Development and application of a multiplexable flow cytometry‐based assay to quantify cell‐mediated cytolysis

Although target cell cytolysis has been widely employed to describe effector function of cells, cytolysis assays as commonly employed do not generate quantitative data. In this report we describe the development and application of a statistically supported flow cytometry‐based assay to quantify cell‐mediated cytolysis. The assay depends on the use of the fluorescent dye CFSE to distinguish target from effector cells, the DNA intercalating dye 7AAD to distinguish dead from live cell events, and on the establishment of a cytolysis curve that allows for the derivation of statistically robust data. We demonstrate that the cytolysis curve is well described by a four parameter logistic regression model provided that (i) the range of effector to target (E:T) ratios studied allows for full description of the logistic curve, and (ii) an adequate number of data points are collected to estimate the model parameters. We show that the assay is highly reproducible and accurate, and comparable in sensitivity with the standard 51Cr assay. We report on the potential for this assay to generate quantitative data on the cytolytic activity of both CD8 T and NK cells; describe a relationship between the efficiency of effector cell degranulation and target cell cytolysis throughout a range of E:T ratios, and demonstrate the potential to multiplex with other platforms to obtain broader datasets for the effector phenotype of cells. Appropriate use of this assay will enhance the ability to derive quantitative and integrated correlative datasets from basic, translational, and clinical studies. © 2010 International Society for Advancement of Cytometry

Over the past few years a number of assays have been proposed as alternatives to the 51 Cr release assay (2À9). In many cases these assays have involved the use of flow-cytometrybased approaches and fluorescent dyes to distinguish target from effector cells and dead from live target cells. Primarily using NK model systems, such assays have been shown to be specific, reproducible, and to generate datasets that compare favorably with the 51 Cr release assay (3,5,6,8À14). Importantly, in both principle and practice flow-based cytolysis platforms allow for the potential to simultaneously evaluate, on both target and effector cells, other phenotypic and effector parameters such as activation status and cytokine production by effector cells, as well as phenotype and apoptotic status of targets (15,16).
One issue that has not been adequately addressed to-date is the need to establish statistical underpinning for cytolysis assays. Since cytolysis curves are nonlinear in nature and individual cytolysis curves most often have different kinetics, quantitative comparisons of the specific cytolysis values for different E:T combinations within and between experiments are difficult to perform. Datasets across experiments are additionally compromised by the variability inherent in biological assays as well as the lack of standardization for assay set-up. Thus, the absence of statistical underpinnings for cytolysis assays has resulted in datasets that are at best, semi-quantitative. These limitations are particularly problematic for the evaluation of clinical trial-related specimens where quantitative data on modulation of effector function may be critical to be able to guide further product development.
In this report we describe the development of a statistically supported flow-cytometry-based assay to quantify target cell cytolysis. The assay is based on the previously described Guava cytolysis assay (Guava Technologies, Hayward, CA), which we adapted for use on standard flow cytometers. This assay uses the fluorescent dye CFSE (carboxy fluorescein diacetate, succinimidyl ester) to distinguish target cells from effector cells and the DNA binding dye 7AAD (7 amino-actinomycin D) to discriminate live from dead cells. We show that the cytolysis curve is well described by a four parameter logistic regression model which requires a minimum number of data point to generate statistically robust model parameter estimates. Additionally, we show that the assay is reproducible, accurate, comparable to the 51 Cr-based assay with regard to sensitivity, and applicable to a variety of E:T combinations. Importantly, we show that application of the statistical supported cytolysis assay allows for the generation of numerical data that allow for the quantitative assessment of cytotoxicity in biological samples. Finally, we provide examples that highlight an important advantage of this assay by demonstrating the potential to multiplex with other assays and simultaneously obtain information about the phenotype and function of the effector cells.

MATERIALS AND METHODS
The Clinical Immunobiology Correlative Studies Laboratory operates under principles of Good Laboratory Practice (GLP). All instrumentation is maintained and operated under established Standard Operating Protocol (SOP) by qualified personnel. All laboratory procedures were performed by trained and qualified personnel under established SOP and using qualified assays.

