PGE2 limits effector expansion of tumour-infiltrating stem-like CD8+ T cells – Nature

Mice

All mice used in this study were on a C57BL/6J genetic background and purchased from the Jackson Laboratory (JAX). OT-I × CD45.1 mice were generated by crossing OT-I mice (JAX, 003831) to CD45.1 (JAX, 002014) mice. Ptger2^−/−Ptger4^fl/fl mice were generated by crossing Ptger2^−/− mice (JAX, 004376) to Ptger4^fl/fl mice (JAX, 028102) and further crossed to Cd4^cre mice (JAX, 022071) to generate Cd4^crePtger2^−/−Ptger4^fl/fl mice or crossed to Gzmb^cre mice (JAX, 003734) to generate Gzmb^crePtger2^−/−Ptger4^fl/fl mice. Unless stated otherwise, mice were on a CD45.2/CD45.2 background. For some experiments, Cd4^crePtger2^−/−Ptger4^fl/fl mice and Ptger2^−/−Ptger4^fl/fl mice were crossed to OT-I mice to generate Cd4^crePtger2^−/−Ptger4^fl/fl OT-I mice and Ptger2^−/−Ptger4^fl/fl OT-I mice and used on a CD45.1/CD45.2 or CD45.1/CD45.1 background. WT or Rag1^−/− mice (JAX, 002216) on a CD45.2/CD45.2 background were used as recipients in adoptive transfer experiments. In all experiments, mice at 6–12 weeks of age were sex-matched and randomly assigned to control or treatment groups. Mouse experiments with Ptgs1/Ptgs2^−/− BRAF^V600E tumours and T cell depletion were conducted without blinding; all other experiments were performed in a blinded manner. No statistical methods were used to predetermine sample sizes. Mice were killed by cervical dislocation under anaesthesia. All mice were maintained and bred at the Klinikum rechts der Isar, TUM, or at the Klinikum der Universität München, LMU, under specific-pathogen-free, controlled conditions with a 12-h light–dark cycle, ambient temperature of 24 °C and humidity maintained at 55%, and in accordance with the guidelines of the Federation of European Laboratory Animal Science Associations. All animal experiments were performed in accordance with the guidelines of the district government of upper Bavaria (Department 5–Environment, Health and Consumer Protection).

Cell lines

Control and Ptgs1/Ptgs2^−/− BRAF^V600E melanoma cells were generated using the CRISPR–Cas9 system as previously described¹⁴. BRAF^V600E-OVA and Ptgs1/Ptgs2^−/− BRAF^V600E-OVA cells were generated by lentiviral transduction. In brief, OVA cDNA was subcloned into a pHIV-7 transfer vector carrying both the phosphoglycerate kinase (PGK) promoter and IRES-puromycin-resistance sequence. The production of third-generation self-inactivating lentiviral vectors, pseudotyped with VSV.G, was carried out as previously described⁴⁴. Specifically, packaging cells were transfected and, after 2 days, cell supernatants were collected, filtered and used to transduce tumour cell lines in the presence of 8 µg ml^–1 polybrene (Merck). After the incubation period, medium was exchanged for fresh medium, and target cells were passaged at least three times after transduction and selected using puromycin. MC38 cells were provided by A. Krüger, Institute of Experimental Oncology, TUM, and MC38-OVA and Panc02 cells were provided by V. Buchholz, Institute for Medical Microbiology, Immunology and Hygiene, TUM.

