Improving Reliability of Immunological Assays by Defining Minimal Criteria for Cell Fitness Open Access

The last decade has witnessed an explosion of new therapeutics that target specific components of the immune response (1–3) and offer new hope to patients with autoinflammatory and autoimmune diseases. A common approach is to transition from generalized immune suppression to a personalized approach where highly targeted therapies are used only in the subset of patients for whom the targeted pathway is involved in disease etiology (3).

Currently, selection of biologic therapies is still largely left to trial and error (4–7) because stratification based on single genetic risk factors or the activity of individual biomarkers is often ineffective (8). Accordingly, the field has turned to multiomic approaches that combine genetic, transcriptional, and/or protein profiling to define composite biomarker scores that integrate measures of multiple pathways (5, 8, 9). PBMCs are commonly used to measure the activity of various pathways because these are an accessible source of immune cells amenable to shipping and storage. However, PBMCs are subject to heterogeneity introduced by differences in isolation, storage, and cryopreservation practices. Because these factors also further variation caused by differences in protocols, reagents, instruments, analysis, and interpretation of data (10–15), it has been exceedingly challenging to develop reproducible PBMC-based biomarkers.

Many reports have shown that consistent isolation, freezing, and thawing of high-quality PBMCs requires precise standard operating procedures and skilled laboratory personnel (16–18), but an outstanding question is what constitutes “high quality”? The most common parameter used to assess PBMC quality is viability at thaw, typically measured using vitality dyes such as trypan blue, propidium iodide (PI), or 7-aminoactinomycin D (7AAD), which are excluded from cells with intact outer membranes, including those in early apoptosis stages (19–21). Protocols that call for assessment of cell viability after an overnight rest allow cells in early stages of apoptosis to become detectable by vitality dyes, allowing a more accurate assessment of cell health (22–26). However, there have been discrepancies in applying these assessments, including designating viability cutoffs between 66 and 80% (27–29) and using a combination of viability, recovery, and minimum events (29, 30).

The most rigorous standardization of cell-based immunoassays using cryopreserved PBMCs to date has been done for monitoring of Ag-specific T cell responses using ELISPOT or intracellular cytokine staining (12, 14, 30–34). Several authors have reported viability cutoffs above which the T cell response (typically to mitogens) is no longer dependent on cell viability as measured by vitality dye at the time of thaw. Using these criteria, Weinberg et al. (16, 35) reported a viability cutoff of 70–75%. Another study exposed cells to various suboptimal conditions and found that samples with >18% cells in late apoptosis (AnnexinV⁺7AAD⁺) had dramatically decreased assay performance (36). A z score–based QC rating and parameter-specific thresholds can also be used to include only data from optimally treated cells (37). Although this last example effectively controls for variability in batched experiments, it does not identify individual test samples that are suboptimal due to cell handling before and during the cryopreservation process, a variable especially important for multisite studies.

PBMCs for biomarker studies are often from biorepositories with poorly understood history (38) and/or from multiple sites with varying degrees of experience, creating a need for evidence-based standards for cell fitness requirements for biomarker assays. In this study, we used an induced fail approach to develop recommendations for minimum fitness requirements for cryopreserved PBMCs in representative biomarker assays, including immune phenotyping and functional response to inflammatory stimulation. Widespread implementation of minimal cell fitness requirements for thawed PBMCs will increase data reproducibility and help biomarker assays better reveal the characteristics of their host.

We analyzed cells from three different sources. For Figs. 1, 2, 3, 4, and 5, buffy coats from eight healthy volunteers (four males and four females; average age, 38 y; range, 18–70 y) were obtained by Canadian Blood Service’s Blood4research program (study approval 2015.028). For Fig. 6, some blood was collected from consenting people with Crohn’s disease (CD) who had not yet been treated with a biologic therapy; study approval was from the Western Institutional Review Board (protocol 20141347). The CD cohort included 12 males and 12 females with an average age of 36.5 y (range, 19–71 y). At the time of blood collection, five subjects in the CD cohort were taking steroids, and six were taking antimetabolites (methotrexate or azathioprine). In addition, blood samples from six healthy volunteers were collected (two males, four females; age > 19 y; University of British Columbia Clinical Research Ethics Board H18-02553). CD and control blood were collected into three to four 10-ml sodium heparin vacutainers, and PBMCs were isolated the next day (after shipping or overnight incubation at room temperature [RT]).

