T cells are critical mediators of immunity, harboring a diverse repertoire of T cell receptors (TCRs) that facilitate recognition and elimination of both foreign infectious agents and malignant host cells. Differentiation of a naive T cell to an effector T (Teff) cell is orchestrated by metabolic, epigenetic and transcriptional reprogramming events required for effective disease control. Yet the activity of Teff cells is tightly regulated to prevent autoimmunity. Upon antigen clearance, Teff cells contract, maintaining a pool of long-lived memory T (Tmem) cells. In non-resolving inflammatory settings such as cancer, however, T cell fate diverges from memory formation to a metabolically and epigenetically distinct effector lineage, termed exhaustion1,2,3. T cell infiltration in cancer is a well reported positive prognostic marker, yet even patients with immunologically active tumors may progress despite therapeutic strategies targeting inhibitory pathways on exhausted T cells3,4,5. Immunotherapy via antibody-mediated blockade of ‘checkpoint’ receptors, such as PD-1, have highlighted both strengths and limitations of re-igniting antitumor immunity. While checkpoint blockade therapy has been a success, clinical efficacy of these agents can vary dramatically across tumor types, with most patients not achieving sustained remission. Beyond the simple number of infiltrating cells, the makeup of a patient’s tumor-infiltrating lymphocytes (TILs) plays a substantial role in the immunotherapy response3,5. Further, cell-extrinsic factors such as metabolic restriction and interaction with immunoregulatory populations also hinder effective TIL-mediated tumor control6,7. A deeper understanding of mechanisms of immune suppression in cancer is needed to advance development of better immunotherapies.

Exhaustion has broad and somewhat contested terminology. Although it has now been well established that exhaustion in CD8+ T cells is not a singular state but is instead a progressive lineage that differentiates though (at least) two discrete stages. A pool of progenitor exhausted T (pTex) cells, maintained by the transcription factor TCF-1, is held in check by the PD-1 receptor and carries self-renewal capacity, which can seed the predominant TCF-1 terminally exhausted T (tTex) cell pool1. Progression of the pTex lineage toward the tTex cell fate has been defined by progressive loss of effector cytokine production and proliferative potential, upregulation of multiple co-inhibitory and co-stimulatory receptors and metabolic derangements that precipitate these impairments. Exhausted T cells have been further characterized by a distinct epigenic and transcriptional signature2,3,8. Persistent TCR signaling, exaggerated by metabolic and inflammatory stress, is believed to be the primary driver of T cell exhaustion9. Considerable efforts to deconvolute this state have proposed that diminished responsiveness in exhausted T cells is a compensatory mechanism to limit, or suppress, immunopathology in settings of persistent stimulation1,10. Indeed, tTex cells have been identified in numerous chronic pathologic settings but also in the context of homeostatic immune processes, such as in the placenta during gestation11. In both, T cell exhaustion serves to tolerize T cells experiencing continuous antigen exposure. Thus, exhaustion may be a developmentally advantageous program to control persistent T cell activation that arises in various pathologic states.

A high proportion of tTex cells among the total tumor immune infiltrate, for instance, predicts resistance and not response to therapeutic PD-1 blockade, despite these cells expressing the highest levels of PD-1 receptor3,5,12. Functionally, tTex cells are often characterized by what is lost: diminished polyfunctionality, proliferative potential and stunted capacity to lyse target cells. Yet beyond the loss of function associated with terminal exhaustion, tTex cells have also been characterized by gain of functions in the upregulation of effector molecules involved in immune suppression and wound repair (for example IL-10 and CD39).

As cancer progresses, increased secretion of angiogenic factors and heightened oxidative metabolism yield significant oxygen deprivation across the tumor bed and activate hypoxia and energetic stress responses6,7. In CD8+ T cells, sustained exposure to hypoxia rapidly accelerates differentiation to terminal exhaustion and represses effective antitumor immunity13. In both preclinical models and patients undergoing anti-PD-1 immunotherapy, heightened tumor oxidative metabolism and hypoxia staining correlates with poor immunotherapeutic efficacy driven, in part, by CD8+ T cell dysfunction14,15. T cells sense low oxygen tensions through hypoxia-sensitive elements (for example HIF1α) that orchestrate transcriptional and post-translational programs. One such gene is CD39 (Entpd1), a key member of the adenosine pathway and an ectoenzyme regularly cited as a faithful marker of tumor antigen experienced tTex cells in murine and human disease16. CD39 rapidly catabolizes extracellular proinflammatory adenosine tri- and diphosphate (eATP and eADP) to the mono-phosphorylated form. CD73 then acts on AMP to produce adenosine, inhibiting local signaling cascades. It remains unclear, however, whether expression of regulatory molecules such as CD39 on CD8+ T cells engenders meaningful consequence to antitumor immunity.

Here we reveal CD8+ tTex cells as a regulatory population in solid tumors with a similar suppressive capacity to neighboring CD4+Foxp3+ regulatory T (Treg) cells. However, unlike Treg cells, which utilize many nonredundant suppressive mechanisms, tTex cells elicit their inhibitory function solely through CD39. Deletion of CD39 in CD8+ tTex cells augments immunotherapy and its expression on tTex cells is reduced by pharmacological or genetic elevation of intratumoral oxygen tension, dampening the suppressor potential of tTex cells. Thus, we define tTex cells not solely as a hypofunctional state, but one capable of conditionally possessing potent antifunctional characteristics that suppress local T cell immunity.


Intratumoral exhausted T cells are functionally suppressive

Enrichment of tTex cells in solid tumors has been shown to correlate with poor clinical responses to checkpoint blockade therapy3,5; however, it remains unclear whether ineffective re-stimulation of antitumor immunity is simply the result of a loss of T cell functionality or whether tTex cells themselves contribute to the tolerogenic tumor microenvironment (TME). Thus, we sought to determine whether tTex cells were directly immunosuppressive. To begin to delineate alternative effector roles in terminal exhaustion, we employed the aggressive, immunotherapy-resistant B16-F10 melanoma line, orthotopically implanted intradermally in C57BL6 mice. On day 14 (tumor area, 8–10 mm2), PD-1hiTim-3+ tTex cells are the plurality both among CD8+ T cell populations and among total CD3+ T cells (Fig. 1a,b and Extended Data Fig. 1a–c). At day 14, infiltrating CD8+ T cells were sorted into four groups, stratified by the expression of inhibitory receptors, PD-1 and Tim-3 (PD-1, PD-1int, PD-1hiTim-3 and PD-1hiTim-3+;)8,13. Coexpression of PD-1 and Tim-3 has been shown to reliably identify tTex cells with tumor antigen specificity and classic hallmarks of exhaustion such as loss of polyfunctional cytokine production, poor proliferation and elevated TOX expression. Effector pTex cells with antitumor potential are identified by isolated expression of PD-1 (PD-1 and PD-1int) concomitant with Tcf1, Slamf6 (Ly108) and CXCR5. As CD8+ T cells progress to terminal exhaustion and proinflammatory functions are repressed, tTex cells tend to upregulate genes associated with Foxp3+CD4+ Treg cells (Fig. 1c and Extended Data Fig. 1d); these include the transcription factor Helios (Ikzf2) and molecules driving Treg cell suppressor functions: neuropillin 1 (Nrp1), IL-2 receptor-α (Il2ra, CD25), Fas ligand (Fasl), CD39 (Entpd1) and numerous co-inhibitory and co-stimulatory receptors (cytotoxic T lymphocyte antigen 4 (Ctla4), lymphocyte activating 3 (Lag3) and T cell immunoreceptor with Ig and ITIM domains (Tigit)). Notably, by mining well-studied RNA-seq datasets3,8,17, we discovered that beyond upregulation of immunosuppression-related effector molecules, tTex cells upregulate a broader transcriptional signature in common with intratumoral Treg cells. Tumor-infiltrating CD8+ T cell transcriptomes expectedly cluster separately from CD4+Foxp3+ Treg cells and Foxp3 conventional CD4+ T (Tcon) cells from the same environment (Fig. 1d); however, if we utilize the Magnuson et al. tumor-infiltrating Treg cell signature as a template, which defines genes uniquely upregulated in B16-infiltrating Treg cells compared to their lymph node-resident counterparts, we observe significant gene set enrichment in CD8+ tTex cells versus their progenitor counterparts (Fig. 1e and Extended Data Fig. 2a–f). These data suggest that CD4+ Treg cells and CD8+ tTex cells possess some overlap in transcriptional programming in the TME, which may support exhausted CD8+ T cells possessing a suppressive phenotype upon terminal differentiation.