BRAF^V600E, Ptgs1/Ptgs2^−/− BRAF^V600E, BRAF^V600E-OVA and Ptgs1/Ptgs2^−/− BRAF^V600E-OVA cells were cultured in complete RPMI medium (RPMI 1640 medium (Thermo Fisher Scientific) supplemented with 10% FCS (Merck), 50 µM β-mercaptoethanol (Thermo Fisher Scientific), 50 U ml^–1 penicillin (Thermo Fisher Scientific), 50 µg ml^–1 streptomycin (Thermo Fisher Scientific) and 2 mM l-glutamine (Thermo Fisher Scientific). D4M.3A-pOVA cells were generated as previously described⁴⁵ and cultured in DMEM-F12 medium (Thermo Fisher Scientific). MC38, MC38-OVA and Panc02 cells were cultured in DMEM (Thermo Fisher Scientific), with both media supplemented with 10% FCS, 50 µM β-mercaptoethanol, 50 U ml^–1 penicillin, 50 mg ml^–1 streptomycin, 2 mM l-glutamine, 1× MEM non-essential amino acids solution (Thermo Fisher Scientific) and 1 mM sodium pyruvate (Thermo Fisher Scientific). To generate tumour cell conditioned medium (CM), 5 × 10⁶ tumour cells were cultured in 20 ml complete RPMI medium for 48 h and the supernatant was collected, filtered and stored at −20 °C until further use. All cell lines were routinely tested for mycoplasma contamination in-house by PCR. For Ptgs1/Ptgs2^−/− BRAF^V600E cells, the absence of PGE₂ production was routinely confirmed by PGE₂ ELISA (Cayman Chemical). No further cell line authentications were conducted in the laboratory.

Tumour cell injections

Tumour cell lines were detached by trypsinization (Thermo Fisher Scientific) and washed three times in sterile PBS (Thermo Fisher Scientific). Unless stated otherwise, 2 × 10⁶ cells were injected s.c. in 100 µl sterile PBS into the flank of each recipient mouse. Tumour growth was measured using a digital calliper. Tumour diameters stated in the figures refer to the average values of the longest diameter and its perpendicular for each tumour. A maximal tumour diameter of 15 mm served as the humane end point and was not exceeded in any of the experiments.

CD4⁺ and CD8⁺ T cell depletion in vivo

To deplete CD4⁺ and CD8⁺ T cells, mice received intraperitoneal (i.p.) injections of 100 µl anti-mouse CD4 (100 µg per mouse, GK1.5, BioXCell) or anti-mouse CD8β (100 µg per mouse, 53-5.8, BioXCell) antibodies every 5 days, beginning on day 1 following tumour cell inoculation.

FTY720 treatment in vivo

FTY720 treatment was performed by injecting mice i.p. with 100 µl of FTY720 (20 µg per mouse, Merck) on day 1 or day 6 after tumour cell transplantation. Injection with 100 µl sterile isotonic NaCl served as control.

IL-2 receptor blockade in vivo

For blockade of IL-2Rβ and IL-2Rγc, mice received i.p. injections of 150 µl anti-mouse CD122 (300 µg per mouse, TM-Beta 1, BioXCell) and anti-mouse CD132 (300 µg per mouse, 3E12, BioXCell) antibodies on days 6 and 7 after tumour cell transplantation. Injections with 150 µl sterile isotonic NaCl served as control.

Processing of tumour tissue and lymphoid organs

Tumours, tdLNs or spleens of tumour-bearing mice were excised at the indicated time points after cell transplantation. Tumour or organ weight was determined using a microscale. For subsequent analyses by flow cytometry or cell sorting, tumour samples were mechanically dissociated and incubated with collagenase IV (200 U ml^–1, Thermo Fisher Scientific) and DNase I (100 µg ml^–1, Merck) for 40 min at 37 °C and filtered through a 70 µm and a 30 µm cell strainer (Miltenyi) to generate single-cell suspensions. Spleens were passed through a 70 µm cell strainer, followed by red blood cell lysis and a second filtration step using a 30 µm cell strainer. LNs were passed through a 30 µm cell strainer. For the isolation of migratory cDC1s, LNs were processed as described for tumour samples.

