PBMC samples were thawed in RPMI (Gibco) 45% FCS with DNAse I (0.1 mg/mL, Stemcell), washed twice and resuspended in 150 µL PBS with DNAse I (50U, Stemcell) for 10 min at RT. Cells were then washed and resuspended in 50 µL PBS with Fc receptor blocking agent (eBioscience) and live/dead Zombie UV (Invitrogen) for 20 min at 4°C. Cells were washed and stained in 100 µL of staining buffer (PBS, 2 mM EDTA, 2% FCS) containing antibodies for CD3, CD8α, TCRαβ, PD-1 and TIGIT and incubated for 30 min at 4°C. PD-1 detection for on-treatment samples was performed using antihuman IgG4 biotinylated (Sigma, 20 µg.mL).7 34 After two washes, cells were stained in 100 µL of staining buffer containing antibodies for CD28, HLA-DR and CD38 (panel 1) or CD62L, CD95, CD45RO and CCR7 (panel 2) or Tim-3, Lag3 and CXCR5 (panel 3) and secondary antibody (streptavidin) and incubated for 30 min at 4°C. Corresponding isotype antibodies were used as a control. Cells were then washed twice before acquisition on a BD ARIA II cell sorter (BD Bioscience). Data were collected using the BD FACS Diva Software. Four subpopulations were sorted based on the expression of PD-1 and/or TIGIT, among live cells, CD3^pos, CD8^pos, TCRαβ^pos T cells. Approximately 5× 10⁵ cells were directly sorted in PBS and used for RNA and TCR sequencing experiments. For HLA-A2*0201 patients, 5×10⁴ cells were sorted, centrifuged and resuspended in 5 mL of RPMI 1640 (Sigma), 8% human serum, interleukin (IL)-2 (Novartis, 150 U/mL) and PHA-L (1 µg/mL) (Sigma) for in vitro expansion according to a procedure previously described.35 Briefly, sorted T cells (50 µL) were seeded in 96 well plate containing irradiated feeder cells (10×10⁶ PBMC from two donors and EBV-transformed B cell-line). Amplified T cells were used for in vitro experiment after a 14 days amplification period.

CD3 (BV510, SK7), CD8α (BB700, HIT8a), PD-1 (BV421, EH12.1), CD28 (BB515, CD28.2), HLA-DR (PE, L243), CD38 (APC-R700, HIT2), CD62L (BV650, SK11), CD95 (PE, DX2), CD45RO (BB515, UCHL1), Tim-3 (BV650, 7D3), LAG-3 (APC-R700, T47-530), CXCR5 (BB515, RF8B2), CCR7 (APC-R700, 3D12) and streptavidin (BV421) reagents were purchased from BD Bioscience. TCRαβ (APC, IP26) and TIGIT (PE-Cy7, A15153G) antibodies were purchased from Biolegend.

The production of 25 cytokines was measured by Luminex assay (EPX250-12166-901, ThermoFisher Scientific). Briefly, 50 000 sorted T cells were amplified on feeder cells after flow cytometry-based cell sorting (baseline and M1 from five patients in cohort 1). After 14 days of culture, 100 000 cells from each subset were stimulated in duplicate with 0.5 ug.mL anti-CD3 antibody (coated, clone OKT3). Supernatants were collected after 12 hours of culture and analyzed according to manufacturer instructions.

Interferon (IFN)-γ ELISPOT assays was performed to assess tumor-antigen reactivity by sorted and amplified T-cell populations from HLA-A*0201 patients. T cells (1×10⁵) were seeded in triplicate in precoated ELISPOT 96 well plates (Mabtech) either with tumor peptide (10 µg/mL), viral peptide (10 µg/mL) anti-CD3 antibody (positive control) (Mabtech, 1 µg/mL) or without stimulation (negative control) and incubated 24 hours at 37°C and 5% CO₂. Plates were washed five times with PBS and incubated 2 hours at room temperature with 100 µL PBS 0.5% FBS with 1 µg/mL of biotinylated detection IFNγ. Plates were washed again five times with PBS before incubation with 100 µL of PBS 0.5% FBS streptavidin-HRP (Mabtech) for 1 hour. Plates were washed five times with PBS. TMB (tetramethhylbenzidine)-substrate solution (100 µL/well) and ELISPOT were developed for 15 min precisely. IFNγ spots were counted using an ELISPOT reader (Bio-Sys’ Bioreader). The number of spots forming units (SFU)/10⁵ cells was calculated from triplicates after subtraction of the negative control. An IFNγ ELISPOT was considered positive when superior to background and at least >30 SFU/ 10⁵ cells. All peptides were purchased from Genecust. Peptides are listed in online supplemental table S4.

