Regulatory T cells converted from Th1 cells in tumors suppress cancer immunity via CD39

FOXP3-expressing regulatory T (Treg) cells are vital in maintaining immune tolerance and homeostasis through the suppression of excessive immune responses deleterious to the host (Ohkura et al., 2013). However, the suppressive activity of Treg cells in the tumor microenvironment (TME) also enables tumor cells to evade host immunologic surveillance (Chen et al., 2016). Moreover, the TME appears to promote the recruitment and development of Treg cells (Ohue and Nishikawa, 2019; Tanaka and Sakaguchi, 2019). The presence of a large number of tumor-infiltrating Treg cells has been reported in a variety of human cancers, including ovarian cancer (Sato et al., 2005), hepatocellular carcinoma (HCC) (Ormandy et al., 2005), and breast cancer (Liyanage et al., 2002). Importantly, decreased ratios of tumor-infiltrating CD8⁺ T cells to Treg cells were often found to correlate with poor prognosis in cancer patients (Tanaka and Sakaguchi, 2017). It has been demonstrated in mice that CD8⁺ T cell–mediated rejection of tumor was suppressed by adoptive transfer of Treg cells (Antony et al., 2005; Turk et al., 2004). Conversely, upon removal of Treg cells, an increase in tumor-infiltrating CD8⁺ T cells was observed in the mice challenged with tumor cells (Shimizu et al., 1999; Yamaguchi and Sakaguchi, 2006). Therefore, tumor-infiltrating Treg cells are potential targets of cancer immunotherapy (Ellis and Riley, 2020; Tay et al., 2023). However, it is noteworthy that systemic removal of Treg cells may lead to simultaneous autoimmunity (Sakaguchi, 2004). For instance, targeting cytotoxic T lymphocyte-associated antigen 4 (CTLA-4), constitutively expressed by Treg cells, often elicited autoimmune side effects (Nishikawa and Sakaguchi, 2010; Pol and Kroemer, 2018; Walker, 2013). Thus, it is essential to identify and selectively target the predominant subpopulation of Treg cells in the TME.

Based on their origins, Treg cells can be divided into two subpopulations, namely thymus-derived natural Treg (tTreg) cells and periphery-induced Treg (pTreg) cells (Rodríguez-Perea et al., 2016). tTreg cells are generated in the thymus through selections on relatively high-affinity interactions between T cell receptors (TCRs) and self-peptide/major histocompatibility complexes (Jordan et al., 2001). pTreg cells instead arise in the periphery from conventional CD4⁺ T cells as a result of TCR stimulation in the presence of the transforming growth factor (TGF)-β and interleukin (IL)-2 (Rodriguez-Perea et al., 2016). FOXP3 transcription factor is the hallmark of both types of Treg cells and essential for their cell lineage commitment or differentiation in the thymus and the periphery (Fontenot et al., 2003; Hori et al., 2003). It has been shown that neuropilin 1 (NRP1) may serve as a specific marker on tTreg but not pTreg cells (Weiss et al., 2012; Yadav et al., 2012). Treg cells are known to exert suppressive activities via diverse mechanisms. These include secretion of immunosuppressive cytokines TGF-β, IL-10, and IL-35, inhibition of costimulation mediated by CTLA-4, expression of suppressor molecules such as TIGIT (T-cell immunoreceptor with Ig and ITIM domains), and depriving IL-2 from surrounding conventional T (Tconv) cells (Josefowicz et al., 2012). Moreover, Treg cells may express ectonucleotidases CD39 and CD73, which mediate the hydrolysis of adenosine triphosphate (ATP) to adenosine monophosphate (AMP) and AMP to adenosine, respectively (Borsellino et al., 2007; Deaglio et al., 2007). Adenosine may bind to ADORA2A receptor (A2AR) on effector T cells and suppress their proliferation (Nishikawa and Koyama, 2021; Ohta and Sitkovsky, 2014).

Although the accumulation of Treg cells in the TME has long been reported, the origin of tumor-infiltrating Treg cells remains controversial. CD4⁺CD25⁻ T cells cultured in conditioned media derived from murine tumor cells developed into FOXP3⁺ Treg cells, which were reversed by a TGF-β blocking antibody (Liu et al., 2007). However, whether pTreg cells were induced in the TME was not addressed in vivo. In contrast, a previous TCR repertoire analysis of tumor-infiltrating lymphocytes showed that the TCR repertoire of intratumoral Treg cells shared low similarity with that of intratumoral conventional CD4⁺ T cells (Hindley et al., 2011), and so it was concluded that conventional CD4⁺ T cells did not convert to pTreg cells significantly.

In response to environmental signals, Treg cells may show characteristics of plasticity and adopt phenotypes analogous to CD4⁺ helper T cells (Dong, 2021; Shi and Chi, 2019). It was first reported in mice infected with Mycobacterium tuberculosis that T-bet⁺ Treg cells accumulated at the sites of inflammation to suppress Th1-mediated inflammatory responses (Koch et al., 2009). Similarly, T-bet⁺ Treg cells were enriched in secondary lymphoid organs to suppress Th1-type inflammation in mice challenged with Listeria monocytogenes (Levine et al., 2017). T-bet⁺ Treg cells have also been found in the TME of human oropharyngeal cancer (Santegoets et al., 2019), human non-small cell lung carcinoma (NSCLC), and murine lung tumor model (Kachler et al., 2018), as well as in the gut of patients with inflammatory bowel disease and dextran sodium sulfate–induced colitis model (Di Giovangiulio et al., 2019). However, the development and functions of T-bet–expressing Treg cells in cancer remain obscure.

In this study, we have systematically analyzed the development and functional mechanisms of tumor-resident Treg cells. We identified T-bet⁺ pTreg cells as a dominant group of intratumoral Treg cells, which developed from Th1 cells and exerted suppressive functions through CD39. Inhibition of the generation or function of these pTreg cells boosted antitumor immune responses. Taken together, our findings suggest potential strategies for targeting intratumoral pTreg cells as effective immunotherapies, without triggering systemic autoimmune responses.

To date, intratumoral Treg cells are still poorly studied in terms of their origins and functional mechanisms. Therefore, we first characterized Treg cell subsets in multiple murine cancer types. NRP1 has been known to be selectively expressed by tTreg cells, but not by pTreg cells (Weiss et al., 2012; Yadav et al., 2012). Therefore, we employed NRP1 as a marker of tTreg cells for the phenotypic analysis of Treg cell subsets in the TME. We stained Treg cells derived from lymph nodes (LN), spleens (SPL), tumor-draining lymph nodes (TDLN), and tumor-infiltrating lymphocytes (TIL) with NRP1 in E.G7 (Fig. 1, A and C), Hepa1-6 (Fig. 1, B and D), and B16-OVA (Fig. 1 E) murine tumor models. Treg cells derived from LN and SPL of tumor-free mice were included as controls. The results showed that nearly 80% of Treg cells derived from LN, SPL, and TDLN were NRP1⁺, indicating that they were mainly tTreg cells. In contrast, >50% of the intratumoral Treg cells were NRP1⁻ Treg cells, suggesting that there was a substantial accumulation of de novo–induced pTreg cells in the TME of multiple cancer types.

We then validated our observations using an adoptive transfer approach. We sorted tTreg cells from the SPL and LNs of CD45.1/CD45.2 (double-positive) Foxp3^GFP reporter mice and sorted conventional FOXP3⁻ CD4⁺ T cells and CD8⁺ T cells from CD45.2 Foxp3^GFP reporter mice. We then co-transferred tTreg cells, conventional CD4⁺ T cells, and CD8⁺ T cells at ratios of 1:10:8 into TCRbd^−/− mice to mimic the physiological lymphoid compartment of normal mice, followed by the inoculation of E.G7 tumor cells 3 days later (Fig. 1 F). Different congenic markers of CD45 carried by the donor cells were utilized for distinguishing tTreg cells (CD45.1⁺) and pTreg cells (CD45.1⁻). The ratios between tTreg cells and conventional CD4⁺ T cells were close to 1:10 in both LN and SPL, similar to the initial 1:10 ratios to be used for the adoptive transfer of tTreg cells and conventional CD4⁺ T cells sorted from LN and SPL isolated from the donor mice (Fig. 1, G and H). We found that at least half (50%) of the intratumoral Treg cells were composed of pTreg cells (CD45.1⁻) derived from conventional naive CD4⁺ T cells (Fig. 1, I and J). On the contrary, the majority (70–80%) of Treg cells in LNs and SPL were tTreg cells (CD45.1⁺) (Fig. 1, I and J). These results were consistent with those observed in the NRP1 staining. We further confirmed these findings by assessing NRP1 expression in the donor cells. We found that the majority (>70%) of CD45.1⁺ tTreg cells were NRP1⁺ whereas >60% of CD45.1⁻ pTreg cells were NRP1⁻ (Fig. 1, K and L). Thus, pTreg cells were at least comparable with tTreg cells in proportions, if not more dominant, in the TME. These results suggest that TME may be suppressive, at least in part, via induction of pTreg cells.