Antibodies and reagents for flow cytometry and cell sorting

The following antibodies and staining reagents were used for flow cytometry or cell sorting: fixable viability dye eFluor 450 (dilution: 1:500; Thermo Fisher Scientific); fixable viability dye APC-eF780 (1:1,000; Thermo Fisher Scientific); viability dye SYTOX-blue (1:2,000; Thermo Fisher Scientific); APC anti-CD3 (1:100; clone 17A2, Thermo Fisher Scientific); PE anti-CD4 (1:200; GK1.5, Biolegend); AF647 anti-CD4 (1:200; GK1.5, Biolegend); PerCP/Cy5.5 anti-CD4 (1:200; GK1.5, Biolegend); BV421 anti-CD8α (1:200; 53-6.7, Biolegend); FITC anti-CD8α (1:200; 53-6.7, Biolegend); PE-Dazzle594 anti-CD8α (1:200; 53-6.7, Biolegend); PE-Cy7 anti-CD8α (1:200; 53-6.7, Biolegend); BV605 anti-CD11b (1:200; M1/70, Biolegend); PE-Cy7 anti-CD11c (1:200; N418, Biolegend); BV570 anti-mouse/human-CD44 (1:100; IM7, Biolegend); BV711 anti-mouse/human-CD44 (1:100; IM7, Biolegend); FITC anti-mouse/human-CD44 (1:100; IM7, Biolegend); PerCP-Cy5.5 anti-mouse/human-CD44 (1:100; IM7, Biolegend); AF647 anti-CD45.1 (1:100; A20, Biolegend); PE anti-CD45.1 (1:100; A20, Biolegend); PE-Dazzle594 anti-CD45.1 (1:100; A20, Biolegend); PerCP/Cy5.5 anti-CD45.1 (1:100; A20, Biolegend); BV510 anti-CD45.2 (1:100; 104, Biolegend); FITC anti-CD45.2 (1:100; 104, Biolegend); PerCP-Cy5.5 anti-CD45.2 (1:100; 104, Biolegend); FITC anti-CD62L (1:100; MEL-14, Biolegend); PE-Dazzle594 anti-CD62L (1:100; MEL-14, Biolegend); FITC anti-CD103 (1:100; M290, BD Biosciences); APC anti-CD132/IL2Rγc (1:100; TUGm2, Biolegend); PE-Dazzle594 anti-CD186/CXCR6 (1:200; SA051D1, Biolegend); PE anti-CX₃CR1 (1:100; SA011F11, Biolegend); BV605 anti-CD279/PD-1 (1:100; 29 F.1A12, Biolegend); BV421 anti-CD366/TIM-3 (1:200; RMT3-23, Biolegend); PerCP/Cy5.5 anti-TCRβ (1:100; H57-597, Biolegend); AF700 anti-I-A/I-E (1:500; MHC class II) (M5/114.15.2, Biolegend); PE anti-H-2K^b bound to SIINFEKL (1:100; 25-D1.16, Biolegend); APC anti-human GZMB (1:200; GB12, Thermo Fisher Scientific); FITC anti-Ki-67 (1:100; SolA-15, Thermo Fisher Scientific); AF700 anti-Ki-67 (1:100; SolA-15, Thermo Fisher Scientific); PE anti-TCF1/TCF7 (1:40; S33-966, BD Biosciences); AF488 anti-human pSTAT5 (0.03 µg per test, 47/Stat5(pY694); BD Biosciences); eF660 anti-TOX (1:100; TXRX10, Thermo Fisher Scientific); eFluor660 Rat-IgG2a-κ isotype-control (1:100; eBR2a, Thermo Fisher Scientific); APC mouse-IgG1κ isotype-control (1:200; P3.6.2.8.1, Thermo Fisher Scientific); AF488 mouse-IgG1κ isotype-control (0.03 µg per test; MOPC-21, Biolegend); and rabbit-anti-mouse-TCF1/TCF7 (1:100; C.725.7, Thermo Fisher Scientific). These were followed by AF647 donkey-anti-rabbit IgG (1:200; Poly4064, Biolegend) or DL488 donkey-anti-rabbit IgG (1:200; Poly4064, Biolegend). Unless stated otherwise, all antibodies were anti-mouse antibodies.

