For the analysis of CD8 TILs melanoma patients we downloaded processed gene expression data [log₂ (TMP+1)] from NCBI GEO [accession GSE120575, (16)]. Only non-lymph node samples were considered. High-quality cells were filtered based on number of detected genes (between 1,000 and 6,000) and percentage of TPMs mapping to ribosomal genes (<10%) (1482 TILs from 23 patients). PD-1+ CD8 T cells were further filtered from other immune cell infiltrates based on the co-expression of CD2, CD8A, CD8B, PDCD1 [log₂ (TMP+1) > = 1], lack of CD4 expression (TPM = 0) and low expression of non-T cell genes FCER1G, TYROBP, SPI1, IGKC, IGJ, IGHG3 [log₂ (TMP+1) <3]. Next, we performed differential gene expression analysis between TCF7+ [log₂ (TMP+1) > 0] and TCF7- (TPM = 0) cells (cutoff based on the TCF7 expression distribution of Supplementary Figure 1), using Seurat v2.3.4 and MAST v1.8.2 (17) with parameters min.pct = 0.1, min.diff.pct = 0.01, logfc.threshold = 0.1. For visualization, in the human volcano plot (Figure 1A) we truncated log (fold-change) values at 3 and –log (p-value) at 25. For human TIL virtual gating strategy (Figures 1B–D) we used 1 UMI as a cut-off value to define PDCD1+, HAVCR2+, ENTPD1+ and SLAMF6+ cells.

For analysis of murine melanoma CD8 TILs we used processed data (UMI count matrix) from NCBI GEO [accession GSE116390, (18)]. Briefly, gene expression values were normalized and scaled as log(UMI counts / sum UMI count in cell ^* 10,000 + 1). Next, differential expression between memory-like/progenitor exhausted subset vs. terminally exhausted subsets [as defined in Carmona et al. (18)] was conducted using Seurat and MAST with parameters min.pct = 0.1, min.diff.pct=0.1, and logfc.threshold = 0.25. For the murine TIL virtual gating strategy (Figures 1B–D), we used 1 UMI as a cut-off value to define Pdcd1+, Havcr2+, Entpd1+, and Slamf6+ cells.

C57BL/6 (B6) mice were purchased from ENVIGO (Gannat, France) and CD45.1 OT1 mice were bred at the animal facility (Epalinges) of the University of Lausanne. Mice were all females, 7–10 weeks old, and maintained in conventional facilities of the University of Lausanne. This study was approved by the Veterinary Authority of the Swiss Canton of Vaud (authorization VD3360.c) and performed in accordance with Swiss ethical guidelines.

The B16.OVA cell line was generated as previously described [refers to B16.N4 (19)]. Cells underwent a maximum of two passages from thawing to the time of engraftment and were cultured in complete DMEM (10% FCS (10270-106, Gibco), 1% Penicilin/Streptomycin (15070-063, Gibco), 0.1% β-mercapto ethanol (31350-010, Gibco), 1% Hepes (5-31F00-H, Amimed) in DMEM (31966-021, Gibco).

2 × 10⁵ B16.OVA cells were subcutaneously (s.c.) engrafted on the right flank of B6 mice. After 6 days, 10⁵ CD45.1 OT1 T cells were intravenously (i.v.) transferred. One day later, mice were s.c. vaccinated with 10 μg SIINFEKL (N4) peptide (Protein and Peptide Chemistry Facility, UNIL) and 50 μg CpG (ODN 1826, Microsynth).

Tumors were harvested 21 days post tumor engraftment and dissociated with the Tumor Dissociation Kit (130-096-730, Miltenyi Biotec) following manufacturer's instructions before antibody staining for flow cytometry. Tumors of 6–10 individual mice were pooled before marker based cell sorting and transfer to secondary hosts.

Three hundred flow cytometry-sorted naïve CD8+ CD45.1 cells from spleens of OT1 mice, or 300 OT1 cells (PD-1+, PD-1+ Tim3– Slamf6+ CD39–, or PD-1+ Tim3– Slamf6+ CD39+) isolated from tumors of B16.OVA tumor bearing mice, were i.v. transferred to naïve B6 secondary hosts. The same day, mice were infected i.v. with 2000 colony forming units (cfu) of OVA expressing Listeria monocytogenes (Lm-OVA).