Total RNA were extracted from sorted T cells using RNeasy Kit (QIAGEN). Total RNA quality and quantity were assessed using high sensitivity RNA 6000 pico kit (Agilent) and Agilent 2100 Bioanalyzer. RNA with RIN >7 were obtained for all samples. Total RNA were stored at −80°C until TCR and gene expression profiling.

Gene expression analysis by RNAseq was done using 10 ng of total RNA with QIAseq Targeted RNA Custom Panel (CRHS-10563Z-975 and CRHS-10707Z-977) with QIAseq Targeted RNA 96-Index kits (QIAGEN) according to manufacturer’s recommendations. Briefly, RNA is treated with DNase to remove any residual DNA, and the RNA is reverse transcribed into cDNA. In the first reactions a gene-specific primer with Unique Molecular Indexes (UMIs) is used in a single primer extension reaction to capture the gene of interest. Next, a gene-specific reverse primer and a universal primer is used to capture and amplify the region of interest. A unique sample dual index is then added using a PCR reaction.

Sequencing libraries were quantified with QIAGEN’s QIAseq Library Quant System and quality control was performed by capillary electrophoresis on a TapeStation System (Agilent Technologies) using a High Sensitivity D1000 Screen Tape. Each library was diluted and normalized to 4 nM according to the QIAseq Library Quant results prior to pooling equally and denaturing. The denatured library pool with a final concentration of 1.2 pM was run on a NextSeq High Output V2 kit (Illumina) using dual indexing, single end sequencing with 150 cycles and a Custom Read 1 sequencing Primer.

FASTQ files were generated from Illumina’s NextSeq Sequencing run and analyzed on QIAGEN’s GeneGlobe Data Analysis Center (https://www.qiagen.com/us/shop/genes-and-pathways/data-analysis-center-overview-page). GeneGlobe provides the primary read alignment and demultiplexed UMI values.

Raw count data were imported into R (V.3.5.2). The edgeR R package was used to calculate the normalization factors to scale the raw library sizes,36 followed by a voom transformation from the limma R package.37 38 Briefly, it transforms count data to log₂ counts per million (log₂ CPM), and estimates the mean-variance relationship to compute appropriate observation-level weights.

Differential expression analyzes were performed using the limma statistical framework and associated R package.38 39 A linear model was fitted to each gene, and empirical Bayes moderated t-statistics (two tailed) were used to assess differences in expression.39 Contrasts comparing time points, fractions and/or outcomes were tested. When needed, intraclass correlations were estimated to account for measures originating from the same patients.40 An absolute log₂-fold change cut-off of 1 and an false discovery rate (FDR) cut-off of 5% were used to determine DEGs.

Gene set enrichment analysis (GSEA) was performed using the R function Camera implemented within the limma R package.41 The same contrasts as above were investigated. An FDR cut-off of 5% was used to determine significant gene sets. Kyoto Encyclopedia of Genes and Genomes (KEGG), Hallmark, immunological signatures (c7) and curated pathways were used as gene sets. Those gene sets were downloaded from MSigDB database.42

We looked at the gene signature specific of PD-1 +TIGIT+ fraction at M1 in the bulk RNA-seq of 473 skin cutaneous melanoma tumors from TCGA. Using the R code and data provided by Jerby-Arnon et al.43 (https://github.com/livnatje/ImmuneResistance), we computed the overall expression of the upregulated differentially expressed genes between PD-1 +TIGIT+ and PD-1+, TIGIT+and PD-1-TIGIT-; and predicted the overall survival in TCGA melanoma patients using this signature. Kaplan-Meier curves were stratified by high (top 20%), low (bottom 20%) or intermediate (remainder) overall expression.

