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 samplehandling system, designated HyperCyt, by which samplesare aspirated from microplate wells and delivered to theflow cytometer for analysis at rates approaching 100 sam-ples per minute. In this approach, an autosampler andperistaltic pump introduce samples into a tubing line thatdirectly connects to the flow cytometer. Air bubbles areinserted between samples to prevent sample dispersion.In the present work, we compare results of HyperCyt withthose of conventional manual flow cytometric analysis inrepresentative flow cytometric bioassays and describe acell suspension method in which HyperCyt exploits theuse of microvolume wells.Methods: Human eosinophils and neutrophils were treatedwith trypsin to generate a wide (�25-fold) range of mem-brane P-selectin glycoprotein ligand-1 (PSGL-1) expressionand then stained with fluorescent anti–PSGL-1 antibodies.Human peripheral blood mononuclear cells were stainedwith fluorescein isothiocyanate- and phycoerythrin-conju-gated monoclonal antibodies for multiparameter immuno-phenotype analysis. U937 cells labeled with PKH62GL wereused to assess cell settling in microplate wells.Results: Differences in PSGL-1 expression levels were de-tected by HyperCyt autosampling of leukocytes from 96-wellplates at an analysis rate of approximately 1.5 s/well. Hyper-Cyt measurements linearly correlated with parallel manual

measurements (r2 � 0.98). Lymphocyte subpopulationswere accurately distinguished and reproducibly quantified inmultiparameter immunophenotyping assays performed overa range of HyperCyt analysis rates (1.4–5.5 s/sample). Whenassay volumes were reduced to 10 �l/well in 60-well Ter-asaki plates, cells could be maintained in uniform suspensionfor up to 30 min by periodically inverting plates on a rotatingcarousel before HyperCyt analysis. HyperCyt analysis of fivefluorescence-level Cyto-Plex beads sampled from Terasakiplate microwells at 2.5 s/well produced highly reproducibleresults over a wide range of input bead concentrations (from7 � 105 to 20 � 106 beads/ml) that linearly correlated withmanual analysis results.Conclusions: The HyperCyt autosampling system en-abled a 10-fold or greater increase in sample thoughputcompared with conventional manual flow cytometric sam-ple analysis, with comparable analysis results. Assays wereperformed efficiently in 10-�l volumes to enable signifi-cant reagent cost savings, use of quantity-limited reagentsat otherwise prohibitive concentrations, and maintenanceof 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

One of the most powerful aspects of flow cytometry isthe ability to screen thousands of individual cells or par-ticles per second on the basis of optical markers of cellphenotype or physiologic response. In contrast, flow cy-tometry has been slow to develop as an efficient means forrapid screening of multiple, discrete suspensions of cellsor reagents. Such a capability promises to benefit a num-ber of areas of biological investigation. For example, mod-ern drug discovery involves testing of cellular targetsagainst millions of potentially valuable compounds thatmay bind cellular receptors to effect clinically therapeutic

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 NewMexico 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: bedwards@salud.unm.edu or lsklar@salud.unm.eduPublished online in Wiley InterScience (www.interscience.wiley.com).

DOI: 10.1002/cyto.a.10035

© 2003 Wiley-Liss, Inc. Cytometry Part A 53A:55–65 (2003)

cellular responses. High-throughput flow cytometry rep-resents an important multifactorial approach for screeninglarge combinatorial libraries of such compounds. More-over, there are other routinely performed assays thatmight benefit from an automated approach that is fast andrequires only microliter sample volumes.

Although there have been several commercial efforts toautomate flow cytometry, only one previous approach,which we designated Plug Flow Cytometry, permits anal-ysis of as many as 9 to 10 samples per minute (1–4).

We recently reported a novel next-generation auto-mated approach, designated HyperCyt, by which samplesare aspirated from microplate wells and delivered to theflow cytometer for analysis at rates approaching 100 sam-ples/min (5). The sampling probe of the autosampler isconnected directly to the sample input port of a flowcytometer with commercial peristaltic tubing. A peristal-tic pump is used to sequentially aspirate sample particlesuspensions from multiwell microplates and insert air bub-bles between individual samples. This results in the gen-eration of a tandem series of bubble-separated samplesthat are delivered directly to the flow cytometer for serialanalysis. The sample size and air bubble size are deter-mined by the time that the autosampler probe is in amicroplate well or above a well taking in air.

The HyperCyt system differs from our previously de-scribed Plug Flow Cytometry system in its ability to usesmaller sample volumes (�3 �l), its delivery to the flowcytometer of a “train” of tandem samples separated by airbubbles, and its direct peristaltic pumping of sample intothe flow cytometer without an intervening coupling valve(4). Because of the unique features of the HyperCyt mech-anism (in particular intersample air bubbles and directperistaltic pumping), it is imperative to evaluate the use ofHyperCyt in bioassays independently of the previous char-acterization of the Plug Flow system. The initial HyperCytstudies documented important features for optimizing op-tical measurement quality and sample carryover (5). In thepresent study, we show that this autosampling methodproduces results comparable to those of conventionalmanual sample analysis in representative flow cytometricbioassays. We also demonstrate that cells can be efficientlysampled and analyzed multiple times from microplatewells of 10-�l volume capacity and that this autosamplingformat enables a novel solution to the problem of cellsettling.