Depletion of Treg cells evoked effective antitumor immunity (Onizuka et al., 1999; Shimizu et al., 1999; Yamaguchi and Sakaguchi, 2006), which may also result in autoimmune responses (Sakaguchi, 2004). To determine if pTreg cells play a crucial role in tumor immunity, we used an adoptive transfer approach as above that enabled us to selectively deplete pTreg cells. We sorted tTreg cells from CD45.1/CD45.2 (double-positive) Foxp3^GFP reporter mice and sorted conventional CD4⁺ T cells and CD8⁺ T cells from CD45.2 Foxp3^DTR mice. We then co-transferred tTreg cells, conventional CD4⁺ T cells, and CD8⁺ T cells at 1:10:8 ratios into TCRbd^−/− mice. After 3 days, E.G7 tumor cells were inoculated. By using Foxp3^DTR mice as the source of conventional CD4⁺ T cells, we selectively depleted pTreg cells by diptheria toxin (DT) administration without affecting tTreg cells and therefore avoided autoimmune response. DT was administered every 2 days from day 12 until day 20 in E.G7 tumor-bearing mice, which were sacrificed and analyzed on day 22 (Fig. 2 A). We found that tumor growth was suppressed in DT-treated mice as compared with the untreated control mice (Fig. 2, B and C). An absolute number of Treg cells was significantly reduced in LNs, SPL, and tumors derived from DT-treated mice (Fig. 2, D–F). In addition, pTreg cells (CD45.1⁻) were almost completely depleted in LNs, SPL, and tumors upon DT treatments (Fig. 2, G–J). Upon depletion of pTreg cells, there was an increase in the absolute number and frequencies of CD8⁺ T cells in total live cells in TILs (Fig. 2, K and L) leading to an elevation in the ratios of CD8⁺ T cells to total Treg cells in TILs (Fig. 2 M). CD8⁺ T cells were activated and proliferative (CD44⁺Ki67⁺) in these mice, indicating that they were functional (Fig. 2, N and O). Assessment of the expression of interferon-gamma (IFN-γ) and CD107a in CD8⁺ T cells revealed that they displayed increased cytolytic activities (CD107a⁺IFN-γ⁺) upon pTreg cell depletion (Fig. 2, P–S). It has been increasingly recognized that myeloid cells play a critical role in tumor immunity (Dou and Fang, 2021; Mantovani et al., 2021). We, therefore, evaluated the changes in CD45⁺CD11b⁺ myeloid cells upon depletion of pTreg cells (Fig. S1, A and B). CD45⁺CD11b⁺ cells were mainly composed of three subpopulations, namely Ly6C⁺Ly6G⁺ (neutrophils), Ly6C⁺Ly6G⁻ (monocytes), and Ly6C⁻Ly6G⁻Siglec-F⁻ (tumor-associated macrophages, TAMs). We found that the depletion of pTreg cells resulted in a shift in myeloid cell populations from TAMs and monocytes to neutrophils (Fig. S1, C–G). In summary, pTreg cells played a crucial role in promoting tumor growth by modulating both antitumor (such as CD8⁺ T cells) and protumorigenic (such as TAMs and monocytes) immune cell types.

We next evaluated the potential therapeutic effects of pTreg cell depletion. With a similar approach, we established E.G7 tumor models in TCRbd^−/− mice 3 days after co-transfer of tTreg cells (sorted from CD45.1/CD45.2 double-positive Foxp3^GFP reporter mice), conventional CD4⁺ T cells and CD8⁺ T cells (sorted from CD45.2 Foxp3^DTR mice). DT was administered every 2 days from day 18 until day 24. Then, the tumor-bearing mice were sacrificed on day 25 (Fig. S1 H). We found that tumor growth was reduced in the DT-treated group as compared with the PBS control group (Fig. S1 I). By flow cytometry, we confirmed that pTreg cells (CD45.1⁻) were almost completely depleted by DT (Fig. S1, J and K). Ratios of CD8⁺ T cells to total Treg cells were increased upon pTreg cell depletion (Fig. S1 L). Consistently, there were increases in the frequencies of intratumoral CD8⁺ T cells expressing IFN-γ, granzyme B, and CD107a in total live cells in TILs (Fig. S1, M–R). Collectively, these results indicated that depletion of pTreg cells conferred potent therapeutic effects on tumor-bearing mice by increasing effector CD8⁺ T cells with cytolytic functions.

Treg cells contain several functional subsets analogous to CD4⁺ helper T (Th) cells (Sakaguchi et al., 2013; Shi and Chi, 2019). Thus, we analyzed if intratumoral Treg cells were similar to any of the Th subsets in their gene expression. First, we performed transcriptomic analyses on intratumoral Treg cells derived from both Hepa1-6 and E.G7 murine tumor models by RNA sequencing. Given that intratumoral Treg cells were composed of a substantial proportion of pTreg cells whereas Treg cells derived from LNs were mainly composed of tTreg cells (Fig. 1), we included Treg cells purified from tumor-free LNs as a control. As compared with Treg cells derived from tumor-free LNs, intratumoral Treg cells from both models exhibited increased expression of Th1 cell feature genes, such as Tbx21 and Ccr5 (Fig. 3, A and B). On the contrary, intratumoral Treg cells barely expressed the lineage-defining transcription factor of Th2, Gata3, and had reduced expression of Th17 signature genes such as Rorc and Ccr6 (Fig. 3, A and B). In addition, consistent with NRP1 staining analyzed by flow cytometry (Fig. 1, A–D), there were lower levels of Nrp1 expression in intratumoral Treg cells of both Hepa1-6 and E.G7 tumor models as compared with those in Treg cells derived from LNs (Fig. 3, A and B).

Next, we further verified our findings by flow cytometry. We stained T-bet, RORγt, BCL6, and GATA3 in Treg cells derived from LN, SPL, TDLN, and tumors in Hepa1-6 and E.G7 tumor models. Treg cells derived from tumor-free LNs and SPL were included as controls. tTreg cell marker, NRP1, was co-stained with these transcription factors. Consistent with our observations obtained from RNA sequencing data, intratumoral Treg cells highly expressed T-bet (Fig. 3, C–F). Similar results were obtained from the B16-OVA tumor model (Fig. 3, G and H). We further explored the relationship between NRP1 and T-bet expression in intratumoral Treg cells by co-staining. We found that T-bet–expressing intratumoral Treg cells were derived from NRP1⁻ intratumoral Treg cells (pTreg cells). Importantly, T-bet⁺NRP1⁻ Treg cells constituted ∼60% of intratumoral Treg cells (Fig. 3, I and J). By contrast, intratumoral Treg cells barely expressed RORγt (Fig. 3, K and L), BCL6 (Fig. 3, M and N), or GATA3 (Fig. 3, O and P). Taken together, our results showed that T-bet⁺NRP1⁻ Treg cells constituted a dominant group of intratumoral Treg cells in multiple cancer models.

In the global transcriptomics of intratumoral Treg cells, we identified upregulation of Th1-associated transcripts in intratumoral Treg cells (Fig. 3, A and B). Next, we further investigated if these transcripts were upregulated by intratumoral tTreg cells or pTreg cells by comparing their gene expression profiles. We sorted tTreg cells from the spleen and LNs of CD45.1 Foxp3^YFP-cre mice and sorted conventional FOXP3⁻ CD4⁺ T cells and CD8⁺ T cells from CD45.2 Foxp3^YFP-cre mice. We then co-transferred tTreg cells, conventional CD4⁺ T cells, and CD8⁺ T cells at ratios of 1:10:8 into TCRbd^−/− mice, followed by the inoculation of Hepa1-6 tumor cells 3 days later. We generated bulk RNA sequencing data from intratumoral pTreg and tTreg cells. Treg cells purified from tumor-free LNs were included as controls.

As compared with intratumoral tTreg cells, intratumoral pTreg cells showed upregulation of Th1 cell-associated transcripts (Tbx21, Ccr5, Cxcr3, Il12rb2, Il18r1, and Ifngr1) (Fig. 4 A). As compared with LN-derived Treg cells, intratumoral pTreg cells showed upregulation in all of these transcripts whereas intratumoral tTreg cells displayed downregulation of these transcripts, except Ccr5 (Fig. 4, B and C). This suggested that intratumoral T-bet⁺ Treg cells were mainly pTreg cells.

Next, we further compared the gene expression profiles between intratumoral pTreg and tTreg cells. As compared with tTreg cells, pTreg cells showed higher expression levels of genes related to Treg cell suppressive functions such as Entpd1, Tigit, Pdcd1, Tgfb1, Il10, and Ebi3 (Fig. 4, D). pTreg cells also showed upregulation of transcripts involved in activation (Tnfrsf18, Icos, Cd28, Cd69, Klrg1, Prdm1, Tnfrsf4, and Tnfrsf9), regulation of cell cycles (Pclaf, Rrm2, Smc2, Top2a, and Mcm4) as well as chemokine receptors, which facilitate their trafficking (Ccr8, Ccr2, Ccr4, Ccr6, Ccr9) (Fig. 4 D).

It has been known that the majority of Treg cells in the gut are pTreg cells (Harada et al., 2022). Thus, we investigated if the tumor-associated pTreg cells shared a similar genetic profile with the Treg cells in the gut. We compared the differentially expressed genes (DEGs) of intratumoral pTreg cells with the signature genes of colonic Treg cells retrieved from the publicly available single-cell RNA sequencing data (Miragaia et al., 2019). Gene set enrichment analysis (GSEA) revealed that the DEGs of intratumoral pTreg cells had positive correlations with those of colon-derived suppressive Treg cells, lymphoid-tissue (LT)-like Treg cells, and non-lymphoid tissue (NLT) Treg cells (Fig. 4, E–G). Nevertheless, the most significant positive correlation was found between the DEGs of intratumoral pTreg cells and those of colonic NLT Treg cells (Fig. 4 G). The common marker genes included Tnfrsf4 (OX40), Tnfrsf9 (4-1BB), Tnfrsf18 (GITR), Ccr8, Pdcd1, Batf, Hopx, Il2ra, and Klrg1 (Fig. 4 H).

We further compared the genetic profile of intratumoral pTreg cells with the signature genes of Treg cells derived from intestinal lymphoid and non-lymphoid tissue compartments of wild-type (WT) mice retrieved from another publicly available single-cell RNA sequencing data (Gu et al., 2024). The LT compartment was composed of secondary lymphoid organs, including mesenteric lymph nodes (mLN), the caecal patch (CP), and distal colon-organized lymphoid structures (OLS). The intestinal NLT compartment comprised lamina propria (LP) and small lymphoid aggregates (LA), which were present in the caecum and proximal colon. GSEA revealed that the DEGs of intratumoral pTreg cells had a significant positive correlation with those of Treg cells derived from intestinal NLT (LP and LA) (Fig. 4 I). The shared marker genes included Lag3, Tigit, Tnfrsf4 (OX40), Ctla4, Hopx, and S100a4 (Fig. 4 J). Although the correlation between the DEGs of intratumoral pTreg cells and those of intestinal LT Treg cells (mLN, CP, and OLS) showed a positive trend, the P value was on the border line (Fig. 4 K). Collectively, these results indicated that intratumoral pTreg cells shared similar genetic profile with intestinal Treg cells, particularly those derived from NLT, such as LP which contains a high proportion of pTreg cells.

