Building on our previous analysis shedding light on the onco-GPCRome and aberrant GPCR signaling and activity on tumor cells⁸, we investigated the landscape of GPCR expression on each tumor-infiltrating CD8⁺ T cell subtype. We collected data and performed an integrated analysis of scRNA-seq datasets from 19 cancer types (Fig. 1a and Supplementary Table 1a). The cancer types covered in this analysis encompass those in which individuals show higher (for example, ~25–50% in non-small cell lung cancer and cutaneous melanoma) and lower (for example, <20% in ovarian cancer and pancreatic cancer) response rates to immunotherapy¹⁴ and thus provide a comprehensive overview of the expression landscape of tumor-infiltrating CD8⁺ T cells. Using the Seurat SCTransform scRNA-seq integration method, we jointly analyzed 217,953 total CD8⁺ T cells, which were stratified into naive, proliferating, cytotoxic, effector memory and exhausted subtypes based on landmark genes and nearest neighbor analysis, ProjecTIL¹⁵ (Fig. 1b,c, Extended Data Fig. 1a,b and Supplementary Table 1b). This resulted in delineation of functional (cytotoxic and effector memory) and dysfunctional (exhausted) CD8⁺ subtypes, nonspecific naive T cells that stochastically migrate through the tumor and proliferating T cells that express features of both effector memory and exhausted T cells^16,17. When we analyzed the relative expression of 386 non-olfactory GPCR genes, many GPCRs showed a distinct expression pattern in each CD8⁺ T cell subtype (Fig. 1d and Supplementary Tables 1c,d). Terminally exhausted CD8⁺ T cells in both cancer and chronic viral infection most commonly have been shown to express high levels of PD-1, TIM-3, CXCR6, LAG3, CTLA-4 and CXCL13 proteins, all of which are reflected at the transcriptomic level in our combined analysis, as previously documented and validated in the individual datasets (Supplementary Table 1a–c). Within exhausted CD8⁺ T cells, we found that CXCR6, PTGER4, GPR65, ADGRE5 and F2R were among the most highly expressed genes encoding GPCRs, with expression patterns similar to those of signature exhaustion genes (Fig. 1e).

A transcriptional gene module predicting T cell dysfunction programs linked to LAG3 expression (for example, TIGIT, PDCD1, LAG3 and CXCL13) was established in individuals with melanoma¹⁸. To identify GPCRs most relevant to this dysfunction program, we generated dysfunction scores for all CD8⁺ T cells from our integrated analyses (Fig. 1f and Supplementary Table 1e). As expected, exhausted CD8⁺ Tcells across 19 cancer types displayed the highest dysfunction scores, significantly higher than proliferating, cytotoxic and effector memory CD8⁺ T cells (Fig. 1g and Extended Data Fig. 1c). We next investigated the GPCR genes most strongly associated with T cell dysfunction scores. The gene encoding CXCR6, which is expressed on PD-1^hi effector and exhausted CD8⁺ Tcells during chronic viral infection, was the GPCR gene most significantly correlated with Tcell dysfunction¹⁹ (Fig. 1h). In total, 35 GPCR genes were shown to be significantly correlated with the dysfunction score, with GPR56, CCRL2, GIPR and F2R among the top candidates to contribute to Tcell exhaustion (Fig. 1h and Supplementary Table 2a). We next aimed to distinguish GPCR expression patterns based on heterotrimeric G-protein coupling information to begin defining functional gene sets. Based on the International Union of Basic and Clinical Pharmacology (IUPHAR) classification, we grouped all GPCRs into G-protein programs based on their primary G-protein coupling to Gα_12/13, Gα_i, Gα_q/11 or Gα_s (Supplementary Table 2b). Intriguingly, when we calculated the mean correlation of each pathway, we found that the Gα_s program was the most enriched with Tcelldysfunction, while the Gα_i programwasthe leastcorrelated(Fig. 1i). Multiple genes encoding Gα_s-coupled GPCRs were positively correlated with Tcell dysfunction with significant Spearman correlations, including GIPR, ADORA2A, PTGER4, GPR65 and TSHR (Fig. 1j). These data raise the possibility of a GPCR–Gα_s signaling program that correlates with T cell dysfunction in tumor-infiltrating CD8⁺ T cells.

Next, we investigated the relevance of Gα_s–GPCRs to exhausted T cell phenotypes. We analyzed bulk transcriptomic data from the mouse lymphocytic choriomeningitis virus (LCMV) chronic infection model, which represents the experimental model system most central to recent landmark discoveries defining T cell exhaustion²⁰ (Fig. 2a and Supplementary Table 3a). In this model, acute infection with a strain of LCMV (Armstrong) generates robust, proliferative and activated effector CD8⁺ T cells that are able to resolve infection²¹. By contrast, chronic infection with a different strain of LCMV (clone 13) leads to persistent antigen exposure, generating a distinct exhausted T cell state with sustained expression of inhibitory receptors and hierarchical loss of effector functions²¹. We combined three RNA-seq datasets of fluorescence-activated cell sorting-purified effector CD8⁺ T cells from acute LCMV infection and exhausted T cells from chronic LCMV infection and used DESeq2 to directly compare differential expression of GPCRs between the two CD8⁺ T cell subtypes. This revealed that multiple genes encoding Gα_s–GPCRs are upregulated in exhausted T cells compared to in effector T cells, including Glp1r, Ptger2, Ptger4 and Gpr65, which are also significantly correlated with T cell dysfunction in human tumor-infiltrating CD8⁺ T cells (Figs. 1g and 2b and Supplementary Table 3b). This suggests that the Gα_s–GPCRs that are expressed on both exhausted tumor-infiltrating CD8⁺ T cells and exhausted CD8⁺ T cells during chronic viral infection are intrinsically similar.

