Combined Single-Cell Measurement of Cytokine mRNA and Protein Identifies T Cells with Persistent Effector Function

One of the critical features of T cells for clearing pathogens is the production of cytokines, such as IFN-γ, IL-2, and TNF-α. Their production depends on effective transcription and on posttranscriptional events, such as mRNA stabilization and/or changes in translation efficiency (1–3). Intriguingly, increased mRNA levels for cytokines do not always correlate with protein production. In fact, chronically stimulated anergic T cells, as well as tumor-derived exhausted T cells that fail to produce the corresponding cytokines, contain high levels of cytokine transcripts (4, 5). These observations point toward a critical role of posttranscriptional events for T cell effector function.

To unravel T cell responsiveness, studies determined the levels of cytokine transcripts by single-cell RT-PCR or RNA sequencing (6–8) and the levels of protein production by flow cytometry or ELISPOT-based approaches (9–11). However, these approaches do not directly correlate the transcript levels with cytokine production at the single-cell level. Upon activation, not all Ag-specific T cells participate in the response. Alternatively, they do respond to the activation, but do so by following different kinetics, or by producing different cytokines (10, 11). Additionally, because increased mRNA levels do not always result in protein production (4, 5), it is imperative to study mRNA transcripts simultaneously with protein production for an accurate analysis of T cell responses. Efforts have been undertaken to determine cytokine transcript levels by flow cytometry (12, 13). Also, the first studies combined it with protein expression (14, 15). However, effective T cell responses consist of polyfunctional T cells that produce more than one effector molecule, and studying more than one cytokine transcript and protein was previously not feasible. Furthermore, methods for multiparameter stainings in combination with transcript analysis by fluorescence in situ hybridization (FISH) that are required for an in-depth single-tube analysis of T cell responses are still lacking.

In this study, we present the analysis of IFN-γ, IL-2, and TNF-α mRNA levels combined with protein production by flow cytometry in primary human T cells. We demonstrate that flow cytometry–based FISH (Flow-FISH) can be used for multiparameter flow cytometry to determine the relative responsiveness of naive, effector, and memory T cell subsets in one single measurement, without requiring cell sorting. Intriguingly, Flow-FISH revealed a subset of cytokine-producing T cells that maintained high levels of cytokine transcripts throughout stimulation. The mRNA^hi-expressing T cells produced not only more of the corresponding protein, but they also expressed higher transcript and protein levels for other cytokines. We conclude that Flow-FISH can be used to identify T cells with the highest effector potential.

The study was performed according to the Declaration of Helsinki (seventh revision, 2013), with PBMCs from healthy volunteers having been obtained with written informed consent (Sanquin, Amsterdam, the Netherlands). PBMCs were isolated by Lymphoprep (Stemcell Technologies) density gradient separation, and cells were used freshly or stored at liquid nitrogen until further use. Cells were cultured in IMDM (Lonza) supplemented with 10% FCS, 100 U/ml penicillin, 100 μg/ml streptomycin, and 2 mM l-glutamine and maintained in a humidified incubator at 37°C with 5% CO₂.

T cells were used freshly or after activation with plate-bound anti-CD3/soluble anti-CD28. Briefly, 24-well plates (Corning) precoated overnight with 0.5 μg/ml rat anti-mouse IgG2a (MW1483; Sanquin) were incubated for 2 h with 1 μg/ml anti-CD3 (clone Hit3a; eBioscience). Soluble anti-CD28 (1 μg/ml; clone CD28.2; eBioscience) was added together with 1 × 10⁶ PBMCs. After 48 h of activation, cells were harvested and transferred to uncoated T25 flasks at a density of 0.8 × 10⁶/ml and allowed to rest for 3–5 d with 10 ng/ml human recombinant IL-15 (PeproTech). Fresh T cells were used directly for T cell activation or were sorted for effector, memory, and naive CD8⁺ T cells based on the expression of CD27 and CD45RA (clones, see below) using a FACSAria III cell sorter (BD Biosciences) prior to activation. Dead cells were excluded during sorting with a Near-IR Live/Dead cell stain kit (Life Technologies).