Statistical Methods
The full dose response curve for antigen-specific killing by an effector cell population of antigen-positive and -negative target cells follows an S shape which is well described by a four parameter logistic regression model (17), (Fig. 1). Two elements are critical for appropriate implementation of the four parameter logistic model: (i) the range of E:T ratios used must allow for full description of the logistic curve, and (ii) sufficient data points must be collected across this range to estimate the model parameters.
The parameters of relevance for this model are the upper asymptote of the percent specific lysis curve which represents the maximum percent specific lysis, the lower asymptote of percent specific lysis which represents the minimum percent specific lysis, the ED50 which represents the E:T ratio at the midpoint or inflection point between the upper and lower asymptotes for percent specific lysis, and the Hill slope, which represents the change in percent specific lysis for a unit change in dilution of effector cells at the inflection point. Although the ED50 value is by definition the point in the curve with lowest variance (i.e., the most sensitive part of the curve), in principle, depending on the specific assay limitations and the availability of appropriate software, lower and higher ED values can be calculated and used to quantify cytolytic activity in cultures. For most experimental situations the four parame- ter logistic regression model is the preferred choice; however, in experimental situations where the min and max values are known to be fixed, a lower parameter logistic model could be applied if necessary. To allow for assumption of equal variances throughout the curves we first transformed the ratio value for the dilution of effector cells relative to a target cell dilution factor of 1 [E(x):T(1)] to a log e scale and then implemented a model which estimates the log e ED50 values to allow for comparisons of this parameter for two samples on the transformed scale.
To estimate the numeric values for the parameters we exploited nonlinear least squares implemented through the software package GraphPad Prism (version 5.0.1 for Windows, GraphPad Software, San Diego, CA, www.graphpad.com). The estimated numeric values for these four parameters serve to describe the specific lytic activity of an effector cell population. Comparison of the lytic activity across cell populations can be made by assessing if the 95% confidence intervals based on a Student's t distribution for comparable parameter estimates (i.e., the Hill slope for two cell populations) overlap; non-verlapping confidence intervals indicate a statistically significant difference.

Cell Lines and Cell Culture
UPN035 is an oligoclonal population of primary CD31 CD81 T cells transfected to express a chimeric ScFv immunoreceptor that directs specific cytolysis of CD191 target cells. UPN035 is a clinical product and was generated by successive in-vitro expansions using anti-CD3 mAb (OKT3) and rhIL-2, essentially as per (18). UPN035 is [95% CD31/CD81, essentially 100% of cells express the chimeric immunoreceptor on the surface, and the chimeric immunoreceptor directs specific cytolysis of CD19-positive targets. For these studies UPN035 was used under an Institutional Review Board (IRB) approved protocol (COH IRB#08027), was additionally expanded for one cycle as described above with the addition of 10 ng ml 21 rhIL-15, and evaluated either fresh or after freezing and subsequent thaw according to established laboratory SOP.
PBMC from patients and aphaeresis discard samples were obtained according to institutional guidelines and under IRB approved protocols (COH IRB#04152 and COH IRB#05096, respectively). For patient samples, whole blood was collected in lavender top tubes containing potassium EDTA (Becton Dickinson #365974). PBMC samples were processed using Accuspin columns (Sigma), and frozen according to established laboratory SOP.
NS0, a mouse B cell myeloma cell line (ATCC# CRL-2695) and NS0/CD19 (NS0 transfected to stably express a membranetruncated CD19 gene product in [95% of cells), were generous gifts from the Jensen laboratory (COH), and were grown using standard laboratory methodologies in high glucose DMEM medium supplemented with 10% fetal calf serum.
Cytotoxicity Assays Cell preparation. Effector cells were used either fresh from culture or thawed from frozen stock and washed; in the case of PBMC, cultures were used immediately after thaw or rested overnight in rhIL-2 as indicated. Target cells were obtained from cultures maintained at a minimum of 85% viability. Cell counts and viability were assessed using a GuavaPcA96 system (Guava Technologies).
Flow-based cytotolysis assay (FBCA). A detailed SOP that describes this assay is included in the Supporting Information for this manuscript. Briefly, effector cells were harvested, counted, washed, and resuspended to 1 3 10 6 cells ml 21 in Tcell media. Effectors and CFSE-labeled target cells were mixed at a range of E:T, either in sterile 96-well round bottom plates (Corning, Lowell, MA) plating 1 3 10 4 targets/well and variable numbers of effector cells, with triplicate wells/condition, or in sterile FACS tubes (BD Falcon) which contained 1.5 3 10 4 targets/tube and variable numbers of effector cells. Cultures were incubated for 4 h at 378C under 5% CO 2 . The 7-AAD was then added to samples and 1% Saponin to the max samples, and cultures were incubated for 30 min in the dark, washed, and resuspended in 900 ll FACS buffer (PBS, 0.5% BSA, 0.006% Sodium Azide); 100 ll of beads (1 3 10 6 ml 21 ), (Bangs Laboratories, Fishers, IN) were added to each sample.
Data acquisition and analysis. After staining, samples were placed on ice and data collected immediately on an FC500 flow cytometer (Beckman Coulter), with each collection acquiring 20,000 bead events ($ 20% of each sample). Data were analyzed using FCS Express V3.0 software. The gating strategy is described in the SOP included in the Supporting Information, and a graphic description of the gating strategies is presented in Supporting Information Figure 1. Because cytolysis can destroy cell architecture cytolysis, % live cells were recorded for each E:T and % cytolysis was calculated as 1 2 % live cells. Since detergent lysis destroys cell architecture, the max % cytolysis value was established as the maximum possible cytolysis, determined by: 1 2 [(live cell events max / total number cells min ) 3 100].