Fig. 1: Terminally exhausted CD8+ T cells infiltrating tumors are functionally suppressive.
figure 1

a, CD8+ tumor-infiltrating T cells from day 14 B16-F10 murine melanoma tumor and tumor-draining lymph nodes (dLNs). b, Total CD3+ T cells from B16-F10 TILs and dLNs with quantified fold change. c, Expression of CD4+Foxp3+ Treg cell-associated genes among tumor-infiltrating T cell populations. d, Principal-component analysis plot of tumor-infiltrating T cells from RNA-seq mined from data in Magnuson et al.17 and Miller et al.3. e, Gene set enrichment analysis (GSEA) of tumor-infiltrating Treg cell signature on tetramer+SLAMF6+ progenitor and Tim-3+ terminally exhausted CD8+ T cell transcripts from Miller et al. f,g, Model of suppression assay (f) and results from sorted B16-F10 melanoma TILs (g). Statistics are linear regression (g) with P < 0.05, P < 0.01 and P < 0.001.

Source data

To determine whether intratumoral tTex cells indeed possess intrinsic suppressive functions, we next compared tumor-infiltrating CD8+ T cell populations with CD4+ Treg cells directly ex vivo in miniaturized T cell suppression assays18. B16-F10 tumors were intradermally injected into Treg cell reporter mice (Flpo and mAmetrine concomitant with Foxp3; hereafter, Foxp3Ametrine-Flpo) and T cell populations were isolated at day 14. These putative ‘suppressor’ populations were then co-cultured at diminishing ratios with proliferation dye-labeled, congenically mismatched naive CD4+ T cells (responder T cells) and T cell-depleted splenocytes coated with anti-CD3 (Fig. 1f). Notably, on a per-cell basis, CD8+ tTex cells and CD4+Foxp3+ Treg cells, sorted from the same B16-F10 tumor, displayed equivalent suppression of co-cultured responding T cells (Fig. 1g). Progenitor exhausted T cells carried no suppressor capacity; in fact, no other tumor-infiltrating CD4+ or CD8+ T cell population assayed, regardless of their inhibitory receptor expression, displayed detectable suppressive function (Fig. 2a). While suppression assays typically utilize naive CD4+ or CD8+ T cells as responders, in repeat experiments, we found that tTex cells were also capable of suppressing memory or effector CD8+ T cells isolated from the tumor-draining lymph nodes (dLNs) of B16-F10-bearing mice (Fig. 2b) or their intratumoral pTex cell counterparts (Fig. 2c).

Fig. 2: Suppressive function of tTex cells affects many targets and is environment-dependent.
figure 2

a, CD4+ and CD8+ TIL populations from B16-F10 tumors stratified by inhibitory receptor expression and utilized as the ‘suppressor’ group in ex vivo suppression assay. b, Suppression assay utilizing diverse responder populations from dLNs and co-cultured with PD-1hiTim-3+ tTex cells or Foxp3+ Treg cells. c, Proliferation dye-labeled pTex cells co-cultured with tTex cells in repeat suppression assays. df, Suppression assay with TIL from Ptenf/fBRafLSL.V600ETyr2Cre.ER-derived melanoma clone, dubbed clone 24 (d), PD-1-resistant MEER head and neck carcinoma (e) or PD-1-sensitive MC38 adenocarcinoma (f). Statistics used were one-way analysis of variance (ANOVA) (a) and linear regression (df) with P < 0.05, P < 0.01, P < 0.001 and P < 0.0001.

Source data

Taken together, these data suggest that tTex cells harbor a regulatory function unique among CD8+ TILs and, given that tTex cells are significantly enriched in the TME, represent an unappreciated suppressor population within solid tumors. We next validated our ex vivo findings in other size-matched immunotherapy-resistant tumor models including clone 24, a poorly infiltrated melanoma line derived from the autochthonous Ptenf/fBrafLSL.V600ETyr2Cre.ER mouse model15 (Fig. 2d and Extended Data Fig. 3a) and MEER, an immunotherapy-resistant head and neck carcinoma line14,18 (Fig. 2e). Isolated from these diverse environments, tTex cells consistently displayed measurable ex vivo immune suppression, whereas progenitor cells did not; however, not every TME yielded PD-1hiTim-3+ tTex cells with a functional suppressive character. CD8+PD-1hiTim-3+ TILs isolated from anti-PD-1-sensitive adenocarcinoma MC38 failed to suppress T cell proliferation (Fig. 2f and Extended Data Fig. 3b). Notably, we recently reported that the regulatory capacity of Foxp3+ Treg cells is mediated by their local metabolic milieu18. Here, we also observe that suppressive features of tTex cells are tuned by their environment, but within a greater dynamic range. These data suggest there is some plasticity in this antifunctional suppressor program that may be targeted for therapeutic benefit. Further, these data highlight that inhibitory receptor expression is not a requisite for suppressive functionality, as PD-1hiTim-3+ tTex cells were similarly enriched across all tumors assayed (Extended Data Fig. 3c). We thus sought to understand the mechanism of tTex cell suppression to delineate the environmental cues that enforce regulatory functions in CD8+ TILs.

tTex cell apoptosis enhances suppressive function ex vivo

CD4+Foxp3+ Treg cells are potent suppressors that employ multiple, sometimes redundant means to suppress immune function. We dissected various potential inhibitory mechanisms employed by tTex cells. Several reports have identified a subpopulation of CD8+ T cells that secrete IL-10 and play far-reaching regulatory roles in chronic disease states; protecting against autoimmune disease19, orchestrating chronicity in viral infection20 and supporting cancer progression21; however, germline deletion of IL-10 did not perturb tumor growth versus wild-type C57/BL6 mice (Extended Data Fig. 4a,b) and sorted Il10–/– tTex cells had similar suppressive function to wild-type tTex cells (Fig. 3a). A well-characterized IL-10 neutralizing antibody22 also failed to neutralize tTex cell-mediated suppression (Extended Data Fig. 4c). These experiments suggest that while IL-10 has classically been defined by its suppressive character in situ (and also increasing by its influence on T cell metabolism21,23), in this context, IL-10 does not contribute to observed suppression by tTex cells.

Fig. 3: Ex vivo tTex cell suppression is associated with tTex cell-intrinsic apoptosis.
figure 3

a, Suppression assay of B16-F10 TIL from wild-type (WT) C57/BL6 or Il10–/– mice. b, Suppression assay of B16-F10-derived tTex cells co-cultured with activated C57/BL6 or Bcl2tg-responding T cells. c, Cleaved caspase-3 staining in CD8+ T cells from B16-F10 or MC38 tumors and dLNs. d, Suppression assay of B16-F10 TILs from C57/BL6 or Bcl2tg mice. e, Repeat experiments from Fig. 4d with live or mitomycin C (MC)-killed tTex cells. Statistics are linear regression (a,b,d,e) and one-way ANOVA (c) with P < 0.05, P < 0.01, P < 0.001 and P < 0.0001.