MATERIALS AND METHODSCells and Beads

U937 and K562 cell lines were cultured in RPMI-1640medium supplemented with 10% fetal bovine serum (bothfrom Hyclone, Logan, UT) at 37°C, 5% CO2, in T75 Falcontissue culture flasks (Becton Dickinson, Franklin Lakes,NJ). Cyto-Plex 4-�m-diameter red fluorescent beads wereobtained from Duke Scientific Corp. (Palo Alto, CA). Themanufacturer’s designation for the five sets of beads, listedin order of increasing fluorescence intensity, were L2, L4,L6, L8, and L10.

Eosinophils were prepared from peripheral blood ofhealthy volunteers, as previously described (6), by sepa-

ration of granulocytes on Percoll and hypotonic lysis ofcontaminating erythrocytes. Eosinophils were then puri-fied from the granulocytes on a Vario-Max magnetic neg-ative selection column (Miltenyi Biotec, Auburn, CA) us-ing magnetic microbeads coated with monoclonalantibodies (mAbs) CD16 and CD3 (Miltenyi Biotec) toremove neutrophils and the small number of contaminat-ing T lymphocytes, respectively. Such preparations con-sisted of 95% or more eosinophils. Unfractionated granu-locyte preparations typically consisted of 95% or moreneutrophils and only 1% to 3% eosinophils and thus weredesignated as neutrophils for analysis.

Mononuclear cells were prepared from heparinized pe-ripheral blood of a healthy volunteer that was separatedon a discontinuous gradient of Ficoll-Paque (PharmaciaBiotech 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) andstored on ice before use.

Trypsin Cleavage and Labeling of P-SelectinGlycoprotein Ligand-1 (PSGL-1)

Eosinophils and neutrophils were added to V-wells of96-well microplates at 105 cells/well, washed once with200 �l of phosphate buffered saline (GibcoBRL, GrandIsland, NY), and resuspended in 50 �l of ice-cold phos-phate buffered saline supplemented with trypsin and eth-ylene-diaminetetra-acetic acid (GibcoBRL) to final trypsinconcentrations of 0 to 2.5 mg/ml. After 5 min of incuba-tion on ice, 50 �l of HHB/HSA was added to neutralize thetrypsin. Cells were then pelleted and stained as previouslydescribed (7) with phycoerythrin (PE)–conjugated anti–PSGL-1 (CD162) mAbs PL1 (Ancell, Bayport, MN) andKPL1 (Pharmingen, San Diego, CA) and with G1 mAb(anti-CD62P, Ancell) as an isotype-matched negative con-trol. Cells were then resuspended in 200 �l of HHB/HSAfor flow cytometric analysis.

Immunophenotyping

Mononuclear cells were stained in V-well microplates,as previously described (7), with fluorescein isothiocya-nate (FITC)–and PE-conjugated mAbs directed againstCD3, 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 withPKH67GL (Sigma, St. Louis, MO) according to the manu-facturer’s instructions. In some experiments, K562 cellswere labeled for 15 min at 37°C with 5-,6-carboxyfluores-cein diacetate, succinimidyl ester (CFSE; MolecularProbes, Eugene, OR) at indicated concentrations. Labeledcells were washed, resuspended in HHB/HSA buffer, andadded to wells of Terasaki microplates at 20,000 to 25,000cells in 10 �l/well. Microplates were then incubated in themicroassay rotational suspension system (MARSS), in

56 RAMIREZ ET AL.

which they were rotated between inverted and uprightpositions at approximately 4 rpm for the indicated timeintervals at 24°C before HyperCyt flow cytometric analy-sis. The MARSS consisted of the revolving drum of aperistaltic pump, and the microplates were attached tothe surface of the drum. In some experiments, stationaryupright control microplates were incubated in parallel.

HyperCyt Autosampling and Flow Cytometry

The HyperCyt system for automated sampling of micro-plate wells and delivery to a flow cytometer was describedelsewhere (5) (Fig. 1). Briefly, this system consists of aGilson 215 Liquid Handler or 223 Sample Changer au-tosampler (Middleton, WI) equipped with a GilsonMinipuls 3 peristaltic pump and controlled by a laptopcomputer running a Microsoft Visual C�� software pro-gram, CRFSoft, developed in-house. This approach uses airbubbles to separate samples introduced by the autosam-pler and peristaltic pump into a tubing line that directlyconnects to the flow cytometer. The sample size and airbubble size are determined by the time that the autosam-pler probe is in a microplate well or above a well taking inair. A FACScan flow cytometer (BD Biosciences, San Jose,CA) was used for all experiments. Fluorescence of all dyeswas 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-conjugatedmAbs, 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 forthe first sample in a series to travel from its source well to

the laser beam interrogation point. This was typicallyapproximately 15 s under the standard volumetric flowconditions used in the present studies. Sample analysistime was defined as the average time between samplesduring passage through the laser beam interrogationpoint. This differed with respect to programmed autosam-pling 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 detec-tion of the leading edge of the first sample and the trailingedge of the last sample, respectively, and n is the totalnumber of samples in the series. All estimates of sampleanalysis rates were based on average sample analysistimes.