TCR-sequencing analysis cDNA librairies for TCR-seq analysis were prepared from 15 ng of total RNA using Human TCR Panel QIAseq Immune Repertoire RNA Library Kit (QIAGEN) according to manufacturer’s instructions and was described previously.44

Sequencing libraries were quantified with QIAseq Library Quant System (QIAGEN) and quality control was performed by capillary electrophoresis on a TapeStation System (Agilent Technologies). For sequencing, each library was diluted to 4 nM, pooled and denatured. Denatured library pool with 1.0 pM concentration was run on a NextSeq Mid Output V2 kit (Illumina) for asymmetrical pair-end 261×41 base read. FASTQ files were analyzed in the QIAGEN GeneGlobe Data Analysis Center as described previously.44

TCR sequence analysis was performed using R (V.3.5.2). The input data consisted of the observed number of each TCR (alpha or beta) sequence in each sample. Within each fraction, Fisher exact tests to calculate differential abundance of each TRBC (or TRAC) between two time points (T0 vs M1) were performed. The clonality score is derived from the Shannon entropy, which is calculated from the frequencies of all sequences divided by the logarithm of the total number of unique productive sequences. This normalized entropy value was then inverted (1—normalized entropy) to produce the clonality metric. Differentially abundant sequences (FDR 5%) were then divided into four clusters (emerging, expanding, contracting and non-expanding/contracting) according to their signs of the fold-change (negative or positive) between T0 and M1 and whether or not the sequence was present at T0.

Statistical significance was determined using one-way ANOVA or Mann-Whitney test as indicated (Prism V.8, GraphPad Software). Differences were considered statistically significant when p<0.05. Only significant p values are displayed. Unless otherwise specified, experiments were performed without duplicates due to material restrictions.

To assess the phenotype, function and clinical relevance of the peripheral CD8 T cell subset coexpressing PD-1 and TIGIT, we longitudinally isolated PBMCs from two cohorts of melanoma patients receiving anti-PD-1 therapy (13 patients in each cohort), and a cohort of MCC patients (n=15). Blood samples were obtained at baseline (before treatment initiation) and after 1 month of PD-1 inhibitor therapy, and also 2 months for the first melanoma cohort (online supplemental figure S1A). We sorted by flow cytometry four distinct CD8 T cell populations based on the expression of the inhibitory receptors PD-1 and TIGIT (ie, PD-1^-TIGIT^-, PD-1⁺TIGIT^-, PD-1⁺TIGIT⁺ and PD-1^-TIGIT⁺ sub-populations hereafter, respectively, referred as double negative (DNEG), PD-1, DPOS and TIGIT). The gating strategy is depicted in online supplemental figure S1B. Patients with partial and complete responses were classified as responders while non-responders exhibited SD or progressive disease (RECIST V.1.1). In the two melanoma and the MCC cohorts the frequencies of the four T-cell subpopulations within the entire CD8 peripheral T cell population were globally highly conserved across time (figure 1A) with the DNEG population being the most represented (46%±14% and 61%±19% in the two melanoma cohorts and 35%±13% in the MCC cohort). In melanoma cohorts, DPOS and TIGIT fractions are equally represented (globally around 20%), followed by the PD-1 fraction (13%±8% and 8%±7% in the two melanoma cohorts) (figure 1A, left and middle panels). In the MCC cohort, the TIGIT fraction appeared less represented (14%±10%) than the PD-1 (26%±14%) and DPOS (24%±13%) subpopulations (figure 1A, right). Strikingly, when responding and non-responding patients (online supplemental tables S1–S3) were analyzed separately in each patient cohort, the frequency of the DPOS subset was significantly higher in responding patients after 1 month of therapy, in contrast to the other subsets not significantly different according to clinical outcome (figure 1B–D). The frequency of the DPOS subpopulation thus appears predictive of anti-PD-1 efficacy at month 1 in the blood of melanoma and MCC patients.

We performed the same analysis at month 1 by grouping together patients from both melanoma cohorts (17 non-responding patients vs 9 responding patients), and confirmed results obtained on individual cohorts, namely that the frequency of the DPOS fraction at M1 was significantly associated with clinical responses (figure 1E). Finally, we performed receiver operator characteristic (ROC) curve analysis to evaluate the performance of the DPOS subset’s frequency in predicting response to PD-1 therapy with samples from the MCC cohort and the combined melanoma cohorts. The DPOS frequency appeared to predict response to PD-1 blockade significantly well with area under the curve values (AUC) of 0.96 for the MCC cohort (figures 1F and 5 non-responders and 10 responders, p=0.0048) and 0.78 for the melanoma cohorts (figures 1G and even non-responders and nine responders, p=0.0203), respectively. We also defined a cut-off for the frequency of the DPOS subset that could be used to predict response to PD-1 inhibitors. We identified 17.35% for melanoma patients (77.8% sensitivity and 75% of specificity) and 16.25% for MCC patients (100% of sensitivity and 80% of specificity) as potentially relevant values.