Recently, we have determined coupling across 148 human GPCRs for 11 specific human G proteins and used a machine learning approach to augment GPCR coupling predictions for class A GPCRs²². Here, we curated gene sets stratifying for their predictive G-protein coupling as designated by IUPHAR and our reported assay. We found that the Gα_s coupling gene set by IUPHAR had the most significant enrichment with upregulated GPCRs on exhausted T cells in LCMV infection (Fig. 2c). Additional gene sets were also significantly enriched for upregulated GPCRs on exhausted T cells, including GNAQ predicted coupling, G_q/11 primary IUPHAR coupling and GNA14 predicted coupling (Fig. 2d and Supplementary Table 3c). Together, this suggests that, in addition to T cell dysfunction, the expression of these Gα_s-coupled GPCRs may also play a role in driving T cell exhaustion.

To gain a better understanding of how the expression of these Gα_s–GPCRs modulate CD8⁺ T cells, we first sought to model impairment of T cell function resulting from persistent T cell antigen receptor (TCR) signaling by modulating T cell activation in vitro (Fig. 2e). CD8⁺ T cells from splenocytes of C57BL/6 mice were activated with anti-CD3 and anti-CD28 for 48 h. Subsequently, cells were subjected to additional restimulation (chronic stimulation) to model reactivation of T cells after TCR engagement at the tumor. As a control, CD8⁺ T cells were also expanded in culture with interleukin-2 (IL-2; activated) without additional restimulation. Characteristic of terminally exhausted cells, the expression levels of PD-1, CTLA-4, TIM-3 and LAG3 on CD8⁺ T cells were all significantly elevated in chronically versus acutely stimulated CD8⁺ T cells, concomitant with a decrease in IFNγ and tumor necrosis factor (TNF), confirming in this assay that chronically activated CD8⁺ T cells begin acquiring an exhaustion-like phenotype (Extended Data Fig. 2a).

To simulate exposure of Gα_s ligands at the TME, we added ligands stimulating Gα_s–GPCRs to our chronic exhaustion assay (prostaglandin E₂ (PGE₂) for EP₂ and EP₄, dobutamine for β₁ adrenergic receptor (β₁-AR) and β₂AR and CGS-21680 for adenosine 2A receptor (A_2AR)). After 48 h of the initial activation, CD8⁺ T cells were replated for an additional 48 h of activation with or without Gα_s ligands, and we measured the functional capacity by flow cytometric analysis of IFNγ, TNF and granzyme B. Continuous stimulation and the addition of PGE₂, dobutamine or CGS-21680 all significantly reduced IFNγ, TNF and granzyme B single positivity and highly cytotoxic polyfunctional CD8⁺ T cells, as measured by cells expressing both IFNγ and TNF (Fig. 2f,g and Extended Data Fig. 2b). A similar reduction was seen in the viability and proliferative capacity of CD8⁺ T cells, as indicated by the reduction in Ki-67 positivity (Extended Data Fig. 2c). Moreover, stimulation with PGE₂ and dobutamine significantly elevated PD-1 and TIM-3 expression (Fig. 2g and Extended Data Fig. 2d). The simultaneous reduction in T cell function and proliferation and increase of inhibitory receptor expression suggest that Gα_s ligands augment the dysfunctional phenotypes in CD8⁺ T cells polarizing toward exhaustion.

To evaluate the functional suppression by Gα_s ligands on cytotoxic T cell killing, we activated purified T cells from OT-1 transgenic mice whose TCRs are specific for the ovalbumin (OVA) peptide SIINFEKL (OVA_257–264) and cocultured them with MC38 tumor cells expressing OVA (MC38-OVA) at a 1:5 effector:target ratio (Fig. 2h). Gα_s ligands significantly diminished the cytotoxicity of OT-1 CD8⁺ T cells toward MC38-OVA cells as measured by tumor cell viability after 2 d (Fig. 2i). Together, our in vitro experiments identify an inhibitory effect of ligands for Gα_s–GPCRs on T cell function and cytotoxicity.