In summary, our findings showed that intratumoral T-bet⁺ pTreg cells displayed enhanced activation, proliferation and suppressive functions. This observation further strengthened the roles of T-bet⁺ pTreg cells in the TME.

Next, we investigated the function of T-bet expression in intratumoral pTreg cells. We generated Foxp3^YFP-creTbx21^fl/fl mice to specifically delete Tbx21 in Treg cells. As previously described (Di Giovangiulio et al., 2019), these mice were vital and did not show any signs of autoimmune pathology. The body weights were similar between the Foxp3^YFP-creTbx21^fl/fl mice (designated KO group) and Foxp3^YFP-cre mice (designated WT group) (Fig. S2 A). We then established B16-OVA and E.G7 tumor models in KO and WT groups. We observed a reduction in tumor growth of B16-OVA and E.G7 in the KO group as compared with those in the WT group (Fig. 5, A and B). The tumor-bearing mice were sacrificed on days 18–21. LN, SPL, and TIL were isolated and processed for flow cytometry analyses. We obtained very similar results in B16-OVA and E.G7 tumor models, which showed that total Treg cell frequencies and absolute numbers were drastically reduced by at least 50% in tumors after knockout of T-bet. However, total Treg cell frequencies were not significantly affected in LN and SPL (Fig. 5, C–E and Fig. S2, B–E). Consistently, the frequencies of proliferative (Ki67⁺) intratumoral Treg cells among total live cells were also significantly lower in the KO group (Fig. S2 F), implying that there were less proliferative Treg cells in tumors after knockout of T-bet.

On the other hand, the expression of IFN-γ, tumor necrosis factor alpha (TNF-α), and granzyme B was increased in intratumoral CD8⁺ T cells in these mice (Fig. 5, F and H; and Fig. S2 G). Consistently, TIM-3⁺PD-1⁺ intratumoral exhausted CD8⁺ T cells were significantly reduced (Fig. 5, G and I; and Fig. S2 H), whereas the frequencies of stem-like CD8⁺ T cells were increased in intratumoral CD8⁺ T cells, shown by significantly increased percentages of CD8⁺ T cells expressing Ly108 and/or TCF1 (Fig. 5, J and K; and Fig. S2 I). Moreover, the proliferation of intratumoral CD8⁺ T cells was increased, with a significantly higher expression level of Ki67 (Fig. S2 J). This was accompanied by an increase in the ratios of CD8⁺ T cells to Treg cells in tumors of the KO group (Fig. S2 K). In addition, in Foxp3^YFP-creTbx21^fl/fl mice, we observed significantly higher frequencies and absolute numbers of IFN-γ–expressing CD4⁺FOXP3⁻ T cells in the TME than those of WT Foxp3^YFP-cre mice (Fig. S2, L–O).

Based on the data shown in Fig. 3, I and J; and Fig. 4, A–C, we speculated that T-bet plays an essential role in pTreg cells but not tTreg cells. Indeed, the staining of NRP1, the tTreg cell marker, showed that the frequencies of NRP1⁻ Treg cells (pTreg cells) were reduced in Treg cell populations in all LNs, SPL, and tumors of Tbx21 conditional KO mice (Fig. S2 P). Next, we used an adoptive transfer approach to further validate the role of T-bet in pTreg cells. We sorted conventional CD4⁺ T cells and CD8⁺ T cells from either CD45.1/CD45.2 (double-positive) Foxp3^YFP-creTbx21^fl/fl (designated KO group) or CD45.1/CD45.2 (double-positive) Foxp3^YFP-cre mice (designated WT group). tTreg cells were isolated from CD45.1 Foxp3^YFP-cre mice. Then, conventional CD4⁺ T cells, CD8⁺ T cells, and tTreg cells were co-transferred into TCRbd^−/− mice. After 3 days, E.G7 tumor cells were inoculated in the TCRbd^−/− recipient mice. The tumor-bearing mice were sacrificed for flow cytometry analyses on day 24 (Fig. S2 Q). We observed a reduction in tumor growth in KO group (Fig. S2 R). Consistent with the observation obtained from Tbx21 conditional knockout mice, the frequencies of intratumoral Treg cells were reduced by at least 50% in the KO group (Fig. S2 S). When the donor conventional CD4⁺ T cells were gated, the formation of pTreg cells was almost completely abolished in the KO group (Fig. S2 T). In parallel with these findings, we found that the ratios of CD8⁺ T cells to Treg cells were significantly higher in the KO group (Fig. S2 U). The frequencies of both total CD8⁺ T cells and tumor antigen-specific (OVA-tetramer⁺) CD8⁺ T cells were significantly higher in the KO group (Fig. S2, V–Y). In summary, our findings indicate that T-bet was not only highly expressed by intratumoral pTreg cells but also required for their functions in the TME.

Next, we evaluated the therapeutic effect of Treg cell–specific deletion of Tbx21 in combination with anti-PD-1 treatment. We established B16-OVA tumor model in Foxp3^YFP-cre (WT) mice and Foxp3^YFP-creTbx21^fl/fl (KO) mice. Anti-PD-1 antibody was injected every 3 days from day 9 onwards. Tumor growth and survival were assessed in the tumor-bearing mice. Either anti-PD-1 treatment alone or Treg cell–specific deletion of Tbx21 alone significantly reduced tumor growth in B16-OVA tumor models. Nevertheless, Treg cell–specific deletion of Tbx21 in combination with anti-PD-1 treatments showed the most significant reduction in tumor growth as compared with WT mice (Fig. 5 L). Moreover, Treg cell–specific deletion of Tbx21 significantly prolonged the survival in B16-OVA–bearing mice to an extent comparable to that of anti-PD-1 treatment (Fig. 5 M).

Thus far, we have established that a large proportion of intratumoral Treg cells were peripherally induced and expressed T-bet. However, it was not clear if they were directly induced from naive CD4⁺ T cells or converted from Th1 cells in the TME. It was reported that FOXP3 expression could be induced in Th1 cells by TGF-β in vitro (Kachler et al., 2018). However, it is still unknown if the conversion of Th1 cells into Treg cells could occur in vivo, especially in tumors. We thus utilized IFN-γ fate-mapping (Ifng^icreRosa26^YFP) mice to establish E.G7 and B16-OVA tumor models. The mice were sacrificed on days 16–18 after tumor inoculation. LN, SPL, TDLN, and tumors were isolated and processed for flow cytometry analyses. Tumor-free LN and SPL were included as controls. As compared with CD4⁺ T cells derived from LN, SPL, and TDLN, there were significantly higher frequencies of YFP⁺ cell population in tumor-infiltrating CD4⁺ T cells (Fig. 6, A and B). Most CD4⁺YFP⁺ cells (nearly 80%) highly expressed T-bet (Fig. 6, C and D). Among them, 20–30% exhibited FOXP3 expression (Fig. 6, C and E). Thus, FOXP3⁻T-bet⁺ (single-positive) and FOXP3⁺T-bet⁺ (double-positive) cells constituted a ratio of 3:1 in CD4⁺YFP⁺ cells (Fig. 6 C). The percentages of T-bet⁺FOXP3⁺ double-positive cells were significantly higher than those of FOXP3⁺ single-positive cells (Fig. 6 E).

In the CD4⁺YFP⁺ cell population, the majority (>80%) of these Th1-like Treg cells were NRP1⁻, indicating that they were pTreg cells (Fig. 6, F and G). Importantly, we found that co-staining of NRP1 and T-bet in the FOXP3⁺YFP⁺ cell population gave rise to an expression pattern (T-bet⁺NRP1⁻) (Fig. 6, H and I) similar to that observed in the predominant subset of the intratumoral Treg cells (Fig. 3, I and J). Interestingly, the majority (∼90%) of T-bet⁺FOXP3⁺ cells did not have IFN-γ expression (Fig. 6, J and K).

We gained similar findings in the B16-OVA tumor model. There was an accumulation of the YFP⁺ population in tumor-infiltrating CD4⁺ T cells in the B16-OVA tumor model (Fig. S3 A). This CD4⁺YFP⁺ population was highly associated (>70%) with T-bet expression (Fig. S3 B). More than 30% of these Th1-like (CD4⁺T-bet⁺YFP⁺) cells had FOXP3 expression (Fig. S3 C). The majority of YFP⁺ Treg cells (>80%) were NRP1⁻, indicating that they were pTreg cells (Fig. S3 D). In addition, most (∼90%) of T-bet⁺FOXP3⁺ T cells lacked IFN-γ expression (Fig. S3 E). These observations hinted that there was a conversion of IFN-γ–expressing Th1 cells into pTreg cells in the TME and constituted the NRP1⁻T-bet⁺ dominant subset of intratumoral Treg cells.

Since the development of Th17 and Treg cells has been linked and is thought to be reciprocally regulated (Lee, 2018), we also performed fate-mapping of Il17f-producing CD4⁺ T cells by establishing E.G7 tumor model in Il17f^CreRosa26^tdTomato mice. The results showed no significant frequency of IL-17F–producing cells in either FOXP3⁻ or FOXP3⁺ CD4⁺ T cell compartments (Fig. S3 F). This result was consistent with the observation that RORγt was barely expressed by intratumoral Treg cells in the E.G7 tumor model (Fig. 3, K and L). Collectively, these findings suggest that the TME was not associated with a strong Th17 response and there was no evidence of Th17 cell conversion into Treg cells in the E.G7 tumor model.