Flow cytometry and cell sorting

For staining of surface markers and viability dyes, cells were stained for 15 min at 4 °C in FACS buffer (PBS with 1% FCS and 2 mM EDTA). Staining of SIINFEKL–MHC class I complexes on cDC1s for analysis of OVA cross-presentation was performed for 40 min. For intracellular staining of GZMB, TCF1, Ki-67 and TOX, cells were fixed and permeabilized using the True-Nuclear Transcription Factor Buffer Set (Biolegend) according to the manufacturer’s protocol. Intracellular staining was performed overnight in permeabilization buffer at 4 °C. For intracellular staining of pSTAT5, cells were fixed and permeabilized using BD Cytofix (BD Biosciences) and BD Phosflow Perm Buffer III (BD Biosciences) according to the manufacturer’s instructions (protocols II and III, BD Biosciences). For the detection of EdU incorporation, EdU was added to the culture at a final concentration of 15 µM for the last 3 h of the experiment, and analysis was performed using an EdU Proliferation kit (iFluor 488, Abcam) according to the manufacturer’s protocol.

Flow cytometry analyses were performed using a LSR Fortessa Cell Analyzer (BD Biosciences, BD FACSDiva software v.8.0.1 and v.9.0.1), a SP6800 Spectral Cell Analyzer (Sony Biotechnologies, spectral analyser software v.2.0.2.14140) or a SA3800 Spectral Cell Analyzer (Sony Biotechnologies, spectral analyser software v.2.0.5.54250). For flow cytometric quantification of cell numbers, CountBright Absolute Counting Beads (Thermo Fisher Scientific) were added to samples before analyses. For some experiments, CD8⁺ TILs (live CD45⁺CD3⁺CD8⁺ cells), stem-like Cd4^crePtger2^−/−Ptger4^fl/fl OT-I TILs (live CD45.1⁺CD8⁺CD44⁺TIM-3⁻CXCR6⁻) or differentiated effector Cd4^crePtger2^−/−Ptger4^fl/fl OT-I TILs (live CD45.1⁺CD8⁺CD44⁺TIM-3⁺CXCR6⁺) were sorted using a FACS Aria III Cell Sorter (BD Biosciences, BD FACSDiva software v.9.0.1). Naive OT-I T cells (CD45.1⁺CD8⁺CD62L⁺CD44^–) used in adoptive transfer experiments were sorted from blood using a SH800S Cell Sorter (Sony Biotechnologies, cell sorter software v.2.1.6). All flow cytometric data were analysed using FlowJo (BD Biosciences, v.00.8.1 and v.10.8.2).

Adoptive T cell transfer

For adoptive T cell transfer of naive T cells, 1 × 10³ congenically marked naive CD8⁺ T cells from OT-I, Ptger2^−/−Ptger4^fl/fl OT-I or Cd4^crePtger2^−/−Ptger4^fl/fl OT-I donor mice were injected i.v. in sterile PBS into sex-matched recipient WT mice 6 h before tumour cell transplantation s.c. For adoptive transfer of CRISPR-Cas9-edited T cells, 1 × 10³ cells congenically marked OT-I T cells from in vitro T cell cultures were injected i.v. into recipient mice at day 2 after tumour cell transplantation s.c. For re-transfer of CD8⁺ TILs, 7 × 10³ congenically marked stem-like (TIM-3^–CXCR6^–) or differentiated effector (TIM-3⁺CXCR6⁺) Cd4^crePtger2^−/−Ptger4^fl/fl OT-I TILs were sorted from MC38-OVA tumours from WT mice and injected i.v. in sterile PBS into sex-matched recipient Rag1^−/− mice inoculated with MC38-OVA tumour cells 2 days before T cell re-transfer.