Source data

We next investigated whether tTex cell-mediated cytotoxicity could explain suppressor function in ex vivo assays. In both mice and humans, CD8+ T cells have repeatedly been shown to induce tolerance in autoimmune disease through lysis of autoreactive CD4 T cells24. Indeed, cytotoxicity also contributes to CD4+Foxp3+ Treg cell suppression25; however, in ex vivo suppression assays, we do not observe meaningful changes in cell viability among responding CD4+ T cells or antigen-presenting cells (APCs) regardless of the co-cultured putative ‘suppressor’ population, or the ratio of suppressors to responders (Extended Data Fig. 4d, e). These data suggest CD8+ tTex cells are not enforcing immune suppression by direct cell lysis ex vivo. Despite little evidence that tTex cell cytotoxicity was responsible for the observed immunosuppression, we sought a more robust study to validate our initial observations. To accomplish this, we derived the responding T cell and APC pools from the well-established model of apoptosis resistance via BCL-2 overexpression (Bcl2tg). Here too, we observed no diminishing of tTex cell-mediated suppression and, in fact, by improving cell viability in the responding population (Extended Data Fig. 4f), tTex cells seemed more suppressive (Fig. 3b). Of note, viability of APCs was dispensable for the assay: if apoptosis was directly induced in APCs by mitomycin C treatment before their addition in suppression assay, tTex cells retained suppressive capacity (Extended Data Fig. 4g); however, tTex cell suppression was overcome by replacing the typically used T cell-depleted splenocyte APCs with anti-CD3/anti-CD28-coated microbeads, suggesting that local immune cells may be directly influencing the capacity of tTex cells to suppress T cell responses.

While direct cytotoxicity by tTex cells was not apparent in our ex vivo experiments, numerous studies have suggested that tTex cells themselves are apoptosis prone2,13,20. Indeed, directly ex vivo, suppressive tTex cells from B16-F10 display elevated cleaved caspase-3 staining versus progenitor counterparts or non-suppressive MC38 tTex cells (Fig. 3c). In ex vivo co-culture too, tTex cells are significantly less viable than progenitors after 3 days, despite initiating the assay with sorted live populations (Extended Data Fig. 4h). Intriguingly, apoptosis-resistant Bcl2tg tTex cells from B16-F10 tumors displayed markedly reduced suppressive capacity ex vivo (Fig. 3d and Extended Data Fig. 4i), suggesting that there may be some role for cell death in mediating tTex cell control of local immune functions. Further, when we triggered apoptosis in Bcl2tg tTex cells with mitomycin C treatment before assay, we could augment their suppressive function (Fig. 3e). Evidence suggests that CD4+Foxp3+ Treg cells retain suppressive functions as they undergo apoptosis26. Thus, tTex cells may not suppress local immunity by killing, but rather further exert their regulatory capabilities though dying.

CD39 is required for tTex cell suppressor function

Apoptosis bolsters Treg cell suppressive function through the release and subsequent CD39/CD73-mediated hydrolysis of extracellular ATP (eATP) to adenosine27. CD39 faithfully marks the most terminally exhausted CD8+ T cells16,28 both in human (Extended Data Fig. 5a,b) and murine (Fig. 1c) cancers. Notably, in our ex vivo system, higher CD39 expression among tTex correlated with increased suppression (9.64% increase per 0.1 log CD39 mean fluorescence intensity (MFI); P < 0.001; Fig. 4a). To determine the importance of CD39 density in suppression, we flow cytometrically purified CD39hi or CD39lo tTex cells from either B16-F10 (harboring highly suppressive tTex) or MC38 (harboring poorly suppressive tTex cells). This revealed that within MC38-derived tTex, CD39hi cells carried some suppressor function compared to CD39lo cells (Fig. 4b and Extended Data Fig. 5c). These data highlight that density of surface CD39 expression (rather than simply whether it is expressed) likely plays a role in observed ex vivo suppressive function.

Fig. 4: tTex cells suppress through CD39-mediated eATP depletion and adenosine production.
figure 4

a, Pearson correlation of percent tTex cell-mediated suppression in 1:4 ratio in suppression assay versus fold change in CD39 MFI in tTex cells between murine tumor lines. b, Repeat experiments from Figs. 1g and 2f, in which tTex cells from B16-F10 and MC38 tumors were stratified by CD39 expression. c, Percent CD8+ T cell populations from B16-F10-expressing CD39 or CD73. d, Representative bi-variant plots of B16-F10 infiltrating T cell populations. e, Suppression assay of B16-F10 TIL co-cultured with activated CTV-labeled C57/BL or Nt5e– ‘responding’ T cells. f, Suppression assay of B16-F10 infiltrating CD8+ tTex cells co-cultured with dimethylsulfoxide (DMSO) vehicle or A2AR/A2BR small-molecule inhibitor AB928 at 3 μg ml−1. g, Suppression assay of B16-F10 TIL from Entpd1f/f or Cd4CreEntpd1f/f mice. Statistics are linear regression (a,eg) and one-way ANOVA (b,c) with P < 0.05, P < 0.01, P < 0.001 and P < 0.0001.

Source data

Foxp3+ Treg cells generate adenosine through coordinate action of CD39 and CD73, which they coexpress. Notably, unlike Treg cells, CD8+ tTex cells largely do not coexpress CD73 and during differentiation to exhaustion ‘switch’ from being CD73hiCD39 to CD73loCD39+ (Fig. 4c,d). Analysis of CD8+ T cells across numerous lymphoid and non-lymphoid organs in a tumor-bearing mouse reveals that CD39 expression is essentially restricted to intratumoral tTex cells and is undetectable in cells from non-inflamed healthy tissues (Extended Data Fig. 5d). Thus, while tTex cells may be poorly capable of producing adenosine in isolation, as they are CD73lo, elevated CD39 expression may contribute to immune suppression by supporting the generation of an adenosine-rich environment with local CD73+ cells. Indeed, circulating human CD73loCD39+ Treg cells have been shown to associate with CD73-expressing cells or exosomes to elicit their suppressive function29, supporting the notion that CD39+ tTex cells may be suppressive only when in proximity to CD73+ cells.

To model the influence of CD73+ cells in tTex cell-mediated suppression, we next derived responder T cell and APC populations from CD73-deficient (Nt5e–/–) mice. Indeed, wild-type tTex cells were significantly less suppressive when APCs and responding T cells were CD73-deficient (Fig. 4e); however, tTex cell suppressive capacity was not wholly absent in this setting, suggesting that the residual CD73 expression on tTex cells may mediate some control of local immune activation via the eATP–adenosine axis. To further elucidate the role of adenosine signaling in tTex cell-mediated suppression, we treated cells with a small-molecule inhibitor specific to the adenosine receptors, A2AR and A2BR (AB928; Arcus Biosciences), which showed that blockade of adenosine receptor signaling is sufficient to completely ablate tTex cell suppression ex vivo (Fig. 4f). Notably, neither CD73 deficiency (Extended Data Fig. 5e) nor adenosine receptor blockade (Extended Data Fig. 5f) prevented CD4+Foxp3+ Treg cell-mediated suppression in ex vivo co-culture, revealing that unlike the numerous inhibitory mechanisms employed by Treg cells, tTex cells employ a single dominant method of constraining local immune responses, via the generation of a local environment primed for generating adenosine.