To elucidate the underlying mechanisms of Gα_s-mediated CD8⁺ T cell dysfunction, we next characterized the pathways downstream of Gα_s/cAMP that drive immune suppression and focused on PGE₂, as this inflammatory mediator led to the most pronounced inhibition of function from all Gα_s–GPCR ligands tested. Addition of PGE₂ and forskolin (fsk), a direct adenylyl cyclase-cAMP activator, during chronic stimulation led to a significant increase of the cAMP response element-binding protein (CREB), which becomes phosphorylated (pCREB) following increased intracellular cAMP and protein kinase A (PKA) activation (Fig. 2j). EP₂ or EP₄ inhibitors (EP₂i and EP₄i, respectively) significantly decreased PGE₂-induced, but not fsk-induced, activation of pCREB, supporting the receptor-mediated effects of PGE₂ on PKA (Fig. 2j). Similarly, addition of EP₂i or EP₄i antagonists restored production of IFNγ and TNF secretion following inhibition by PGE₂ but not in response to fsk (Fig. 2k). Interestingly, the addition of EP₂ and EP₄ antagonists together more significantly reduced PGE₂-mediated IFNγ and TNF inhibition than each antagonist alone, suggesting that concomitant blockade of Gα_s activation by multiple Gα_s receptors may afford better rescue from immune-suppressive effects on CD8⁺ T cells. Next, we sought to determine whether inhibitory effects by PGE₂/cAMP-induced inhibition could be rescued by activation of an endogenously expressed Gα_i–GPCR CXCR3 with its ligand CXCL10. Although CXCL10 alone did not significantly affect IFNγ and TNF secretion, the addition of CXCL10 with PGE₂ alleviated inhibition by PGE₂ (Extended Data Fig. 2e). Together, these data strongly suggest that Gα_s activation leads to dampening of CD8⁺ T cell function, which can be rescued by either blocking Gα_s–GPCRs or by stimulating the Gα_i axis.

To begin exploring the role of the Gα_s–PKA axis in PGE₂-mediated inhibition of T cell function, we generated conditional CD8-specific Gnas-knockout (KO) mice (Fig. 2l). We used mice with CD8⁺ T cell-restricted Cre recombinase temporally controlled with tamoxifen²³ (E8i-Cre^ERT2) crossed with mice with loxP sites flanking Gnas exon 1 (referred to as CD8-Gnas KO; Fig. 2l and Supplementary Table 4). E8i-Cre^ERT2 mice express the reporter gene in mature CD8⁺ T cells but not in CD4⁺ T cells, B cells, myeloid cells, dendritic cells or natural killer (NK) cell populations²⁴. CD8⁺ T cells were isolated and chronically stimulated following administration of tamoxifen. As expected, CD8⁺ T cells with functional Gα_s expressed less IFNγ and TNF after treatment with PGE₂ (Fig. 2m), but this inhibition was rescued in CD8-Gnas KO mice (Fig. 2m). In search of the potential mechanisms, we examined the involvement of PKA, a key downstream signaling effector of Gα_s. For these studies, we leveraged our protein kinase inhibitor (PKI) peptide mouse model²⁵. Tet-GFP-PKI mice were crossed with E8i^CreERT2-ROSA26^LSLrtTA mice to generate a conditional, tetracycline-inducible PKI system to specifically inhibit PKA in CD8⁺ T cells (referred to as CD8-PKI; Extended Data Fig. 3a–c and Supplementary Table 5). CD8⁺ T cells were isolated from the CD8-PKI mice following tamoxifen and doxycycline administration and were chronically stimulated as described above. When green fluorescent protein (GFP)–PKI was expressed in chronically activated CD8⁺ T cells, it prevented the decrease in IFNγ⁺TNF⁺ double-positive CD8⁺ T cells caused by PGE₂ (Extended Data Fig. 3d), which mirrors Gnas deletion (Fig. 2m). This strongly suggests that Gnas and its downstream signaling target, PKA, are necessary for PGE₂-mediated inhibitory effects on CD8⁺ T cells.

The use of designer receptors exclusively activated by designer drugs (DREADDs), which are GPCRs engineered to be non-responsive to endogenous ligands but responsive to the designer drug clozapine-N-oxide, can be exploited to achieve G-protein activation in a tissue-specific fashion²⁶. To bypass potential off-target effects elicited by clozapine-N-oxide, we used the recently developed deschloroclozapine (DCZ), which affords higher affinity and more selective agonist activity for DREADDs²⁷. To specifically interrogate the function of the Gα_s signaling axis in CD8⁺ T cells, we expressed Gα_s–DREADD by crossing E8i^CreERT2 with ROSA26^LSLGsDREADD mice (Fig. 3a, Extended Data Fig. 4a and Supplementary Table 6). The resulting mice, referred to as CD8-GsD mice, were dosed with tamoxifen every day for 3 d, and Gα_s–DREADD expression and activation was subsequently verified (Fig. 3b). As assessed by quantitative PCR (qPCR), only in mice treated with tamoxifen was there demonstrable expression of Gα_s–DREADD in CD8⁺ T cells and not in CD4⁺ T cells (Fig. 3c). This confirmed tamoxifen-inducible recombination by the Cre recombinase for CD8-restricted Gα_s–DREADD expression.