CD107 Degranulation Assay
The CD107 degranulation/mobilization assay was performed as previously described using only the CD107a antibody (9), either using aliquots from the cytolysis assays or using frozen PBMC rested overnight. For surface marker staining following degranulation, after the 4-h incubation period for the cytotoxicity assay cells were washed twice with 2 ml staining buffer and stained with mAbs (CD3 and CD8b or CD56) for 30 min at room temperature in the dark, and washed again before flow analysis. Samples were acquired on a FC500 (Beckman Coulter). A total 100,000 cell events were acquired and samples were analyzed using FCS Express v3.0 software. The analysis was performed on gated cells that fell within the lymphocyte population with subsequent gating on the CD3 1 or CD8b 1 populations.

Luminex Bead Array Cytokine Analysis
Following the 4-h incubation period for cytotoxicity assays, 50 ll supernatant were collected from each condition for Luminex-based array cytokine analysis. A Bio-Plex Luminex 100 XYP instrument was used for data acquisition and Bio-Plex Manager 5.0 software (Bio-Rad Laboratories, Hercules, CA) was used for analysis. Cytokine muliplex analysis was performed using a human cytokine 30-Plex antibody bead kits (Invitrogen, Camarillo, CA) as per the manufacturer's protocol. A standard curve cut-off value at 70% observed/expected values was employed. With the exception of IL1RA (48 pg ml 21 ), MCP-1 (52 pg ml 21 ), and IL-7 (18 pg ml 21 ), the lower limit of quantification for all cytokines was less than 11 pg ml 21 . Each sample was evaluated in duplicate and the average value determined; % CV in all cases was less than 16.7%.

Evaluation of the FBCA Curve
Cytolysis curves have been shown to conform to a four parameter logistic regression model (17). Two elements are critical for appropriate implementation of the four parameter logistic model, (i) the range of E:T ratios used must allow for full description of the logistic curve, and (ii) sufficient data points must be collected across this range to estimate the model parameters.
To evaluate the impact of reducing the number of data points in the FBCA assay, we generated a dataset that described the full S-shaped curve using UPN035 effector cells and NS0/CD19 target cells at 10 E:T ratios ( Fig. 1) and used that data to calculate the parameter estimates. We then calculated the relative variance inflation of the parameter estimates when selected data points were removed from the upper, lower, or middle portions of the curve. The formula for relative variance inflation is: VIF reduced schedule /VIF full schedule where (VIF) 5 (X'X) 21 and X is the gradient or 1st derivative matrix for the four parameter logistic model based on the parameter estimates from the full dataset (19). VIF is the component of the variance of the parameter estimates that is determined by the number and placement of the data points used to generate the curve. Table 1 reports the relative variance inflation for each of the studied sampling schedules derived from the experiment presented in Figure 1. Since the relative variance inflation is a ratio of the VIF for the dataset studied versus the full dataset, when the full set (all 10 points) are used to generate the cytolysis curve the relative variance inflation for the resulting estimates is 1 (Row 1). The estimates of VIF, and relative variance inflation were calculated using the R:nlme package (20,21). Rows 2 through 4 show the consequences of removing points from the top and bottom of the curve on the relative variance inflation for the calculated log e ED50 and Hill slope values, with 1, 2, and 3 points removed at each end for Rows 2, 3, and 4, respectively. In each of these cases, log e ED50 values continue to be estimated with relative variance inflation near 1. However, the relative variance inflation for estimating Hill slope values is significantly impacted, with the variance increasing by 20% (1.2 vs. 1) and 80% (1.8 vs. 1), respectively. Rows 5 through 7 show the consequences of removing points from the middle part of the curve on the relative variance inflation for the calculated log e ED50 and Hill slope values, with 2, 4, and 6 points removed from the middle por- Relative efficiency was determined by comparing the Variance Inflation Factor (VIF) described by Gabrielsson and Weiner (19). The formula for relative variance inflation5 VIF reduced model /VIF full model where (VIF) 5 (X'X) 21 and X is the gradient or 1st derivative matrix. For Rows 2À4, 5À6, and 8À9, data points removed from the subsequent row are highlighted in bold.