Our fate-mapping experiments suggested that Th1 cells could convert into T-bet⁺ pTreg cells in tumors. Therefore, we further validated by performing in vitro differentiation of Th1 cells from naive CD4⁺ T cells sorted from CD45.1/CD45.2 (double-positive) OT-II Ifng^YFP mice. On the fourth day of culture, we sorted IFN-γ–producing CD4⁺ T cells with the YFP signal and transferred them into E.G7-bearing C57BL/6J (B6) mice on day 7 after tumor inoculation. The tumor-bearing mice were sacrificed on day 21 for flow cytometry analyses (Fig. S3 G). The results showed that ∼10% of the donor Th1 cells (CD45.1⁺CD4⁺ T cells) were converted into T-bet⁺FOXP3⁺ Treg cells in the TME (Fig. 6, L and M). As expected, most (>80%) of the donor Th1 cells expressed T-bet (Fig. 6, L and N). Consistent with the findings of fate-mapping studies, T-bet⁺FOXP3⁺ cells were mostly NRP1⁻ (Fig. S3 H). Moreover, most of the T-bet⁺FOXP3⁺ cells no longer expressed IFN-γ (Fig. S3 I). Therefore, the adoptive transfer of in vitro differentiated Th1 cells further demonstrated a conversion from Th1 cells into pTreg cells in the TME.

Since TGF-β has been known to regulate the generation of pTreg cells (Liu et al., 2007), we next tested whether the conversion of Th1 cells into pTreg cells required TGF-β signaling. As previously described (Marie et al., 2006), Cd4^CreTgfbr2^fl/fl mice developed autoimmune pathology leading to 100% fatality within 3 wk after birth. Therefore, we generated OT-II Cd4^CreTgfbr2^fl/fl mice, which only showed mild immune pathology and therefore were able to survive up to a few months after birth, some even longer. We sorted naive CD4⁺ T cells from CD45.1/CD45.2 (double-positive) OT-II Cd4^CreTgfbr2^fl/fl (designated KO group) or CD45.1/CD45.2 (double-positive) OT-II Tgfbr2^fl/fl mice (designated WT group). CD8⁺ T cells and tTreg cells were sorted from CD45.2 Foxp3^GFP reporter mice. The sorted naive CD4⁺ T cells, tTreg cells, and CD8⁺ T cells were co-transferred to TCRbd^−/− mice. After 3 days, B16-OVA tumor cells were inoculated in the TCRbd^−/− recipient mice. The mice were sacrificed on day 16 for flow cytometry analyses (Fig. S3 J). Tumor growth was significantly reduced in the KO group (Fig. 6 O and Fig. S3 K). We evaluated the capability of the donor cells in infiltration, activation, and proliferation after the deletion of Tgfbr2. The results showed that knocking out Tgfbr2 did not affect the infiltration of total CD45.1⁺CD4⁺ T cells (Fig. S3 L) or the activation of CD45.1⁺CD4⁺ T cells in tumors (Fig. S3, M and N). However, the proliferation of CD45.1⁺CD4⁺ T cells was slightly reduced in the KO group (Fig. S3, M and O). Consistent with our hypothesis, deficiency in Tgfbr2 impaired the conversion from donor CD45.1⁺ conventional CD4⁺ T cells to pTreg cells (Fig. 6, P and Q). Moreover, the absence of Tgfbr2 led to an increase in the frequencies of Th1 cells (T-bet⁺FOXP3⁻) (Fig. 6, R and S). The increase in the frequencies of Th1 cells (T-bet⁺FOXP3⁻) resulted in an elevated level of IFN-γ expression in CD4⁺ T cells and increased numbers of IFN-γ⁺CD4⁺ T cells (Fig. 6, T and U; and Fig. S3, P–S). As a result of impaired pTreg cell generation in the TME, we observed an increase in the absolute numbers of CD8⁺ T cells in the KO group (Fig. S3 T). In parallel with these findings, we observed no changes in body weight and gut morphology in both WT and KO groups (Fig. S3, U and V). Taken together, our findings indicate that pTreg cells could be generated in the TME from Th1 cells, dependent on the TGF-β signaling pathway.

To further elucidate the ontogeny of intratumoral T-bet⁺ Treg cells and the role of T-bet in regulating this specific Treg cell population in the TME, we performed single-cell transcriptomics on Treg cells sorted from Hepa1-6 tumors of Foxp3^YFP-cre and Foxp3^YFP-creTbx21^fl/fl mice. We sorted Treg cells (CD4⁺YFP⁺) from Hepa1-6 tumors of Foxp3^YFP-cre and Foxp3^YFP-creTbx21^fl/fl mice. Conventional CD4⁺ T cells (CD4⁺YFP⁻) and Treg cells (CD4⁺YFP⁺) sorted from Hepa1-6 tumors and tumor-free LN of Foxp3^YFP-cre mice, respectively, were included for comparison (only data on Treg cells were shown).

Based on these single-cell gene expression data, we identified five transcriptionally distinct cell clusters (Fig. 7 A). Clusters 0 (C0) and 4 (C4) were enriched in WT intratumoral Treg cells, whereas cluster 2 (C2) was enriched in Tbx21-deficient intratumoral Treg cells. Cluster 1 (C1) was enriched in WT LN-derived Treg cells. Cluster 3 (C3) was enriched in WT intratumoral conventional CD4⁺ T cells (data not shown) and showed a comparable proportion in WT intratumoral Treg cells and Tbx21-deficient intratumoral Treg cells (Fig. 7, B and C).

C0, marked by Th1 cell features (Tbx21, Ccr5, Cxcr3, Il12rb2, and Il18r1) was the most dominant cell cluster, contributing up to 71% in intratumoral WT Treg cells (Fig. 7, B and C). C0 was enriched with molecules associated with suppressive activities of Treg cells, including Entpd1, Ctla4, Tigit, Tgfb1, Il10, and Ebi3, as well as activation markers such as Icos, Klrg1, Prdm1, Tnfrsf4 (OX40), and Tnfrsf9 (4-1BB) (Fig. 7 D). Taken together, this cluster represents tumor-infiltrating Th1-like effector Treg cells with enhanced activation and suppressive function.

C1 cells were characterized by increased expression of Foxp3 and a tTreg cell marker, Nrp1. This cluster also displayed increased expression of genes that support Treg cell differentiation and suppressive functions (Foxp1, Foxo1, Bach2, and Ikzf2). This cluster expressed transcripts of naive Treg cells, including Sell, Slamf6, Tcf7, Lef1, and Ccr7 (Fig. 7 D). Collectively, this cluster belonged to naive/resting Treg cells. The majority of cluster 1 cells were possibly tTreg cells.

C2 was the most enriched cluster in Tbx21-deficient Treg cells. This cluster was marked by Th2/Th17 cell features. C2 upregulated Th2 cell–specific transcription factor (Gata3) and Th2 cytokine transcripts (Il4, Il13, and Il5). This cluster also upregulated Th17 cell lineage–defining transcription factor (Rorc), Th17 cell cytokine transcript (Il17a), and Irf4, a transcription factor that promotes differentiation of both Th2 and Th17 cells (Fig. 7 D).

C3 was defined as a Th1 cell cluster. C3 shared signature genes of C0, including Tbx21, Ccr5, Il12rb2, Il18r1, and Entpd1 (Fig. 7 D). C3 also highly expressed transcripts of Th1 effector cytokines (Ifng, Tnf1) and granzyme family members (Gzmd, Gzme, and Gzmk), accompanied by cytotoxic-associated molecules (Prf1, Nkg7). This cluster also upregulated the signature genes associated with activated effector cells, such as Icos, Prdm1, and Cx3cr1 (Fig. 7 D). Taken together, C3 marks the most differentiated effector Th1 cells.

C4 was the most proliferative cluster, indicated by the increased expression of the proliferation marker gene, Mki67. This cluster was enriched with a series of genes involved in DNA replication, transcription, and cell cycle regulation, including Exo1, Rrm1, Rrm2, minichromosome maintenance (MCM) gene family (Mcm3, Mcm5, Mcm10), Pclaf, Smc2, and Top2a. Clusters 4 and 0 shared expression of Treg cell activation markers Icos, Tnfrsf4 (OX40), and Tnfrsf9 (4-1BB) (Fig. 7 D). Thus, this cluster was defined as activated proliferative Treg cells.

Next, we further investigated the role of T-bet in intratumoral Treg cells. Given that the cell proportion of Th1-like Treg cells (C0) showed the most profound decrease in intratumoral Tbx21-deficient Treg cells as compared with that in intratumoral WT Treg cells (KO versus WT, 0.28 versus 0.71), we further analyzed their DEGs in C0. As expected, as compared with intratumoral WT Treg cells, Th1-associated molecules (Tbx21, Ccr5, Cxcr3, Il12rb2, Il18r1) were drastically downregulated in Tbx21-deficient Treg cells (Fig. S4 A). Moreover, there was downregulation of transcripts involved in activation (Icos, Cd28, Klrg1, Prdm1, Tnfrsf4, Tnfrsf9, and Tnfrsf18) and proliferation (Mki67, Rrm1, Rrm2, Smc2, Top2a, Tk1, Mcm9, and Mcm6) in Tbx21-deficient Treg cells (Fig. S4 A). We further found that there was downregulation of genes which support the Treg cell suppressive functions (Lrrc32, Cd27, Cd74, Ikzf2, Ikzf4, Foxo1, Foxo3, Nr4a1, Nr4a2, Nr4a3, Itgb8, Itgav, Itgae, Myb, and Cd86), and cell trafficking (Ccr2, Ccr7) in Tbx21-deficient Treg cells (Fig. S4 A). Consistently, Tbx21-deficient Treg cells downregulated most of the transcripts involved in Treg cell suppressive activities, including Tgfb1, Ebi3, Entpd1, Tigit, CTLA4, and Lag3 (Fig. S4 A). In addition to Entpd1 (encoding CD39), Tbx21-deficient Treg cells downregulated other genes involved in nucleotide metabolic process (Gda, Ada, Upp1, Nudt4, Nudt16, Enpp1, Enpp5, Art2b, Ncf2, Pnp, Nt5c2, and Entpd7) (Fig. S4 A). By contrast, there was upregulation of Th2/Th17-associated transcripts (Gata3, Il4, Il13, Il5, Rorc, Il17a, Irf4) in Tbx21-deficient Treg cells, as observed in cluster 2 (Fig. S4 A). Moreover, Tbx21-deficient Treg cells upregulated genes involved in the regulation of inflammatory response, such as Cd40lg, Pparg, Il1r1, Ilrl1, Ilr12, Jun, Il6, Il6ra, Fn1, Thbs1, Tnfsf4, Tnfsf11, Vegf, Fosl1, Arg1, and Arg2 (Fig. S4 A).