All reagents and centrifuges were used at RT (21°C) until the final cryopreservation step. PBMCs were isolated as described by Ivison et al. (39). For “optimal” samples, isolated PBMCs were resuspended in freezing medium (10% DMSO; 90% FCS) at 20 × 10⁶ cells/ml for immediate cryopreservation in 0.5 ml (10 × 10⁶ cell) aliquots. Cells were first transferred to a Mr. Frosty container and placed at −80°C for 1–3 d before transfer to liquid nitrogen for storage. To thaw cells, we placed aliquots in a 37°C water bath for 1 min, then vial contents were transferred to 37°C thawing medium (15% FCS, 25 mM HEPES, 0.0375% sodium bicarbonate in RPMI). Cells were spun down at 453 × g for 10 min; the pellet was washed once in PBS to remove excess DMSO and serum, and then resuspended in an assay-specific fashion for counting.

For the “induced fail” treatment, isolated PBMCs were resuspended at 10 × 10⁶ cells/ml in 2% FBS in PBS. For heat stress–induced fails, cells were aliquoted into volumes of 8–10 ml per tube and incubated in a water bath at the indicated temperature and time before being transferred to a second water bath at RT for 10 min. Cells were then frozen in aliquots of 10 × 10⁶, without adjusting cell numbers after treatment. Real-world induced fails involved subjecting cells to common temperature and timing variations including temperature variation before freezing (RT for 18 h, or 4°C for 24 or 48 h); freezing in different DMSO concentrations (7.5 or 15% instead of 10% DMSO); or exposure to freeze-thaw cycles (either two freeze-thaw cycles using a Mr. Frosty or one freeze-thaw cycle in which the cells were thawed at RT and then returned directly to liquid nitrogen).

Cells were counted using the Cellometer Auto 2000 (Nexcelom Bioscience, Lawrence, MA) after staining with acridine orange and PI (catalog no. [cat#] CS2-0106) according to the manufacturer’s instructions using the default setting “Immune Cells, low RBCs.” Cell fitness was determined by a flow cytometry–based live cell/apoptosis assay (cat# ab176749; Abcam, Cambridge, UK), performed according to the manufacturer’s instructions. Specifically, 10⁵ cells were resuspended in 100 µl of kit assay buffer in a 96-well V-bottom plate (cat# 82050-652; VWR, Radnor, PA), centrifuged for 453 × g for 5 min, and incubated at 37°C for 30–60 min with 1/100 diluted stock solutions of Apopxin-Green, 7AAD, and CytoCalcein Violet 450 (CC-450) in assay buffer. After incubation, cells were diluted in another 150 µl of assay before acquisition. Cells were acquired on a CytoFLEX using a vial of PBMCs with a known viability of ∼50% as a compensation control. Live, apoptosis-negative (LAN) cells were defined as CC-450 positive, Apopxin-Green negative (see Supplemental Fig. 1A and 1B for gating). For comparison of LAN values with various biomarkers, values obtained from two repeat apoptosis assay sets were correlated for each PBMC sample, and those with r² < 0.9 were discarded.