Generation of repetitively activated antigen-experienced TCF1⁺CD8⁺ T cells

TCF1⁺CD8⁺ T cells were differentiated from splenic naive CD8⁺ T cells by repetitive activation as previously described³⁵, with minor modifications. In brief, 1 × 10⁶ naive CD8⁺ T cells were seeded in complete RPMI medium supplemented with 1× MEM non-essential amino acids solution and 1 mM sodium pyruvate. Low-dose IL-2 (85 U ml^–1) and mouse anti-CD3/CD28 microbeads were added to the culture while maintaining a CD8⁺ T cell concentration of 1 × 10⁶ cells per ml for multiple (re-)activation cycles over a course of 4 days, followed by purification of live cells by gradient centrifugation (Pancoll).

T cell effector differentiation

Effector differentiation of TCF1⁺CD8⁺ T cells was performed as previously described³⁵, with minor modifications. In brief, cells were cultured with mouse anti-CD3/CD28 microbeads in the presence of high-dose IL-2 (350 U ml^–1). Where indicated, PGE₂ (100 ng ml^–1, unless indicated otherwise in the figure legend; Thermo Fisher Scientific), tumour cell CM, IL-7 (10 ng ml^–1, Miltenyi), IL-12 (10 ng ml^–1, Biolegend) or IL-15/15Rα (1 ng ml^–1, Thermo Fisher Scientific) was added to the culture. To assess T cell expansion, the numbers of live CD45⁺CD3⁺CD8⁺ T cells were quantified by flow cytometry 72 h after the incubation period.

Gene deletion by CRISPR–Cas9–gRNA complex electroporation

Cd4^crePtger2^−/−Ptger4^fl/fl OT-I T cells were purified from spleen and cultured in complete RPMI supplemented with IL-2 (10 U ml^–1) and IL-7 (5 ng ml^–1) in the presence of mouse anti-CD3/CD28 microbeads. After 24 h, anti-CD3/CD28 microbeads were removed by magnetic separation and cells were electroporated (4D-Nucleofector, Lonza; pulse program CM137)⁴⁶ in P3 electroporation buffer supplemented with the Cas9 electroporation enhancer (IDT), Cas9 protein (IDT) and Cd122-targeting or non-targeting gRNAs. gRNAs were generated by hybridizing trRNA (IDT) with Cd122-targeting (sequences TATGTCAAGGAGGTCCACGG and CTGGGAACGACCCGAGGATC, generated using CHOPCHOP; ref. ⁴⁷) or non-targeting crRNA (IDT) (GCCTGCCCTAAACCCCGGAA; ref. ⁴⁸) as mock control. Cells were rested in complete RPMI supplemented with IL-7 (5 ng ml^–1, Miltenyi) at 37 °C for 48 h and validated for specific knockout by CD122 surface staining before injection into recipient mice.

Analysis of IL-2Rγc expression and IL-2 signalling

TCF1⁺CD8⁺ T cells from in vitro cultures were rested for 20 h in complete RPMI supplemented with low-dose IL-2 and purified by gradient centrifugation. Cells were stimulated with mouse anti-CD3/CD28 microbeads and low-dose IL-2 for 24 h in the absence or presence of PGE₂ (100 ng ml^–1). After 24 h, IL-2Rγc chain expression was analysed by flow cytometry. For analysis of IL-2-induced STAT5 signalling, anti-CD3/CD28 microbeads were removed by magnetic separation, cells were rested for 30 min at 37 °C in complete RPMI and stimulated for 30 min with different concentrations of IL-2 (10–100 U ml^–1, as indicated). After the incubation period, fixation buffer was directly added to the culture to terminate the signalling process and cells were stained for flow cytometry analysis.

PGE₂ measurements

Tumours and organs of tumour-bearing mice were excised 11 days after tumour cell transplantation, directly frozen in liquid nitrogen and stored at −80 °C until further processing. Samples were homogenized in homogenization buffer (0.1 M PBS, 1 mM EDTA and 10 µM indomethacin (Merck), pH 7.4) using a gentleMACS Dissociator (Miltenyi) followed by a freeze–thaw cycle. PGE₂ concentrations were measured by ELISA (Cayman Chemical) according to the manufacturer’s protocol.