While the cell proportion of Th1-like Treg cells (C0) was decreased in intratumoral Tbx21-deficient Treg cells, the proportion of Th2/Th17-like Treg cells (C2) was increased as compared with that of intratumoral WT Treg cells (KO versus WT, 0.45 versus 0.02). Thus, we further compared the DEGs and enriched pathways between Th1-like Treg cells (C0) and Th2/Th17-like Treg cells (C2). The DEGs and pathway analyses showed similar results as those observed in the comparison between intratumoral WT and Tbx21-deficient Treg cells in C0. As compared with Th2/Th17-like Treg cells (C2), Th1-like Treg cells (C0) showed upregulation of Th1-associated molecules (Tbx21, Ccr5, Cxcr3), proliferation (Mki67), and activation marker genes (Icos), as well as genes involved in the mechanisms of Treg cell suppression (Entpd1, Nt5e, Tgfb1, Ebi3, Ctla4, Lag3, Tigit, and Gzmb). Meanwhile, Th2/Th17-like Treg cells upregulated marker genes of Th2 cells (Gata3, Il4, Il5, and Il13) and Th17 cells (Rorc and Il17a) (Fig. 7 E). By flow cytometry, we detected an increased expression of Th2/Th17 transcription factors and cytokines in intratumoral Treg cells derived from Tbx21 conditional KO mice (Fig. S4, B and C). These findings suggested that T-bet⁺ Treg cells prevented Th2/Th17 cell response in Treg cells in the TME.

Next, we delineated the pathways regulated by T-bet in intratumoral Treg cells. As compared with Th1-like Treg cells (C0), Th2/Th17-like Treg cells (C2) showed downregulation of pathways related to Treg cell suppression mechanisms, including regulation of cell–cell adhesion, CD39-mediated nucleotide catabolic process, and TGF-β production, as well as Th1 type immune response. Meanwhile, Th2/Th17-like Treg cells upregulated pathways involved in Th2/Th17 cell differentiation and cytokine production, as well as regulation of inflammatory response (Fig. S4, D and E). Collectively, these findings indicated that there was a dynamic change in the transcriptomic program resulting in a deviation from Th1-like Treg cells to Th2/Th17-like Treg cells in the TME after deletion of Tbx21 in Treg cells.

In contrast to Th1-like Treg cells (C0), which showed enhanced activated and suppressive protumorigenic characteristics, Th2/Th17-like Treg cells (C2) displayed less suppressive characteristics. It is likely that the decrease of C0 and increase of C2 in cell proportion created a less suppressive TME in Tbx21 conditional KO mice as compared with WT mice. Consistently, our flow cytometry analyses showed a drastic decrease in intratumoral Treg cell frequencies and absolute number in Tbx21 conditional KO mice as compared with those of WT mice (Fig. 5, C–E and Fig. S2, B–E). Thus, it is conceivable that the proportion of suppressive protumorigenic Treg cell population was generally decreased in Tbx21 conditional KO mice.

Given that Th2/Th17-like Treg cells (C2) showed downregulation in various transcripts involved in Treg cell suppression mechanisms as compared with Th1-like Treg cells (C0) (Fig. 7 E), we further analyzed the co-expression between these transcripts and Tbx21 (Fig. 7 F). Among these transcripts, Entpd1 and Tigit showed relatively higher fold-change between C0 and C2 (Fig. 7 E), both of these transcripts co-expressed with Tbx21. By contrast, both Entpd1 and Tigit showed no co-expression with Gata3 or Rorc (Fig. 7 F). In addition, our flow cytometry results verified that both CD39 and TIGIT expression were reduced in Tbx21-deficient Treg cells. However, as compared with TIGIT expression, CD39 expression showed a relatively more profound loss in Tbx21-deficient Treg cells in both percentage and mean fluorescence intensity (MFI) (Fig. 7, G–I). These findings suggested that CD39-mediated metabolic disruption could be a specific suppression mechanism exerted by Th1-like Treg cells in the TME. In summary, T-bet plays an essential role in supporting the activation and proliferation of Treg cells in tumors, as well as Treg suppressive functions, particularly, CD39-mediated metabolic disruption via nucleotide catabolic process, and prevention of Th2/Th17 cell response.

To further delineate the ontogeny of intratumoral T-bet⁺ Treg cells, we used RNA velocity to reveal the temporal dynamics of WT Treg cells in the TME (Fig. 7 J). We found that the Th1-like Treg cell cluster (C0) was the main terminally differentiated cell population. C0 was mainly initiated from C3, a cell cluster that differentially upregulated the Th1 cell signature genes, indicating a transition from C3 to C0. Most of the proliferating cells from C4 terminally differentiated into Th1-like Treg cells (C0). Taken together, these findings further strengthened the evidence that intratumoral Th1-like Treg cells were derived from pTreg cells converted from Th1 cells in the TME.

Although multiple suppression mechanisms have been identified in Treg cells (Schmidt et al., 2012), their exact functions in tumors remain unclear. Through our single-cell transcriptomics and flow cytometry analyses on intratumoral Tbx21-deficient Treg cells, we have found that CD39-mediated metabolic disruption could be a Th1-like Treg cell–specific suppression mechanism. Therefore, we further investigated the roles of the CD39-mediated suppression mechanism in intratumoral Treg cells. First, we turned to our global transcriptomics of Treg cells. The DEGs and pathway analyses on the RNA sequencing data of murine E.G7 and Hepa1-6 tumor models gave rise to six different clusters (Fig. 8, A–E). Clusters 1 and 2 consisted of genes upregulated in intratumoral Treg cells of both E.G7 and Hepa1-6 (Fig. 8, A–C). Clusters 5 and 6 consisted of genes upregulated in intratumoral Treg cells of E.G7 and Hepa1-6, respectively (Fig. 8, A, D, and E). The pathway analyses on the upregulated genes in clusters 1, 2, 5, and 6 revealed similar enriched pathways of nucleotide metabolism (Fig. 8, B–E). In parallel with these findings, we further unraveled that intratumoral Treg cells of both E.G7 and Hepa1-6 upregulated most of the genes involved in purine metabolism. These included genes encode ectonucleoside triphosphate diphosphohydrolase family (Entpd1, Entpd7), purine-nucleoside phosphorylase (Pnp), nudix hydrolase family (Nudt1, Nudt4, Nudt9, Nudt21), adenosine kinase family (Ak1, Ak2, Ak3, Ak4, Ak5, Ak6, Ak8), NME/NM23 nucleoside diphosphate kinase family (Nme1, Nme4), and hypoxanthine guanine phosphoribosyl transferase (Hprt) (Fig. 8, F and G). Besides, enriched pathways of response to hypoxia and oxidative stress were also identified in these gene clusters (Fig. 8, B–E). Consistent with these results, the comparison of gene expression profiles between intratumoral pTreg and tTreg cells in the Hepa1-6–bearing TCRbd^−/− mice that were co-transferred with tTreg and Tconv cells unraveled that intratumoral pTreg cells upregulated the genes involved in nucleotide catabolism (Fig. 8 H).

It has been known that CD39 co-operates with CD73 in adenosine signaling pathway. CD39 converts ATP into AMP whereas CD73 further converts AMP into adenosine (Borsellino et al., 2007; Deaglio et al., 2007; Kobie et al., 2006). Thus, we determined if there is co-expression of CD39 and CD73 on intratumoral Treg cells. We established E.G7 and Hepa1-6 tumor models in WT C57BL/6J mice. On day 21, the tumor-bearing mice were sacrificed for flow cytometry. We found that CD39 was highly expressed by intratumoral Treg cells. While CD39 was barely expressed by Treg cells derived from LN, SPL, and TDLN (Fig. S5, A–C), CD73 was highly expressed by Treg cells derived from LN, SPL, TDLN, and tumors (Fig. S5, A, D, and E). Therefore, in the Treg cell populations examined in these tissues, CD39 and CD73 were only co-expressed in Treg cells in the TME (Fig. S5, A, F, and G). This observation was consistent in both tumor models.

CD39 (encoded by ectonucleoside triphosphate diphosphohydrolase-1, ENTPD1 gene) is an ectonucleotidase that hydrolyzes the phosphate bond in ATP and converts ATP into AMP (Borsellino et al., 2007; Deaglio et al., 2007). CD39-mediated phosphate bond hydrolysis is essential for nucleotide metabolism, namely purine metabolism and pyrimidine metabolism (Wu et al., 2022). Purine metabolism is crucial in the production of adenosine which may suppress the functions of CD8⁺ T cells through binding to the A2AR (Sitkovsky et al., 2008; Sorrentino et al., 2019; Timperi and Barnaba, 2021). Therefore, we examined the expression of A2AR on CD8⁺ T cells. We established E.G7 tumor models in WT C57BL/6J mice. On day 21, the tumor-bearing mice were sacrificed for flow cytometry. We found that A2AR expression in intratumoral CD8⁺ T cells was significantly higher than that in LN-derived CD8⁺ T cells (Fig. S5, H and I). We further characterized intratumoral A2AR⁺ CD8⁺ T cells by co-staining A2AR with PD-1, ICOS, CX3CR1, TIGIT, TIM-3, and Ly108. We found that intratumoral A2AR⁺CD8⁺ T cells showed effector phenotypes (PD-1^hi ICOS^hi CX3CR1^hi) and expressed significantly higher levels of exhaustion-associated molecules (TIM-3⁺TIGIT⁺Ly108⁻) as compared with their counterpart (Fig. S5, J–L).