After thawing and counting, 0.25 × 10⁶ or 1 × 10⁶ cells were added to a V-bottom plate for the overview or Th1/2/17 panel, respectively. All incubation steps were done in the dark. Cells were washed once in 150 µl of PBS, centrifuged at RT, 974 × g for 5 min, and plates were flicked and blotted to remove supernatant. Pellets were resuspended in 1:1000 diluted fixable viability dye (FVD) eFluor 780 (65-0865; eBioscience) so that the cell concentration was 5 × 10⁶/ml (50 µl for overview, 200 µl for Th1/2/17) and incubated for 25 min at 4°C. After washing, cells were blocked with 4 µl of Fc Receptor block (14-9161; eBioscience) in 20 µl of FACs buffer (2% FBS in PBS) at 4°C for 10 min. Abs were added in a total volume of 30 µl including 10 µl of BD Horizon Brilliant Stain buffer (566349; BD Biosciences, Franklin Lakes, NJ) and 1 ml of each Ab (1/50 dilution). See Supplemental Table I for a list of all Abs, including clones, vendor, and catalog numbers. Staining was done for 30 min at 4°C. After washing cells in 150 µl of FACS buffer, pellets were resuspended in 300 µl of IOTest3 fix buffer (A07800; Beckman-Coulter, Brea, CA) and stored at 4°C in the dark until acquisition on an LSR-Fortessa (within 24 h) (see Supplemental Fig. 1C and 1D for gating).

After thawing, cells were counted and resuspended in OpTmizer medium (cat# A1048501; ThermoFisher Scientific, Waltham, MA) at 2 × 10⁶ cells/ml, and 100 µl aliquots were plated in a 96-well round-bottom plate. After 1–3 h rest at 37°C, 100 µl of OpTmizer with 2× stimulant concentration was added, and cells were returned to 37°C. The final concentration of stimulants was 10 ng/ml LPS (cat# tlrl-3pelps; Invivogen, San Diego, CA), 10 ng/ml PMA, and 0.5 µg/ml ionomycin (cat# P8139 and cat# I0634, respectively; both from Sigma-Aldrich, St. Louis, MO). After 24 h, cells were briefly mixed and then spun down at 974 × g for 5 min, and the supernatant was frozen at −80°C.

For ELISA, used to detect IL-6, 96-well plates were coated with 1 µg/ml IL-6 capture Ab (cat# 14-7069-85; eBioscience, San Diego, CA) and blocked with 1% BSA (A9418; Sigma-Aldrich) in PBS. Supernatants were diluted 1/30 in blocking buffer, and the IL-6 standard was from eBioscience (14-8069-80); incubation was 2 h at RT or overnight at 4°C. Detection of biotinylated anti–IL-6, added at 1 µg/ml for 1 h (13-7068-85; eBioscience), was with 1:1000 diluted SA-HRP (554066; BD Biosciences) and 75 µl of 3,3′, 5,5′ tetramethylbenzidine solution set (555214; BD Biosciences); reaction was stopped with 75 µl of 2N H₂SO₄, and plates were immediately read on a microplate reader at 450 nm. Concentrations of IL-6 were calculated from the log-transformed standard curve using MS Excel’s TREND function. Quantification of cytokines other than IL-6 was done using a Legendplex custom kit, which detected 12 proteins (BioLegend, San Diego, CA). The assay was done according to the manufacturer’s instructions (cat# 92919, lot# B25788) with the exception that 10 µl of supernatant/standard was used in each reaction. After resuspending stained beads in 120 µl/well, they were acquired on the CytoFLEX, at 90 µl/min for 60 s/well. Concentrations of each cytokine were calculated from standard curves using the Legendplex software version 7.1 for OSX. The three cytokines (TNF-α, IL-10, and IL-1β) that yielded concentrations greater than the assay detection limit in the majority of samples were evaluated.

After thawing, cells were resuspended at 1 × 10⁶ cells/ml in X-VIVO15 serum-free media and 100 µl/well placed in a 96-well V-bottom plate, which was then cultured overnight at 37°C and 5% CO2. The next day, 100 ng/ml IL-6 (14-8069; eBioscience) or an equal volume of PBS was added, and samples were shaken for 15 min on a heated plate shaker at 1000 rpm after which 100 µl of prewarmed BD Cytofix buffer (554655; BD Biosciences) was added to each well and the plate shaken for a further 15 min. The cells were then centrifuged for 5 min at 974 × g and then flicked and blotted to completely remove supernatant. The plate was placed on ice and pellets resuspended in 100 µl of ice-cold BD Perm III solution (558050; BD Biosciences). After incubation for 30 min on ice, the plate was centrifuged again, and the pellets were washed twice in 150 µl of FACS buffer. Cells were stained with Abs against CD3-PE (1/50, UCHT1; 555333; BD), CD4-FITC (1/50, RPA-T4; 11-0049; eBioscience), and pY705-STAT3-AF647 (1/25, 4/PSTAT3; 557815; BD Biosciences) for 45 min at 4°C in the dark. After washing, cells were resuspended in 150 µl of FACs buffer and acquired immediately (within ∼4 h).