Next, to specifically interrogate the consequences of CD39 activity on tTex cells, we bred Entpd1f/f (encoding CD39) mice to Cd4Cre mice (hereafter, Cd4CreEntpd1f/f)30, generating T cell-specific deletion of CD39 (Extended Data Fig. 6a). In large B16-F10 tumors (8–10 mm; day 14), CD39 expression was dispensable for expression of inhibitory receptors, PD-1 and Tim-3 (Extended Data Fig. 6b). While the percent of PD-1hiTim-3+ tTex among CD8+ TILs did not change with CD39 deletion in T cells, tTex cells were slightly but significantly reduced by cell number in Cd4CreEntpd1f/f mice (Extended Data Fig. 6c); however, CD39-deficient tTex cells displayed a complete loss of inhibitory capacity compared to CD39-competent counterparts (Fig. 4f). T cell-specific deletion of CD39 was also sufficient to slow tumor growth (Extended Data Fig. 6d); however, in this Cre-lox system, we cannot distinguish the contributions of CD39 deletion in CD4+ Treg cells versus CD8+ tTex cells to the improved tumor response.

It is noteworthy that CD39, like many other T cell exhaustion-associated markers (for example PD-1 and Tim-3), is rapidly and transiently upregulated in the immediate response to antigen recognition. Acute infection models have revealed the temporal nature of CD39 upregulation in activated CD4+ and CD8+ T cells. To determine whether activation-induced CD39 expression, or activated Teff cells in general, were capable of suppressing local T cell responses, we generated Teff cells from a model of acute infection, transferring 1 × 105 naive Thy1.1+ OT-I TCR transgenic CD8+ T cells from Cd4CreEntpd1f/f or Entpd1f/f mice into wild-type Thy1.2+ hosts infected with OVA-expressing Vaccinia virus (VacciniaOVA). Regardless of genotype, Thy1.1+CD44+ OT-I Teff cells isolated at 8 days after transfer failed to suppress co-cultured responding T cell proliferation, (Extended Data Fig. 6e) and displayed minimal CD39 expression in ex vivo culture (Extended Data Fig. 6f).

We next assessed the requirement for TCR stimulation in tTex cell suppression. When OT-I CD8+ T cells were used as responders, polyclonal intratumoral tTex cell-mediated suppression was observed when cultures were stimulated with anti-CD3 (engaging the TCR of both tTex cells and responder cells), typically performed in suppression assay protocols; however, when TCR signaling in tTex cell was bypassed by activating OT-I responders with cognate peptide, tTex cell-mediated suppression was lost (Extended Data Fig. 6g). Notably, while PD-1 and Tim-3 expression remained elevated in tTex cells between these two stimulation conditions (Extended Data Fig. 6h, i), CD39 was absent in tTex cells in SIINFEKL co-cultures, suggesting that TCR signaling was required to maintain CD39 expression on tTex cells (Extended Data Fig. 6j).

CD39 overexpression restricts local T cell activation

We next sought to better understand the mechanism by which CD39 may be limiting antitumor immunity with a gain-of-function approach. We retrovirally overexpressed CD39 concurrent with mCherry (pMSCV-Entpd1-IRES-mCherry; Fig. 5a). Transduced polyclonal CD8+ T cells bore CD39 density similar to B16-F10-derived tTex cells in situ (Fig. 5b) and possessed functional enzymatic activity (Fig. 5c). As eATP has been defined as a crucial autocrine signaling molecule in activated T cells, we first investigated the effects of CD39 overexpression on the overexpressing cell itself. Compared to empty vector control CD8+ T cells (pMSCV T cells), CD39-overexpressing T cells displayed significant reduction in TCR signaling via phosphorylated ZAP-70 (Y319) and Akt (S473) (Fig. 5d,e), consistent with the role of adenosine on T cell activation. CD39-overexpressing T cells also displayed both diminished and delayed switch to glycolysis when acutely stimulated during extracellular flux assays as previously described31 (Extended Data Fig. 7a–d). Reduced TCR signaling in CD39-overexpressing T cells was also evident in diminution of downstream calcium flux (Extended Data Fig. 7e) and cytokine assays (Extended Data Fig. 7f,g). These data highlighted CD39-mediated autoregulation, yet did not reveal the ability of CD39 to impose suppression of nearby T cells. Thus, we next investigated the effect of CD39 overexpression on neighboring cells, as the presence of ectonucleotidase would presumably generate an adenosine-rich microenvironment. We repeated re-stimulation experiments in 1:1 co-culture using congenically mismatched CD39-overexpressing and empty vector control T cells. Compared to control T cells cultured alone, CD39-overexpressing T cells displayed a significant reduction in cytokine production (Fig. 5f) with correspondingly reduced proliferation (Fig. 5g); however, when control transduced T cells were cultured in the presence of CD39-overexpressing T cells, the control T cells suffered an equivalent loss of cytokine production and proliferation compared to the CD39-overexpressing cells alone (Fig. 5f,g), suggesting that enforced CD39 expression is sufficient for suppressor functions even in in vitro-generated Teff cells.

Fig. 5: Enforced CD39 expression in Teff cells inhibits intrinsic and neighboring T cell function.
figure 5

a, Model of activated T cells transduced with pMSCV-Entpd1-IRES-mCherry and then rested for 7 d before assay. b, CD39 expression in day 7-transduced T cells or day 14-B16-F10 TILs. c, Percent of spiked in ATP consumed in transduced T cells. Entpd1oe, Entpd1 overexpression. d,e, Immunoblot of phosphorylation events following of anti-CD3/anti-CD28-coated microbead stimulation in transduced T cells (d) and quantified band intensity normalized to t = 0 in empty vector T cells (e). f, Cytokine production of transduced T cells in 24-h co-culture experiments following anti-CD3/anti-CD28-coated microbead stimulation. OE, overexpression; EV, empty vector. g, Division on CTV-labeled transduced T cells over 72 h. Statistics used were linear regression (c) and one-way ANOVA (eg) with P < 0.05, P < 0.01, P < 0.001 and P < 0.0001.

Source data

tTex cells are functionally important suppressors in tumors

Several immune cells within the TME, as well as tumor cells and stroma, can express some level of CD39. Both extracellular and intracellular factors regulate the expression of CD39, including activation signals and hypoxia-sensitive programs. So, we next sought to determine the specific contribution of tTex cell-mediated immunosuppression to the TME. Within the CD8+ compartment of a tumor-bearing mouse, CD39 expression is essentially restricted to tTex cells (Extended Data Fig. 5d); therefore, to determine the significance of CD39 expression of tTex cells alone, we employed a tamoxifen (TAM)-inducible, CD8-specific mouse, in which transgenic Cre-ERT2 is under the control of the CD8a enhancer (E8i) and a Rosa26-driven recombination reporter, crossed with Entpd1f/f (E8iGFP-Cre-ERT2-Rosa26LSL-TdTomatoEntpd1f/f; hereafter, E8iCre-ERT2Entpd1f/f). This mouse was used to induce Entpd1 deletion specifically in CD8+ T cells before B16-F10 tumor engraftment (Extended Data Fig. 8a). Indeed, similar to our initial findings in the Cd4Cre system, inducible CD8-specific deletion of CD39 resulted in significantly enrichment of Slamf6+Tim-3 pTex cells and fewer in Slamf6Tim-3+ tTex cells in B16 tumors compared to TAM-treated controls (Fig. 6a). Notably, in these advanced tumors, we did not observe significant changes in ex vivo cytokine production in TILs (Extended Data Fig. 8b).