ORIGINAL ARTICLE
tion of Rows 5, 6, and 7, respectively. In each of these cases, the ability to estimate log e ED50 values is dramatically impacted, with relative variance inflation values starting at 150%). The ability to estimate Hill slope values is also significantly impacted, particularly when more than two central points are removed. Rows 8 and 9 show the consequences of removing intermittent points from the curve on the relative variance inflation for the calculated log e ED50 and Hill slope values, with 2 and 4 points removed for Rows 8 and 9, respectively. When two intermittent points are removed from this analysis, both log e ED50 and Hill slope values are calculated with efficiencies near 1 (Row 8). When four intermittent points are removed both log e ED50 and Hill slope values are significantly impacted (variance values increasing 40% (1.4 vs. 1) and 70% (1.7 vs. 1), respectively (Row 9). Relative variance inflation values for the lower and upper curve parameters were significantly impacted in most cases when more than two data points were removed (Rows 3, 4, 6, 7, and 9).

Specificity of FBCA
To evaluate the ability to use the FBCA to measure antigen-specific cytotoxicity we performed a series of experiments using UPN035 as effector cells and a series of CD19-expressing and nonexpressing target cells. Figure 2A presents the cytotoxic activity of UPN035 cells against Daudi (CD191) (closed squares) and Jurkat (CD19-) targets (open squares), and Figure 2B presents the specific cytotoxicity of UPN035 against a matching pair of NS0 targets, expressing the CD19 antigen (NS0/CD19, closed diamonds), and lacking CD19 (NS0, open diamonds). As assessed by flow cytometry, both target cell populations expressed the CD19 antigen on [95% of cells and at comparably high levels (not shown). In both cases specific cytotoxicity was restricted to target cells that expressed the CD19 molecule. The ability to specifically detect cytolysis of antigen-positive targets was confirmed by evaluating T cells with different antigenic specificities and corresponding target cells, including other CAR-re-directed T cells; T cells that recognized targets through canonical TcR binding to MHC class I:peptide, as well as ADCC mediated by bulk PBMC (data not shown).

Reproducibility of FBCA
To evaluate the reproducibility of the FBCA we performed a series of experiments using as effectors UPN035 cells expanded by OKT3 stimulation in the presence of IL-2 and IL-15, and frozen 23-days poststimulation. Effector cells were independently tested in three assays over a course of 6 weeks against NS0/CD19 and NS0 targets. Figure 3 presents the average percent specific cytolysis for the three experiments. For cytolysis of the NS0/CD19 targets, the maximal Standard Deviation (SD) was12.3% (for ratio 0.78:1 E:T), and for most data point was less than 8%. Background cytolysis of NS0 targets was minimal and highly reproducible at all E:T ratios. These results demonstrate that when utilized appropriately this assay shows high reproducibility.

Accuracy of FBCA
To evaluate how accurately the FBCA could quantify antigen-specific cytolytic activity UPN035 effector cells were titrated into allogeneic PBMC at two different ratios and cytolysis of NS0-CD19 and NS0 parental target cells was measured. For these experiments, UPN035 effector cells were titrated to comprise either 50 or 10% of the effector cell population. Representative results from these experiments are presented in Figure 4. A difference in CD19-specific cytolysis curves could be observed qualitatively for the 50 and 10% UPN035 curves, with the curve generated using the 50% UPN035 effector cell population showing a higher specific cytolysis essentially throughout the titration range; very low cytolysis could be detected against the NS0 parental targets by the 50% UPN035 effectors (closed triangles) or the 10% UPN035 effectors (not shown). Use of the four parameter logistic regression model allowed for a quantitative measurement of the cytotoxicity for both the effector cell populations. Although an additional dilution point would have been optimal to define the lower parts of these cytolysis curves, as presented in Table 2 the obtained data points were adequate to sufficiently estimate the logED50 and Hill slope values. As is also shown in Table 2, although the Hill slope values for the two curves (a measure of effector cell potency) were statistically indistinguishable, comparison of the log e ED50 values for the two specific cytolysis curves demonstrated a 4.3-fold higher activity for the 50% UPN035 effector cell population compared to the 10% UPN035 effector cell population. This 4.3-fold difference is within 14% of the actual five-fold difference in the specific effector cell composition of the two populations and demonstrate that the data from FBCA analyzed using nonlinear least squares can be used to accurately quantify specific effector cell activity in cell populations.