A computer program, FCSQuery, was developed in-housefor efficiently analyzing the time-resolved data clusters gen-erated by rapid multiwell sampling. Software algorithms au-tomatically identified data clusters representing cells fromindividual sample wells and calculated mean and medianfluorescence intensities and event number for each cluster.Data also were input automatically to Microsoft Excel spread-sheets for more detailed analysis. All analyses were doneoffline on flow cytometry list-mode data files stored in FCS2.0 format.

RESULTSQuantitative Analysis of Cellular Membrane

Protein Expression

We first evaluated the HyperCyt autosampling systemfor its ability to accurately discriminate different levels ofcellular membrane molecule expression. The adhesion

FIG. 1. HyperCyt configuration. Systemcomponents used in the present studies in-clude: (a) a PC laptop computer, Windows95/98 operating system, CRFSoft software(in-house); (b) a Gilson autosampler (223Sample Changer or 215 Liquid Handler); (c) astainless steel sampling probe (0.01 in. innerdiameter, 0.02 in. outer diameter; Small Parts,Inc., Miami Lakes, FL); (d) a 96 V-well(Costar, Cambridge, MA) or 60-well Terasakimicroplate (Nalge Nunc International, Roch-ester, NY); (e) polyvinyl chloride tubing(PVC; 0.01 in. inner diameter; Spectra Hard-ware, Westmoreland City, PA); (f) a GilsonMinipuls 3 peristaltic pump; (g) junctionbetween the PVC tubing and the flow cytom-eter sampling port; (h) FACScan flow cytom-eter (BD Biosciences); (i) Macintosh com-puter, OS9 operating system, Cell Questsoftware (BD Biosciences); and (j) microassayrotational suspension system.

57MICROVOLUME HT FLOW CYTOMETRY

molecule PSGL-1 is a trypsin-sensitive molecule that isexpressed at approximately twofold higher levels on eo-sinophils than on neutrophils (6,8). To produce a widerange of membrane PSGL-1 expression, neutrophils andeosinophils were briefly exposed to different concentra-tions of trypsin. The trypsin reactions were performed inV-bottom wells of a 96-well microplate, and the cellssubsequently were stained immunofluorescently withmAbs that recognize two distinct PSGL-1 epitopes, PL1and KPL1 (9,10). The cells in each well were then ana-lyzed in the flow cytometer by three different methods.First, cells were sampled with the HyperCyt autosamplingsystem with a 0.9-s sample aspiration time per well and a0.9-s sampling probe wash step between wells. Second,the cells were sampled by the HyperCyt system with thewash step omitted. Third, after two rounds of samplingwith each HyperCyt protocol, the remaining contents ofeach well were transferred to separate tubes for conven-tional manual sample analysis.

In representative results from one round of HyperCytno-wash autosampling, 29 wells were analyzed in approx-imately 44 s (Fig. 2, top). Each of the 29 discrete clustersof dots represents the log fluorescence distribution ofindividual cells from one well. Because of the relativelywide dynamic range of PSGL-1 fluorescence intensity,flow cytometric measurements were made on a log scale.The median log fluorescence intensity of each cluster wascalculated and converted to a linear scale value for use insubsequent quantitative analyses (Fig. 2, bottom).

When similar operations were performed on data fromeach of the three sampling methods, the PSGL-1 trypsinhydrolysis profiles obtained from both HyperCyt autosam-pling protocols were similar to each other and to theprofiles obtained by conventional manual sampling (Fig.3A, B). Response linearity was evaluated by plotting man-ual versus HyperCyt median fluorescence intensity results(Fig. 3C). The lines fitted by regression of the original logfluorescence intensity data had slopes of 0.97 � 0.01 and0.95 � 0.01 (mean � standard deviation) for rinse andno-rinse HyperCyt protocols, respectively. No significantdeparture from linearity was detectable in a runs test ofthe data (P � 0.25 and 0.80 for rinse and no-rinse proto-cols, respectively). Response reproducibility was evalu-ated as the coefficient of variation (C.V.) for duplicatemeasurements made in each assay condition. This corre-sponded to a 2.2 � 3.2% C.V. for the rinse protocol and a1.9 � 3.7% C.V. for the no-rinse protocol. Sample car-ryover from the HyperCyt rinse protocol results was esti-mated at 3.8 � 2.2%. Carryover was calculated as 100 �R/(S � R), where R is the number of cells detected duringthe rinse step and S is the number of cells detected in thesample just before the rinse.

These results thus demonstrated a strong linear relationbetween HyperCyt analyses and data obtained by conven-tional manual analysis. Moreover, results of the HyperCytno-wash protocol, in which effects of sample carryovershould have been maximally manifested, were compara-ble to results of the wash protocol. This observation sug-gested that sample carryover between wells is sufficientlylow for this type of quantitative fluorescence analysis so

that a sampling-probe wash step is not required. More-over, the small sample volumes (2.5 �l), the rapid sam-pling rates, and the use of air bubbles to separate samplesin the autosampling protocols did not appear to signifi-cantly compromise the quality of the data.