Fig. 6: CD39 on endogenous intratumoral tTex cells suppresses newly infiltrating T cells.
figure 6

a, Tim-3 and Slamf6 staining on CD8+ T cells from day 14 B16-F10 tumors. b, Schematic of adoptive cell therapy protocol. Entpd1f/f or E8iCre-ERT2Entpd1f/f mice were treated with TAM starting 3 days before B16-F10 tumor inoculation then continued thrice weekly thereafter. The 3–5 million activated pmel-I T cells were transferred retro-orbitally on day 7. T.I.W., three times a week. c, Tumor growth from adoptive T cell transfer experiment. d,e, Percent infiltration and absolute numbers of adoptively transferred pmel-I T cells into B16-F10 tumor. f, IFN-γ production in pmel-I T cells restimulated ex vivo with gp100. Statistics used were Mann–Whitney U-test (a,e,f) and two-way ANOVA (c) with P < 0.05, P < 0.01, P < 0.001 and P < 0.0001.

Source data

We next sought to determine, using the E8iCre-ERT2Entpd1f/f system, whether CD39+ tTex cells suppressed tumor-infiltrating T cells in vivo. To do this, we inoculated B16-F10 cells into TAM-treated E8iCre-ERT2Entpd1f/f or control mice and at day 7, adoptively transferred a subtherapeutic dose of activated, congenically marked pmel-I T cells (carrying a transgenic TCR specific for the melanocyte antigen, gp100, expressed in B16-F10) (Fig. 6b), allowing us to determine whether CD39 expression on tTex cells mediated suppression of a wild-type T cell population in tumors. Ten days after pmel-I T cell transfer, B16-F10 tumors on E8iCre-ERT2Entpd1f/f mice were significantly smaller compared to controls (Fig. 6c). Further, while the number of transferred T cells in dLNs was equivalent between groups (Extended Data Fig. 8c), pmel-I T cells within the tumor were on average four times more numerous in E8iCre-ERT2Entpd1f/f mice compared to controls (Fig. 6d,e and Extended Data Fig. 8d). Moreover, the pmel-I T cells possessed significantly elevated interferon (IFN)-γ production over controls (Fig. 6f and Extended Data Fig. 8e). Thus, our data suggest that CD39-expressing tTex cells are capable of suppressing tumor-specific T cell populations in vivo.

Evidence of in situ suppression by tTex cells encouraged further investigation of the therapeutic benefit of targeting CD39 expression. Thus, we orthotopically inoculated E8iCre-ERT2Entpd1f/f mice and controls with B16-F10 as before, but delayed TAM treatment (and thus CD39 deletion) in all groups to when tumors were palpable (~2 mm, day 5; Fig. 7a). After 3 days of TAM treatment, we treated half of the cohort now carrying advanced tumors (~5 mm, day 8) with anti-PD-1 and anti-CTLA-4 immunotherapy. At this time point, checkpoint blockade immunotherapy is incapable of causing complete regressions in B16-F10 tumors and mice progress with little therapeutic benefit (Fig. 7b,c). As described earlier, CD8+ T cell-restricted deletion of CD39 itself was sufficient to significantly slow tumor progression (Fig. 7b,c) to a degree similar to combination checkpoint blockade alone; however, checkpoint blockade in the context of CD39 deletion further improved the therapeutic response, resulting in 37.5% of mice achieving a partial response (3 of 8; static or decreased tumor size after 2 days) and 12.5% with complete tumor remission (1 of 8; grossly undetectable disease; Fig. 7c).

Fig. 7: tTex cell-restricted CD39 deletion bolsters immunotherapeutic efficacy.
figure 7

a, Schematic of experimental design. TAM treatment began at day 5 after tumor inoculation then continued thrice weekly thereafter; immunotherapy (IMT) treatment began on day 8 and was administered thrice weekly. b,c, Tumor growth over time and survival in Entpd1f/f or E8iCre-ERT2Entpd1f/f mice bearing B16-F10 with or without combination immunotherapy. d, CD8+ TIL restimulated with PMA and ionomycin at day 8 of tumor growth, day 3 of TAM treatment. e,f, PD-1 and Tim-3 or TOX expression in CD8+ T cells on day 8 of tumor growth, day 3 of TAM treatment. Statistics used were log-rank test (b), one-way ANOVA (d) and Mann–Whitney U-test (e,f) with P < 0.05, P < 0.01, P < 0.001 and P < 0.0001.

Source data

Given the absence of improved cytokine production in intratumoral CD39-deficient T cells (Extended Data Fig. 8c) despite a diminished proportion of these being Slamf6Tim-3+ tTex cells (Fig. 6a), we sought to interrogate the immune milieu early in tumor progression. Here, we treated E8iCre-ERT2Entpd1f/f mice with a 3-day loading course of TAM beginning at day 5 after tumor engraftment as before, then isolated CD8+ TILs when tumors between groups were still similarly sized (~5 mm, day 8). At this time point, CD39 expression was significantly depleted in E8iCre-ERT2Entpd1f/f TIL as compared to controls (~60% decrease from control; Extended Data Fig. 8f). Notably, B16-F10-infiltrating CD39-deficient CD8+ T cells displayed significantly improved cytokine competency when restimulated ex vivo as compared to control mice, with ~two-fold increase in IFN-γ+ and IL-2+ T cells by flow cytometry (Fig. 7d). Further, at this early time point, CD8+ T cell-specific deletion of CD39 also resulted in delayed accumulation of hallmark features of exhaustion, including inhibitory receptor (PD-1 and Tim-3) expression as well as expression of the exhaustion-associated transcriptional repressor, TOX (Fig. 7e,f and Extended Data Fig. 8g,h). Thus, preventing suppressor activity in tTex cells creates an environment favorable for immunotherapy response.

Hypoxia mitigation weakens suppressor capacity of tTex cells

While the hypoxia–Hif1α–von Hippel-Lindau (VHL) axis has been implicated in T cell activation and proinflammatory effector functions32, the repressive role that hypoxia plays in the tumor microenvironment has been well documented by our group and others7,13,14,15. Disruption of this pathway through genetic or pharmacologic approaches reinvigorates antitumor immunity and sensitizes immunotherapy-resistant tumors. Hypoxia, oxidative stress and proinflammatory mediators have all been shown to induce CD39 expression7. Hypoxia-selective pimonidazole (commercially known as Hypoxyprobe) labeling15, expectedly shows CD8+ TILs experiencing low oxygen tension compared to dLNs (Extended Data Fig. 9a). As we have previously shown13, tTex cells experience significant hypoxia (pimonidazolehi) relative to progenitors (Fig. 8a) and exposure to hypoxia is associated with enrichment of surface CD39 staining (Fig. 8b). Thus, we asked whether expression of CD39 could be driven by hypoxia in isolation, via in vitro culture at oxygen tensions similar to what is observed in tumors33 (~1.5% O2). Indeed, CD8+ T cells cultured in tumor hypoxic conditions displayed significantly higher CD39 staining compared to those maintained in atmospheric oxygen tensions (~20%), in a Hif1α-dependent manner (via Cd4CreHif1αf/f; Fig. 8c), supporting the hypoxia–Hif1α–VHL axis as a key contributor to the suppressor fate of CD39+ tTex cells. MC38 tumor-infiltrating tTex cells, which we show are largely non-suppressive due to being CD39lo (Extended Data Fig. 9b), experience significantly lower hypoxia exposure versus their B16-F10-resident counterparts34 (Extended Data Fig. 9c).