To confirm activation of Gα_s–DREADD, mice were first dosed with tamoxifen, and then 0.01 mg per kg (body weight) DCZ was administered intraperitoneally (i.p.; Fig. 3b). After administration of tamoxifen and DCZ, mice were subsequently bled to assess pCREB induction by flow cytometry. Whereas the frequencies of pCREB⁺NK1.1⁺ NK cells, pCREB⁺CD11b⁺ myeloid cells and pCREB⁺CD4⁺ T cells were low and did not differ between DCZ-treated and untreated mice, the frequency of pCREB⁺CD8⁺ cells significantly increased in DCZ-treated mice (Fig. 3d). Mice treated with DCZ did not show significant changes in frequency of CD8⁺ or CD4⁺ T cells, NK cells or CD11b⁺ myeloid cells in the peripheral blood (Extended Data Fig. 4b). When we isolated peripheral blood from mice dosed with tamoxifen, the addition of DCZ in vitro also increased pCREB exclusively in CD8⁺ T cells (Fig. 3e). These results confirm tamoxifen-inducible CD8-specific expression and activation of Gα_s–DREADD by DCZ in CD8-GsD mice. In agreement with our data for Gα_s–GPCR ligands, only in mice with tamoxifen-induced expression of Gα_s–DREADD did CD8–Gα_s activation lead to significant decreases in IFNγ, TNF, Ki-67 and granzyme B concomitant with increases in PD-1 and TIM-3 expression (Fig. 3f–h and Extended Data Fig. 4c). To gain an understanding of downstream transcriptional modulations stimulated by Gα_s signaling in CD8⁺ T cells, we assessed gene expression of dual-specificity phosphatase 1 (Dusp1), a CREB target shown to negatively regulate T cell activation and function through inactivation of JNK and reduced NFATc1 (refs. 28,29; Fig. 3i). Stimulation of Gα_s–DREADD significantly increased the expression of Dusp1, in addition to Tigit and Tox, both of which are highly expressed on terminally exhausted CD8⁺ T cells^30,31 (Fig. 3j). Together, activation of Gα_s–DREADD on CD8⁺ T cells was sufficient to exacerbate exhaustion-related phenotypes, as indicated by decreased cytotoxic function in tandem with increased expression of exhaustion-related genes.

To assess biological relevance of the inhibitory Gα_s signaling in CD8⁺ T cells in the tumor setting, we used the OVA tumor model system to investigate the effect of Gα_s signaling on recruitment and function of antigen-specific CD8⁺ T cells. To study native T cell trafficking and function, we took advantage of an orthotopic head and neck cancer syngeneic mouse model, 4MOSC1, which recapitulates human head and neck cancer mutational signatures with ~93% similarity³². CD8-GsD or littermate control mice were given three doses of tamoxifen before tumor implantation with 4MOSC1-OVA, and DCZ was given every day starting 1 d after tumor implantation (Fig. 3k). CD8-GsD mice dosed with tamoxifen and with or without DCZ were killed at an early time point to quantify tumor-infiltrating SIINFEKL-tetramer⁺CD8⁺ T cells. There was a significant decrease in antigen-specific CD8⁺ T cells infiltrating the tumor, and IFNγ and TNF from the bulk CD8⁺ T cell population were both significantly decreased in DCZ-treated mice compared to in untreated mice (Fig. 3l,m). Our in vitro and in vivo data collectively indicate sufficiency of Gα_s activation to drive exhaustion-related phenotypes and prevent trafficking of tumor-specific CD8⁺ T cells.

We next investigated the effect of CD8-specific Gα_s signaling on immunotherapy response and tumor killing in vivo. CD8-GsD mice were dosed with tamoxifen, and on day 0, 4MOSC1 cells or MC38-OVA cells were implanted into the tongues or flanks of mice, respectively (Fig. 4a). DCZ (0.01 mg per kg (body weight)) was administered daily to mice starting 1 d after tumor implantation. In the absence of agonist stimulation CD8-GsD mice with 4MOSC1 partially responded to anti-PD-1, similar to wild-type (WT) mice, as previously reported³²; but activation of Gα_s-–DREADD by DCZ administration abolished any antitumoral responses (Fig. 4b). Additionally, although anti-PD-1 afforded a survival advantage in tumor-bearing mice, activation of the Gα_s signaling axis in CD8⁺ T cells led to survival rates not significantly different from untreated mice (Fig. 4c). As controls, neither DCZ nor tamoxifen affected tumor growth in littermate control mice (Extended Data Fig. 5).

These findings prompted us to determine whether activation of Gα_s signaling in CD8⁺ T cells was sufficient to limit the long-term antitumor immunity from combined PD-1 and CTLA-4 ICB³². In the absence of DREADD ligand stimulation, 90% of CD8–Gα_s–DREADD mice responded to dual ICB with durable tumor regression (Fig. 4b,c). This was abrogated by the administration of DCZ (Fig. 4b,c and Extended Data Fig. 6), indicating that activation of the Gα_s signaling axis on CD8⁺ T cells is sufficient to limit T cell responses to ICB. Intriguingly, this was paralleled by a significant decrease of tumor growth in CD8-Gnas KO mice compared to in littermate control mice (Extended Data Fig. 7a). In these mice, there was a significant decrease in CD8⁺ T cell exhaustion markers, such as PD-1⁺TIGIT⁺ tumor-infiltrating and experienced CD8⁺ T cells lacking Gnas (Extended Data Fig. 7b,c).

Next, we determined whether diminished response to immunotherapies following activation of Gα_s signaling was associated with diminished T cell functionality. Similar to our observations in the 4MOSC1 model, anti-PD-1 provided significant antitumor activity in CD8-GsD mice with MC38-OVA tumor cells in the absence of stimulation, but DCZ-treated mice led to failed responsiveness and worse overall survival (Fig. 4d,e). Treatment with anti-PD-1 in CD8-GsD mice led to a significant increase in tetramer⁺CD8⁺ T cells and IFNγ and granzyme B secretion (Fig. 4f), which were abolished by DCZ treatment concurrent with anti-PD-1 (Fig. 4f). These data suggest that activation of Gα_s signaling exclusively on CD8⁺ T cells leads to functional impairment and decreased recruitment of antigen-specific CD8⁺ T cells at the tumor, which ultimately leads to failure to respond to immunotherapies.