For Flow-FISH or quantitative PCR (qPCR) analysis, 4 × 10⁵ T cells per well seeded in a 96-round bottom plate were stimulated for indicated times with 10 ng/ml PMA and 1 μM ionomycin (Sigma-Aldrich), with plate-bound anti-CD3 (0.1–10 μg/ml as indicated) and soluble anti-CD28 (1 μg/ml), with anti-CD3 alone, or with 1 μg/ml of an MHC class I–restricted CMV or EBV peptide pool (ProImmune, Oxford, U.K.).

Single-molecule FISH probes of 20 nt for IFN-γ, TNF-α, and IL-2 were designed according to the manufacturer’s guidelines (Biosearch Technologies). Probes with predicted high affinity for secondary target genes (identified by BLASTN) were discarded when the human gene skyline in the Immunological Genome Project (http://www.immgen.org) indicated gene expression in T cells. This resulted in 37 Quasar 570–labeled probes for IFN-γ, 27 Quasar 670–labeled probes for IL-2, and in 45 Quasar 670–labeled probes for TNF-α (sequences are available upon request). Binding competition assays were performed with identical unlabeled probe sets (Sigma-Aldrich).

Combined cytokine mRNA and protein staining was performed by adapting the protocol of the Cytofix/Cytoperm kit (BD Biosciences) for Flow-FISH. Briefly, activated T cells were harvested and transferred into 1.5 ml of LoBind Eppendorf tubes. Cells were labeled in PBS supplemented with 0.1% FCS for 10 min at room temperature in the dark with the following Abs: anti-CD8α (RPA-T8) and anti-CD45RA (HI100) from BioLegend; anti-CD45RA (HI100), anti-CD4 (RPA-T4), and anti-CD8α (RPA-T8) from BD Biosciences; and anti-CD27 (CLB-27; PeliCluster) from Sanquin. A Near-IR Live/Dead marker was included for dead cell exclusion. Cells were washed with PBS, and intracellular cytokine staining (ICCS) was performed with the Cytofix/Cytoperm kit according to the manufacturer’s protocol (BD Biosciences). For Flow-FISH, cells were fixed with 100 μl of Cytofix for 15 min at room temperature in the dark, centrifuged at 500 × g for 4 min in a swing-out rotor after fixation, and 150 μl of Cytoperm was added for 20 min in the dark at 4°C. All buffers were pretreated with 1:1000 RNAse inhibitor (murine RNAse inhibitor; New England BioLabs). Samples were washed with 800 μl of wash buffer containing 12.5% formamide and 2× SSC in RNAse-free demineralized water supplemented with 1:10,000 RNAse inhibitor. Hybridization with 15 nM Stellaris FISH probes (Biosearch Technologies) was performed according to the manufacturer’s protocol with 100 μl of hybridization buffer containing 10% formamide, 1× SSC, and 0.1 g/ml dextran sulfate salts (Sigma-Aldrich) in RNAse-free demineralized water containing 1:1000 RNAse inhibitor. Cells were incubated for 16 h at 37°C with FISH probes and the following Abs: anti–IL-2 (MQ1-17H12) and anti–TNF-α (MAb11) from BioLegend, and anti–IFN-γ (4S.B3) from eBioscience. Cells were washed once with wash buffer and once with PBS prior to acquisition by flow cytometry on an LSRFortessa (BD Biosciences) using FACSDiva (v.8.0; BD Biosciences). Data analysis was performed with Flowjo X (Tree Star).

mRNA was extracted using TRIzol/chloroform according to the manufacturer’s protocol (Invitrogen). cDNA was generated with SuperScript III reverse transcriptase (Invitrogen). qPCR was performed with SYBR Green (Applied Biosystems) with the following primers: IFN-γ (5′-AGCTCTGCATCGTTTTGGGTT-3′ and 5′-GTTCCATTATCCGCTACATCTGAA-3′), TNF-α (5′-GCCAGAGGGCTGATTAGAG-3′ and 5′-TCAGCCTCTTCTCCTTCCTG-3′), and IL-2 (5′-CAAGAATCCCAAACTCACCAG-3′ and 5′-CGTTGATATTGCTGATTAAGTCC-3′). 18S was used as reference gene (5′-AGACAACAAGCTCCGTGAAGA-3′ and 5′-CAGAAGTGACGCAGCCCTCTA-3′). Samples were acquired using ABI StepOne or ABI 7500 (Applied Biosystems) with respective software.