Based on this cut-off value, we further explored whether DPOS frequency was associated with overall survival in the two melanoma cohorts. As shown on figure 1H, the overall survival of patients with circulating DPOS frequencies >17% (green curve, n=11) is significantly better than that of patients with DPOS frequencies >17% (red curve, n=15, p=0.0019).

Thus, in three independent cohorts of patients and across two distinct pathologies, we identified the frequency of PD-1 +TIGIT+ CD8 T cells in the blood as a promising and early, cellular based, immune marker of response to PD-1 therapy and of overall survival.

We next characterized by flow cytometry the levels of PD-1 and TIGIT expression associated with the DPOS subpopulation, in the three patients’ cohorts. Notably, the median level of PD-1 expression was significantly higher on the DPOS population compared with the PD-1 single positive population at baseline and after initiation of the therapy (figure 2A, not significant at baseline for the first melanoma cohort p=0.11). The median level of TIGIT expression was similar between the DPOS and the TIGIT populations (online supplemental figure S2A). Coexpression of TIGIT with PD-1 on CD8 T cells was restricted to cells with high level of PD-1 expression, possibly due to strong activation. This finding is consistent with a previous report in non-small cell lung cancer (NSCLC) patients describing a correlation between TIGIT expression on CD8 TILs and peripheral lymphocytes expressing high levels of PD-1.45 A recent report also identified high levels of PD-1 expression on CD8 T cells in TILs in NSCSLC patients as a feature predictive of PD-1 therapy outcome, consistent with our data in other cancer types.15 Tim-3 and Lag-3 expression were weakly detected in the four different populations (assessed on the first melanoma cohort) confirming previous reports19 23 (online supplemental figure S2B and S2C). Our results highlight again the different expression patterns for inhibitory receptors on peripheral CD8 T cells vs TIL, and the special relevance of monitoring PD-1 and TIGIT coexpression on circulating CD8 T lymphocytes.

Observation of the immunological response to PD-1 blockade in the blood of cancer patients has notably been described by a proliferative burst of CD8 T cells expressing the intracellular proliferation marker Ki67.26 34 46 The combined expression of the ectoenzyme CD38 and HLA-DR at the T-cell membrane strongly correlates with Ki67 expression on vaccine-induced T cells34 47 and was used to determine what T-cell fraction contributes to the proliferative burst in vivo following anti-PD-1 therapy. We found that HLA-DR/CD38 coexpression was largely restricted to the DPOS T-cell fraction in the three cohorts at baseline and we observed a marked increase in frequency of HLA-DR⁺CD38⁺ cells following PD-1 blockade (figure 2B and online supplemental figure S1C, upper panel). Furthermore, for the MCC cohort of patients, the frequency of HLA-DR⁺CD38⁺ cells was significantly higher within the DPOS subset compared with the three other populations after only one cycle of therapy (figure 2B, right panel). Thus, PD-1 and TIGIT coexpression, rather than PD-1 alone, in the blood of melanoma and MCC patients receiving anti-PD-1 therapy identifies a CD8 T cell subset enriched for HLA-DR and CD38 coexpression that increases markedly in frequency in the very first weeks of therapy, and this increase is associated with clinical outcome.26 34 46

Recent studies identified a CXCR5⁺ population of CD8 T cells as the pendant of CD4 Tfh named cytotoxic Tfc that localizes in secondary/tertiary lymphoid organs.25–31 We, thus, investigated longitudinally CXCR5 expression on the 4 T-cell subpopulations from the three cohorts of cancer patients. Again, CXCR5⁺ cells were largely confined to the DPOS population with significantly higher frequencies in comparison to the DNEG and PD-1 populations regardless of the time point for the three cohorts (figure 2C and online supplemental figure S1C, lower panel) and to the TIGIT single positive population for the melanoma validation cohort and the MCC cohort after initiation of PD-1 therapy (figure 2C, middle and right panels). While described as very transiently detectable in the blood of mice in another study (present at D8 and undetectable at D3026), here the increased frequency of CXCR5⁺ cells within the DPOS T cell population within the blood remained stable until M2 (figure 2C, left panel). Nonetheless, the expression of these markers (HLA-DR/CD28 and CXCR5), while appearing to be a characteristic of this subpopulation, only occurs in a fraction of these cells, which also suggests that this DPOS T cell subpopulation is heterogeneous, possibly consisting of a mixture of activated/exhausted T cells and of Tfc like T cells.

