ranging in concentration from 7 ⫻ 105 to 20 ⫻ 106 beads/ml. Each mixture contained five sets of beads with discrete levels of red fluorescent dye.

When analyzed manually under optimal conditions of flow (FACScan low-flow setting, nominal sample flow rate of ⬃0.2 ␮l/s), the beads were resolved on a log scale as five distinct populations spanning a three-decade range of fluorescence intensity (Fig. 7A). HyperCyt analysis was performed by sampling 12 replicate wells at each bead concentration. Sample aspiration time was 2 s for each well, with no rinsing of the sample probe between wells, a


protocol that achieved a sampling rate of approximately 24 wells/min. The number of beads analyzed in each sample ranged from 1,000 at the low end of bead concentration to 12,000 at the high end (Table 2). Although there was a broadening of the fluorescence intensity distributions of the beads under these conditions of HyperCyt analysis (Fig. 7B), the median fluorescence intensities of

63

each bead set under each condition of bead concentration showed a linear correlation with measurements made manually in parallel (Fig. 7C, Table 2). Likewise, the percentages of beads detected in each population peak were nearly identical under manual and all HyperCyt analysis conditions. DISCUSSION In the initial characterization of the HyperCyt system (5), we demonstrated sampling of beads and cells from 96-well plates at rates of up to 100 samples/min and carryover between samples of 1% to 3%. The present studies extended these findings to validate the HyperCyt autosampling system as a high-throughput method for cell and particle fluorescence measurements of comparable quality to conventional manual flow cytometry. We demonstrated the ability of HyperCyt to accurately measure variations in cell and bead immunofluorescence over a four-decade range of fluorescence intensity and at input particle concentrations of up to 20 ⫻ 106/ml. This was accomplished at analysis rates of 24 to 40 samples/min (1.5 to 2.5 s/sample) and the use of source well sample volumes as small as 8 ␮l. The HyperCyt system also permitted reproducible quantification of lymphocyte subpopulations when multiparameter immunophenotyping assays were performed in 96-well microplates. An important feature of HyperCyt autosampling is the requirement for minimal sample volumes. In the present studies, the sampling probe was typically programmed to aspirate from each well for approximately 1 s. This resulted in samples averaging approximately 2.5 ␮l and sample analysis rates averaging 40 samples/min. Even at fivefold longer aspiration times, implemented to increase the total cell counts in each sample (Table 1), the total sample volume removed from each well was much smaller than feasible with manual flow cytometry (⬃13 ␮l), and analysis rates were still in excess of 10 samples/ min. Another approach by which to increase the cell counts in each aspirated sample is to increase cell concentrations in source wells (Tables 1 and 2). At higher cell concentrations, smaller sample volumes are required to achieve a targeted cell count range, and sample throughput can be increased proportionately. However, limiting

4™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™ FIG. 7. Analysis of a five fluorescent peak array of Cyto-Plex beads. A stock mixture of Cyto-Plex beads was prepared from five sets of beads with discrete levels of red fluorescent dye. The final concentration of each set was approximately 4 ⫻ 106 beads/ml for a total concentration of 20 ⫻ 106 beads/ml. A: The stock was diluted to approximately 1.1 ⫻ 106 beads/ml for manual analysis at a sample flow rate of approximately 0.2 ␮l/s that resulted in optimal resolution of the five fluorescence intensity peaks. B: Alternatively, 10-␮l aliquots of the 20 ⫻ 106 beads/ml stock were added to each of 12 replicate wells of a Terasaki plate (2 ⫻ 105 beads/well) and analyzed by HyperCyt with an aspiration time of 2 s/sample and a peristaltic pump rate of 15 rpm. The illustrated fluorescence intensity profile represents pooled data from all 12 samples. Statistics for intersample variation are presented in Table 2. C: Median log fluorescence intensity data are plotted for each bead set as obtained in the manual and HyperCyt analysis protocols illustrated in A and B. The fitted line had a slope of 0.997 and a correlation coefficient (r2) of 1.0. Similar linear correlations were obtained when HyperCyt analysis results from more dilute input concentrations of beads were compared with the optimal manual analysis results (Table 2).

64

RAMIREZ ET AL.