Data were analyzed with Prism 6 (GraphPad Software). A Student t test (paired and unpaired) was used for comparison of two measurements as indicated. When more than two groups were compared, a one-way ANOVA followed by a Tukey multiple comparison test was performed. Data are presented as mean ± SD.

We first established the simultaneous detection of IFN-γ mRNA and protein levels by flow cytometry. We activated primary human PBMCs with PMA-ionomycin. Similar to classical ICCS, we blocked protein secretion with monensin throughout T cell activation to measure IFN-γ protein. For mRNA detection, we adjusted the ICCS procedure after surface Ab staining, fixation, and permeabilization by including an overnight hybridization step with FISH probes. Of note, IFN-γ mRNA levels are a snapshot at the time point measured and therefore reflect changes in mRNA levels during stimulation (Fig. 1A). We found that at 1 h after activation with PMA-ionomycin, CD4⁺ and CD8⁺ T cells express detectable levels of IFN-γ mRNA (Fig. 1A). This expression of IFN-γ transcripts preceded that of IFN-γ protein, which was detectable from 2 h of activation onward (Fig. 1A). In particular, at 4 and 6 h after activation, the IFN-γ mRNA levels correlated well with those of IFN-γ protein expression upon T cell stimulation with PMA-ionomycin (Fig. 1A).

We next validated whether the FISH probe signal was specific by spiking in unlabeled probes to the labeled probe mix. We found that the signal of labeled FISH probes dropped in a concentration-dependent manner, with a nearly complete loss of signal at a ratio of 100:1 of unlabeled probes to labeled probes (Fig. 1B). We also compared the mRNA measurements of Flow-FISH with semiquantitative RT-PCR of the total T cell population. We found that Flow-FISH displayed a similar curve of IFN-γ geometric mean fluorescence intensity (GeoMFI) that closely correlated with the values obtained by qPCR (Fig. 1C), which shows that the signal obtained with Flow-FISH is specific and representative. Importantly, also the percentage of IFN-γ protein–expressing T cells measured by Flow-FISH was comparable to classical ICCS staining (Fig. 1D). Therefore, we conclude that Flow-FISH allows for the direct comparison of IFN-γ mRNA and protein on a single-cell level and in a high-throughput manner.

We next determined whether Flow-FISH allowed for the detection of Ag-specific T cells. To this end, we activated human PBMCs with 26 pooled MHC class I–restricted EBV peptides or 14 pooled CMV peptides for 6 h. This resulted in low but clearly detectable responses, indicating that Flow-FISH can identify CD8⁺ T cell responses of low magnitude (Fig. 2A). T cell activation with anti-CD3 alone or anti-CD3 stimulation in combination with anti-CD28 also induced detectable levels of IFN-γ mRNA and protein expression in CD8⁺ T cells (Fig. 2A). Intriguingly, the pattern of mRNA and protein expression differed between stimuli. Whereas activation with PMA-ionomycin resulted in T cell responses that showed a strong correlation of transcript and protein expression, only a fraction of T cells responding to CMV/EBV peptides or to TCR triggering with anti-CD3 (1 μg/ml) Abs contained IFN-γ mRNA at 6 h after activation. Combining anti-CD3 triggering with anti-CD28 costimulation increased transcript-expressing T cells (Fig. 2A, Supplemental Fig. 1A). We also stimulated CD8⁺ T cells with a range of anti-CD3 concentrations and found that from a concentration of 0.3 μg/ml onward, the amount of total IFN-γ–producing T cells and the levels of mRNA-expressing and protein only–expressing IFN-γ–producing T cells were similar (Supplemental Fig. 1B–E). Our data therefore suggest that the nature of stimulation determines the magnitude of T cell responses by differently modulating mRNA and protein expression. In particular, stimulation with PMA-ionomycin induces the highest levels of mRNA that correlate with higher protein output.