We performed a more complete analysis of the differentiation status of the T cell subsets on the 13 melanoma patients from the original cohort. Naïve T cells (CD45RO^-CCR7⁺CD62L⁺CD95^low) were almost exclusively present in the DNEG population at all time points as expected (online supplemental figure S5D, left panel). The distribution of TEM (CD45RO⁺CCR7^-CD62L^-CD95⁺) and TEMRA (CD45RO^-CCR7^-CD62L^-CD95⁺) is totally reversed in the 4 T cell-subpopulations, with the majority of TEM present in the PD1+ and DPOS fractions (>60% at baseline), and the majority of TEMRA detected in the TIGIT+and DNEG fractions (around 60% at baseline) (online supplemental figure S2D) middle and right panel). CD28 costimulatory receptor was described as the primary target of PD-1 mediated T-cell inhibition and T-cell reinvigoration on PD-1 blockade.46 48 Consequently, we assessed CD28 expression longitudinally on the 4 T-cell subpopulations of interest on the first melanoma patients’ cohort and the MCC patients’ cohort. Before therapy, DNEG and PD-1 populations expressed CD28 molecules at very high frequencies (around 80% online supplemental figure S2E and S2F). We observed a slightly reduced frequency of cells expressing CD28 on the DPOS population on the two cohorts (not significant) and another marked and significant reduction for the TIGIT population (around 30%). Interestingly, CD28 expression was not altered on the different T-cell subpopulations on PD-1 blockade. Altogether, analysis of the differentiation status and CD28 expression profile suggest that the TIGIT population could represent a more differentiated T-cell population.

Altogether, these findings describe PD-1⁺TIGIT⁺ CD8 T cells as the main proliferative T-cell population in response to PD-1 therapy and exhibit an effector-like phenotype, a greater number of PD-1 molecules at the membrane, a maintained CD28 expression and a higher frequency of T-cells expressing the Tfc marker CXCR5. Strikingly, the increase in DPOS frequency at M1 within the entire peripheral CD8 T cell population was predictive of anti-PD-1 therapeutic efficacy in three independent cohorts and across two distinct cancer pathologies. These findings prompted us to characterize the DPOS CD8 T-cells in greater depth

We then performed targeted RNA-sequencing of 975 genes related to gene activation on the different T-cell subsets sorted before therapy and at M1 and M2 from the blood of our initial cohort of 13 melanoma patients. Multidimensional scaling markedly distinguished the DNEG population from the three other T cell subsets showing that the single expression of at least one of the inhibitory receptors PD-1 or TIGIT was sufficient to distinguish gene expression profiles from DNEG CD8 T cell population within the blood of melanoma patients (figure 3A). The PD-1 population also clustered independently from the others, even from the DPOS subset, suggesting particular features of DPOS cells that have significantly higher level of PD-1 expression by flow cytometry. In contrast, transcriptomic profiles from the two populations sorted based on the expression of TIGIT (ie, DPOS and TIGIT populations) were largely overlapping suggesting acquisition of a common gene expression program in parallel of TIGIT expression.