Table 2 Comparison of HyperCyt and Manual Analysis of Cyto-Plex Beads* HyperCyt Input beads/ml (⫻106) Beads detected/sample % Totala Peak 1 Peak 2 Peak 3 Peak 4 Peak 5 MFIb Peak 1 Peak 2 Peak 3 Peak 4 Peak 5

Manual (high)

Manual (low)

20 12,000 ⫾ 350

6.7 6,000 ⫾ 490

2.2 2,200 ⫾ 250

0.7 1,000 ⫾ 160

1.1 1,100 ⫾ 90

1.1 390 ⫾ 70

17 ⫾ 0.4 21 ⫾ 0.3 23 ⫾ 0.5 19 ⫾ 0.3 20 ⫾ 0.3

17 ⫾ 0.6 22 ⫾ 0.7 23 ⫾ 0.6 19 ⫾ 0.5 20 ⫾ 0.8

17 ⫾ 1.0 22 ⫾ 0.5 23 ⫾ 0.5 18 ⫾ 0.7 19 ⫾ 0.8

17 ⫾ 1.2 22 ⫾ 0.3 23 ⫾ 1.2 19 ⫾ 0.8 19 ⫾ 1.6

17 ⫾ 0.9 21 ⫾ 1.4 23 ⫾ 1.7 19 ⫾ 1.2 20 ⫾ 1.2

16 ⫾ 1.9 21 ⫾ 3.6 24 ⫾ 2.0 19 ⫾ 2.6 20 ⫾ 1.9

240 ⫾ 1 403 ⫾ 1 544 ⫾ 1 675 ⫾ 1 772 ⫾ 1

240 ⫾ 1 402 ⫾ 1 542 ⫾ 1 672 ⫾ 2 770 ⫾ 1

241 ⫾ 1 402 ⫾ 1 542 ⫾ 1 673 ⫾ 2 770 ⫾ 1

241 ⫾ 2 403 ⫾ 2 543 ⫾ 2 672 ⫾ 2 770 ⫾ 1

243 ⫾ 1 405 ⫾ 1 546 ⫾ 1 678 ⫾ 1 773 ⫾ 1

245 ⫾ 1 406 ⫾ 1 548 ⫾ 1 679 ⫾ 1 775 ⫾ 1

*Cyto-Plex beads were added to Terasaki plate wells at the indicated total input bead concentrations and sampled with the HyperCyt system or added to a tube at 1.1 ⫻ 106 beads/ml and sampled manually at the high and low FACScan sample flow rate settings (nominally 1 and 0.2 ␮l/s, respectively). Twelve replicate wells were sampled to obtain a mean and standard deviation for each HyperCyt measurement. The continuous sample data stream from each manual analysis was divided into 12 sequential 2.5-s segments (the average time span of HyperCyt samples) to obtain mean and standard deviation estimates for each manual measurement. a Percentage of total beads (mean ⫾ standard deviation) detected in each discrete bead fluorescence intensity population; peaks are numbered in order of increasing intensity. b Median fluorescence intensity (mean ⫾ standard deviation, log scale units) of each bead population.

this approach is the ability of the flow cytometer’s detection electronics to efficiently detect and process all events as they become more closely spaced in time at progressively higher cell concentrations. This is evident in data shown in Table 2, in which threefold increases in input particle concentrations resulted in only approximately 2.3-fold average increases in the numbers of detected events. Despite this apparent loss of data, presumably due to detection circuit “dead time” during processing of individual events, data from successfully processed events were of high quality. This was indicated by the tight linear relation between data obtained manually under optimal conditions of sample flow and data from HyperCyt analysis performed even at the highest tested input sample concentration of 20 ⫻ 106 particles/ml (Fig. 7C). In a flow cytometer with faster event-processing capabilities, it is expected that a desired sample event number endpoint should be attained with lower input particle concentrations and/or shorter sample aspiration times. Thus, improving event-processing efficiency represents one approach by which it should be possible to minimize sample quantity requirements and maximize sample throughput. It is also worth noting that, with multiplexing of beadbased assays on multilevel fluorescent bead sets, each set providing a platform for a discrete assay may enable an additional 5- to 10-fold increase in assay throughput. Because of the reduced sample volume requirement with HyperCyt, it was feasible to perform flow cytometric assays on cells seeded in 10-␮l Terasaki plate wells. It is expected that this microwell format will enable quantitative multiparameter flow cytometric measurements to be made rapidly on small numbers of cells so as to maximize the efficient use of, e.g., limited numbers of cells from clinical specimens. Likewise, quantity-limited reagents