Seven days post-infection, spleens from infected mice were collected, smashed through a 70 μm diameter cell strainer (352350, Falcon) and red blood cells were lysed with RBC lysis buffer (158904, Qiagen) before antibody staining for flow cytometry analysis of OT1 cells.

Mouse splenocytes were stimulated with 10 μg/ml SIINFEKL peptide or 10 ng/ml PMA (Sigma-Aldrich) and 500 ng/ml ionomycin as positive control (Sigma-Aldrich) for 4 h at 37°C in complete DMEM. Unstimulated cells were used as negative control. BD GolgiPlug and GolgiStop (BD Biosciences) were added to the cells 30 min upon starting the stimulation assay.

Surface antibody staining was performed in darkness for 20 min at 4°C in FACS buffer (2% FCS, 2 mmol/L EDTA in PBS). Cells were stained with the following antibodies: CD3-AF700 (cl. 17A2, eBioscience), CD8-PerCp-Cy5.5 (cl. 53-6.7, eBioscience), CD45.1-BV650 (cl. A20, BioLegend), PD-1-BV711 (cl. 29F-1A12 BioLegend), Tim-3-BV421 (cl. RMT3-23, BioLegend), CD39-PE-Cy7 (cl. Duha59, BioLegend), and Slamf6-APC (cl. 330-AJ, BioLegend). LIVE/DEAD fixable Aqua fluorescent reactive dye (L34957, Thermo Fisher Scientific) or Fixable Viability Dye eFluor 506 (65-0866-14, Thermo Fisher Scientific) was used for dead cell discrimination. For intracellular Tcf1 staining, cells were fixed and permeabilized with Foxp3/Transcription Factor Staining Buffer Set (00-5523-00, eBioscience) according to the manufacturer's instructions and stained for 1 h at 4°C with Tcf-1-Rabbit (2203, Cell Signaling) followed by 15 min Fab aRabbit-PE-TexasRed staining (cl. Poly464, BioLegend), both in Perm/wash buffer. For intracellular cytokine staining, cells were fixed (Fixation buffer; 420801 BioLegend), and permeabilized (Perm/wash buffer; 421002 BioLegend) before being stained with IFNγ-PerCp-Cy5.5 (cl. XMG1.2, eBioscience) and TNFα-APC-Cy7 (cl. MP6-XT22, BioLegend). Samples were analyzed with LSRII flow cytometer (BD Biosciences). For flow cytometry cell sorting, cells were sorted with FACS Aria (BD Biosciences) following the gating strategy from Supplementary Figure 3A and collected in complete DMEM.

Flow cytometry data were analyzed with FlowJo (TreeStar). Graphs and statistical analysis were made with Prism (GraphPad Software) except scRNAseq data which was analyzed with R (described above). Specific statistical analyses are described in figures' captions. P-values are coded as ^*p > 0.05, ^**p > 0.01, ^***p > 0.001, and ^****p > 0.0001.

To identify surface markers of tumor-specific Tcf1+ PD-1+ Tpe that would allow their selection for ACT, we performed a differential gene expression analysis of Tcf7+ Pdcd1+ (Tpe) vs. Tcf7- Pdcd1+ (Tex) from scRNAseq data of CD8 TILs from murine B16 (18) and human primary melanoma tumors [publicly available data from Sade-Feldman et al. (16)] (Supplementary Tables 1, 2). In both mouse and human melanoma tumors we observed that Tcf7+ Pdcd1+ Tpe were enriched in Slamf6 and depleted of Havcr2 and Entpd1, which encode Slamf6, Tim3, and CD39 cell surface proteins, respectively (Figure 1A). Thus, we hypothesized that the expression patterns of Slamf6, Tim3, and CD39 may help discriminating Tpe from Tex among tumor-specific TILs.