Comparison of FBCA and 51 Cr-Based Cytolysis Assays
To directly compare the results obtained using the FBCA with a standard 51 Cr-based assay we used the same populations of effector cells (UPN035 frozen at Day 23 postexpansion and freshly thawed), and the same culture of target cells (NS0/ CD19 and NS0 parental) either labeled overnight with 51 Cr or labeled with CFSE. As shown in Figure 5, comparable specific cytolysis curves were obtained for the two methodologies using both antigen-positive (CFSE assay closed squares, 51 Cr assay closed triangles) and negative target cells (CFSE assay open squares, 51 Cr assay open triangles). As shown in Table 3, for all four parameter estimates the 95% confidence intervals overlap, indicating the models for the two methods are not statistically different. The standard errors for the parameter estimates from four parameter logistic regression model were notably higher for the 51 Cr assay, a phenomenon we consistently observed over multiple experiments using different effector and target combinations. These results demonstrate that compared to the standard 51 Cr assay, the FBCA demonstrates similar sensitivity and lower variability across the titration curve.

Quantification of NK Cytolytic Activity Directly in Patient PBMC
We next sought to evaluate the potential to apply the FBCA to quantify NK cytolytic activity directly ex vivo in PBMC samples by measuring cytolysis of the NK-sensitive target cell line K562. For these analyses we obtained under IRB approval whole blood samples from a renal cell carcinoma Figure 3. Reproducibility of the FBCA. Frozen UPN035 effector cells were thawed, washed, and evaluated in three independent assays for cytotoxic activity against CFSE-painted NS0 transfected to express the CD19 gene product: ÀnÀ and CFSE-painted NS0 parental À&À target cells. Average cytolysis and standard deviations at each E:T are shown. Standard deviation values ranged from 6.9 to 12.3%. Figure 4. Accuracy of FBCA. Frozen UPN035 cells were thawed, washed, titrated into bulk PBMC and evaluated for specific cytolysis of CFSE-painted NS0 target cells transfected to express the CD19 gene product. 50% UPN035: ÀnÀ; 10% UPN035: À&À. Cytolytic activity of the 50% UPN035 culture against CFSE-painted NS0 parental targets: À~À. Data are representative of three independent assays. Estimated values, the standard error (S.E.) and the 95% confidence interval (95% C.I.) are based on a t-distribution test for each parameter and cell population.

ORIGINAL ARTICLE
patient undergoing standard care high dose IL-2 therapy, preor 14 days post IL-2 treatment; in each case, PBMC were processed immediately and stored in liquid nitrogen. Prior to the cytolysis assay, samples were thawed and the percentage of NK cells in the samples was determined by flow cytometry by recording the percentage of CD32/CD561 cells in each sample. Bulk PBMC (effectors) were mixed with K562 target cells at E:T ratios ranging from 250:1 to 0.5:1 and a FBCA was performed. Specific cytolysis of K562 cells could be detected in both the pre and post IL-2 treatment samples (Fig. 6). As shown in Figure 6A, higher specific cytolysis was observed in the post-IL2 treatment samples; however, as shown in Figure  6B, when normalizing for the percentage of NK cells in each sample (6.5% pre-and 17% post-IL2 administration) the cytolytic activity of the two samples across the cytolysis curves was essentially identical. Although limitations in the available volume of clinical samples obtained for this study precluded the ability to generate an upper asymptote for specific cytolysis and therefore a full S-shaped cytolysis curve to generate logED50 values in this analysis, the essentially absolute overlap of the two curves throughout the range which would in all probability include the logED50 value demonstrates that the NK-specific cytolytic activity in the pre-and post-IL-2 treatment samples are indistinguishable. Similar results were obtained in vitro using healthy donor PBMC incubated overnight with ''low'' (10 IU) or ''high'' (100 IU) rhIL-2 (data not shown).