Fig. 8: Hypoxia reversibly drives CD39-dependent suppression by tTex cells.
figure 8

a, PD-1 and Tim-3 staining in pimonidazole fractions; negative, low and high. b, CD39 staining in PD-1hiTim-3+ tTex cells from pimonidazole fractions. c, Percent of CD8+ T cells expressing CD39 following 72-hour stimulation at atmospheric or tumor hypoxic conditions (1.5% O2). d, CD39 staining of day 6 chronic stimulation and hypoxia (CS + H) cells. e, Proliferation of CTV-labeled murine CD4+ T cells in suppression assay experiments with CS + H cells. f, Proliferation of three human healthy donor CTV-labeled CD8+ T cells in suppression assay experiments with syngeneic human CS + H cells. g, Histogram overlay displaying pimonidazole staining in intratumoral CD8+ T cells from parental B16-F10 or Ndufs4 (B16-F10ND4–) tumors and dLNs. h, CD39 staining intratumoral T cells B16-F10 or B16-F10ND4– tumors. i, Biological replicate suppression assays utilizing B16-F10 or B16-F10ND4– TILs. j, Pimonidazole staining in intratumoral CD8+ T cells infiltrating B16-F10 tumors or dLNs following two treatments of anti-hypoxia agents axitinib (Ax) or metformin (Met). k, CD39 staining in intratumoral T cells from various treatment groups. l, Biological replicate suppression assays of B16-F10 TIL following two treatments of either vehicle, Ax or Met. Statistics used were one-way ANOVA (bd,f,j,k), linear regression (e,i,l) and Mann–Whitney U-test (g,h) with P < 0.05, P < 0.01, P < 0.001 and P < 0.0001.

Source data

To further examine the role of hypoxia in the generation of suppressor tTex cells, we next asked whether our recently published in vitro protocol (Extended Data Fig. 9d,e) for generating tTex-like cells de novo could also recapitulate our observed suppressor functions in co-culture assay. As we have previously reported13, the combination of continuous TCR stimulation and hypoxia (CS + H) exposure resulted in significantly higher CD39 expression than either signal alone (Fig. 8d). Thus, we can leverage this system to generate T cells with sustained CD39 expression and to test their capacity to suppress effector responses in co-culture assay. CD39hi T cells that had previously experienced continuous TCR stimulation under tumor hypoxic conditions displayed potent suppressive capacity in suppression assay relative to CD39lo T cells acutely stimulated under hypoxia (Fig. 8e). Further, these findings could be recapitulated in human healthy donor-derived CD8+ T cells cultured under similar conditions ex vivo (Extended Data Fig. 9f–h): continuous TCR signaling under hypoxia was sufficient to generate human CD39hi T cells that were capable of suppressing co-cultured activated syngeneic T cells (Fig. 8f).

We have previously demonstrated that disruption of tumor oxygen consumption through deletion of mitochondrial complex I subunit, NADH:ubiquinone oxidoreductase subunit S4 (Ndufs4; B16-F10ND4–) significantly reduces tumor hypoxia13,15 (Fig. 8g and Extended Data Fig. 10a). Further, B16-F10ND4–-infiltrating CD8+ T cells display a more activated and less exhausted signature compared to those derived from the parental B16-F10 line (Extended Data Fig. 10b). In direct comparison, PD-1hiTim-3+ tTex cells that differentiate in the less hypoxic B16-F10ND4– tumor display significantly reduced surface CD39 staining (Fig. 8h) and notably, when assayed in ex vivo co-culture, B16-F10ND4– tTex cells failed to display any functional suppression (Fig. 8i). Taken together, these data propose tumor hypoxia as an enforcer of tTex cell suppressor functions via elevated expression of CD39; however, within this system we cannot assess whether this antifunctional state is irreversible or whether alleviation of tumor hypoxia may itself rescue tTex cells from a suppressive state.

To assess the plasticity of tTex cell functions in vivo, we treated Foxp3Ametrine-Flpo mice bearing advanced wild-type B16-F10 tumors (~6 mm; day 10;) with one of two tumor hypoxia-targeting agents, axitinib or metformin (Extended Data Fig. 10c). We have previous shown that when treated early, both axitinib and metformin display a mild therapeutic benefit and sensitize hypoxic tumors to checkpoint blockade immunotherapy13,34. Two treatments of axitinib (2 mg kg−1) or metformin (50 mg kg−1), delivered intraperitoneally (i.p.) and late in B16-F10 melanoma disease course, were sufficient to significantly mitigate tumor hypoxia (Fig. 8j and Extended Data Fig. 10d). As with our B16-F10ND4– model, infiltrating CD8+ T cells from either axitinib or metformin treatment groups displayed lower inhibitory receptor expression (Extended Data Fig. 10e) and further, when subset by PD-1 and Tim-3 expression, tTex cells from treatment groups expressed lower levels of CD39 (Fig. 8k). Moreover, axitinib or metformin-treated tTex cells were reliably less suppressive ex vivo than their vehicle-treated counterparts (Fig. 8l). These findings further support tumor hypoxia as a critical mediator of suppressor function in tTex cells and reveal a degree of reversibility of this state. In contrast, while tTex cell suppressor functions were significantly compromised by targeting tumor hypoxia, Foxp3+ Treg cells from the same environments (B16-F10ND4– or axitinib/metformin-treated tumors) maintained their potent inhibitory character ex vivo (Extended Data Fig. 10f,g). These data reveal that the therapeutic benefit derived from anti-hypoxia agents in combination with checkpoint blockade is not primarily derived from altered suppressive capabilities of Foxp3+ Treg cells, but may result from changes in CD8+ T cell functionality, by (1) limiting metabolic stress experienced by tumor-specific T cells and (2) preventing progression of tTex cells to an antifunctional state that directly counteracts antitumor immunity.


T cell exhaustion has been described in numerous inflammatory contexts; in both pathological settings such as cancer, chronic viral infection and autoimmunity1,10, but also in states requiring immune homeostasis and tolerance, such as in gestation11. The commonality within these inflammatory contexts is that the differentiation program of exhaustion is dependent upon chronic exposure of T cells to their cognate antigen. While the impact and consequence of T cells exhaustion in pathologic or homeostatic settings are still being untangled, it is likely that the evolutionary advantages of this functional state and how this program is co-opted in disease settings such as cancer is context dependent10. Indeed, recent studies have highlighted a critical role for suppressor CD8+ T cells in mitigating pathologic autoreactive CD4+ T cells via perforin-mediated lysis24. In tumor-infiltrating tTex cells, however, this pathway has been shown to be defective35,36. Notably, therapeutic strategies that enforce terminal exhaustion in T cells have also been shown to dampen autoimmune disease progression37. In this study, using gold-standard assays, we clearly demonstrate CD8+ tTex cells, a common cellular fate of CD8+ T cells in solid tumors, upregulate the ectonucleotidase CD39 to levels sufficient to create a suppressive microenvironment. CD39 is a well-characterized immunosuppressive molecule and indeed recent efforts to therapeutically block CD39 and CD73 in combination with checkpoint blockade have yielded encouraging results38,39.

We have described a transcriptional and post-translational circuit that allows terminally exhausted T cells to suppress conditionally. Specifically, tTex cells mediate their suppressor function through (1) CD39 expression driven principally through the TCR and hypoxic sensing, (2) apoptotic generation of eATP and (3) CD73 expression on the responding cell population. This triad likely evolved to prevent immunopathology in inflamed and damaged tissues, promoting immune tolerance, but only to newly activated, CD73hi T cells; however, in the deregulated TME, this drives immune dysfunction and immunotherapy resistance. While tTex cells have been characterized by high cell turnover via apoptosis2,20, several groups have reported that maintenance of T cell viability through overexpression of BCL-2 not only augments antitumor responses, but also lessens immune tolerance in the TME40,41. Indeed, enforced expression of BCL-2 in chimeric antigen receptor T cells improved persistence and response in both preclinical models and in human trials. These data are supported by our findings that maintaining tTex cell viability through BCL-2 expression restricts their suppressor functionality. In several models tested, the suppressive capacity of apoptosis-prone tTex cells was equivalent to Foxp3+ Treg cells isolated from the same tumor environment. These suppressive functions can be inhibited upon BCL-2 overexpression, genetic deletion of CD39 or pharmacologic blockade of adenosine signaling. Further, we delineate a specific in vivo role of CD39 expression on tTex cells, revealing that CD8+ T cell-restricted deletion of CD39 results in retained polyfunctionality in both the endogenous tumor-infiltrating T cell compartment as well as transferred wild-type tumor antigen-specific T cells. This improvement to CD8+ TIL effector functionality via endogenous CD39 deletion resulted in better control of tumor progression and enhanced therapeutic efficacy of adoptive T cell transfer and checkpoint blockade strategies. Notably, targeting this pathway is insufficient to prevent T cell exhaustion in cancer, but we do show evidence that CD39 deletion can slow terminal differentiation. These data not only improve our understanding of the breadth of functions attributed to exhausted T cells, but also supports the eATP–adenosine axis as an attractive target to subvert resistance to checkpoint blockade by exhausted T cells.