Acquisition of the overview, Th1/2/17, and STAT3 phosphorylation panels was done using an LSR-Fortessa X20 (BD Biosciences) using BD FACSDiva software version 8.0. Optimal voltages were set up for a 12-color panel, and apparent voltages were captured as fluorescent intensities for recalibration using Flow Set Pro beads (A63492; Beckman-Coulter). These settings were fixed on the LSR-Fortessa relative to the cytometer setting and tracking (CS&T) research beads (650621; BD Biosciences) using the application settings function; CS&T was run before every experiment. Apparent voltages were recalibrated after CS&T bead lot and/or baseline changes using Flow Set Pro beads. Single-cell controls were run with every experiment using cells stained with Abs against CD4 in every fluorochrome. Live-cell/apoptosis assays and Legendplex analyses were done on the CytoFLEX (Beckman-Coulter) using fixed default gains. Analysis of exported FCS3.0 files from both the LSR-Fortessa and the CytoFLEX (including compensation) was done with FlowJo version 10 (BD Biosciences). Analysis of Legendplex data was done with the LEGENDplex Data Analysis Software Suite from BioLegend.

All statistical analyses were performed in R 4.2.1 (40). The effect of cell damage on the observed parameters can be systematic, causing an increase or decrease in values compared with optimally treated samples. However, it may also introduce additional “noise,” resulting in loss of precision, reducing the statistical power and increasing the spread of confidence intervals (CIs). To better capture this type of information loss, statistical analyses focused on effect size estimates, distributions, and correlations, rather than reporting p values from statistical testing of the null hypothesis (no systematic effects).

For each treatment and donor, a ratio to “optimal condition” was computed and illustrated as boxplots. In the absence of effect from induced cell damage, the expected median of ratios would be 1, with a boxplot of minimal spread centered on this value. Inference of a systematic effect of each treatment, relative to optimal, was also represented by a parametric (t distribution) 95% CI, computed for the mean of the log-transformed ratios, and then converted back to the original scale. Under the hypothesis of log-normal distribution for the ratios, the resulting interval represents inference on the median relative effect (fold change [FC]). The effect of each treatment versus optimal was evaluated independently, because we observed clear deviations from sphericity in the complete experiment. Wilcoxon nonparametric analyses were conducted to confirm robustness of the results to deviation from normality.

The combined effect of each treatment on viability (as measured by LAN) and parameters was illustrated using 95% CI ellipses based on covariance matrix and t statistics (R function ellipses, package psych v. 2.2.9 [41]). The width of each ellipse represents the 95% CI for the mean of the LAN after each treatment. The height of each ellipse represents the 95% CI for the relative effect of the treatment, computed on the log scale. The angle of each ellipse is proportional to the covariance between LAN and relative effect of treatment. Individual data points were added to the plots. The 95% CI for the LAN after optimal treatment was illustrated as a grayed vertical zone. Statistical analyses on the parameters were conducted after log transformation, thus reflecting the multiplicative effect observed. Analyses for the LAN were performed on the original scale, because the distribution was more consistent with linear effects.

Unless otherwise stated, boxplots are shown bounded by first and third quartiles with whiskers indicating the range up to ±1.5 interquartile range distance and the median depicted as a black bar. The 95% CIs for the mean are indicated beside each box.