Tracking CD107 Degranulation and Cytolysis Kinetics
One of the advantages of the FBCA is the potential to multiplex this assay with other platforms and assays, thus providing a more comprehensive clinical correlative dataset. Cytolysis of target cells is associated with the release of cytotoxic granules which contain perforin and granzyme B by effector cells. The release of the cytotoxic granules can be tracked by accumulation of LAMP-1 and -2 proteins (CD107a and b) on the surface of effector cells (9). The flow-cytometrybased nature of this assay allowed us to quantify both target cell cytolysis and the efficiency of the degranulation by effector cells. For these experiments we examined the degranulation properties of effector subsets in PBMC incubated overnight in 100 U ml 21 IL-2 and then cocultured with NK-sensitive K562 cells; as shown in Figure 7, top panel, cytolytic activity in this assay increased with increasing E:T ratio. Because of detection limitations with the instrumentation used for these analyses, it was necessary for us to perform these analyses in replicate tubes; to this end, a portion of the cytolysis cultures was removed at the start of the cytolysis assay and subjected to a CD107 degranulation assay, followed by surface staining for CD56 and CD3 to identify NK cells (CD32/CD561), NK T cells and activated T cells (CD31/CD561), and CD3 lymphocytes (CD562/CD31), Figure 7, middle panel. As shown in Figure 7, bottom panels (AÀC) using the 6.25:1 E:T as a representative example, the lytic activity in the PBMC cultures is a result of specific degranulation of the NK cell population (CD561/CD3), while the NK T cell and CD3 lymphocyte subsets do not demonstrate specific degranulation in response to K562 targets.
In the course of evaluating the degranulation properties of effector subsets, we observed that the percentage of degranulating effector cells decreased as the E:T ratios of cultures increased. To examine this issue in more detail we tracked the degranulation kinetics of effector cells in response to targets over a broad range of E:T ratios using as effector cells UPN035 and as targets NS0/CD19 and NS0 cells. As can be seen in Fig-Figure 5. Comparison of specific cytolysis measured by FBCA and 51 Cr-based cytolysis assays. Target cells [NS0 transfected to express the CD19 gene product (NSO/CD19) and NS0 parental] were either labeled overnight with 51 Cr or cultured overnight and painted with CFSE for 30 min prior to mixing with effector cells (freshly thawed UPN035 cells) at the indicated E:T ratios; cultures were incubated for 4 h at 378C and specific cytolysis was determined. CFSE-based cytolysis: UPN035 vs. NS0/CD19: ÀnÀ; UPN035 vs. NS0: À&À. 51 Cr-based cytolysis: UPN035 vs. NS0/ CD19: À~À; UPN035 vs. NS0: À~À. Data are representative of three independent experiments. ure 8A, the efficiency of effector cell degranulation is strongly dependent on the E:T ratio. Maximum efficiency of degranulation was observed at E:T ratios below 0.16:1. The efficiency of degranulation was very sensitive to E:T ratios from 0.3:1 up through 2.5:1 (reflecting the linear nature of the cytolysis curve at those ratios), and at higher E:T the efficiency of degranulation was substantially reduced. A subset of the degranulation of effector cells is additionally presented in the form of flow cytometry dot plot data in Figure 8B. These results suggest that optimal effector function is likely to occur under conditions where target cells are in significant excess; this observation, not possible using 51 Cr assays, should be taken into account when using cytolysis values from high E:T to make quantitative assessments of the specific cytotoxic activity of effector cell populations. Supporting Information Figure 2 further supports this point by presenting an overlay of cytolysis and degranulation curves from cytolysis and degranulation assays preformed in parallel using UPN035 effector and NS0/CD19 target cells.
We performed this analysis in two assays systems. First we collected and evaluated supernatants from the CD8 T cell cytolysis assay presented in Figure 5 (UPN035 vs. NS0/CD19 and NS0 parental targets). As summarized in Table 4, IL-12p40/p70, TNF-a, MIP-1a, and MIP-1b were specifically detected upon coculture of UPN015 with NS0/CD19 but not NS0 parental targets. Additionally, high levels of RANTES (112À125 pg ml 21 ) were detected in both cultures. None of the remainder of factors (IL-1b, IL-1RA, IL-2, IL-2R, IL-4, IL-5, IL-6, IL-7, IL-8, IL-10, IL-13, IL-15, IL-17, IFN-a, IFN-c, GM-CSF, IP-10, MIG, Eotaxin, MCP-1, VEGF, G-CSF, EGF, FGF-basic, and HGF) were detected under the conditions of this assay. Because NS0 cells are murine in origin and the antibodies specific for IL-12 p40/p70 and TNF-a do not recognize the murine homologues (Invitrogen 30-plex package insert) it is clear that these factors were produced by the effector and not the target cells; since it is not known whether the MIP1a and MIP1b antibody reagents distinguish between the human and murine species, it is theoretically possible that the detected factors are produced by target cells. Second, we collected and evaluated supernatants from an NK cytolysis assay with healthy PBMC cultured overnight with 100 IU human IL-2, and K562 targets. As can be seen in Table 5, IL-1b, IL-12, IP-10, and HGF could be specifically detected in the coculture conditions but not in cultures comprised of PBMC or K562 alone. Additionally, levels of RANTES (42À84 pg ml 21 ), MIP1b (11À18 pg ml 21 ), IL-8 (36À170 pg ml 21 ), IL1Ra (42À150 pg ml 21 ), and IL-2 (15À22 pg ml 21 ) were detected in the effector 1 target or effector alone cultures but not in the target alone cultures; the remainder of cytokines were not detected under the conditions of this assay. Qualitatively similar results were obtained using PBMC cultured overnight in lower concentrations of IL-2. Since both effector and target cells are human in origin in this instance, it is theoretically possible that the measured cytokines are produced by the target and not the effector cells; intracellular cytokine staining in conjunction with surface staining for effector and target-specific markers would allow for this discrimination to be made conclusively. It should be emphasized that this dataset is presented as illustrative for the potential to multiplex the flowbased cytolysis assay with cytokine detection; clearly, kinetics  of cytokine production and secretion will not in most cases optimally conform with the time-course of a 4-h cytolysis assay. Furthermore, these types of data have the potential to be complemented by ICS analyses of effector and target cell populations over a range of E:T ratios to provide insights into the numbers and absolute amounts of cytokines produced by individual cell subsets in the cytolysis assay.