Multiparameter Immunophenotype Analysis

A second set of studies evaluated the performance ofHyperCyt in a multiparameter fluorescence analysis for-mat. Prototypic of such a format is the conventional im-munofluorescence analysis of leukocyte phenotype, inwhich multiple leukocyte membrane determinants areprobed in parallel by using antibodies conjugated withdistinct fluorophores. Mononuclear cells were stainedwith fluorescent mAbs in 96 V-well microplates and resus-pended in 200 �l of HHB per well for subsequent Hyper-Cyt analysis. The mAbs were used in FITC- and PE-conju-gated pairs to distinguish T lymphocytes from naturalkiller (NK) lymphocytes (CD3-PE, CD16-FITC), T lympho-

FIG. 2. HyperCyt analysis of P-selectin glycoprotein ligand-1 (PSGL-1)adhesion molecule expression after graded cleavage with trypsin. In wellsof a 96-well plate, eosinophils and neutrophils were incubated on ice for5 min with the indicated trypsin concentrations. Cells were stained withfluorescent antibodies directed against PSGL-1 epitopes, PL1 and KPL1,and with G1 antibody as a negative control. Twenty-nine wells wereanalyzed in approximately 44 s by the HyperCyt at a 0.9-s sample aspira-tion time per well and no rinsing between wells (top). Because of thewide range of PSGL-1 expression, measurements were made on a logscale, with each discrete cluster of dots representing individual cells fromone well. The median log fluorescence intensity of each cluster wascalculated and converted to a linear scale value for use in subsequentquantitative analyses (bottom).

58 RAMIREZ ET AL.

cytes from B lymphocytes (CD3-PE, CD19-FITC), and CD4T lymphocytes from CD8 T lymphocytes (CD4-PE, CD3-FITC).

Light scatter distributions characteristic of lymphocytesand monocytes were distinguishable under HyperCyt anal-

ysis conditions, but were shifted to the left on the forwardlight scatter intensity axis as compared with mononuclearcells analyzed under conventional manual conditions (Fig.4, top left panel). This finding was consistent with previ-ously observed effects of HyperCyt analysis on light scat-ter distributions of cells (5). To evaluate lymphocyte sub-set measurement reproducibility, the HyperCyt systemwas programmed to aspirate triplicate samples from eachwell, with an air bubble inserted between each sample. Ina representative experiment (Fig. 4, top right and bottompanels), triplicate samples were taken from each of threewells with an aspiration time of 2 s/sample. The samplingprobe was not washed between samples. The nine totalsamples were resolved as temporally discrete data clusterswhen FITC fluorescence intensity was plotted as a func-tion of time (Fig. 4, top right panel). The two-color fluo-rescence intensity dot plots for each of the nine dataclusters are illustrated in the bottom panel of Figure 4.Lymphocyte subset percentage measurements were gen-erally reproducible within each triplicate group.

Cells from the same donor also were seeded in V-wellsat 1.5 � 106 and 3 � 106 cells/ml and immunostained asbefore, and the aspiration time for each sample waschanged. At sample aspiration times of 0.9, 2, and 5 s, theaverage times required to analyze cells were 1.4 � 0.3,2.5 � 0.3, and 5.5 � 0.3 s/sample (mean � standarddeviation), respectively. The number of cells available foranalysis in each sample ranged from 460 to 4430, depend-ing on the sample aspiration time and number of cellsinitially seeded in each well (Table 1). The lymphocytesubset percentage measurements were remarkably similarover the entire range of experimental variation. Theseresults illustrate how sample size can be significantly in-creased by relatively simple means to enable, e.g., moreaccurate measurements of low-frequency cell subpopula-tions. It is noteworthy in the above set of experiments thata 10-fold increase in sample size was achieved whilesustaining analysis rates in excess of 10 samples/min.

In a previous study (11), the mononuclear cells of thisdonor were rigorously characterized in 30 separate deter-minations over a 19-month period. Edwards et al. detected58 � 4% CD3, 9 � 1% CD20 (which detects most but notall B cells detected by CD19), and 39 � 3% CD3/CD4double-positive lymphocytes. These results correspondedquite well to the 58 � 2% CD3, 13 � 1% CD19, and 37 �1% CD3/CD4-positive lymphocytes detected by HyperCytin the present study (mean � standard deviation of pooleddata from Table 1). Accounting for the relatively highpercentage of CD3/CD19 double-negative lymphocyteswas the presence of a high frequency of NK cells in thisdonor. This subset corresponded to the 19% CD16-posi-tive NK cells (Fig. 4, Table 1) and an additional 10% or soof NK cells detected with CD56 mAbs (data not shown).Another point of interest is that the two CD3 mAb conju-gates detected similar frequencies of T cells despite therelatively 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 likelyreflected the use of a suboptimal mAb preparation be-