There was no DEG detected for a given sorted population across time points (exception of MYO7A gene in the DNEG population between T0 and M1, (online supplemental figure S3A and S3B). The major differences in gene expression profiles were then observed between the different subpopulations disregarding the time point (online supplemental figure S3C). At M1, we first confirmed that the expression profiles of the PDCD1 and TIGIT genes were consistent with the phenotypic characteristics of the four subpopulations (figure 3B). The DPOS population displayed higher expression of MHC-II related genes (HLA-DPA1, HLA-DQA1, HLA-DRA, HLA-DRB1, CD74) and genes associated with cell proliferation, cell cycle and cell division (CDKN3, CDK1, CCNA2, CCNB2, UBEC2C) hence describing a population of highly activated and proliferating T cells (figure 3B,C). The DPOS population expressed high transcript levels of IFNG and GZMK gene transcripts but intermediate levels of GZMH, GZMB, GZNA and tumor necrosis factor (TNF) transcripts. IL-10 transcript was overexpressed by the DPOS population, compared with the DNEG subset, suggesting acquired altered functional profile (figure 3C). Consistent with higher frequency of HLA-DR⁺CD38⁺ cells observed by flow cytometry, the proliferation marker MKI67 was upregulated at the mRNA level on the DPOS population compared with the three other subsets (figure 3B,C). Regarding costimulatory molecules, CD28 expression was similar between DPOS, PD-1 and DNEG population and was lower on TIGIT population confirming flow cytometry data (figure 3C). CD27 expression was also maintained on DPOS and PD-1 populations but reduced on TIGIT population (figure 3C). CD27 was notably described as a key feature of a CAR T-cell population responsible for tumor control49 and was overexpressed by PD-1^high TILs associated with response to PD-1 blockade in NSCLC.15 Finally, 4-1BB (TNFRSF9) was overexpressed on the DPOS population and was also higher on the TIGIT population compared with PD-1 and DNEG ones (figure 3C). Tumor-reactive T cells, including neoantigen-specific T cells, have been successfully enriched from the blood or the tumor of cancer patients using 4-1BB, PD-1 and CD39 markers.50–54 The DPOS population overexpressed all of these markers again suggesting recent tumor-antigen encounter. Furthermore, several genes related to immune T-cell trafficking (CCR5, CXCL13, CCL4L2, CCL3) were also upregulated on the DPOS population compared with PD-1 and TIGIT populations (figure 3B). Interestingly, CXCL13 and its receptor CXCR5 were significantly upregulated on the DPOS population, compared with all the other subsets, consistent with cytometry results for CXCR5 expression (figures 3C and 2C). CXCR3, a chemokine receptor for the IFN-inducible chemokines CXCL9, CXCL10 and CXCL11 and demonstrated as required on CD8 TILs for effective antitumor response under PD-1 therapy55 56 was upregulated only on the peripheral DPOS population in comparison to the DNEG population (figure 3C). The DPOS population also over-expressed the transcription factor TOX (figure 3B), recently described as hallmark of T cell exhaustion.57–59 These results highlight that the DPOS T cell subset is likely to be a heterogeneous population consisting of relevant T cells with features described as restricted to progenitor stem-cell like cells but also activated/exhausted T cells.

In contrast, PD-1 and TIGIT populations exhibited an intermediate transcriptomic profile regarding these gene clusters. Surprisingly, TIGIT population highly upregulated several immunoregulatory receptors classically expressed on NK cell subset (KLRC2, KLRC3, KLRC4, KLRD1, KLRF1, KIR3DL1) and the transcription factor TBX21. These results, together with the large proportion of TEMRA cells observed in this population (online supplemental figure S2D) is consistent with a more terminally differentiated state (figure 3B).

Two recent reports described a population of dysfunctional TILs with predictive potential, identified by high PD-1 and CXCL13 expression, with impaired T cell function, high proliferative potential and a distinct transcriptional profile.14 15 Given the high level of similarity with the DPOS population described here, we performed GSEA using the 23-gene signature upregulated in these TILs populations. The DPOS population was strongly enriched for the two different gene signatures in contrast to PD-1 and TIGIT populations (figure 4A). Consequently, the DPOS population from the blood of melanoma patients contained T cells that can be described as the pendant of PD-1^high TIL population found in NSCLC patients.

We next defined a gene signature specific to the DPOS population based on DEGs that overlapped when comparing DPOS transcriptomic profile to DNEG, PD-1 and TIGIT subsets at M1 (figure 4B). This DPOS-specific program contained 12 upregulated genes with notably PDCD1, CXCL13, CXCR5 and MKI67 (complete list online supplemental table S5). Despite being expressed at similar levels between the DPOS and TIGIT populations (online supplemental figures S2A and S3D), we decided to include TIGIT within this gene signature as this marker was used to sort the different cell subsets. We tested the ability of this DPOS gene signature to predict clinical outcome in melanoma patients with bulk RNA-seq from TCGA. Our DPOS gene signature was highly predictive of survival in 473 melanoma patients, even out of any immunotherapy context (p=2.2×10^-4, log rank=5.4×10^-5) (figure 4C).

Furthermore, analysis according to the clinical response (NR vs R) described CXCL13 as the only gene differentially upregulated in responding patients within the DPOS population at T0 strengthening the crucial role of this chemokine in driving an effective anti-PD-1 immune response (p=0.040, figure 4D). Moreover, despite the limited number of genes analyzed, gene enrichment pathway analysis revealed notably the cytokine-cytokine receptor interaction pathway as the most up-regulated pathway within the DPOS subset in responding patients compared with non-responding ones (p=9.88×10^-5, FDR=0.011) suggesting that high migration capacities for the DPOS population constituted an important feature for therapeutic efficacy (figure 4E).