Pearson correlations were computed between each condition and the optimally treated cells, on log-transformed data. The Pearson correlation measures the linear relationship between the measures; if between-donor differences are conserved (i.e., low noise induced by cell damage), the Pearson correlation is expected to be 1.

To assess how low PBMC viability affects functional responses, we used heat shock to create aliquots of PBMCs with a range of viability. Cells were exposed to 46°C for different lengths of time (1–20 min) and then frozen. Upon thawing, viability was assessed using PI either immediately after thawing or after 24-h incubation at 37°C. Viability decreased with the length of heat exposure and was further diminished after 24-h incubation (Fig. 1A, lines). To assess how responses to an innate stimulus were affected by viability, we stimulated heat-treated PBMCs with LPS for 24 h and measured levels of IL-6 in the supernatant (Fig. 1a, bars). Although PBMCs heat-shocked for 5–10 min had high viability at thaw (>90%), they lost the ability to respond to LPS, suggesting that viability measured on the basis of PI incorporation does not reliably predict cell fitness.

A limitation of measuring cell viability using PI (or other membrane exclusion–based methods) is these dyes do not stain intact cells in early phases of apoptosis (19–21), so we asked whether measuring viability with an assay that differentiated early and late stages of apoptosis could more reliably predict function. Cells were analyzed using a flow cytometry–based assay for metabolic activity (measured by CytoCalcein-Violet, which detects esterase activity), a marker of early apoptosis (Apopxin Green, which binds phosphatidylserine), and the cell permeability dye 7AAD, either upon thaw or after 24-h incubation at 37°C. After resting, some cells shifted toward later apoptotic stages, shown by the increase in cells lacking metabolic activity and uptake of 7AAD (Fig. 1B). The proportion of healthy or “fit” cells (apoptosis negative, positive for marker of metabolic activity, hereafter referred to as LAN) was similar at thaw and after 24 h. When stimulated with LPS, production of IL-6 showed a higher correlation with the proportion of LAN cells than with 7AAD-negative cells (Fig. 1C). These data suggest that measuring early apoptosis and metabolic activity is more predictive of cell fitness than solely measuring membrane permeability directly after thawing.

To further explore the impact of cell viability on downstream assays, we subjected PBMCs to conditions that could occur during the collection and storage of clinical samples. These so-called induced fail conditions included prolonged incubation at different temperatures, suboptimal freezing protocols, and additional freeze-thaw cycles (Fig. 2A). For all samples, cell viability was determined on the basis of PI exclusion and the apoptosis assay, immediately after thaw and after 24-h incubation at 37°C. PI-based analysis showed that the viability decreased in the 24-h incubation condition, except for cells that had been incubated at RT for 18 h before cryopreservation, which produced cells with lower viability upon thawing. The percent LAN at thaw correlated more strongly with PI-based viability in samples rested for 24 h than at thaw; Pearson r (with 95% CI) = 0.86 (0.76–0.92) versus 0.54 (0.34–0.70), respectively (Fig. 2B). Therefore, measuring LAN at the time of thawing can predict response to LPS and viability after 24 h of culture.

Because biomarker assays may target different cell types within PBMCs, we examined the effect of damage on main immune cell subsets using a basic immune phenotyping panel (Supplemental Fig. 1C), with representative flow cytometry data from specific populations in Supplemental Fig. 2A. As shown in Fig. 3A, data are graphed as a ratio to the average of optimally treated samples from the same donor, i.e., the further the data extend from y = 1, the larger the impact of cell damage. Raw data are provided in Supplemental Table II.

Overall, DMSO 7.5% and DMSO 15% did not cause notable cell damage in the analyzed cell types. T cell proportions were least affected by the different types of treatment, as reflected by narrow boxplots for the ratios and medians close to 1. However, CD4⁺ and CD8⁺ T cells were differently susceptible to various types of damage. In comparison with CD8⁺ T cells, CD4⁺ T cell proportions were stable after freeze-thaw but reduced after 18-h incubation at RT before cryopreservation (Supplemental Fig. 2B).