Adenosine rapidly accumulates in response to tissue injury and ischemia, subverting effective T cell activation and pushing the balance to tissue repair42. In cancer, pioneering studies have identified many pathways upregulated in immune and stromal cells generating adenosine through consumption of immunostimulatory metabolites. Among these, the ectonucleotidase, CD39, is increasing reported as a faithful marker of tumor antigen experienced CD8+ T cells16,28, particularly among terminally exhausted3,5. An increasing number of preclinical studies have revealed the therapeutic potential of CD39 inhibition–in addition to other targets in adenosine metabolism (spurring several phase I/II trials in solid tumors and lymphoma with antibody or inhibitor therapy43). In solid tumors, the metabolic features of tumor cells15 and the resulting hypoxic13,14 and nutrient dearth microenvironment6 accelerates CD8+ T cell dysfunction and diminishes efficacy of checkpoint blockade therapy. We demonstrate that tTex cells encountering tumor hypoxia gain a functional suppressor program principally via expression of CD39 and can limit effective tumor control in a manner independent of CD39 expression on other infiltrating immune cells, tumor cells or stroma. Our study supports a model in which chronically stimulated tumor resident CD8+ T cells sensing their hypoxic microenvironment respond by rapidly progressing to a terminally differentiated exhausted fate and upregulating immunosuppressive programs to repress local antitumor immunity.

In human cancer, CD39 expression in CD8+ T cells is associated with tumor antigen specificity and a terminally exhausted signature16,28, with some reports stating enrichment of CD39 prevents effective antitumor immunity44,45. Of note, infiltration of CD39+CD103+ T cells (likely identifying tissue-resident memory populations) has also been shown to correlate with improved outcomes46,47. Taken together, these data suggest that while CD39 restricts inflammation, enrichment of tumor-specific T cells remains a dominant prognostic factor. Our data also suggest that density of CD39, rather than its expression per se, is critical for suppressor function. Empowering those specific T cells through CD39 blockade may be an attractive target to circumvent the suppressor program enforced upon terminal exhaustion38,39.

A noteworthy result was that IL-10 was dispensable for tTex cell suppression, although the diversity of roles of IL-10 in cancer continues to widen. Numerous reports have highlighted a role for this pleiotropic cytokine, both in the progression of cancer and restriction of antitumor immunity48, in enforcement of terminal exhaustion21,49 and in the cytotoxicity of CD8+ T cells21,50. We investigated whether exposure to IL-10 within the tumor or secretion of IL-10 by tTex cells19 could contribute to suppression of T cell responses. Contrary to studies highlighting a role for IL-10 in CD8+ T cell-mediated control of inflammation19,51, neither genetic deletion of IL-10 nor in vitro blockade diminished tTex cell-mediated suppression in our system. Moreover, several recent reports have highlighted that engineered IL-10 molecules can metabolically reprogram tTex cells, improving their oxidative capacity and inflammatory functions in cancer21,52. Thus, IL-10 does not seem to be a dominant tolerogenic mechanism utilized by tTex cells in cancer.

Exposure to hypoxia accelerates T cell exhaustion via metabolic byproducts such as reactive oxygen species and alleviation of hypoxia sensitizes tumors otherwise resistant to immunotherapy13. We posit differentiation to terminal exhaustion results not only in dysfunctional but in antifunctional cells, coordinated primarily through CD39 expression and adenosine-mediated suppression of T cell activation. Unlike many features of exhaustion, we show that this suppressor state retains plasticity; while T cell exhaustion may not be capable of being fully reversed by checkpoint blockade immunotherapy, the inhibitory functions of tTex cells may be diminished through targeting tumor hypoxia or CD39 enzymatic activity. We propose multiple tenable therapeutic avenues to restrict tTex cell-mediated immune regulation and propose a mechanism to explain how anti-hypoxia therapies may promote improved CD8+ T cell effector functions (by dampening tumor hypoxia, CD39 expression is diminished on tTex cells and their regulatory potential is limited). Sequencing checkpoint blockade after tumor hypoxia mitigation and/or blockade of the eATP–adenosine axis may bring the curative promise of immunotherapy to additional patients.



All experiments requiring live animals or ex vivo immune cells were performed in accordance with the Institutional Animal Care and Use Committee of the University of Pittsburgh. All mice were housed in specific-pathogen-free conditions before use. Mice were maintained at room temperature on a 12–12-h light–dark cycle in boxes containing five (for female) or four (for male) mice. Experimental mice received Purina Prolab Isopro RMH 3000 (5P75 and 5P76) chow ad libitum. Both male and female mice were used and mice were 6–8 weeks old at the start of experimentation. Wild-type C57BL/6 mice and mice strains on the C57BL/6 background (including Cd4Cre, Hif1αf/f, Nt5e–/–, Il10–/–, Bcl2tg and Thy1a) were obtained from the Jackson Laboratory. Entpd1f/f mice on the C57BL/6 background were obtained from S.C. Robson (Beth Israel Deaconess Medical Center). E8iGFP-Cre-ERT2-Rosa26-LSL-TdTomato and Foxp3mAmetrine-Flpo mice on the C57BL/6 background were generated by D.A.A. Vignali (University of Pittsburgh). Tumor experiments for ex vivo functional assays and analysis were typically completed on day 14, when tumors were 8–10 mm in diameter. For immunotherapy experiments, mice were killed when tumors reached 15 mm in diameter or 1,500 mm3 or were moribund.

In vivo treatments and immunotherapy were performed as previously described13,34 with Bio X Cell-supplied anti-PD-1 (clone J43), anti-CTLA-4 (clone 9H10) and host-appropriate isotype controls. Cayman Chemical supplied axitinib (2 mg kg−1 i.p.; suspended in DMSO (10%), Tween 80 (10%) and 30% captisol (80%)) and metformin (50 mg kg−1 i.p.; saline (100%)). Sigma supplied TAM (50 mg kg−1 orally; 100% ethanol (5%) and corn oil (95%)).

Tissue culture and T cell isolation

B16-F10, MC38 and MEER were originally purchased from American Type Culture Collection. B16-F0OVA was obtained from P. Basse and L. Falo (University of Pittsburgh). Platinum-E (Plat-E) was obtained from L. Kane (University of Pittsburgh). Some immortalized tumor lines were generated inhouse as previously described; B16-F10ND4– (ref. 13), clone 24 (CL24)15 and MEERPD-1res0 (ref. 14). All cell lines were maintained in complete DMEM (10% heat-inactivated fetal bovine serum, 1 mM sodium pyruvate, 2 mM l-glutamine, 5 mM HEPES buffer, 100 U penicillin, 100 µg ml−1 streptomycin and 500 µg ml−1 gentamycin) at no greater than 80% confluence.