RNA isolation and quantitative real-time PCR

RNA was isolated using an Arcturus PicoPure RNA isolation kit (Thermo Fisher Scientific) and cDNA was generated using a SensiFAST cDNA synthesis kit (Bioline). Quantitative real-time PCR was carried out on a LightCycler 480 (Roche, LightCycler 480 software v.1.5.1) using a TAKYON No ROX SYBR MasterMix dTTP Blue kit (Eurogentec) according to the manufacturer’s protocol. Ptger4 expression was determined using the ΔCt method, with Hprt serving as reference gene. Primer sequences were from a previous study³⁸. All primers were purchased from Eurofins.

scRNA-seq and scTCR-seq

CD8⁺ TILs were sorted from BRAF^V600E tumours 11 days after tumour cell transplantation. A combination of cell hashing and DNA barcoding during library preparation was used for sample multiplexing, which enabled the simultaneous sequencing of four biological replicates from each group. For cell hashing, unique TotalSeq-C anti-mouse hashtag antibodies were used for hashing of cells from each experimental group as follows: WT: hashtag 1; Ptger2^−/−Ptger4^fl/fl: hashtag 2; Cd4^crePtger2^−/−Ptger4^fl/fl: hashtag 3; and Gzmb^crePtger2^−/−Ptger4^fl/fl: hashtag 4 (1:250 each, Biolegend). Hashtagged cells from one tumour-bearing mouse of each group were pooled and loaded on a Chromium Next GEM Chip (10x Genomics). RNA-seq libraries were generated using Chromium Next GEM Single Cell 5′ Reagent kits v.2 User Guide with Feature Barcode technology for Cell Surface Protein (Rev D) according to the manufacturer’s protocol (10x Genomics). Quality control was carried out using a High Sensitivity DNA kit (Agilent), a Bioanalyzer 2100 and a Qubit dsDNA HS Assay kit (Thermo Fisher Scientific). For sequencing, libraries were pooled and analysed by paired-end sequencing (2 × 150 bp) on a NovaSeq6000 platform using S4 v.1.5 (300 cycles) sequencing kits (Illumina). Libraries were sequenced to a depth of at least 2 × 10⁴ reads per cell for gene expression libraries and 5 × 10³ reads per cell for T cell receptor libraries.

Initial scRNA-seq analyses were performed for all samples from the groups Ptger2^−/−Ptger4^fl/fl, Cd4^crePtger2^−/−Ptger4^fl/fl and Gzmb^crePtger2^−/−Ptger4^fl/fl, with data from the WT group being added at a later stage for validation of Ptger2 and Ptger4 read coverage (see below). Alignment of gene expression libraries and demultiplexing were performed using cellranger multi (Cell Ranger (v.6.1.1)⁴⁹; 10x Genomics) against the pre-built mouse reference v2020-A (10x Genomics, mm10/GRCm38, annotation from GENCODE Release M23) with the number of expected cells equals 21.000 as input argument. The BAM files were converted to FASTQ files using the tool bamtofastq with the argument –reads-per-fastq set to the total number of reads in the BAM file plus 10,000. After that, gene expression and TCR analysis were combined by running cellranger multi separately for each demultiplexed sample, disabling library concordance reinforcement. The algorithm was forced to find the number of cells identified in the first step of demultiplexing, and sample-specific FASTQ files were used as input for the gene expression analysis pipeline. The pre-built Ensembl GRCm38 Mouse V(D)J Reference v.5.0.0 was used for TCR analysis.