FIG. 3. Comparison of HyperCyt with manual flow cytometric analysis.Eosinophils and neutrophils were treated with trypsin and stained as inFigure 2 with antibodies PL1 (A) and KPL1 (B). Cells were suspended in200 �l per well and sampled first by HyperCyt at a 0.9-s sample aspirationtime 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 sameaspiration time per well but no intervening rinse step (squares). Remain-ing cells in each well were then sampled manually (triangles). Resultswere plotted after transformation of log fluorescence data to a linearscale. C: A plot of manual versus HyperCyt measurements for bothmonoclonal antibodies in the original log fluorescence intensity unitsindicated a significant linear correlation for the rinse (circles) and no-rinse(triangles) protocols (r2 � 0.98, P � 0.0001 for each). Results are repre-sentative of two separate experiments.

59MICROVOLUME HT FLOW CYTOMETRY

cause the other FITC-conjugated mAbs produced well-resolved staining.

HyperCyt Analysis inSmall-Volume Microplate Wells

Because the autosampling method worked reliably withaspirated sample volumes of 2.5 �l or smaller, we inves-tigated the use of microplates with smaller wells. Terasakimicroplates have truncated conical wells of approxi-

mately 10-�l volume capacity arranged in a 6-row, 10-column, 60-well format. A 10-�l fluid volume was consid-ered large enough that evaporative fluid loss was unlikelyto be a significant problem under most assay conditionsbut small enough to substantially reduce total amounts ofcells and reagents required for assays.

An additional feature of fluids in microwells of thisgeometric scale is the strong influence of liquid surfacetension forces. Terasaki microplates can be completely

FIG. 4. Immunophenotype analysis withHyperCyt. Peripheral mononuclear cells froma healthy donor were stained with fluores-cent monoclonal antibodies to distinguish B,T, and natural killer (NK) lymphocyte sub-populations. Top, left: Light scatter analysisof cells under manual (top) and HyperCyt(bottom) analysis conditions. Enclosedwithin circles are the putative lymphocytepopulations on which fluorescence analyseswere gated. Top, right: Three wells of a 96-well microplate were sampled in triplicate byHyperCyt, with an aspiration time of 2 s/sam-ple. In post-data acquisition analysis withFCSQuery software, the resulting nine dataclusters were resolved automatically on thehorizontal time axis into separate groups forfluorescence analysis. Each rectangle en-closes a distinct cluster of events represent-ing cells labeled with CD16-FITC plus CD3-PEmAbs (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 ofeach rectangle, the time domain boundaries,were used to gate the fluorescence analysis;the top and bottom rectangular boundariesare visual aids to help distinguish how thedata were partitioned on the time axis. Thesolid bar beneath the horizontal time axisrepresents a 20-s span. Bottom: Dot plots ofFITC and PE fluorescence intensity data foreach of the time domain-gated data clustersA1–3 (top row), B1–3 (middle row), andC1–3 (bottom row), distinguished as de-scribed above. Numbers in dot plot quad-rants represent the percentage of total eventsdetected in the quadrant. Approximately1,530 � 160 (mean � standard deviation)events were analyzed in each dot plot. FITC,fluorescein isothiocyanate; mAbs, monoclo-nal antibodies; PE, phycoerythrin.

60 RAMIREZ ET AL.

inverted without spilling of well contents. We expectedthat this latter feature could be exploited as a novelmeans to avoid the problem of cell settling, a confound-ing factor in flow cytometric automation. To test thisconcept, Terasaki microplates containing 10-�l suspen-sions of U937 cells were attached to the surface of avertically revolving wheel on which they continuouslyorbited the axis of rotation, reciprocating between in-verted and upright orientations, at a rate of approxi-mately 4 rpm. When microplates were incubated up to20 min on this MARSS, the numbers of U937 cellsrecovered from 10 replicate well suspensions by Hyper-Cyt were consistently comparable to initial samplesaspirated at 0 min (Fig. 5). In contrast, cell recoveriesprogressively decreased over 20 min in stationary con-trol microplates (Fig. 5). Thus, the use of small micro-volume wells in combination with periodic plate inver-sion was a simple and practical means for prolongedmaintenance of uniform cell suspensions.

To assess HyperCyt analysis of cell fluorescence inTerasaki microplates, K562 cells were labeled with CFSEat two concentrations (200 nM and 2 �M). The resultingcell preparations were strongly fluorescent in comparisonwith unlabeled cells, falling in the third and fourth de-cades of the log intensity scale, respectively (Fig. 6A).Labeled cells from each preparation were dispensed intoalternating wells in the first five wells of each six-wellmicroplate row. Unlabeled cells (autofluorescence con-trol) were dispensed into the wells at the end of each row.When all 60 wells were sampled by HyperCyt and ana-lyzed within a 90-s span, 60 clearly resolved data clusterswere detected (Fig. 6B). The three levels of fluorescenceintensity were reproducibly detected over the entire spanof the microplate when the cells were sampled immedi-