We next examined how different T cell subsets respond to stimulation. We sorted naive, effector, effector memory, and memory CD8⁺ T cells based on their expression levels of CD45RA and CD27 (16) (Fig. 2B, left panel). As expected, whereas naive T cells showed a very limited IFN-γ signal after 4 h stimulation with anti-CD3/anti-CD28, we found that effector, effector memory, and memory CD8⁺ T cells clearly produced IFN-γ protein (Fig. 2B). We then established that Flow-FISH also allowed for single-tube analysis of different T cell subsets without the necessity to sort the individual populations. Although the CD45RA and CD27 signal intensity of the same donor was reduced compared with classical ICCS, the signal sufficed to gate on the individual T cell subsets with a similar distribution (Fig. 2B, 2C, left panels). Of note, the overall responsiveness of nonsorted T cell subpopulations was higher than that of sorted T cells (Fig. 2B, 2C). Flow-FISH revealed an interesting feature of mRNA expression in different T cell subsets. Irrespective of the preparation of samples, only a minority of effector T cells coexpressed IFN-γ mRNA and protein at 4 h after stimulation (Fig. 2D, 2E). In contrast, memory T cells displayed the most coexpression of mRNA and protein (Fig. 2D, 2E). The higher responsiveness of memory T cells is also reflected by a higher expression level of CD28 (16) (data not shown). Collectively, these findings suggest that the levels of IFN-γ mRNA are not uniformly regulated in all T cell subsets.

We next extended Flow-FISH to the two other critical effector cytokines of T cells, that is, TNF-α and IL-2. Also for these two cytokines, we detected substantial transcript and protein levels upon 6 h of PMA-ionomycin stimulation (Fig. 3A). Of note, we detected differences in fluorescence intensity for each cytokine. These differences were also reflected by the differential fold increase over nonactivated T cells of individual cytokine mRNA levels measured by qPCR (Fig. 3B).

We also determined the kinetics of transcript and protein expression of IFN-γ, IL-2, and TNF-α. To synchronize T cells, we activated PBMCs for 48 h with anti-CD3/anti-CD28 and allowed them to rest for 4 d in the presence of low levels of human rIL-15. We then activated CD4⁺ and CD8⁺ T cells together with anti-CD3/anti-CD28 Abs to measure the response kinetics by Flow-FISH. IFN-γ mRNA was measurable already at 1 h with little protein staining. Transcript and protein levels significantly increased from 2 h onward, with the highest levels at 6 h after activation (Fig. 3C). For IL-2, mRNA detection clearly preceded protein production, but overall production levels were lower and primarily found in CD4⁺ T cells (Fig. 3D; see below). In contrast, the kinetics of TNF-α transcripts completely overlapped with those of protein expression already at 1 h after stimulation, suggesting that transcript levels directly correlated with protein expression (Fig. 3E). Interestingly, despite the greater number of FISH probes for TNF-α compared with IL-2 (45 versus 27 probes), we found that the mRNA signal of TNF-α was lower than that of IL-2 (Fig. 3A). This finding correlated well with increases in transcript levels measured by qPCR (Fig. 3B), suggesting that the lower transcript signal of TNF-α was not due to technical limitations, but rather a biological feature.

Flow-FISH also allowed us to simultaneously measure the kinetics of transcripts and proteins for these three cytokines. We found that CD4⁺ and CD8⁺ T cells activated with PMA-ionomycin show similar expression kinetics as described for mRNA measured by qPCR and protein measured by ELISA or ICCS (Fig. 3F–H) (11, 17). Taken together, these data demonstrate that Flow-FISH can be used to study the kinetics of cytokine production in T cells.

Whereas previous analyses with qPCR and flow cytometry provided critical information on the kinetics of cytokine production, it did not allow the direct comparison of transcript levels with protein production. Interestingly, when we used Flow-FISH on CD8⁺ T cells that were activated with anti-CD3/anti-CD28 for 6 h, T cells did not uniformly respond to activation. Rather, we could distinguish several types of responders based on their transcript levels. One population coexpressed IFN-γ transcript and protein (Fig. 4A, left panel; mRNA^hi), whereas another population lacked detectable IFN-γ mRNA but had produced protein during T cell activation (Fig. 4A, left panel; mRNA^lo). A third type of responders had intermediate levels of IFN-γ mRNA (RNA^int) with no detectable levels of IFN-γ protein.