Thus, the transcriptomic analysis revealed a specific gene signature for the circulating DPOS T cell population that is predictive of response to PD-1 therapy, strongly associated with survival in melanoma patients, and comparable to that of TILs with predictive potential in NSLCC.

Transcriptomic analysis of DPOS T cells highlighted the expression of genes associated with exhaustion. We, therefore, compared the reactivity of the four sorted fractions from five patients from the first cohort (three NR and two R), at baseline and M1, based on their ability to produce cytokines on anti-CD3 stimulation. We detected the production of 10 cytokines among the 25 tested, GM-CSF being the most produced cytokine by these CD8⁺ T cells subpopulations. IFN-γ, TNF-α, IL-13 and IL-4 were also strongly produced by all the T cell subpopulations (online supplemental figure S4). No significant differences were observed between the different subsets, although cytokine production by DPOS T cells appeared slightly lower than that of other T cell subpopulations. These results describe, at the bulk level, that T cells isolated from the four fractions contain functional and reactive T cells and could be classified as classical pro-inflammatory CD8⁺ T cells nonetheless this method doesn’t resolve intrasubset heterogeneity. Therefore, the possibility that DPOS T cells contains a fraction of exhausted-like T cells with altered cytokine production remains.

Nonetheless, the association between the high frequencies of the DPOS population in the blood after 1 month of therapy and clinical responses prompted us to hypothesize that this subset could be enriched in recirculating tumor-specific T cells. As tumor material was not accessible for this study, we assessed the tumor reactivity of the four sorted subpopulations from 7 HLA-A*0201 patients at T0 and M1 after in vitro expansion against a panel of 11 peptides derived from melanoma antigens (online supplemental table S4).60 The total number of responses was increased following anti-PD-1 therapy (19 tumor-specific responses detected at T0 and 29 at M1) (figure 5A). Notably, the number of responses observed in the DPOS T cell population was more than doubled (7 vs 15 responses at T0 and M1, respectively, figure 5A,B). An example of ELISPOT images is shown in online supplemental figure S4C, from P19 patient. PD-1 blockade also diversified the antigens recognized within this subset (6 vs 9 epitopes at T0 and M1, among them melanocytic differentiation, tumor-specific and the overexpressed antigen PRDX5) (figure 5B). Nonetheless, tumor reactivity was not restricted to the DPOS fraction as PD-1 and TIGIT populations also depicted antigen recognition but less amplified and diversified on PD-1 blockade (figure 5B). Notably, diversity and numbers of responses detected within the PD-1 population moderately decreased following PD-1 therapy (6 vs 4 recognized tumor epitopes). This could result to a conversion of the tumor-reactive T cells from PD-1 to DPOS population exemplified in P15 and P19 patients for Melan-A_A27L, gp100_YLE, SSX2_KAS and NA17-A_VLP specific responses (figure 5B). The DNEG population displayed two responses pretherapy and three at M1 all modest in magnitude and specific for the melanocyte differentiation antigen melan-A or the tumor-specific antigen LAGE-1 (figure 5B,C). The TIGIT population displayed a more diverse range of antitumor responses with three new antigens recognized after PD-1 blockade but both the magnitude and number of these responses remained smaller than for the DPOS population (figure 5A,B). One patient among the 7 HLA-A*0201 developed a CR (P8) and we observed the development of 4 newly detected antitumor responses at M1 within the DPOS population (figure 5A,B).

Our data shows that antigen-specific T cell responses within DPOS T cells were diversified on PD-1 blockade that could be the reflection of the effective stimulation of new highly reactive tumor-specific T cells, associated with clinical benefit. This prompt us to assess dynamic changes within the TCR repertoires of the DPOS and other 3 T-cell populations distinguished by different inhibitory receptor expression patterns. We performed TCRα and TCRβ sequencing analysis of the four sorted T-cell populations (11 patients from the first cohort: 7 NR and 4 R) at T0 and M1. Clonality analysis revealed the DNEG repertoire as the most diverse at both timepoints. PD-1 and DPOS populations exhibited higher and similar clonality while TIGIT population displayed a further reduction in T-cell repertoire diversity (figure 6A). Frequency analysis of the most common TRAC and TRBC clonotypes again demonstrated greater clonal expansion within the DPOS, PD-1 and TIGIT repertoires (online supplemental figure S5A). At baseline, the top 10 TRBC clonotypes represented 18,8% of the DNEG repertoire but 45.4%, 44.8% and 53.8% of the PD-1, DPOS and TIGIT repertoires, respectively, with very similar results at M1 and for TRAC repertoires. This clonal enrichment is likely to be driven by clonal expansion following TCR engagement and these global analyzes could not reveal any effect of PD-1 blockade on these characteristics.