Monocyte proportions (of live, CD45⁺ cells) were reduced (FC = 0.3) by incubation for 18 h at RT but slightly increased (FC = 1.5) after incubation at 4°C. We also observed increased noise for the harsh freeze-thaw condition, as indicated by the large spread of ratios, low correlation with optimal (Pearson r = 0.23), and overall low agreement with optimal (Intraclass correlation coefficient = 0.11). The proportion of B cells increased after 18-h incubation at RT (FC = 3.6), whereas proportions of T and NK cells decreased (FC = 0.7 and 0.8, respectively). Freeze-thaw–induced damage had the most negative impact on NK cells, whereas B cells and T cells were less affected.

Plotting ratios to optimally treated samples against LAN values (Fig. 3B) showed that the proportions of T cells were minimally affected by changes in the percent of LAN⁺ cells, whereas other cell types showed increased variability when LAN was <60%. The impact on LAN was also treatment specific: optimally treated cells had LAN values between 61 and 80%. These data indicate that, with the exception of T cells, thawed PBMC samples with LAN < 70% may not be optimal for basic immune cell quantification or phenotyping.

Many immune phenotyping assays rely on cell-surface marker expression to define proportions of functionally distinct cell subsets. Therefore, we tested the predictive value of the LAN assay for effects of cell damage on proportions of Th1, Th17, Th17.1, and Th2 CD4⁺ T cell subsets, quantified by flow cytometry (Supplemental Fig. 1D) according to expression of CXCR3, CCR6, CCR4, CD161, and CRTH2 (42–45). We found significant variability in the effect of damage on different Th subsets (Fig. 4A). Th17 and Th17.1 cells showed the most variation, manifested as decreased proportions, especially after 48 h at 4°C (FC = 0.07 and 0.13, respectively). In contrast, Th1 cell proportions remained similar or even slightly increased. These changes were in part driven by unaffected or increased CXCR3 expression in response to damage, with a parallel reduction in CCR6⁺ proportions (Fig. 4B) and levels of expression (Supplemental Fig. 3). Notably, change compared with optimally treated samples is dependent both on the treatment and the subset/marker: some forms of damage induced significant changes in cell proportions despite having LANs in the range of optimally treated samples (e.g., for the 4°C at 24 h condition, the LAN was >70%, but the proportions of Th17 and Th17.1 were significantly lower). Thus, flow cytometry–based measures are affected by some types of damage, and deviation from “true” may not always be completely mitigated by the use of minimal LAN values.

We next investigated the relationship between LAN values and two functional assays: (1) cytokine secretion in response to LPS and (2) IL-6–stimulated phosphorylation of STAT3. For most conditions, production of TNF-α in response to LPS became much noisier when LAN was <60% (Fig. 5A). Analysis of individual PBMC donors revealed that, although in some samples, responses may remain strong despite low LANs, the ability to differentiate between individuals may be lost (e.g., the increased LPS response in the PBMC donor indicated by filled symbols; Fig. 5B). The impact of lower LAN on biomarker quantity is also donor dependent (e.g., the PBMC donor represented by hollow triangles shows comparable TNF-α production to optimal even at LAN < 40%). Data from other detected cytokines were strongly correlated with those from TNF-α: Spearman’s correlation for TNF-α versus IL-1β and IL-10 was 0.87 and 0.88, respectively (p < 0.0001).

In the assay measuring IL-6–stimulated p-STAT3 in CD3⁺ T cells, the baseline levels of p-STAT3 were more affected by cell damage than were p-STAT3 levels after stimulation. Compared with IL-6–stimulated p-STAT3 levels, which were fairly stable at LAN > 60%, unstimulated samples showed increased p-STAT3 variability at higher LAN values, even at LAN > 70% (Fig. 5C).