Then, we explored the relationships among CD39, T-bet, and NRP1 expression in intratumoral Treg cells. In parallel with our single-cell transcriptomics (Fig. 7 F), our flow cytometry results indicated that intratumoral Treg cells in the Hepa1-6 tumor model showed co-expression of T-bet and CD39 (Fig. 8, I and J). We further characterized the molecular phenotypes of intratumoral Treg cells derived from Foxp3^YFP-creTbx21^fl/fl mice (designated KO group) in comparison with the control Foxp3^YFP-cre mice (designated WT group). Strikingly, we found that T-bet expression was significantly higher in CD39⁺NRP1⁻ intratumoral pTreg cells in WT mice (Fig. 8, K and L). Moreover, CD39 was selectively expressed by NRP1⁻ intratumoral Treg cells in WT mice. This was indicated by a significantly higher level of CD39 expression in NRP1⁻ intratumoral pTreg cells in WT mice, which did not exist in Treg cell–specific T-bet KO mice (Fig. 8, M and N). Thus, the expression of T-bet and CD39 was highly correlative. T-bet expression was always associated with NRP1⁻ intratumoral pTreg cells (Fig. 3, I and J), whereas CD39 was also selectively expressed by NRP1⁻ intratumoral pTreg cells (Fig. 8, K, M, and N). These findings further strengthened the evidence that NRP1⁻T-bet⁺ intratumoral pTreg cells exert the suppressive function through CD39 expression.

To determine if CD39 expression in Treg cells is tumor-specific, we assessed and compared CD39 expression on Treg cells derived from various tissues. These tissues included inguinal LNs (iLN), thymus, Peyer’s patches (PP), SPL, mesenteric LNs (mLN), colonic intraepithelial lymphocytes (IEL), small intestine IEL, colonic lamina propria lymphocytes (LPL), and small intestine LPL, isolated from tumor-free mice. Staining of CD39 expression in intratumoral Treg cells was used as a positive control. Strikingly, we found that similar to tumor-derived Treg cells, CD39 was also highly expressed by Treg cells derived from tissues containing a high proportion of pTreg cells, including small intestine IEL, colonic LPL, and small intestine LPL (Fig. S5, M and N). We further validated these observations by co-staining CD39 and NRP1. Indeed, we found that CD39 was selectively expressed by NRP1⁻ Treg cells (Fig. S5, O and P). These observations were consistent in tumors, small intestine IEL, colonic LPL, and small intestine LPL (Fig. 8, O and P). Taken together, our findings concluded that T-bet⁺ pTreg cells had a high expression level of CD39, a key mediator of the adenosine signaling pathway in nucleotide metabolism. CD39 may serve as a therapeutic target for selectively targeting intratumoral T-bet⁺ pTreg cells without provoking systemic autoimmune responses.

Next, we further verified if nucleotide metabolism was an important mechanism utilized by intratumoral Treg cells to exert their suppressive functions in human cancers. First, we used publicly available RNA sequencing data of Treg cells in human breast cancer to perform pathway analyses (Plitas et al., 2016). GSEA showed that the DEGs of intratumoral Treg cells in both E.G7 and Hepa1-6 murine tumor models had positive correlations with those in intratumoral Treg cells of human breast cancer (Fig. 9, A and B). As expected, pathways of nucleotide metabolism were enriched in intratumoral Treg cells of human breast cancer (Fig. 9 C). We further validated that intratumoral Treg cells in human breast cancer upregulated most of the genes involved in nucleotide metabolism, including the ENTPD1 gene (Fig. 9 D). The ENTPD1 gene was upregulated in the small molecule catabolic process and nucleobase-containing small molecule catabolic process, the two important pathways in nucleotide metabolism (Fig. 9, E and F). Some of the genes that were involved in nucleotide metabolism, such as TYMP, NUDT5, PGM2, HPRT1, GLRX, GPX1, and ENTPD1, were also upregulated in intratumoral Treg cells of other human cancer types (Fig. 9 G). These cancer types included HCC (Zheng et al., 2017), colorectal cancer (CRC), and NSCLC (De Simone et al., 2016). Among these genes, ENTPD1 is the key upregulated gene in intratumoral Treg cells found across different types of human cancers (Fig. 9 G). Volcano plots of HCC intratumoral Treg cells further revealed an upregulated expression of the ENTPD1 gene (Fig. 9 H), as determined by single-cell RNA sequencing in a previous report (Zheng et al., 2017).

By using the publicly available single-cell RNA sequencing data of human cancers, we further investigated if there was co-expression of T-bet (encoded by TBX21) and CD39 (encoded by ENTPD1) in intratumoral Treg cells of human cancers. The TBX21 expression was identified in intratumoral Treg cells across various types of human cancers. There was co-expression of TBX21 and ENTPD1 in intratumoral Treg cells of various human cancers, including nasopharyngeal carcinoma (NPC), breast cancer (BRCA), melanoma, squamous cell carcinoma (SCC), HCC, and CRC (Fig. 9, I–N). Consistent with this observation, our flow cytometric results indicated that intratumoral Treg cells in HCC patients highly co-expressed T-BET and CD39 (Fig. 9, O–S). Taken together, these findings were consistent with those observed in our murine tumor models.

To further elucidate the function of CD39 expressed in intratumoral Treg cells, we generated Foxp3^YFP-creEntpd1^fl/fl mice. To our knowledge, this is the first description of this strain. Flow cytometry confirmed the deletion of Entpd1, indicated by a lack of CD39 expression (a reduction of 92–96%) in Treg cells derived from colonic LPL and E.G7 tumors of Entpd1^fl/flFoxp3^YFP-cre mice as compared with those of Foxp3^YFP-cre mice (Fig. 10, A and B). These mice did not show any sign of autoimmune pathology. The body weights of Foxp3^YFP-creEntpd1^fl/fl (designated KO) mice were comparable with those of Foxp3^YFP-cre (designated WT) mice in both male and female mice (21–24 wk) (Fig. 10 C). Moreover, Entpd1 KO mice showed no morphological changes in guts and SPL (Fig. 10, D and E). Histological analyses on both colons and small intestines of Entpd1 KO mice showed no signs of inflammation (Fig. 10 F). No significant differences were observed in the frequencies of activated cells (CD44^hi) in either FOXP3⁻ or FOXP3⁺ CD4⁺ T cells derived from colonic LPL and small intestine LPL between Entpd1 KO and WT mice (Fig. 10, G and H). Consistently, similar frequencies of effector (CD44⁺CD62L⁻) CD4⁺ and CD8⁺ T cells were observed in LNs and SPL between Entpd1 KO and WT mice (Fig. 10, I–K). In addition, no significant difference was found in the frequencies of Treg cells in LN and tumors between E.G7-bearing Entpd1 KO and WT mice (Fig. 10 L). In the TME, the frequencies of T-bet⁺NRP1⁻ pTreg cells were not affected after the deletion of Entpd1 in Treg cells (Fig. 10 M). These observations indicated that CD39 expression in Treg cells was not required for maintaining immune homeostasis.

Then, we first conducted an in vitro study to investigate the function of CD39 expressed by intratumoral Treg cells. Given that CD39 was selectively expressed by NRP1⁻ pTreg cell subset, we sorted intratumoral CD4⁺YFP⁺NRP1⁻ Treg cells from Hepa1-6–bearing Foxp3^YFP-cre (WT) and Foxp3^YFP-creEntpd1^fl/fl (KO) mice. The sorted Treg cells were co-cultured with naive CD8⁺ T cells in 96-well culture plates for 72 h. CD8⁺ T cells co-cultured in the absence of Treg cells were included as negative controls. The cell division index (DI) of CD8⁺ T cells co-cultured with intratumoral Treg cells of the KO group was significantly higher than that of the WT group (Fig. 10, N and O). The percentage of suppression in intratumoral Treg cells of the KO group was therefore significantly lower than that of the WT group (Fig. 10 P). Thus, as compared with intratumoral WT NRP1⁻ Treg cells, Entpd1-deficient NRP1⁻ Treg cells showed a significantly lower suppressive capacity toward CD8⁺ T cells in vitro.

To further study the role of CD39 expressed by intratumoral Treg cells in the TME, we established E.G7 and B16-OVA tumor models in Foxp3^YFP-creEntpd1^fl/fl mice. The tumor-bearing mice were sacrificed on days 18–21 for flow cytometry analyses. We observed a reduction in the tumor growth in both E.G7 and B16-OVA tumor models (Fig. 10, Q and R). Upon deletion of Entpd1 in Treg cells, CD8⁺ T cell function was enhanced, as evidenced by increased expression of TNF-α and IFN-γ (Fig. 10 S), along with reduced expression of PD-1 and TIM-3 (Fig. 10 T).

Next, we evaluated the therapeutic effects of Treg cell–specific deletion of Entpd1 in combination with anti-PD-1 treatments. We established B16-OVA tumor models in Foxp3^YFP-cre (WT) mice and Foxp3^YFP-creEntpd1^fl/fl (KO) mice. Anti-PD-1 antibody was injected every 3 days from day 9 onwards. We evaluated tumor growth and survival in B16-OVA tumor-bearing WT and KO mice upon anti-PD-1 treatments. Either anti-PD-1 treatment alone or Treg cell–specific deletion of Entpd1 alone significantly reduced tumor growth in B16-OVA tumor models. However, Treg cell–specific deletion of Entpd1 in combination with anti-PD-1 treatments showed the most significant reduction in tumor growth as compared with those of WT mice (Fig. 10 U). In addition, the combination of anti-PD-1 treatment and Treg cell–specific deletion of Entpd1 significantly enhanced the survival in B16-OVA–bearing mice (Fig. 10 V).

Although intratumoral Treg cells are potential targets for antitumor immunotherapy, their ontogeny and functional mechanisms remain poorly understood. Therefore, our study aimed to identify the origins of intratumoral Treg cells and elucidate their roles in the TME. We found that T-bet–expressing pTreg cells, which are converted from Th1 cells in response to TGF-β signaling within the TME, suppress the function of antitumor CD8⁺ T cell function through CD39.