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

Table 1Effect of Cell Concentration and Sample Aspiration Time

Cell phenotypea Cells/mlTime of

aspirations

No. cellsanalyzedin sample

Cells in quadrant (%)

Upper left Upper right Lower left Lower right

CD16-FITC/CD3-PE 1.5 � 106 0.9 590 60 2 18 202 980 58 6 19 175 2,740 59 2 20 19

3.0 � 106 0.9 940 56 2 22 212 1,940 58 2 22 195 4,290 61 2 19 18

CD19-FITC/CD3-PE 1.5 � 106 0.9 460 58 1 28 142 1,230 60 1 26 135 2,780 61 1 25 13

3.0 � 106 0.9 800 59 1 26 142 1,860 60 1 26 135 3,950 60 1 26 13

CD3-FITC/CD4-PE 1.5 � 106 0.9 590 3 38 40 192 1,310 2 39 39 205 3,170 4 36 40 19

3.0 � 106 0.9 750 4 37 38 212 1,730 5 38 39 195 4,430 5 37 39 20

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

FIG. 5. Use of small-volume microplate wells to enable prolonged sus-pension of cells. Terasaki plates containing 10 replicate 10-�l U937 cellsuspensions were attached to the microassay rotational suspension sys-tem so that the microplates periodically alternated between inverted andupright positions at a rate of approximately 4 rpm. Surface tensionprevented sample loss from inverted wells. For up to 20 min on therotating carousel, cell recoveries were consistently comparable to theinitial samples (circles). In contrast, cell recoveries significantly declinedin control plates that were kept upright and stationary (triangles).

61MICROVOLUME HT FLOW CYTOMETRY

HyperCyt Analysis Over a Broad Range ofParticle Concentrations

To more rigorously assess the effect of input particleconcentration on HyperCyt sampling performance, wellsof Terasaki plates were seeded with mixtures of fluores-cent Cyto-Plex beads ranging in concentration from 7 �105 to 20 � 106 beads/ml. Each mixture contained fivesets 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 of0.2 �l/s), the beads were resolved on a log scale as fivedistinct populations spanning a three-decade range of flu-orescence intensity (Fig. 7A). HyperCyt analysis was per-formed by sampling 12 replicate wells at each bead con-centration. Sample aspiration time was 2 s for each well,with no rinsing of the sample probe between wells, a

FIG. 6. Quantitative fluorescence analysisof cells in Terasaki microplates. A: Manualfluorescence analysis of K562 cells that wereunlabeled (light gray histogram) or labeledwith 200 nM (dark gray histogram) and 2 �M(black histogram) of 5-,6-carboxyfluoresceindiacetate, succinimidyl ester (CFSE). B: La-beled and unlabeled cells were dispensed in10-�l aliquots into a 60-well Terasaki plate,with unlabeled cells in every sixth well andthe two levels of CFSE-labeled cells alternat-ing in the remaining wells. All 60 wells weresampled by HyperCyt and analyzed within a90-s span (solid bar, horizontal time axis).Each discrete cluster of events representscells from a separate well. C: The mediangreen fluorescence intensity of cells fromeach well when the microplate was sampledimmediately after dispensing cells into wells.D: The median green fluorescence intensityof cells sampled from the same microplateafter it was incubated for 30 min in the micro-assay rotational suspension system (MARSS;4 rpm, 24°C). The means � standard devia-tions of the number of cells sampled fromeach well were 844 � 210 in immediatelysampled wells and 938 � 242 in wells sam-pled after 30 min in the MARSS.

62 RAMIREZ ET AL.

protocol that achieved a sampling rate of approximately24 wells/min. The number of beads analyzed in eachsample ranged from 1,000 at the low end of bead concen-tration to 12,000 at the high end (Table 2). Although therewas a broadening of the fluorescence intensity distribu-tions of the beads under these conditions of HyperCytanalysis (Fig. 7B), the median fluorescence intensities of

each bead set under each condition of bead concentrationshowed a linear correlation with measurements mademanually in parallel (Fig. 7C, Table 2). Likewise, the per-centages of beads detected in each population peak werenearly identical under manual and all HyperCyt analysisconditions.

DISCUSSIONIn the initial characterization of the HyperCyt system

(5), we demonstrated sampling of beads and cells from96-well plates at rates of up to 100 samples/min andcarryover between samples of 1% to 3%. The presentstudies extended these findings to validate the HyperCytautosampling system as a high-throughput method for celland particle fluorescence measurements of comparablequality to conventional manual flow cytometry. We dem-onstrated the ability of HyperCyt to accurately measurevariations in cell and bead immunofluorescence over afour-decade range of fluorescence intensity and at inputparticle concentrations of up to 20 � 106/ml. This wasaccomplished at analysis rates of 24 to 40 samples/min(1.5 to 2.5 s/sample) and the use of source well samplevolumes as small as 8 �l. The HyperCyt system also per-mitted reproducible quantification of lymphocyte sub-populations when multiparameter immunophenotypingassays were performed in 96-well microplates.