To test the clinical applicability of LAN measurement, we tested the quality of PBMCs from 24 patients with inflammatory bowel disease (IBD), collected from multiple sites as part of a pilot study to find biomarkers of treatment response. After collection, blood was shipped overnight to one of three locations where PBMCs were isolated and cryopreserved according to standardized protocols within 24 h of blood draw. Aliquots of cryopreserved cells were shipped in batches to a single site for analysis of viability, apoptosis, and biomarkers. Blood from six healthy volunteers was collected and cryopreserved using the same protocol. We found that viability (defined as PI negative) at thaw ranged from 30 to 95%, and that approximately one third of the IBD samples had <75% viability (Fig. 6A). Determination of % LAN showed that, although the majority (10/15) of samples with LAN < 60% also had low PI-based viability (<80%), some (16/44) samples with higher PI-based viability (>80%) showed LAN values < 70%.

All samples were then subjected to a subset of the biomarker assays described earlier, specifically phenotyping for Th subsets and p-STAT3 response to IL-6 stimulation. Assays were repeated on two to three different replicate aliquots on different thaw batches when cell numbers allowed. Plotting the biomarker values against LAN (Fig. 6B–D) showed a tendency of Th1 proportions to increase and Th17 and p-STAT3 proportions to decrease when LAN was <70%, consistent with findings in Figs. 4 and 5. Likewise, repeat data were particularly variable in the Th17 assay even at LAN > 70%, further illustrating that this surface receptor-based method of measuring Th17 cells is highly variable. These data indicate that measuring LAN in patient samples can be a useful metric from which to judge PBMC sample quality, and that suitable LAN thresholds are assay dependent.

The lack of reproducibility in biomarker studies impedes the translation of discovery research to clinical utility (46–48). Many studies reported recommendations for various aspects of blood collection, PBMC isolation, cryopreservation, and thawing (17, 25, 28, 49–54), but even stringent standardization of assays will not overcome variability related to sample quality. Cryopreservation is a well-known source of cell injury due to ice crystal formation, membrane damage, and osmotic shock (55), but there is nevertheless a need to cryopreserve samples to allow batch testing or retrospective analysis. Standardization efforts for immunological assays have largely concentrated on T cells, because monocytes and B cells are known to be more variable. These cells have a shorter life span than T cells (56–59) and may be more susceptible to damage-induced apoptosis.

We saw significant variation in how different cell types are affected by different types of damage. For example, NK cells were particularly susceptible to freeze-thaw–induced damage, whereas B cells seemed resistant to prolonged incubation at RT. Differences in PBMC subpopulations after prolonged incubation have been observed in blood, with lymphocytes being more stable than myeloid cell types, but these effects are likely dependent on the activation and differentiation status of these cells. For example, unlike B cells, plasmablasts do not survive a 24-h incubation in unfractionated blood (39). Th17 and Th17.1 cells were more susceptible than Th1 cells to a wide variety of cell damage type cells; however, it should be noted that changes in expression of lineage-defining markers (i.e., CCR4 and CCR6) cannot be separated from actual cell loss in the experiments presented in this article.

We sought to develop evidence-based guidelines that would allow systematic assessment of PBMC cell fitness and subsequent cutoffs for sample inclusion/exclusion. We found that viability as determined by measuring the % of PI or 7AAD⁺ cells directly after thawing was not a reliable indicator of cell fitness. However, samples can be reliably assessed using this method after 24-h incubation at 37°C. This method can be easily incorporated into protocols that use overnight resting before assaying or be used retrospectively to filter data collected from stimulated cells. If samples are to be assessed directly after thaw and/or in short-term assays, then more detailed analysis of cell fitness is required. In this study, we used a commercially available assay that combines measurement of metabolic activity and early apoptosis, the latter determined by detection of externalized phosphatidyl serine. Use of fluorescently labeled pan-caspase binding proteins or compounds that become fluorescent after caspase-mediated cleavage may also detect these cells (60, 61). The latter may be more amenable to incorporation into cell counters with fluorescent detectors, because outer membrane staining alone may not be easy to reliably detect.