To determine whether the mRNA^lo T cell population continuously produced protein during the course of T cell activation, we added monensin at different time points, that is, for the entire activation period as before, or for the last 4, 2, and 1 h of activation. This analysis revealed that the protein production of IFN-γ mRNA^lo CD8⁺ T cells was substantially reduced when monensin was added at later time points, showing that this T cell population ceased to produce protein in the last 1–2 h of activation (Fig. 4). Even though mRNA^hi–expressing CD8⁺ T cells also lost some IFN-γ protein signal when protein accumulation was only during the last hour of activation, the percentage of IFN-γ protein–expressing mRNA^hi T cells remained constant throughout the experiment (Fig. 4). Additionally, this mRNA^hi population could also benefit from some newly protein-producing T cells derived from the protein-negative mRNA^int T cells. However, provided that the loss of overall responding T cells in the last hour of monensin primarily derives from the loss of mRNA^lo protein–producing cells, it may not be the major contributor to the mRNA^hi T cell population. In conclusion, although mRNA^lo T cells lose their protein production at later time points, mRNA^hi-expressing T cells can maintain their protein production for a longer period.

We next examined whether mRNA^hi-expressing CD8⁺ T cells were also better cytokine producers. To this end, we used multiparameter staining to combine IFN-γ and TNF-α, or IFN-γ and IL-2, transcripts and protein staining in a single tube. This resulted in robust signals for both IFN-γ and TNF-α in PMA-ionomycin–activated CD3⁺ T cells (Fig. 5A). Also, the combined analysis of IFN-γ and IL-2 showed significant staining for mRNA and protein for both cytokines (Fig. 5B).

We next used anti-CD3/anti-CD28 stimulation to compare the protein production of T cells that have high versus low transcript levels for the cytokines. We first turned to CD8⁺ T cells. Because CD8⁺ T cells produced very limited amounts of IL-2 upon anti-CD3/anti-CD28 stimulation (0.84 ± 0.47%), we refrained from further analysis for this cytokine in CD8⁺ T cells. When we compared the IFN-γ protein levels of IFN-γ mRNA^hi and mRNA^lo T cells (Fig. 6A), we found that IFN-γ mRNA^hi CD8⁺ T cells generated significantly more IFN-γ protein than did mRNA^lo T cells (Fig. 6A, 6B). Likewise, TNF-α mRNA^hi CD8⁺ T cells also produced more TNF-α protein (Fig. 6C, 6D). Provided that transcript levels are a snapshot analysis after 6 h of stimulation and that the protein measurements detect all accumulated proteins produced throughout the entire stimulation, we conclude that mRNA^hi T cells are more potent cytokine producers.

The multiparameter analysis we employed in the present study also allowed us to determine the cytokine production of mRNA^hi CD8⁺ T cells for another cytokine. This comparison revealed that IFN-γ mRNA^hi CD8⁺ T cells also contained significantly more TNF-α transcripts and protein than did IFN-γ mRNA^lo CD8⁺ T cells (Fig. 6E, 6F). Similarly, TNF-α mRNA^hi CD8⁺ T cells expressed more IFN-γ transcripts and protein (Fig. 6G, 6H). Therefore, we conclude that for IFN-γ and TNF-α, mRNA^hi-expressing CD8⁺ T cells produce more cytokines than do their mRNA^lo counterparts.

We also determined the cytokine production profile of CD4⁺ T cells after anti-CD3/anti-CD28 stimulation. Because CD4⁺ T cells produced all three cytokines (Fig. 7A–C, left panels), we could also include IL-2 in this analysis.

Interestingly, CD4⁺ T cells displayed a similar behavior as CD8⁺ T cells in that high cytokine transcript levels translated into the highest levels of protein production of the respective cytokine. This was not only true for IFN-γ and TNF-α mRNA^hi CD4⁺ T cells (Fig. 7A, 7B), but also for IL-2 mRNA^hi CD4⁺ T cells (Fig. 7C).

Furthermore, when we determined the production of TNF-α and IL-2 in IFN-γ mRNA^hi CD4⁺ T cells, transcript and protein levels were significantly higher than in IFN-γ mRNA^lo T cells (Fig. 7D, 7G, Supplemental Fig. 2A–D). Similarly, the IFN-γ and IL-2 production was higher in TNF-α mRNA^hi T cells (Fig. 7D, 7H, Supplemental Fig. 2E–G), and the TNF-α production was higher in IL-2 mRNA^hi CD4⁺ T cells (Fig. 7F, 7I). Intriguingly, in accordance with our data (Fig. 5B) and with literature (16), IL-2 mRNA^hi CD4⁺ T cells do not produce more IFN-γ transcripts or protein (Fig. 7F, 7I, Supplemental Fig. 2H–J). This finding implies a differential regulation of IL-2 and IFN-γ production in this population. In conclusion, we show in the present study that Flow-FISH identifies CD4⁺ and CD8⁺ T cells with sustained and superior cytokine production.