To begin exploring the clinical relevance of Gα_s–GPCR expression, we first investigated the correlation of expression between PD-1 and various Gα_s–GPCRs. In the The Cancer Genome Atlas skin cutaneous melanoma cohort, we found that expression of the GPCR genes GPR65, PTGER2, PTGER4, ADRB2 and ADORA2A were all significantly positively correlated with PD-1 expression in bulk tumors (Fig. 5a). This suggests that PD-1 expression is likely concurrent with the expression of these GPCRs. We next asked whether the expression of these GPCRs could predict immunotherapy response. We analyzed a cohort of 32 individuals with metastatic melanoma (total of 48 biopsies) treated with anti-PD-1, anti-CTLA-4 or combination therapy and where scRNA-seq was performed on the tumors before and after therapy³³. We observed that the expression levels of GPR65, PTGER2, PTGER4, ADRB2 and ADORA2A align with those of non-responders to immunotherapy, and the expression of PTGER2 and ADORA2A was significantly higher in CD8⁺ T cells from non-responders than in those from responders, with GPR65 nearing significance (Fig. 5b,c). We found that four of five of these Gα_s–GPCRs (PTGER4, GPR65, ADORA2A and PTGER2) have significant predictive power to identify individuals with melanoma who would not respond to immunotherapy. In this analysis, PTGER2 ranked at the top, with the highest area under the curve (AUC) of 0.78 (Fig. 5c). To test the extent of the role of immune suppression in immunotherapy response by Gα_s–GPCR pathway across cancers, we next computed the correlation between mean Gα_s–GPCR pathway levels in a cancer type and immunotherapy objective response rate (ORR) observed across 16 cancer types^14,34,35. Remarkably, in agreement with our hypothesis, we found that the mean Gα_s–GPCR pathway levels in a cancer type are most negatively correlated with the immunotherapy ORR out of all the G-protein signaling programs (Fig. 5d).

In summary, we propose a Gα_s–GPCR signaling axis that, when activated in CD8⁺ T cells, is sufficient to decrease cytotoxic function, exacerbate exhaustion-related phenotypes and abolish responses to immunotherapy (Fig. 6). Thus, the concomitant blockade of Gα_s–GPCRs with other inhibitory receptors, including PD-1 and CTLA-4, may be needed to garner a more effective and durable response to immunotherapy.

The sample size for each experiment was selected based on historical data and previous publications using the same tumor models^32,42. Mice from in vivo experiments were randomized based on tumor volume before initiation of treatment or data collection. Data collection and analysis were not performed blind to the conditions of the experiments. Data reported for all experiments were not subjective but rather based on quantitative analyses. All data collected were included and represented in the main figures or supplementary materials.

The 4MOSC1 cell lines were previously generated in-house by our lab. The MC38 cell line was generously gifted by A. Sharabi (University of California, San Diego). The MC38-OVA cell line was generated by retroviral transduction with pMSCV-OVA (gift from A. Sharabi). The 4MOSC1-SIINFEKL cell line was generated by lentiviral transduction with the pLenti-CMV GFP DEST vector. The 4MOSC1 cell lines were grown in keratinocyte medium with growth supplement, cholera toxin, epidermal growth factor and antibiotics. MC38-OVA cells were grown in DMEM supplemented with 10% fetal bovine serum, 1% antibiotics/anti-mycotics and 1 μg ml^–1 blasticidin. All cell lines were grown at 37 °C and 5% CO₂. PD-1 antibody (clone J43, BE0033-2; clone RMP1-14, BE0146), CTLA-4 antibody (clone 9H10, BP0131), isotype antibody (Armenian hamster IgG isotype control, BE0091; rat IgG isotype control, BE0251; Syrian hamster IgG isotype control, BE0087) and CD8 depletion antibody (clone YTS 169.4, BE0117) were obtained from Bio X Cell. DCZ was purchased from Tocris (7193).

All animal experiments used in this study were approved by the Institutional Animal Care and Use Committee (IACUC) of the University of California, San Diego, with protocol ASP S15195. Mice at Moores Cancer Center, University of California, San Diego, were housed in a microisolator and individually ventilated cages supplied with acidified water and were fed 5053 irradiated Picolab rodent diet 20 from LabDiet. Temperature for laboratory mice in our facility is mandated to be between ~18 and 23 °C with 40–60% humidity, and a 12-h light/12-h dark cycle was maintained for the facility. All animal manipulation activities were conducted in laminar flow hoods. All personnel are required to wear scrubs and/or a lab coat, mask, hair net, dedicated shoes and disposable gloves after entering the animal rooms. All animal studies conducted in this study were approved by the IACUC of the University of California, San Diego, with protocol ASP S15195.