NRP1 was previously identified as a specific marker for thymus-derived tTreg cells (Weiss et al., 2012; Yadav et al., 2012). By employing NRP1 staining to analyze Treg cells derived from E.G7, Hepa1-6, and B16-OVA murine tumor models, we found that at least half of the intratumoral Treg cell population was NRP1-negative, suggesting that these Treg cells were de novo–induced pTreg cells. Similarly, through NRP1 staining, Weiss and colleagues found that the proportion of pTreg cells in the TME of 4T1 tumor model was at least 40–50%, and it was as high as 90% in the MCA38 tumor model (Weiss et al., 2012). Using an adoptive transfer approach, we further verified our findings that at least half of the intratumoral Treg cells were composed of pTreg cells converted from conventional CD4⁺ T cells. In addition, we further confirmed that NRP1 was indeed more selectively, though not exclusively, expressed by tTreg cells. Taken together, pTreg cells were either more dominant than, or at least comparable to, tTreg cells in the TME of multiple tumor models.

In both murine and human cancers, analyses of TCR repertoires suggested that intratumoral Treg cells and conventional CD4⁺ T cells shared little similarity in TCR clonotypes (Ahmadzadeh et al., 2019; Hindley et al., 2011; Plitas et al., 2016). A study by Hindley et al. showed that the TCR repertoires of tumor-infiltrating Tconv cells and Treg cells in carcinogen-induced tumors in mice were largely distinct (Hindley et al., 2011). Thus, the authors proposed that tumor-infiltrating Tconv cells and Treg cells arise from distinct populations with unique TCR repertoires. However, there is another possible explanation. The induction of pTreg cells occurs through the activation of TCR signaling in a manner dependent on both antigens and cytokine signaling (Sauer et al., 2008). Thus, it is possible that pTreg cells converted from conventional CD4⁺ T cells could possess a substantially skewed repertoire and subsequently undergo clonal expansion (Stockis et al., 2019).

A more recent study by Xydia and colleagues investigated the clonal origin of Treg cells in breast cancer patients using TCR and single-cell transcriptome sequencing (Xydia et al., 2021). Their study showed that the TCR repertoire of intratumoral Treg cells was distinct from that of Treg cells in peripheral blood, but shared a common clonal origin with intratumoral antigen-experienced Tconv cells. Their pseudotime analysis identified two major differentiation trajectories within the tumor, both originating from early activated Tconv cells, developing either into Th1 cells or into Treg cells, with genes essential for the conversion of Tconv cells into pTreg cells being expressed along this trajectory. The majority (up to 65%) of intratumoral Treg cells were derived from Tconv, activated Tconv, and Treg cells, which shared highly expanded clones. Collectively, their findings suggested the possibility that antigen-experienced effector T cells convert into pTreg cells upon TCR-dependent activation of a few highly expanded Tconv cells after secondary antigen encounter within the TME. The pseudotime analysis in our tumor mouse model has revealed a major differentiation trajectory in the TME, initiating from Th1 cells, terminally differentiating into Th1-like Treg cells, the most dominant intratumoral Treg cell cluster (71%), which highly expressed genes related to their suppressive functions, particularly Entpd1 (encoding CD39). Thus, in parallel with these results, our findings contribute to elucidating the underlying mechanisms of pTreg cell induction and accumulation, as well as their roles in the TME. Ultimately, our data suggest that CD39 could be a potential therapeutic target for intervening in the suppressive functions of these cells in cancer patients.

Our findings clearly demonstrated that intratumoral pTreg cells play a crucial role in suppressing the function of CD8⁺ T cells. However, we do not rule out the potential functional significance of tumor-infiltrating tTreg cells. While depletion of tTreg cells may also provoke tumor progression, their systemic removal may lead to autoimmune responses. Thus, selective targeting of pTreg cells may be considered a safer approach.

To selectively target tumor-infiltrating pTreg cells, it is crucial to investigate the heterogeneity of Treg cell populations in the TME. It has been increasingly recognized that the plasticity of Treg cells allows them to display gene expression patterns similar to those of CD4⁺ Th cells (Sakaguchi et al., 2013; Shi and Chi, 2019). Through global transcriptomic analyses on intratumoral Treg cells in Hepa1-6 and E.G7 murine tumor models, we showed that intratumoral Treg cells upregulated feature genes of Th1 cells such as Tbx21 and Ccr5. We further verified that Treg cells in the TME highly expressed T-bet by flow cytometry. In addition, we found that this T-bet⁺ Treg cell population was NRP1⁻ pTreg cells. Therefore, our results indicated that T-bet⁺NRP1⁻ Treg cells constituted a dominant group of intratumoral Treg cells in multiple cancer models. In parallel, through single-cell transcriptomics, we identified Th1-like Treg cells as the most dominant cell cluster among tumor-infiltrating Treg cell populations.

By using DT-administered Foxp3-Cre/Tbx21-Flp/VeDTR loxP-FRT (FLP recombination target) mice, Okamoto and colleagues reported that accumulation of tumor-infiltrating T-bet⁺FOXP3⁺ cells was universal across various types of murine tumors. Inducible depletion of T-bet⁺FOXP3⁺ cells did not induce autoimmunity but still gave rise to antitumor effect to an extent nearly similar to inducible depletion of total Treg cells. Upon depletion of T-bet⁺FOXP3⁺ cells in the B16F10 melanoma model, they detected an increase in the percentages of tumor-infiltrating CD4⁺ T cells expressing CD69 and PD-1, but not in CD8⁺ T cells. Besides, they characterized T-bet⁺FOXP3⁺ Treg cells in the spleens by transcriptome and in vitro suppressive activity and showed that splenic T-bet⁺FOXP3⁺ Treg cells were more suppressive than splenic Tbet⁻FOXP3⁻CD4⁺ T cells (Okamoto et al., 2023). However, the function of T-bet in intratumoral Treg cells and how T-bet⁺FOXP3⁺ Treg cells exert their suppressive activities in the TME remain obscure. In our current study, we observed a reduction in tumor growth of B16-OVA and E.G7 in Foxp3^YFP-creTbx21^fl/fl mice as compared with those in Foxp3^YFP-cre control mice. Consistently, we detected higher frequencies of functional tumor-infiltrating CD8⁺ T cells in Foxp3^YFP-creTbx21^fl/fl mice implanted with B16-OVA or E.G7 cells as compared with those of Foxp3^YFP-cre control mice. Our findings further revealed that T-bet is essential for the suppressive functions of intratumoral Treg cells, particularly their suppression toward CD8⁺ T cells through CD39-mediated suppression mechanism.

Through characterization of tumor-bearing Foxp3^YFP-creTbx21^fl/fl mice by flow cytometry, we further demonstrated that Treg cell–specific deletion of T-bet reduced the cellularity of intratumoral pTreg cells. This observation was consistent with our single-cell transcriptomic data, which showed a drastic loss of Th1-like Treg cell cluster in the TME upon Tbx21 knockout in Treg cells. In addition, we observed an increase in IFN-γ–expressing CD4⁺FOXP3⁻ Th1 cells in the TME after Tbx21 deletion in Treg cells. This supports the idea that T-bet expression in Treg cells is crucial for suppressing Th1 cell–mediated responses (Koch et al., 2009). In our single-cell transcriptomics, we detected the formation of a Th2/Th17-like cell cluster in intratumoral Tbx21-deficient Treg cells. The DEGs and pathway analyses showed that there was a transcriptomic change resulting in a deviation from Th1-like Treg cells to Th2/Th17-like Treg cells in the TME following Tbx21 deletion in Treg cells. These findings suggested that T-bet⁺ Treg cells play a role in preventing Th2/Th17 cell responses in tumors.

Our findings are in contrast with the previous results reported by Colbeck and colleagues, where they found no difference in the incidence of methylcholanthrene (MCA)-induced tumors (or percentage of tumor-free) between Tbx21^fl/flFoxp3^-Cre mice and WT C57BL/6 mice, though they detected T-bet expression in 30–40% of intratumoral Treg cells (Colbeck et al., 2015). It is important to note that the research mentioned cannot be directly compared to our current study due to a few critical parameters (tumor model used, strain of control mice, and reporting tumor-free percentage versus reporting tumor growth curve). Whilst the reason for the discrepancy is unclear, it is possible that there was a limited role of T cells in controlling MCA-induced tumors (reviewed in Schreiber and Podack [2009]). This notion was supported by previous studies which have shown no difference in the rate and frequency of MCA-induced tumor formation between immunocompetent mice and immunodeficient mice lacking functional T cells (Noguchi et al., 1996; Qin et al., 2002; Stutman, 1974). In contrast to the MCA-induced tumor model, which is unable to track antigen-specific T cells, our study utilized mouse models with transplantable tumors, including E.G7 and B16-OVA expressing the OVA protein, which contains immunogenic T cell epitopes, making it a potent tumor antigen which aids in enhancing and tracking tumor-specific T cell responses. Given the significance of tumor-specific antigens in eliciting antitumor immune responses, investigating the ontogeny of T-bet expressing Treg cells and understanding their characteristics in a system where antigens may be well-defined during tumor development could help reconcile discrepancies. Other variables that may confound the reported disparity between studies may have resulted from reporting the tumor-free mice (or tumor incidence) percentage compared with reporting tumor growth curves. Given that the mentioned study did not provide an assessment of tumor volume, it is not possible to determine whether there was a difference in the tumor size or kinetics of growth in MCA-induced tumors between Tbx21 conditional KO mice and WT control mice.

Of note, not all the pTreg cells in the tumors were positive for IFN-γ fate-mapping. There are two possibilities that might cause this phenomenon. First, it is conceivable that the Ifng^icre system may have a labeling efficiency issue mainly due to the low efficiency of Cre recombinase. Second, the differentiation of T-bet⁺ pTreg cells could occur independent of IFN-γ expression. It is known that FOXP3 has the capacity to directly repress Ifng expression (Ono et al., 2007), which suggests that FOXP3 suppresses the Ifng expression in CD4⁺ T cells concurrently upregulating T-bet and FOXP3 during differentiation. Nevertheless, our fate-mapping findings were further validated by the adoptive transfer of in vitro differentiated Th1 cells, revealing that Th1 cells could convert into pTreg cells in the TME. We observed that the majority of T-bet⁺FOXP3⁺ CD4⁺ T cells lost IFN-γ expression, indicating that IFN-γ production was shut off in the “ex-Th1 cells” in the course of the conversion. Our findings were in line with other reports that showed that IFN-γ production was reduced when FOXP3 was co-expressed by T-bet⁺ CD4⁺ T cells in the tumoral region of patients with NSCLC (Kachler et al., 2018).