DISCUSSION
In this report we describe the development, evaluation, and application of a statistically supported, flow-based assay to quantify antigen-specific cytolysis. The methodology as described is shown to be comparable to the 51 Cr release assay in terms of sensitivity, and to generate reproducible and accurate data that can be used to quantify the cytolytic activity of both CD8 T and NK effector cells. We have demonstrated the utility to use this methodology to evaluate and quantify cellmediated cytolysis of suspension cells; in principle the same methodology can be applied to evaluate cytolysis of adherent cell types, providing cell integrity is maintained upon detachment. Importantly, use of this assay has the potential to significantly enhance the quality of data obtained from analysis of clinical samples, by allowing for the generation of quantitative  Values are from the average of duplicate measurements and are expressed as picograms detected/ml serum. %CV in all cases was less than 16.7%. ND 5 below the limit of detection. Lower limits of detection were: IL-12: 4.7 pg ml 21 ; TNF-a: 2.5 pg ml 21 ; MIP1a: 3.1 pg ml 21 ; MIP1b: 5.4 pg ml 21 . Data is shown from the 6.25:1 E:T condition. values for cytolytic activity and simultaneously the evaluation of other functional and effector functions in samples. The present work expands on previously published flow-cytometrybased cytotoxicity assays (3,5,6,8À14) in two significant ways by demonstrating the methodological approach and potential to quantify cytolytic activity and the significant potential and advantage for multiplexing the FBCA assay with other platforms to obtain more comprehensive and integrated datasets.
The integration of statistical underpinnings and the FBCA results in the critical advantage of generating quantifiable data for each of the parameters (log e ED50, Hill slope, maximum and minimum cytolysis values) that can be measured in a cytolysis assay. However this integration also has a number of important consequences with regard to assay setup. Specifically, (i) assays need to be set-up so that a minimum of eight data points are obtained per E:T combination, and (ii) data need to be generated so that they span the entire lysis curve. By selectively eliminating data points from a 10point full cytolysis curve, we demonstrate that the ability to accurately calculate estimates for log e ED50 and Hill slope values is significantly impacted by removing as few as two internal data points, while the removal of external data points has similar significant impact on estimating maximum and minimum cytolysis values.
We demonstrate that the flow-based platform can be used to quantify antigen-specific cytolysis mediated by CD8 T cells as well specific cytolysis mediated by NK cells in multiple settings. By quantifying NK cytolytic activity in patient samples we showed that the higher NK activity observed in PBMC samples obtained from a patient post high dose IL-2 therapy could be attributed to a higher percentage of NK cells in the sample post IL-2 treatment, not to increased cytolytic activity by NK cells as previously reported (22À24). These results are at least in part discordant with the proposed function of IL-2 to enhance NK cell functionality in vivo, and highlight the important advantage of being able to quantify NK cytolytic activity on a per cell basis in cytolysis assays.
An important issue related to the use of dyes such as CFSE to label and distinguish target from effector cell populations is the potential for the dyes to be released by labeled target cells either as a consequence of cytolysis or passively and subsequent uptake by effector cells. To address this issue, we evaluated the uptake of CFSE by effector cells at the conclusion of an FBCA, taking advantage of the CD8b surface marker on UPN035 cells. As presented in Supporting Information Figure 3 and summarized in Supporting Information Table 1, cells that stain positively for CD8b can in fact be detected in the CFSE1 quadrant at the conclusion of the cytolysis assay for both target cells. Although the absolute number of CFSE1/CD81 events does not increase dramatically across the evaluated E:T ratios, the percentage of CFSE1/CD8b cell events increases significantly with increasing E:T for both the NSO-CD19 (2.9% at E:T of 0.2:1 to 21% at E:T of 12.5:1) and NS0 (0.6% at E:T of 0.2:1 to 9.4% at E:T of 12.5:1) targets. These data also indicate that robust target cell lysis is not required for the generation of increased numbers of CD81/ CFSE1 cells. Figure 2, back-gating of the CFSE1/CD81 cell populations and evaluation of the forward and side scatter properties reveals that these cells to have size properties distinct from the effector cells and very similar to that of the target cell population. These results clearly demonstrate that the use of dyes such as CFSE to stain target cells has the potential to introduce a degree of error, most apparent at high E:T ratios, and supports the use, if available, of alternative cell identification methodologies such as surface markers to distinguish target from effector cells in FBCA. Notably, although the contamination of the CFSE1 gate by small numbers of effector cells could be responsible for the observed difference in % cytolysis at high E:T ratios in the FBCA vs. 51 Cr assays observed in Figure 5, as shown in Table 3 the two assays generated statistically indistinguishable log ED50 and Hill slope values, supporting the validity of the applying the FBCA assay as an alternative to 51 Cr release assays to quantify target cell cytolysis. These results highlight the need to optimize CFSE or other dye staining of targets, carefully establish gates that maximally separate stained unstained cells, and maimize viability of both target and effector cells that go into the assay.