The initial downstream analysis was performed in R (v.4.0.4) with the R package Seurat (v.4.0.1)⁵⁰. Only cells with more than 1,000 genes detected, less than 10% of mitochondrial genes and with UMI counts less than 3 standard deviations above the mean were kept. The data were filtered for genes detected in at least three cells in one of the samples. Filtered read counts from each sample were normalized independently using sctransform (v.0.3.2)⁵¹ with the glmGamPoi method⁵². Anchors between cells from different replicates were identified on the top 1,000 highly variable genes using canonical correlation analysis and 30 canonical vectors. Data integration was performed on first 20 PC analysis (PCA) dimensions. PCA was calculated for the integrated data on the top 1,000 highly variable genes and both k-nearest neighbour graph and UMAP were computed on the 30 nearest neighbours and first 20 PCA dimensions. Louvain clusters were identified using the shared nearest neighbour modularity optimization-based algorithm at resolutions 0.9, 0.65 and 0.9 for the groups Ptger2^−/−Ptger4^fl/fl, Cd4^crePtger2^−/−Ptger4^fl/fl and Gzmb^crePtger2^−/−Ptger4^fl/fl, respectively. Contaminating myeloid cells were identified based on the average cluster expression of the marker genes Cd14, Lyz2, Fcgr3, Ms4a7, Fcer1g, Cst3, H2-Aa, Ly6d, Ms4a1 and Ly6d. Cycling cells were identified based on expression of Cdk1, Mcm2, Pclaf, H2afz, Birc5 and Mki67.

The integrative analysis between groups was performed in R (v.4.2.1) with the R package Seurat (v.4.1.1)⁵⁰. After general data pre-processing and regression of contaminating cells as mentioned above, filtered read counts from each sample were normalized independently using sctransform (v.0.3.2)⁵¹ with glmGamPoi method⁵². Anchors between cells from all groups and all their replicates were identified using a more conservative approach, which led to weaker batch correction. For that purpose, reciprocal PCA was applied on the top 1,000 highly variable genes of each sample and anchors were picked using the first 20 dimensions and 1 neighbour only. PCA was performed on the integrated data on the top 1,000 highly variable genes. A k-nearest neighbour graph and UMAP (spread of 0.4, minimum distance of 0.01) were computed on the first 20 PCs and 30 nearest neighbours. A resolution of 0.6 was used for Louvain clusters identification using the shared nearest neighbour modularity optimization-based algorithm. DEGs between two groups were identified using the Wilcoxon rank-sum test and Bonferroni correction. Gene set expression scores at single-cell level were calculated using the AddModuleScore function, including only the detected genes. Similarity scores with reference datasets were calculated using the R package SingleR (v.1.10.0)⁵³ with the top 200 DEGs. The processed transcriptome profiles of naive CD8⁺ T cells, memory stem cell CD8⁺ T cells and central memory CD8⁺ T cells were from a previous study⁵⁴. For tumour antigen-specific CD8⁺ T cells in tdLNs, tumour-infiltrating stem-like CD8⁺ T cells and their naive counterparts, data from a previous study³ were processed using the R package DESeq2 (v.1.36)⁵⁵. Gene set expression scores at the single-cell level were calculated using the AddModuleScore function, including only the detected genes. The effector T cell gene signature was from a previous study⁵⁶ (M3013: KAECH_NAIVE_VS_DAY8_EFF_CD8_TCELL_DN). The CD8⁺ T cell proliferation signature was obtained from MSigDB (GO:2000566). Transcriptional trajectories were inferred using the R package slingshot (v.2.4.0)⁵⁷ over the UMAP calculated on the integrated data, approximating the curves by 150 points. The pseudotime was calculated as a weighted average across lineages, weighted by the assignment weight.

TCR analysis of clonotype was performed using the R package scRepertoire (v.1.6.0)⁵⁸. Clonotypes were called based on a combination of VDJC genes comprising the TCR and the nucleotide sequence of the CDR3 region. Whenever the clonotype distribution is shown for individual groups, the cell number was downsampled, so that cluster 1 from all groups had the same maximum size. TF activity was inferred using the weighted mean method of decoupleR (v.2.2.2)⁵⁹ and TF–target interactions available through dorothea (v.1.8.0)⁶⁰, with confidence levels A to C. Normalization to Ptger2^−/−Ptger4^fl/fl was achieved by subtracting its scores from the scores of the other groups. The top 100 variable TFs between clusters within each group were used to draw a network graph with tidygraph (v.1.2.1)⁶¹ based on common targets with same defined mode of regulation as defined on the database. Only TFs with at least two common targets were kept for visualization. Louvain clusters were identified using igraph (v.1.3.2)⁶² at a resolution of 0.5.