The E8i^CreERT2 mice were obtained from D. Vignali (University of Pittsburgh)²³.The Gnas-exon 1^fl/fl mice were obtained from R. Iglesias-Bartolome (National Institutes of Health)²⁵. CD8-Gnas KO mice were generated by crossing E8i^CreERT2 with Gnas-exon 1^fl/fl mice. Information regarding genotyping of CD8-Gnas KO mice is listed in Supplementary Table 4. The Tet-GFP-PKI mice were obtained from R. Iglesias-Bartolome (National Institutes of Health). Transgenic mice were generated as previously described using a codon-optimized sequence for the 1–24 amino acids from PKIA fused to GFP²⁵. The ROSA26^{rtTA-IRES-EGFP} mice were obtained from A. Nagy (Samuel Lunenfeld Research Institute)⁴³. Information regarding genotyping of CD8-PKI mice is listed in Supplementary Table 5. OT-1 mice (C57BL/6-Tg(TcraTcrb)1100Mjb/J, 003831) were purchased from the Jackson Laboratory and bred in-house. The ROSA26^LSLGsDREADD mice were obtained from R. Berdeaux (The University of Texas, Houston)⁴⁴. ROSA26^GsDREADD mice were generated by crossing E8i^CreERT2 mice with ROSA26^LSLGsDREADD mice. Information regarding genotyping of CD8-GsD mice is listed in Supplementary Table 6.

Tamoxifen was purchased from Sigma-Aldrich. Where indicated, CD8-GsD mice were dosed with tamoxifen at 75 mg per kg (body weight). A stock solution of 15 mg ml^–1 was prepared by dissolving 75 mg of tamoxifen in 5 ml of miglyol and dissolved at 37 °C. After dissolving, the solution was stored at −20 °C protected from light. Mice were given 1.5 mg in 100 μl by i.p. injection for 3 d consecutively before tumor implantation or cell isolation. Mice were delivered 25 mg per kg (body weight) doxycycline in the diet (Bio-Serv) beginning the same day as initial tamoxifen administration. Mice consumed doxycycline for 5 d consecutively before cell isolation. DCZ was purchased from Tocris. Mice were dosed with DCZ at 0.01 mg per kg (body weight). A stock solution of 10 mg ml^–1 was prepared by dissolving 10 mg of DCZ into 1 ml of DMSO. Subsequently, a working concentration of 0.002 mg ml^–1 DCZ in PBS was prepared, and mice were given 0.0002 mg of DCZ in 100 μl i.p. daily.

For 4MOSC1 and 4MSOC1-SIINFEKL tumor xenografts, WT female C57BL/6 mice (4–6 weeks old) were purchased from Charles River Laboratories. WT mice or age-matched CD8-GsD mice were first given tamoxifen i.p. every day for 3 d. On the fourth day, 5 × 10⁵ tumor cells were injected into the tongues of mice, and when tumors reached ~30 mm³ (4–5 d after implantation), mice were treated by i.p. injection with isotype control antibody, anti-PD-1 (clone J43) or anti-CTLA-4 (clone 9H10; i.p. 10 mg per kg (body weight) three times a week). Where indicated, DCZ (i.p. 0.1 mg per kg (body weight)) was started on the fifth day and given daily.

For MC38-OVA tumor xenograft studies, WT female C57BL/6 mice (4–6 weeks old) were purchased from Charles River Laboratories. WT mice or age-matched CD8-GsD mice were first given tamoxifen i.p. every day for 3 d. On the fourth day, 1 × 10⁵ cells were injected subcutaneously into the flanks of mice. When tumors reached ~100 mm³ (8–10 d after implantation), mice were treated by i.p. injection with either isotype control antibody or anti-PD-1 (clone RMP1-14; i.p. 10 mg per kg (body weight) three times a week). Where indicated, DCZ treatment was started on the fifth day and given daily.

All mice were killed after the completion of the treatment, when control-treated mice succumbed to tumor burden or when tumors ulcerated. Mice were also killed when tongue tumors reached 8 mm in diameter or when flank tumors reached 1,500 mm³, as determined by the ASP guidelines.

Tumors were dissected as previously described⁴². In brief, tumors were isolated, minced and resuspended into a tumor dissociation kit (Miltenyi Biotec) and were processed with the gentleMACS Octo dissociator, according to the manufacturer’s protocol for tumor dissociation. Digested tumors were then passed through 70-μm strainers to generate single-cell suspensions and were processed further for flow cytometric analysis.

Splenocytes were isolated from 4- to 6-week-old mice and were mechanically disrupted. Red blood cells were lysed in red blood cell lysis buffer (BioLegend) according to manufacturer’s instructions. CD8⁺ T cells were isolated with an EasySep CD8 isolation kit by negative selection. For determining the expression of transgenes in CD4⁺ or CD8⁺ T cells, CD8⁺ T cells were first isolated through positive selection with CD8a MicroBeads (Miltenyi Biotec), followed by positive selection of the prior eluted fraction with CD4 MicroBeads. For activation, cells were then cultured at 1 × 10⁶ cells per well in 1 ml in 24-well plates with Dynabeads Mouse T-Activator CD3/CD28 beads at a 1:1 cell:bead ratio with 25 U ml^–1 human IL-2 (hIL-2; PeproTech) for 48 h. Naive CD8⁺ T cells were cultured with 25 U ml^–1 hIL-2 alone. After 48 h, activated cells were collected, counted and split into the ‘acute’ or ‘chronic’ activation groups. For acute activation, CD8⁺ T cells were cultured at 5 × 10⁴ cells per well in 200 μl in 96-well round-bottom plates with 25 U ml^–1 hIL-2 without beads. For chronic activation, CD8⁺ T cells were cultured at the same cell density with 25 U ml^–1 hIL-2 and with CD3/CD28 beads at a 1:2.5 cell:bead ratio. Where indicated, GPCR agonists were added at the following concentrations: 1 μM 16,16-dimethyl PGE₂ (Tocris), 5 μM dobutamine hydrochloride (Tocris) and 5 μM CGS-21680 hydrochloride (Tocris). For PGE₂ experiments, where indicated, inhibitors or CXCL10 were added at the following concentrations: 1 μM fsk (Tocris), 1 μM PF 04418948 (EP₂i; Tocris), 1 μM ONO AE3 208 (EP₄i; Tocris), 10 nM CXCL10 (gifted by T. Handel, University of California San Diego, La Jolla); PGE₂ and inhibitors were added in the last 24 h of culture. Inhibitors were first added for 30 min, and then 1 μM PGE₂ was added in the last 24 h of culture. For CD8-GsD in vitro experiments, 2 μM 4-hydroxytamoxifen (Sigma-Aldrich) was included in culture. Additionally, DCZ was added at a concentration of 0.002 mg ml^–1.