Tumors, dLNs and spleens were surgically resected, then enzymatically and mechanically digested to a single cell suspension. Cells were purified by either magnetic bead isolation as previously described13 or fluorescence-activated cell sorting (FACS) on a Beckman Coulter Mo-Flo Astrios High Speed Cell Sorter or Sony MA900 Cell Sorter or a combination of the two. All T cells were cultured in complete RPMI (10% heat-inactivated fetal bovine serum, 1 mM sodium pyruvate, 2 mM l-glutamine, 5 mM HEPES buffer, 100 U penicillin, 100 µg ml−1 streptomycin, 500 µg ml−1 gentamycin and 5 nM 2-mercaptoethanol) plus 25–100 U of murine IL-2 (PeproTech).

Ex vivo miniaturized suppression assays were performed as previously described18. To measure cytokine production, cells were incubated at 37 °C for 6 h or overnight in GolgiPlug Protein Transport Inhibitor (BD) with either 2:1 bead:cell ratio of Dynabeads T-Activator CD3/CD28 beads (Thermo Fisher) or phorbol 12-myristate 13-acetate (PMA) and ionomycin (Sigma). Metabolic assays were performed as previously described13,31. In vitro continuous stimulation under hypoxia protocol was performed as previously described13.

Human peripheral blood and TIL isolation

Human blood was obtained from Vitalant; tumor preparations were obtained via University of Pittsburgh Medical Center, Hillman Cancer Center Institutional Review Board protocol 96-099. Each patient was consented as outlined in our protocol, then subject to peripheral blood collection and tumor biopsy. Tumor samples were mechanically digested to a single cell suspension, then passaged through 70-µm cell strainers (BD). Lymphocytes were purified from bulk suspension by density centrifugation and/or cell sorting.

Flow cytometry and FACS

Viability staining with Zombie dyes (BioLegend; 1:1,000 dilution as per manufacturer’s protocol) and surface staining (see below; 1:50–1:1,000 dilution) were performed in PBS at 4 °C for 20 min. Samples were then washed and immediately analyzed on a BD LSRFortessa flow cytometer. For fixation of samples for intracellular hypoxia/transcription factor staining, following viability/surface staining, cells were incubated in 4% paraformaldehyde at room temperature for 5 min, washed in PBS, permeabilized with eBioscience Foxp3/Transcription Factor Staining kit (Thermo Fisher) as per manufacturer’s protocol, washed in permeabilization buffer, then stained in permeabilization buffer at 4 °C overnight. For fixation of samples for intracellular cytokine staining, following viability/surface staining, cells were fixed in BD Cytofix/Cytoperm Fixation/Permeablization kit as per manufacturer’s protocol, washed in permeabilization buffer, then stained in permeabilization buffer at 4 °C overnight.

BioLegend supplied Zombie Fixable Viability Dyes (UV, Aqua, Green and NIR); anti-mouse antibodies including, anti-CD8a (clone 53.6.7; 1:500 dilution), anti-CD4 (GK1.5; 1:500 dilution), anti-PD-1 (29F.1A12; 1:500 dilution), anti-Tim-3 (RMT3-23; 1:250 dilution), anti-Helios (22F6; 1:200 dilution), anti-Nrp1 (3E12; 1:250 dilution), anti-CD25 (PC61; 1:250 dilution), anti-CTLA-4 (UC10-4B9; 1:250 dilution), anti-Lag-3 (C9B7W; 1:200 dilution), anti-CD39 (Duha59; 1:1,000 dilution), anti-FasL (MFL3; 1:500 dilution), anti-TIGIT (1G9; 1:250 dilution), anti-ICOS (C398.4A; 1:250 dilution), anti-CD73 (TY/11.8; 1:500 dilution), anti-IL-2 (JES6-5H4; 1:100 dilution), anti-IFN-γ (XMG1.2; 1:250 dilution), anti-TNF (MP6-XT22; 1:250 dilution), anti-Thy1.1 (OX-7; 1:1,000 dilution) and anti-human antibodies, including, anti-CD8a (RPA-T8; 1:100 dilution), anti-CD4 (RPA-T4; 1:100 dilution), anti-PD-1 (EH12.2H7; 1:50 dilution), anti-Tim-3 (F38-2E2; 1:50 dilution), anti-CD73 (AD2; 1:50 dilution), anti-CD39 (TU66; 1:50 dilution), anti-IL-2 (MQ1-17H12; 1:50 dilution) and anti-IFN-γ (4S.B3; 1:50 dilution). Invitrogen supplied anti-mouse antibody, anti-Foxp3 (FJK-16s; 1:100 dilution) and anti-human antibodies, anti-Lag-3 (3DS223H; 1:50 dilution) and anti-TNF (Mab11; 1:50 dilution). R&D Systems supplied anti-Hif1α (241812; 1:50 dilution). Thermo Fisher supplied the CellTrace Violet Cell Proliferation kit (1:1,000 dilution per manufacturer’s protocol), anti-TOX (TXRX10; 1:100 dilution) and anti-Blimp-1 (5E7; 1:250 dilution).

Immunoblotting analysis

Immunoblotting was performed as previously described13. Santa Cruz Biotechnology supplied anti-actin (clone C4). Cell Signaling supplied Zap70 (D1C10E), phospho-Zap70 Y319/Y352 (65E4), Akt (C67E7) and phosphor-Akt S473 (D9E). Immunoblots were detected via standard secondary detection and chemiluminescent exposure to film. Digitally captured films were analyzed densitometrically using ImageJ software.

Retroviral vector construction and T cell transduction

Entpd1 (encoding CD39) was originally generated by and cloned into a murine stem cell virus (MSCV)-driven retroviral expression vector encoding an internal ribosome entry site (IRES)–GFP cassette, from DAA Vignali (Addgene plasmid no. 52114;; RRID: Addgene_52114). For T cell transductions, T cells were activated with plate-bound anti-CD3 (5 µg ml−1), soluble anti-CD28 (1 µg ml−1) and IL-2 (100 U) for 24 h. Retroviral supernatants were collected, filtered and supplemented with 6 μg ml−1 polybrene. Prepared retrovirus was spun onto activated T cells (2,200 r.p.m., 2 h, 37 °C) then the cell/virus culture was rested in a tissue culture incubator for another 2 h. Cells were then washed and cultured in fresh complete RPMI plus IL-2 (50 U) for 3 d to allow for expansion and expression of vector cassette. After 3 d, cells were purified via FACS by mCherry expression. CD39 expression was validated by flow cytometry and enzymatic activity verified by ENLIGHTEN ATP Assay System (Promega).

Software and code

Data collection utilized BD FACSDiva v.9.0 for flow cytometry and Seahorse Wave Controller Software v.2.4.2 for extracellular flux analysis. Data were analyzed with FlowJo v.10, Prism v.9.2.0, ImageJ and Stata v17.0. For RNA-seq analysis, technical replicates were concatenated into a single fastq file and processed using standard methods, sequencing reads were trimmed with adapters using Cutadapt before being aligned to Musmusculus reference genome (mm10) using the RNA-seq aligner HISAT2. Subread featureCounts function was used for gene level quantification and results were normalized for transcripts per kilobase million. Using the raw quantification, differential genes were found with the R package DESeq2 and GSEA program developed by the Broad Institute.


If the data seemed to follow a normal distribution, we calculated the P values in GraphPad Prism using one-way ANOVA with Dunnett’s multiple comparison test, two-way ANOVA with Sidak’s multiple comparison test, unpaired Student’s t-test or paired Student’s t-test. For data that did not follow a normal distribution, we used the nonparametric Kruskal–Wallis and Mann–Whitney U-tests as indicated. The ex vivo suppression assays were analyzed using linear regression, treating the percent suppression as the dependent continuous variable, the type of cells as a categorical independent variable and the dilution as a continuous independent variable. Interactions were tested and included in the linear regression model if statistically significant. Values of P < 0.05 were considered significant and ranked as *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001.

Reporting summary

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