In this study, we show that Flow-FISH can be used for multiparameter analysis of several cytokine transcripts and proteins levels in primary human T cells. It also allows the analysis of the cytokine transcript and protein profile in distinct T cell subsets in a single tube, without the need of FACS sorting.

The Flow-FISH protocol we present in this study entails several improvements to previously reported strategies (13–15). Fixing T cells with formaldeyde instead of methanol now allows the use of fluorochromes such as allophycocyanin and FITC. This is very valuable when Abs are only available with limited dyes, or require these fluorochromes to reach measurable signal strength. Furthermore, we found that Brilliant Violet (BV) and Brilliant Ultraviolet (BUV) dyes can be used with a very limited loss of signal. More fluorochromes may be suitable for Flow-FISH but will have to be tested. This optimized strategy allowed us to use eight parameters for analysis, but this panel can be extended to more parameters. Additionally, the Flow-FISH method presented in this study is rapid. Similar to classical ICCS, sample preparation requires ∼2 h after activation the first day. After the overnight hybridization with FISH probes, another 30 min is needed to prepare the samples for analysis. Because it also allows for single-tube analysis, it is suitable for diagnostic purposes, such as determining the status of Ag-specific T cell responses to chronic infections, or the efficacy of T cell vaccination strategies to tumor Ags.

Flow-FISH also showed the responsiveness of T cells based on cytokine transcript expression. Memory T cells maintain higher transcript levels for IFN-γ than do naive or effector T cells. Several mechanisms can drive the increased transcript levels in memory T cells. First, changes in transcription factors and gene accessibility drive increased cytokine transcription (18). Second, costimulatory signals such as CD28 and LFA-1 can stabilize cytokine mRNA (19, 20). In line with that, memory T cells also express higher CD28 levels (16). Third, primed T cells respond stronger to T cell signaling (21–23), which can lead to increased transcription and mRNA stability. However, whether these mechanisms act in concert or whether one pathway is the primary driver for maintained cytokine expression in memory T cells is yet to be determined.

Importantly, we discovered a dichotomy in T cells in their capacity to maintain high levels of cytokine transcripts. Whereas one responding T cell population maintained high mRNA levels during stimulation, the other one lost transcript levels and stopped producing the respective cytokine. Remarkably, high mRNA^hi T cells not only produced more of their respective protein, but they also expressed higher transcript and protein levels for other key cytokines. Therefore, Flow-FISH identified the most potent effector T cells. Interestingly, memory T cells contained more mRNA^hi T cells than did naive or effector T cells (Fig. 2B, 2C), suggesting that the differentiation status of T cells determines the transcript levels. Nevertheless, in vitro–activated, synchronized T cells also contain both mRNA^hi and mRNA^lo T cell populations. Therefore, cell-intrinsic differences such as a different metabolic state, or varying expression levels of transcription factors, RNA binding proteins, and miRNA, could contribute to this differential T cell response. As proposed for developmental cells (24), these cell-intrinsic differences could be imprinted or, alternatively, derive from oscillating expression patterns due to a cell-intrinsic clock. Whereas the underlying mechanisms mediating sustained transcript levels is to date unknown, driving T cell responses to contain these high transcript levels will help to increase T cell functionality.

Simultaneous cytokine production of IFN-γ, TNF-α, and IL-2 defines the most effective T cell responses against pathogens and tumors (25–27). We show that T cells that express mRNA^hi T cells are the most potent effector T cells. Using Flow-FISH to assess T cell response should therefore help to identify the most optimal feature for inducing T cell responses.

We thank E. Mul, M. Hoogenboezem, and S. Tol for FACS sorting, and M. Nolte, J. Freen, and F. Salerno for critical reading of the manuscript.

The authors have no financial conflicts of interest