As presented in Panel B of Supporting Information
One of the emerging themes in correlative science is the increasing appreciation for the need to evaluate biological samples in terms of multiple functionalities, as demonstrated by the example of polyfunctional T cells are associated with positive outcomes (16,25À27). A clear advantage for the flow-based cytolysis assay is the ability to multiplex this assay with other assays and platforms to provide a more comprehensive data set to evaluate effector and other cell functionality. We provide two examples of this multiplexing ability. First we demonstrate the ability to multiplex the FBCA with a flow-based CD107 degranulation assay and thus obtain information on the surface phenotype and effector functionality of effector cells in the cytolysis cultures. An important observation from these analyses is that at E:T ratios that are typically used to evaluate and describe cytolytic activity in T cell cultures (E:T [ 5:1), the degranulation of effector cells is substantially less efficient than at lower E:T values, suggesting that at the higher E:T cytolytic activity may be underestimated. This has implications both for the ability to accurately quantify cytolytic activity, but also should be considered when comparing two cultures with significantly different numbers of specific effector cells. Although not illustrated in this report, the combination of CD107 degranulation assay and the cytolysis assay could be further used to analyze the cytotoxic potential of subpopulations of effector cells (for example, central vs. effector memory vs. naÿve CD81 T cells) and to correlate the phenotype of degranulating cells with specific cytolytic activity. Analyses that seek to correlate the phenotype of cells with specific cytolytic activity need to keep in mind and consider the possibility that surface markers may be modulated as a result of target cell recognition and cytolysis.
Finally we demonstrate the ability to multiplex the flowbased cytolysis assay with the Luminex bead array platform to ORIGINAL ARTICLE quantify secreted factors in the cytolysis culture. We demonstrate that in a 4-h cytolysis assay, both CD8 T cells and NK cells produce low levels of IL-12 in response to specific targets, and beyond that a mostly divergent set of immune factors, with the CD8 T cells studies producing most notably TNF-a, MIP1a and MIP1b, and NK cells IL-1b and IP-10. As discussed above, at least for the case of CD81 T cells, the production of IL-12 cannot be attributed to the target cells which are murine in origin. The biological significance of this observation is unclear at this point but may be relevant in the post target cell engagement activation, proliferation, and differentiation process for T and NK cells. Although this demonstration is illustrative in nature, we believe it highlights the significant advantage of this assay to allow for the generation of additional data sets to more comprehensively obtain insights into the biology of systems under evaluation.