could be used in smaller amounts to achieve otherwise prohibitively high concentrations. The small well volume also enabled a solution to the problem of particle settling. Capillary forces due to liquid surface tension are sufficient to prevent sample loss from inverted wells, so that uniform cell suspensions are maintained for extended periods by periodically inverting microplates in the MARSS. The present study demonstrated that suspensions can be maintained over a span of 30 min, but practical upper limits remain to be defined. Our cell settling data (Fig. 5) suggested that optimal results are obtained by analyzing wells in packets that can be processed within 1 to 2 min. The 60-well Terasaki microplate represents a convenient packet of wells commensurate with this requirement. We anticipate that, when large numbers of assays are to be performed, it will be practical to prepare multiple microplates that then can be incubated on the MARSS while awaiting subsequent sequential HyperCyt processing. Routine practical application of this high-throughput flow cytometric analysis technology will require additional considerations that we are now beginning to address. One such issue is synchronization between the autosampler and the flow cytometer so that analyzed samples can be unambiguously assigned to their source wells. Misregistration between the sequence of sampled source wells and the resulting sequence of time-resolved event clusters might occur, for example, if a sample is missed (e.g., empty well), merges with another sample, or subdivides during transit. A general solution to detecting such occurrences is to develop coding systems by which proper registration is reestablished at fixed sampling intervals. A simple example is illustrated in Figure 6B, in which the negative control sample of unlabeled cells is

65


always added to the last well of each six-well row. Offline analysis software can easily detect control well data, count the number of event clusters in between, and flag for reanalysis any row in which the count differs from five. More sophisticated coding schemes can easily be envisioned that involve markers such as multilevel fluorescent beads mixed in different proportions to denote multiple discrete levels of source well/event cluster registration. We recently developed a system involving fluorescent marker beads and custom bead detection circuitry with which a first level of online synchronization between the autosampler and the flow cytometer has been achieved (manuscript in preparation). A second issue in the routine practical application of HyperCyt is the implementation of a data management system to facilitate the storage, retrieval, and quality control of large volumes of rapidly accumulating multiparameter data. This will be of concern especially in extended screening activities involving the analysis of thousands to tens of thousands of samples per day. Fortunately, there are sophisticated commercial data management software packages that are used routinely for managing high-throughput screening data from a diversity of analytical instruments. Of particular interest are packages capable of importing data from spreadsheet programs (e.g., IDBS ActivityBase) as we have already implemented the output of formatted time-resolved data to Microsoft Excel spreadsheets as a standard feature of our in-house FCSQuery software. ACKNOWLEDGMENTS This study was performed with technical and instrument support from the University of New Mexico Shared

Flow Cytometry Resource, University of New Mexico Health Science Center, and CRTC, Albuquerque, NM. LITERATURE CITED 1. Edwards BS, Kuckuck F, Sklar LA. Plug flow cytometry: an automated coupling device for rapid sequential flow cytometric sample analysis. Cytometry 1999;37:156 –159. 2. Edwards BS, Kuckuck FW, Prossnitz ER, Okun A, Ransom JT, Sklar LA. Plug flow cytometry extends analytical capabilities in cell adhesion and receptor pharmacology. Cytometry 2001;43:211:216. 3. Ransom JT, Edwards BS, Kuckuck F, Okun A, Mattox DK, Prossnitz ER, Sklar LA. Flow cytometry systems for drug discovery and development. SPIE Proc 2000;3921:90 –100. 4. Edwards BS, Kuckuck FW, Prossnitz ER, Ransom JT, Sklar LA. HTPS Flow cytometry: a novel platform for automated high throughput drug discovery and characterization. J Biomol Screen 2001;6:83–90. 5. Kuckuck FW, Edwards BS, Sklar LA. High throughput flow cytometry. Cytometry 2001;44:83–90. 6. Edwards BS, Curry MS, Tsuji H, Larson RS, Brown D, Sklar LA. Expression of P-selectin at low site density promotes selective recruitment of eosinophils over neutrophils. J Immunol 2000;165:404 – 410. 7. Edwards BS, Shopp GM. Efficient use of monoclonal antibodies for immunofluorescence. Cytometry 1989;10:94 –97. 8. Symon FA, Lawrence MB, Williamson ML, Walsh GM, Watson SR, Wardlaw AJ. Functional and structural characterization of the eosinophil P-selectin ligand. J Immunol 1996;157:1711–1719. 9. Li F, Erickson HP, James JA, Moore KL, Cummings RD, McEver RP. Visualization of P-selectin glycoprotein ligand-1 as a highly extended molecule and mapping of protein epitopes for monoclonal antibodies. J Biol Chem 1996;271:6342– 6348. 10. Snapp KR, Ding H, Atkins K, Warnke R, Luscinskas FW, Kansas GS. A novel P-selectin glycoprotein ligand-1 monoclonal antibody recognizes an epitope within the tyrosine sulfate motif of human PSGL-1 and blocks recognition of both P- and L-selectin. Blood 1998;91:154 – 164. 11. Edwards BS, Altobelli KK, Nolla HA, Harper DJ, Hoffman RR. A comprehensive quality assessment approach for flow cytometric immunophenotyping of human lymphocytes. Cytometry, 1989;10:433– 441.