Splenocytes were isolated from OT-1 mice (4–6 weeks old) and activated with 100 nM OVA peptide (257–264; GenScript) and 50 U ml^–1 hIL-2 (PeproTech) for 48 h. After 48 h, fresh medium with 50 U ml^–1 IL-2 was added to the culture and incubated for an additional 24 h. Cells were then collected, replated and expanded at 1 × 10⁶ cells per ml.

For the specific cytotoxicity assay, tumor cells (target) were plated at 50,000 cells per well in a 24-well plate. OT-1 T cells (effector) were then added at a 1:5 target:effector ratio. Where indicated, GPCR agonists were added at the following concentrations: 1 μM 16,16-dimethyl PGE₂ (Tocris), 5 μM dobutamine hydrochloride (Tocris) and 5 μM CGS-21680 hydrochloride (Tocris). The coculture was left for 36 h, and cell viability was assessed by flow cytometric staining with Zombie Aqua viability dye (BioLegend).

The following flow cytometry antibodies (mouse) were purchased from BioLegend: CD45 (30-Fll; 1:100), CD3 (145-2Cll; 1:200), CD8a (53–6.7; 1:200), PD-1 (29F.1A12; 1:100), TIM-3 (RMT3-23; 1:100), IFNγ (XMGl.2; 1:100), granzyme B (GBll; 1:100), TNF-α (MP6-XT22; 1:100), CTLA-4 (UC10-4B9; 1:100), LAG3 (C9B7W; 1:100) and Ki-67 (16A8; 1:100). Anti-CREB (pS133)/ATF-1 (pS63; 1:20) was purchased from BD Biosciences. For viability staining of CD8⁺ T cells in vitro, cells were washed once with PBS and stained with Zombie Aqua viability dye (BioLegend) according to manufacturer’s instructions. Cell surface staining was done for 30 min at 4 °C. For intracellular and transcription factor staining, cells were stimulated with 1× cell activation cocktail with brefeldin A (BioLegend) in medium for 4–6 h at 37 °C before viability staining. Unstimulated cells were used as a control. After cell surface staining, cells were fixed with a FOXP3/transcription factor buffer set and stained with intracellular antibodies for 45 min at room temperature.

For flow cytometry acquisition using the Agilent NovoCyte Advanteon, the voltage and photodetector gains were automatically determined by calibration and quality control of the instrument. Automatic compensation was then performed using single-stained samples of OneComp eBeads compensation beads (Thermo Fisher Scientific), and voltage parameters were automatically determined based on predetermined photodetector gains from the calibration. Optimal dilutions for antibodies used in this study were determined based on titrating the antibody amount and adjusting the cell number used. The gating strategy for flow cytometry experiments is shown in Extended Data Fig. 8a,b, and gates were drawn based on non-activated and unstimulated control cells.

For CD8-GsD pCREB activation experiments in vivo, CD8-GsD mice were first given tamoxifen i.p. every day for 3 d. DCZ (i.p. 0.01 mg per kg (body weight)) was started on the fifth day and given daily for 5 d. Blood was collected from mice by retro-orbital bleeding, and samples were lysed and fixed with Lyse/Fix buffer (BD Biosciences). Cells were then permeabilized with Perm Buffer II (BD Biosciences) and stained with anti-CREB (pS133)/ATF-1 (pS63; BD Biosciences) according to the manufacturer’s instructions.

For CD8-GsD pCREB activation experiments in vitro, CD8-GsD mice were first given tamoxifen i.p. every day for 3 d. On the fourth day, blood was collected from mice by retro-orbital bleeding. Red blood cells were lysed in red blood cell lysis buffer (BioLegend) according to manufacturer’s instructions; 0.002 mg ml^–1 DCZ was added to the cultures for 15 min, and cells were fixed with CytoFix buffer (BD Biosciences) according to manufacturer’s instructions. Cells were then permeabilized with Perm Buffer II (BD Biosciences) and stained with extracellular antibodies and with anti-CREB (pS133)/ATF-1 (pS63; BD Biosciences) according to manufacturer’s instructions.