An important feature of HyperCyt autosampling is therequirement for minimal sample volumes. In the presentstudies, the sampling probe was typically programmed toaspirate from each well for approximately 1 s. This re-sulted in samples averaging approximately 2.5 �l andsample analysis rates averaging 40 samples/min. Even atfivefold longer aspiration times, implemented to increasethe total cell counts in each sample (Table 1), the totalsample volume removed from each well was muchsmaller 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 cellcounts in each aspirated sample is to increase cell con-centrations in source wells (Tables 1 and 2). At higher cellconcentrations, smaller sample volumes are required toachieve a targeted cell count range, and sample through-put 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 beadswith discrete levels of red fluorescent dye. The final concentration of eachset 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 intensitypeaks. B: Alternatively, 10-�l aliquots of the 20 � 106 beads/ml stockwere added to each of 12 replicate wells of a Terasaki plate (2 � 105

beads/well) and analyzed by HyperCyt with an aspiration time of 2s/sample and a peristaltic pump rate of 15 rpm. The illustrated fluores-cence intensity profile represents pooled data from all 12 samples. Statis-tics for intersample variation are presented in Table 2. C: Median logfluorescence intensity data are plotted for each bead set as obtained in themanual and HyperCyt analysis protocols illustrated in A and B. The fittedline had a slope of 0.997 and a correlation coefficient (r2) of 1.0. Similarlinear correlations were obtained when HyperCyt analysis results frommore dilute input concentrations of beads were compared with theoptimal manual analysis results (Table 2).

63MICROVOLUME HT FLOW CYTOMETRY

this approach is the ability of the flow cytometer’s detec-tion electronics to efficiently detect and process all eventsas they become more closely spaced in time at progres-sively higher cell concentrations. This is evident in datashown in Table 2, in which threefold increases in inputparticle concentrations resulted in only approximately2.3-fold average increases in the numbers of detectedevents. Despite this apparent loss of data, presumably dueto detection circuit “dead time” during processing of in-dividual events, data from successfully processed eventswere of high quality. This was indicated by the tight linearrelation between data obtained manually under optimalconditions of sample flow and data from HyperCyt analy-sis performed even at the highest tested input sampleconcentration of 20 � 106 particles/ml (Fig. 7C). In a flowcytometer with faster event-processing capabilities, it isexpected that a desired sample event number endpointshould be attained with lower input particle concentra-tions and/or shorter sample aspiration times. Thus, im-proving event-processing efficiency represents one ap-proach by which it should be possible to minimize samplequantity requirements and maximize sample throughput.It is also worth noting that, with multiplexing of bead-based assays on multilevel fluorescent bead sets, each setproviding a platform for a discrete assay may enable anadditional 5- to 10-fold increase in assay throughput.

Because of the reduced sample volume requirementwith HyperCyt, it was feasible to perform flow cytometricassays on cells seeded in 10-�l Terasaki plate wells. It isexpected that this microwell format will enable quantita-tive multiparameter flow cytometric measurements to bemade rapidly on small numbers of cells so as to maximizethe efficient use of, e.g., limited numbers of cells fromclinical specimens. Likewise, quantity-limited reagents

could be used in smaller amounts to achieve otherwiseprohibitively high concentrations.

The small well volume also enabled a solution to theproblem of particle settling. Capillary forces due to liquidsurface tension are sufficient to prevent sample loss frominverted wells, so that uniform cell suspensions are main-tained for extended periods by periodically inverting mi-croplates in the MARSS. The present study demonstratedthat suspensions can be maintained over a span of 30 min,but practical upper limits remain to be defined. Our cellsettling data (Fig. 5) suggested that optimal results areobtained by analyzing wells in packets that can be pro-cessed within 1 to 2 min. The 60-well Terasaki microplaterepresents a convenient packet of wells commensuratewith this requirement. We anticipate that, when largenumbers of assays are to be performed, it will be practicalto prepare multiple microplates that then can be incu-bated on the MARSS while awaiting subsequent sequentialHyperCyt processing.

Routine practical application of this high-throughputflow cytometric analysis technology will require addi-tional considerations that we are now beginning to ad-dress. One such issue is synchronization between theautosampler and the flow cytometer so that analyzed sam-ples can be unambiguously assigned to their source wells.Misregistration between the sequence of sampled sourcewells and the resulting sequence of time-resolved eventclusters might occur, for example, if a sample is missed(e.g., empty well), merges with another sample, or subdi-vides during transit. A general solution to detecting suchoccurrences is to develop coding systems by whichproper registration is reestablished at fixed sampling in-tervals. A simple example is illustrated in Figure 6B, inwhich the negative control sample of unlabeled cells is

Table 2Comparison of HyperCyt and Manual Analysis of Cyto-Plex Beads*

HyperCytManual(high)

Manual(low)

Input beads/ml (�106) 20 6.7 2.2 0.7 1.1 1.1Beads detected/sample 12,000 � 350 6,000 � 490 2,200 � 250 1,000 � 160 1,100 � 90 390 � 70% Totala