As expected, T-cell repertoires were largely patient specific as measured by Morisita’s overlap index (online supplemental figure S5B). But more surprisingly, when we analyzed the overlap for the TRAC and TRBC repertoires between the four sorted subsets in individual patients, T-cell repertoires were also mostly private with few shared clonotypes between fractions (figure 6B and online supplemental figure S5C). Indeed, less than 7% of TRBC clonotypes were shared between at least two different fractions at M1 clearly underlying the distinct origins of the majority of these four populations. Nonetheless, the highest number of shared clonotypes were found between DPOS and TIGIT subsets (figure 6B and online supplemental figure S5C), consistent with the highest similarity between the DPOS and the TIGIT fractions observed through RNAseq analyzes (figure 3A).

We next defined four different clusters of clonotypes within each T-cell repertoire based on their statistically significant expansion or contraction profile on PD-1 therapy between T0 and M1: cluster 1 of emerging clonotypes (not detected at baseline but detected at M1), cluster 2 of expanding clonotypes (present at baseline and expanded in frequency on PD-1 blockade), cluster 3 of contracting clonotypes (with a decreased frequency between baseline and M1) and cluster 4 of non-expanding/non-contracting clonotypes that are not affected during the first month of PD-1 therapy. Typical signatures for each cluster are illustrated in figure 6C, upper panel. Overall, the global distribution of clonotypes within these clusters was similar for the 4 T-cell subsets. Cluster four was the most frequent class for all fractions while expanding clonotypes (cluster 2) represented the less frequent one (figure 6C). Emerging and contracting clonotypes (clusters 1 and 3, respectively) depicted similar frequencies illustrating typical behavior for a given repertoire where new clonotypes emerge replacing former ones within the entire repertoire (either following standard homeostasis mechanisms or T-cell priming and recirculation). Interestingly, the absolute numbers of these two classes of clonotypes appeared slightly higher in the DPOS population compared with the PD-1 and TIGIT repertoires (online supplemental figure S6A). Consistent with its higher diversity, the absolute number of clonotypes was higher in clusters 1, 3 and 4 for the DNEG repertoire (online supplemental figure S6A). We finally explored the distribution of the 4 clusters of clonotypes in the different subpopulations according to clinical outcome (figure 6D). Strikingly, cluster 1 of emerging clonotypes depicted significantly higher frequency for the DPOS subset in patients responding to PD-1 therapy (p=0.040). The importance for newly detected tumor-resident clonotypes in the peripheral blood was notably recently highlighted to prevent relapse in melanoma patients receiving ipilimumab and nivolumab in neoadjuvant settings.61 Accordingly, the absolute number of emerging clonotypes in the DPOS population was also higher in responding patients (p=0.06) (online supplemental figure S6B). A similar trend was observed for the PD-1 fraction, although not significant, nor in frequency or absolute number, illustrating again the preponderant importance of the DPOS population for anti-PD-1 clinical efficacy rather than PD-1 single expressing repertoire. Balancing this emergence of clonotypes, the frequency and absolute number of contracting clonotypes were slightly higher in the DPOS repertoire in responding patients (figure 6D and online supplemental figure S6B). Finally, the frequencies of non-expanding/non-contracting clonotypes (cluster 4) were lower in DNEG, PD-1 and DPOS repertoires in responding patients while TIGIT repertoire distribution remained largely unchanged. This change in DNEG repertoire was balanced within the entire repertoire by a higher frequency of contracting clonotypes and could delineate the conversion of DNEG T cells to PD-1 or DPOS repertoire following antigen encounter as suggested in a recent study.62

In conclusion, we found that cluster 1 of emerging clonotypes within the DPOS subset identified by the coexpression of PD-1 and TIGIT were associated with clinical responses. As the DPOS fraction was enriched in tumor-antigen specific T lymphocytes, we hypothesize that these emerging clonotypes could contain a fraction of antigen-specific T cells recirculating to the periphery after activation following anti-PD-1 therapy.