First, we explored the plausibility of this hypothesis in silico. A previous report already showed that tumor-specific (identified by tetramer staining) Tim3- Slamf6+ cells were enriched in Tcf7+ cells compared to Tim3+ Slamf6- counterparts (7). Here we used Pdcd1 expression (which encodes for PD-1) as a surrogate marker of tumor specificity (14) and observed that indeed, Pdcd1+ Havcr2– Slamf6+ cells were enriched in Tcf7+ cells compared to bulk Pdcd1+ or Pdcd1+ Havcr2– cells (Figure 1B). However, among Pdcd1+ Tim3– Slamf6+ cells we found heterogeneous expression of Entpd1, a marker of Tex, both in murine and human melanoma tumors (Figure 1C).

In fact, Pdcd1+ Tim3– Slamf6+ Entpd1− cells showed increased enrichment for Tcf7+ cells compared to their Entpd1+ counterparts (Figure 1D). Thus, at the mRNA level, Entpd1– (CD39) Havcr2– (Tim3) Slamf6+ Pdcd1+ cells show the highest enrichment for Tcf7+ cells.

Next we sought to verify whether the surface expression of these markers could be used to identify Tpe cells by flow cytometry. To this end, we analyzed Tcf1 expression on total PD-1+ endogenous and transferred OT1 CD8 TILs from B16.OVA tumors before and after marker-based cell selection. We confirmed that CD39– Tim3– Slamf6+ PD-1+ cells showed the highest enrichment for Tcf1+ cells in both endogenous and transferred tumor-specific OT1 cells (Figure 1E), which accounted for the ~3.5% of total endogenous or OT1 TILs (Figure 1F). CCR7 was also highly expressed in Tpes compared to Tex cells (Supplementary Tables 1, 2) but antibody staining of this chemokine receptor was poor and did not select Tcf-1+ CD8 T cells in B16.OVA tumors (Supplementary Figure 2A).

In conclusion, selection of CD39– Tim3– Slamf6+ PD-1+ TILs by surface antibody staining allows for maximum enrichment of Tcf1+ tumor-specific Tpe.

Tpe cells have been described as those cells that maintain antigen-specific responses in the context of chronic infection and tumors (8, 9). Therefore, highly homogenous Tpe cell products would be ideal in the context of ACT to improve persistence of transferred cells in the hosts. Thus, we investigated whether our newly described CD39– Tim3– Slamf6+ PD-1+ TIL subset display better recall responses and persistence in vivo compared to bulk TIL populations. To this end, either 300 PD-1+, PD-1+ Tim3– Slamf6+ CD39–, or PD-1+ Tim3– Slamf6+ CD39+ OT1 CD8 TILs cells were isolated from B16.OVA tumors 21 days post-tumor engraftment following the gating strategy of Supplementary Figure 3A and transferred to a secondary host that was infected on the same day with OVA-expressing Listeria monocytogenes (Lm-OVA). Naïve OT1 cells were also transferred as control (Figure 2A). Seven days post-infection (PI), we observed increased frequencies of transferred OT1 cells in mice receiving PD-1+ Tim3– Slamf6+ CD39– cells compared to those receiving their CD39+ counterparts or total PD-1+ OT1 CD8 TILs (Figure 2B), indicating their enhanced recall response and persistence. At 7 days PI, most transferred cells were Tcf1– (Figure 2C), consistent with previous reports showing that tumor-specific Tcf1+ cells differentiate into Tcf1– negative cells (7, 8).

In addition to the enhanced expansion capacity, PD-1+ Tim3– Slamf6+ CD39– cells showed enhanced functionality in the recall response compared to their CD39+ counterparts, with increased frequency in IFNγ, TNFα and IFNγ+TNFα double expressing cells (Figure 2D) as well as increased expression of these molecules (Figure 2E).

Altogether, selection of CD39– Tim3– Slamf6+ cells among PD-1+ TILs led to an enhanced persistence and functionality of transferred TILs in a recall response compared to their CD39+ counterparts or bulk PD-1+ TILs.

PR receives speaker honoraria from Bristol-Myers Squibb and Roche and is the recipient of research grants in immune-oncology from Roche pRED, Zurich, CH. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.