For PGE₂ pCREB activation experiments in vitro, chronically activated CD8⁺ T cells were serum starved for 1 h. PGE₂ (1 μM) was then added for 15 min in the presence or absence of inhibitors, and cells were fixed with CytoFix buffer (BD Biosciences) according to manufacturer’s instructions. Cells were then permeabilized with Perm Buffer II (BD Biosciences) and stained with extracellular antibodies and with anti-CREB (pS133)/ATF-1 (pS63; BD Biosciences) according to manufacturer’s instructions.

RNA was extracted from naive CD8⁺ T cells, activated CD8⁺ T cells, chronically stimulated CD8⁺ T cells and chronically stimulated CD8⁺ T cells treated with DCZ with an RNeasy mini kit following manufacturer’s instructions (Qiagen). RNA (100 ng) was converted to cDNA using a SuperScript VILO cDNA synthesis kit (Thermo Fisher Scientific). qPCR was performed using SYBR Select master mix (Thermo Fisher Scientific). Actb was used for normalization. The following primers were used for qPCR: Dusp1 forward 5′-GTTGTTGGATTGTCGCTCCTT-3′; Dusp1 reverse 5′-TTGGGCACGATATGCTCCAG-3′; Tox forward 5′-GCTCCCGTTCCATCCACAAA-3′; Tox reverse 5′-TCCCAATCTCT TGCATCACAGA-3′; Tigit forward 5′-GAATGGAACCTGAGGAGTCTCT-3′; Tigit reverse 5′-AGCAATGAAGCTCTCTAGGCT-3′; Actb forward 5′-GG CTGTATTCCCCTCCATCG-3′; Actb reverse 5′-CCAGTTGGTAACAA TGCCATGT-3′.

For scRNA-seq integration, we used the Seurat scRNA-seq SCTransform integration method⁴⁵. We collected 30 datasets with publicly available raw counts across 19 cancer types. Each dataset was transformed individually with the SCTransform function to regress out mitochondrial percentage, S score, G2M score and DIG score. The S, G2M and DIG scores were obtained from a recent publication. The data were then prepared for integration with the SelectIntegrationFeatures, Prep-SCTIntegration and FindIntegrationAnchors functions and integrated with the IntegrateData function. Principle-component analysis and dimensionality reduction were performed with the RunPCA and RunU-MAP functions, respectively. CD8⁺ T cell states were identified using the ProjecTILs package, the multimodal reference mapping workflow in Seurat and key marker genes, resulting in five major functional states: naive, cytotoxic, effector memory, proliferative and exhausted. To examine the relative expression of GPCRs between CD8⁺ T cell subtypes, the average expression of each GPCR was calculated for each of the five groups, row normalized for each GPCR and visualized on a circular heat map using circlize v0.4.12 and ComplexHeatmap v2.6.2. The package versions used were circlize v0.4.12, ComplexHeatmap 2.6.2, ggplot2 3.3.3 and Seurat 4.0.1.

For calculation of the T cell dysfunction score, the average expression of all genes in the T cell dysfunction gene set was calculated and plotted for each CD8⁺ T cell from the integrated CD8⁺ T cell dataset. Spearman correlations for each GPCR and the dysfunction score were calculated by correlating expression of the GPCR with mean expression of the dysfunction gene set for each CD8⁺ T cell from the melanoma dataset (GSE120575).

For differential expression analysis, three datasets of LCMV RNA-seq data were batch corrected with ComBat (v3). Subsequently, DESeq2 (version 1.32) was used to analyze all effector CD8⁺ T cells versus all exhausted CD8⁺ T cells. P values from this analysis were adjusted for multiple hypothesis correction and false discovery rates with Benjamini–Hochberg testing^46,47.

For gene set enrichment analysis, we created GPCR gene sets on the basis of their G-protein coupling mechanisms. We considered either G-protein family-level transduction mechanisms from IUPHAR or individual G-protein couplings from a recent experimental TGFα shedding assay that we augmented through machine learning-based predictions^22,48. For each G-protein family coupling from IUPHAR, we created gene sets based on primary and secondary mechanisms either in isolation or combined. For the experimental TGFα shedding assay, we considered as couplings binding with log (relative intrinsic activity, RAi) values greater than −1.0. We defined the predicted couplings by considering either a looser (0.5) or more stringent (0.9) cutoff of the coupling probabilities outputted from PRECOG⁴⁸. For each individual G-protein coupling, we created gene sets by considering the experimental and predicted couplings either in isolation or in combination. We created gene sets in the .gmt format by considering corresponding Entrez IDs, and we performed gene set enrichment analysis through ClusterProfiler (v.4.6.2)⁴⁹, giving as an input the list of genes ranked according to log₂ (fold change) values from differential expression analysis.

For analysis of ORR, expression profiles of tumors from 2,277 individuals across 14 cancer types treated with immune check-point inhibitors were collected^14,35,50. Mean expression of each GPCR was calculated and classified based on coupling information from IUPHAR.

Graphs were plotted using GraphPad Prism v9.2.0 (GraphPad Software) and R (v4.2.2). Where indicated, data are expressed as mean ± s.e.m. The following tests were performed (see figure legends for details): correlation, unpaired two-tailed Student’s t-test, one- or two-way ANOVA and log-rank (Mantel–Cox) tests for Kaplan–Meier survival curves. Statistical significance was determined as *P < 0.05, **P < 0.01 and ***P < 0.001. Data distribution was assumed to be normal, but this was not formally tested.