For addition of scRNA-seq data from the WT group, samples were pre-processed as described above and mapped to a reference formed by the integrated data of the Ptger2^−/−Ptger4^fl/fl, Cd4^crePtger2^−/−Ptger4^fl/fl and Gzmb^crePtger2^−/−Ptger4^fl/fl groups using the R package Seurat (v.4.1.1)⁵⁰. For that purpose, anchors between cells from the reference and the WT groups along with all replicates were identified using reciprocal PCA on top 1,000 highly variable genes. Anchors were picked using the first 20 dimensions and 1 neighbour only. Annotations were transferred using the function TransferData, and data were integrated using IntegrateEmbeddings. Cells from the added group were then projected onto the coordinates of the reference UMAP calling ProjectUMAP with 30 nearest neighbours. Read coverage was estimated using deepTools (v.3.5.4)⁶³ with bamCoverage and a bin size of 10 bp and normalization by bins per million mapped reads. For coverage analysis on Tcf7/TCF1⁺ and Tcf7/TCF1⁻ clusters, BAM files were split by cell barcodes from clusters 1–2 or clusters 3–8 using samtools (v.1.13)⁶⁴ before coverage estimation. Read coverage on gene tracks was visualized using the R package trackViewer (v.1.32.1)⁶⁵.

RNA-seq

In vitro generated, repetitively activated TCF1⁺CD8⁺ T cells were incubated in the presence or absence of PGE₂ (100 ng ml^–1) for 1 h at 37 °C followed by stimulation with IL-2 or IL-2 plus mouse anti-CD3/CD28 microbeads for an additional 4 h. Total RNA was isolated using Total RNA Miniprep (Monarch). Library preparation was carried out using a NEB Next UltraRNA Library Prep kit with i7 and i5 index reads of 8 bp each for mRNA library preparation and poly A enrichment. Sequencing was performed on a NovaSeq6000 PE150 platform in paired-end mode (read 1: 151 bp, read 2: 151 bp), using S4 (v.1.5) (300 cycles) sequencing kits (Illumina). Reads were aligned to the mouse reference genome (GRCm38/mm10, NCBI) using the Hisat2 (v.2.0.5) mapping tool. To quantify gene expression levels, featureCounts (v.1.5.0-p3) was used to count the reads mapped to each gene, followed by the calculation of fragments per kilobase of transcript sequence per million mapped reads based on gene length and read count. DEGs were identified using the DESeq2 R package (v.1.20.0). Adjusted P values were obtained using Wald test with multiple testing by the Benjamini–Hochberg method, and genes identified by DESeq2 with adjusted P values < 0.05 and fold change ≥ 2 were assigned as DEGs. Volcano plots were visualized using the ggplot2 R package ggplot2 (v.3.4.2), and PCA was conducted using the prcomp function in R and visualized using the R packages ggplot2 and ggrepel (v.0.9.3). DEGs obtained from comparing the groups ‘anti-CD3/CD28 +IL-2’ and ‘PGE₂-treated + anti-CD3/CD28 +IL-2’ were ordered based on their log₂ fold change values and subjected to GSEA using GSEA (v.4.3.2) probing for hallmark genes from mh.all.v2023.1.Mm (MSigDB). The PreRanked tool from GSEA (v.4.3.2) was used to determine the NES and significance by adjusted P values.

Statistical analyses

The GraphPad Prism software (v.9.5.0 and v.9.5.1) was used for statistical analyses. Affinity Designer (v.1.10.6) (Serif) was used to visualize data. Paired or unpaired two-tailed Student’s t-test, one-way ANOVA or two-way ANOVA was used to assess statistical significance, as indicated in in the figure legends. Data are shown as the mean ± s.d., mean ± s.e.m. or box and whiskers plots, as indicated in the figure legends.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.