We found that samples with >60–70% LAN at thaw exhibited less variability in PBMC cell proportions and functional assay responses. It is important to note that, even when there are no significant systematic differences between a treatment group and optimally treated samples, increased variability in samples with lower cell fitness makes it more difficult to detect significant biological differences between groups, especially in studies with small sample sizes. Therefore, the conclusion that “there is no significant difference” between fresh and frozen samples does not mean that both types of sample can be used with equal power. Statistical consideration of the proportion of technical to biological variation such as the intraclass correlation coefficient is an important readout for comparative standardization projects.

A limitation of our induced fail experiments is that they were done with cells collected from blood bank–processed buffy coats and may not represent results from patient-derived samples collected in tubes with different anticoagulants and processed in different ways. Moreover, we did not control for variables such as change in the season, time of day, or location of sample collection. Nevertheless, application of LAN-based cutoffs to samples collected from people with CD confirmed our findings with cells from healthy volunteers. Specifically, proportions of Th1 cells were higher and Th17 cells lower when samples were <70% LAN, as were IL-6–induced p-STAT3 levels. Even at LAN values > 70%, Th17 proportions were variable. Generally, technical variation may obscure differences between individuals or groups in studies with low sample sizes. It is possible that this type of complex immune-based assay should be done on freshly frozen PBMCs, rather than those for which processing has been delayed for 24 h before cryopreservation, to increase LAN values and reduce technical variation caused by cell damage (62, 63). Some blood collection or PBMC processing methods, such as cell preparation tubes or removal of granulocytes before freezing, may reduce damage and increase the power to detect differences between biological groups (62, 64), and fixing at the time of collection can help standardize nonfunctional phenotyping assays (65). Alternatively, some assays may be amenable to immediate whole blood–based protocols, and specialized blood collection tubes have been developed to facilitate and standardize this approach (66).

Our data show that sensitivity of individual assays to cell damage should be considered when selecting biomarkers to test in multisite clinical trials. In addition, to maximize chances of detecting meaningful differences between individuals, assay selection should consider “real-world” sample quality. For example, Th17 cells quantified on the basis of CCR6 can be reduced to half of their original proportions even at >70% LAN; thus, it may be better to choose a different assay to detect these cells (e.g., functional assays). Indeed, the functional assays tested in this study were relatively stable at LAN > 70%, and a comparison of baseline and stimulated p-STAT3 levels shows that the latter is less affected by suboptimal sample handling. It is interesting to note that cells may still produce large amounts of proinflammatory cytokines such as IL-6 or TNF-α despite having low LANs, but differences between donors may be lost.

In summary, determination of LAN values is a rapid and cost-effective method to measure cell fitness for PBMC-based biomarker assays. Our recommended cutoff of 70% LAN cells reduced damaged-induced variation of cell frequencies, phenotype, and function. Incorporation of minimal cell fitness criteria in addition to standardized protocols will increase the likelihood of detecting informative biomarkers in cryopreserved PBMCs.

The authors have no financial conflicts of interest.

M.K.L. receives a salary award from the BC Children’s Hospital Research Institute and is a Canada Research Chair in Engineered Immune Tolerance. J.D.R. holds the Canada Research Chair in Genetics and Genomic Medicine.

Alain Bitton,¹ Gabrielle Boucher,² Guy Charron,² Christine Des Rosiers,² Anik Forest,² Philippe Goyette,² Sabine Ivison,³ Lawrence Joseph,⁴ Rita Kohen,¹ Jean Lachaine,⁵ Sylvie Lesage,⁶ Megan K. Levings,³ John D. Rioux,² Julie Thompson Legault,² Luc Vachon,⁷ Brian White-Guay,⁵ and Sophie Veilleux⁸

¹McGill University Health Centre, Montreal, Quebec, Canada. ²Montreal Heart Institute, Montreal, Quebec, Canada. ³University of British Columbia, Vancouver, British Columbia, Canada. ⁴McGill University, Montreal, Quebec, Canada. ⁵Université de Montréal, Montreal, Quebec, Canada. ⁶Maisonneuve-Rosemont Hospital, Montreal, Quebec, Canada. ⁷Consultant. ⁸Université Laval, Quebec City, Quebec, Canada.