Peak 1 17 � 0.4 17 � 0.6 17 � 1.0 17 � 1.2 17 � 0.9 16 � 1.9Peak 2 21 � 0.3 22 � 0.7 22 � 0.5 22 � 0.3 21 � 1.4 21 � 3.6Peak 3 23 � 0.5 23 � 0.6 23 � 0.5 23 � 1.2 23 � 1.7 24 � 2.0Peak 4 19 � 0.3 19 � 0.5 18 � 0.7 19 � 0.8 19 � 1.2 19 � 2.6Peak 5 20 � 0.3 20 � 0.8 19 � 0.8 19 � 1.6 20 � 1.2 20 � 1.9

MFIb

Peak 1 240 � 1 240 � 1 241 � 1 241 � 2 243 � 1 245 � 1Peak 2 403 � 1 402 � 1 402 � 1 403 � 2 405 � 1 406 � 1Peak 3 544 � 1 542 � 1 542 � 1 543 � 2 546 � 1 548 � 1Peak 4 675 � 1 672 � 2 673 � 2 672 � 2 678 � 1 679 � 1Peak 5 772 � 1 770 � 1 770 � 1 770 � 1 773 � 1 775 � 1

*Cyto-Plex beads were added to Terasaki plate wells at the indicated total input bead concentrations and sampled with the HyperCytsystem or added to a tube at 1.1 � 106 beads/ml and sampled manually at the high and low FACScan sample flow rate settings (nominally1 and 0.2 �l/s, respectively). Twelve replicate wells were sampled to obtain a mean and standard deviation for each HyperCytmeasurement. The continuous sample data stream from each manual analysis was divided into 12 sequential 2.5-s segments (the averagetime span of HyperCyt samples) to obtain mean and standard deviation estimates for each manual measurement.

aPercentage of total beads (mean � standard deviation) detected in each discrete bead fluorescence intensity population; peaks arenumbered in order of increasing intensity.

bMedian fluorescence intensity (mean � standard deviation, log scale units) of each bead population.

64 RAMIREZ ET AL.

always added to the last well of each six-well row. Offlineanalysis software can easily detect control well data, countthe number of event clusters in between, and flag forreanalysis any row in which the count differs from five.More sophisticated coding schemes can easily be envi-sioned that involve markers such as multilevel fluorescentbeads mixed in different proportions to denote multiplediscrete levels of source well/event cluster registration.We recently developed a system involving fluorescentmarker beads and custom bead detection circuitry withwhich a first level of online synchronization between theautosampler and the flow cytometer has been achieved(manuscript in preparation).

A second issue in the routine practical application ofHyperCyt is the implementation of a data management sys-tem to facilitate the storage, retrieval, and quality control oflarge volumes of rapidly accumulating multiparameter data.This will be of concern especially in extended screeningactivities involving the analysis of thousands to tens of thou-sands of samples per day. Fortunately, there are sophisti-cated commercial data management software packages thatare used routinely for managing high-throughput screeningdata from a diversity of analytical instruments. Of particularinterest are packages capable of importing data from spread-sheet programs (e.g., IDBS ActivityBase) as we have alreadyimplemented the output of formatted time-resolved data toMicrosoft Excel spreadsheets as a standard feature of ourin-house FCSQuery software.

ACKNOWLEDGMENTSThis study was performed with technical and instru-

ment support from the University of New Mexico Shared

Flow Cytometry Resource, University of New MexicoHealth Science Center, and CRTC, Albuquerque, NM.

LITERATURE CITED

1. Edwards BS, Kuckuck F, Sklar LA. Plug flow cytometry: an automatedcoupling 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 adhesionand receptor pharmacology. Cytometry 2001;43:211:216.

3. Ransom JT, Edwards BS, Kuckuck F, Okun A, Mattox DK, ProssnitzER, Sklar LA. Flow cytometry systems for drug discovery and devel-opment. SPIE Proc 2000;3921:90–100.

4. Edwards BS, Kuckuck FW, Prossnitz ER, Ransom JT, Sklar LA. HTPSFlow cytometry: a novel platform for automated high throughputdrug 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 recruit-ment of eosinophils over neutrophils. J Immunol 2000;165:404–410.

7. Edwards BS, Shopp GM. Efficient use of monoclonal antibodies forimmunofluorescence. Cytometry 1989;10:94–97.

8. Symon FA, Lawrence MB, Williamson ML, Walsh GM, Watson SR,Wardlaw AJ. Functional and structural characterization of the eosin-ophil 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 extendedmolecule and mapping of protein epitopes for monoclonal antibod-ies. J Biol Chem 1996;271:6342–6348.

10. Snapp KR, Ding H, Atkins K, Warnke R, Luscinskas FW, Kansas GS. Anovel P-selectin glycoprotein ligand-1 monoclonal antibody recog-nizes an epitope within the tyrosine sulfate motif of human PSGL-1and blocks recognition of both P- and L-selectin. Blood 1998;91:154–164.

11. Edwards BS, Altobelli KK, Nolla HA, Harper DJ, Hoffman RR. Acomprehensive quality assessment approach for flow cytometric im-munophenotyping of human lymphocytes. Cytometry, 1989;10:433–441.

65MICROVOLUME HT FLOW CYTOMETRY