Our study further demonstrated that deficiency of Tgfbr2 in conventional CD4⁺ T cells not only impaired the generation of pTreg cells but also led to an accumulation of IFN-γ–expressing Th1 (T-bet⁺FOXP3⁻) cells in the TME. Therefore, our results suggested that Th1 cells converted into pTreg cells in response to TGF-β signaling in the TME. Our results were in keeping with the report that the frequencies of T-bet⁺FOXP3⁺ CD4⁺ T cells were reduced in tumor-bearing hCD2-ΔkTβRII mice, which carried a truncated and dominant-negative form of TGF-βRII on T cells, as compared with WT mice (Kachler et al., 2018). Our findings were also in line with those of Seed and colleagues, who have reported that pTreg cell accumulation was mediated by increased TGF-β activation in the TME as a result of αvβ8 expression on tumor cells (Seed et al., 2021). It has also been demonstrated that FOXP3 expression could be induced in TGF-β–treated Th1 cells in vitro (Kachler et al., 2018). Therefore, our findings of ours and others consistently suggested that TGF-β could be an important player in the induction of T-bet⁺FOXP3⁺ T cells in the TME. Moreover, the conversion of IFN-γ–expressing Th1 cells into immunosuppressive T-bet⁺ pTreg cells could be a mechanism underlying tumor cell immune evasion, especially in the TGF-β–rich TME.

Multiple mechanisms of Treg cell–mediated suppression have been identified (Schmidt et al., 2012), and yet the exact functional mechanisms of Treg cells in the TME remain elusive. Through single-cell transcriptomics, we found that Entpd1 showed a high expression in Th1-like Treg cells as compared with Th2/Th17-like cells and a high co-expression level with Tbx21, while it showed no co-expression with Gata3 or Rorc. Through flow cytometry, we verified that CD39 expression was profoundly reduced in Tbx21-deficient Treg cells. These results suggested that CD39-mediated metabolic disruption could be a Th1-like Treg cell–specific suppression mechanism in the TME. Through pathway analyses on the RNA sequencing data of intratumoral Treg cells in both murine tumor models and human cancers, we identified similar enriched pathways of nucleotide metabolism. In addition, we observed that intratumoral Treg cells upregulate genes involved in purine catabolism, such as genes encode ectonucleoside triphosphate diphosphohydrolase family (ENTPD1, ENTPD7), purine-nucleoside phosphorylase (PNP), and nudix hydrolase (NUDT) family. Importantly, we identified ENTPD1 as the key upregulated gene in intratumoral Treg cells found across different types of human cancers. Our results further revealed a strong correlation between the expression of T-bet and CD39 in NRP1⁻ intratumoral pTreg cells. Moreover, analyses of single-cell RNA sequencing data from cancer patients showed co-expression of TBX21 and ENTPD1 in intratumoral Treg cells of various cancers. Consistent with these findings, our flow cytometric results indicated that intratumoral Treg cells in HCC patients highly co-expressed T-bet and CD39.

Nucleotide metabolism, particularly the adenosine signaling pathway, has been well implicated in contributing to the suppressive TME (Stagg and Smyth, 2010; Vaupel and Mayer, 2016). In response to cellular stress, cell death, and cell damage, ATP is abundantly released in the extracellular space of the TME (Stagg and Smyth, 2010). Extracellular ATP induces cellular repair and anticancer immune response by acting as a proinflammatory signal through the activation of P2X and P2Y receptors expressed by various types of immune cells (Allard et al., 2016). However, the ectonucleotidases CD39 and CD73 may convert extracellular ATP into immunosuppressive adenosine and suppress antitumor immune response (Allard et al., 2016; Guo et al., 2022). Adenosine can activate type1 purinergic (P1) receptors, namely A1, A2A, A2B, and A3 receptors on immune cells. The A2A receptor (A2AR) is the main adenosine receptor subtype expressed by T lymphocytes and is associated with the regulation of CD8⁺ T cells in the TME (Cekic and Linden, 2014). Upon activation of A2AR, immunosuppressive effects are induced through cyclic adenosine monophosphate/protein kinase A (cAMP/PKA)-mediated inhibition of NF-kB, TCR, and JAK-STAT signaling pathways (Palmer and Trevethick, 2008). When cAMP is formed, it induces PKA to suppress the proliferation and function of Tconv cells (Rueda et al., 2016). Therefore, nucleotide metabolism has been increasingly recognized as an underlying mechanism of Treg cell–mediated immunosuppression in the TME (Nishikawa and Koyama, 2021). Our study did not exclude other cells than CD39⁺ Treg cells in the activation of the adenosine signaling pathway in the TME, including CD39-expressing CD8⁺ T cells. However, it is noteworthy that, unlike Treg cells, CD39-expressing CD8⁺ T cells do not co-express CD73 (Vignali et al., 2023). It is conceivable that the suppression of CD8⁺ T cells by CD39⁺ Treg cells through the adenosine signaling pathway is facilitated by the localization of Treg cells in close contact with CD8⁺ T cells in the TME (Curiel et al., 2004). Indeed, the characterization of intratumoral A2AR⁺CD8⁺ T cells in our E.G7 tumor model revealed that this CD8⁺ T cell subset displayed a terminally exhausted phenotype.

It has been reported that CD4⁺CD25⁺ T cells derived from Entpd1-null mice displayed impaired suppressive functions in vitro (Deaglio et al., 2007). A reduction in tumor growth has been observed in hepatic metastatic tumor-carrying Entpd1-null mice and melanoma-bearing Rag2^−/− mice that were co-transferred with natural killer cells and Treg cells derived from Entpd1-null mice (Stagg et al., 2011; Sun et al., 2010). However, the specific roles of CD39 expression in the intratumoral Treg cells remain unclear. By using Foxp3^YFP-creEntpd1^fl/fl mice, our in vivo and in vitro studies consistently demonstrated that CD39 expression in Treg cells mainly contributed to the suppressive functions of intratumoral Treg cells. To the best of our knowledge, we are the first to describe this mouse strain. Through characterization of tumor models using this mouse strain, our study specifically elucidated the roles of CD39 in tumor-associated Treg cells. Our study demonstrated that Treg cell–specific deletion of Entpd1 provoked antitumor immune response by triggering cytolytic activities of CD8⁺ T cells. The therapeutic effects of anti-PD-1 treatments were enhanced by Treg cell–specific deletion of Entpd1 in tumor-bearing mice. In addition, our results showed that CD39 expression in Treg cells was not required for maintaining immune homeostasis. Therefore, CD39 may serve as a potential therapeutic target for selectively targeting intratumoral T-bet⁺ pTreg cells without disrupting immune homeostasis.

Both hypoxia and oxidative stress are hallmarks of the TME (Aboelella et al., 2021; Petrova et al., 2018) and have been well-known to be an important interplay involving various signaling pathways in cancer (Saikolappan et al., 2019; Tan et al., 2016, 2018a, 2018b). Hypoxia has been known to be a key driver for the production of adenosine (Sitkovsky et al., 2014; Vaupel and Mayer, 2016). Hypoxia-inducible factor-1α (HIF1A) regulates the expression of CD39, CD73, A2A, and A2B receptors in the TME. In addition, oxygen supplementation treatment reversed this effect and reduced tumor growth of the B16 melanoma model (Hatfield et al., 2014). Moreover, oxidative stress can also upregulate expression of CD39 (Moesta et al., 2020). Therefore, it is possible that intratumoral Treg cells express CD39 as a response to hypoxia and oxidative stress in the TME. Intriguingly, it has been found that the infiltration of T-bet⁺FOXP3⁺ Treg cells in murine tumors was dependent on antioxidant protein, which confers them high resistance to oxidative stress in the TME (Okamoto et al., 2023).

There are two major limitations to the present study. Unlike murine tTreg cells, NRP1 is not a specific marker of human tTreg cells (Yano et al., 2019). Our transcriptomic analyses and flow cytometry staining in murine tumor models consistently showed that Nrp1 (NRP1) expression in intratumoral Treg cells was lower as compared with that in Treg cells derived from LN and SPL. However, RNA sequencing data in human breast cancer patients showed that NRP1 expression in intratumoral Treg cells was upregulated as compared with that in Treg cells derived from peripheral blood mononuclear cells (PBMC) (Plitas et al., 2016). Therefore, the roles of NRP1 in murine and human Treg cells could be distinct and deserve further investigation. Nevertheless, we found that in mice, Treg cells derived from tissues containing a high proportion of pTreg cells, including colonic LPL and small intestine LPL, were mostly NRP1⁻. By contrast, Treg cells derived from tissues containing high proportions of tTreg cells, such as LNs and SPLs, were mostly NRP1⁺. Given that to date, there is no “all-or-none” marker for tTreg or pTreg cells, NRP1 was the best surface marker so far we could use for distinguishing tTreg cells from pTreg cells in mouse models. Besides, there is an alternative pathway of adenosine production which was less described than the canonical CD39/CD73 adenosinergic pathway. CD73 may collaborate with nicotinamide adenine dinucleotide (NAD⁺) ectoenzymes CD38 and ecto-nucleotidepyrophosphatase CD203a to produce adenosine from extracellular NAD⁺ (Churov and Zhulai, 2021; Horenstein et al., 2013; Wo et al., 2019). The role of this non-canonical adenosinergic pathway in the TME needs to be further elucidated.

Recently, accumulating evidence suggests that CD39 may serve as a promising target for cancer immunotherapy (Battastini et al., 2021; Guo et al., 2022; Nishikawa and Koyama, 2021). Anti-CD39 and anti-CD73 antibodies, such as TTX-30, MEDI9447, and BMS-986179, are currently in clinical trials (NCT03884556, NCT03742102, and NCT02754141). These antibodies may block the activities of ectonucleotidases leading to an increase in the level of proinflammatory ATP and a reduction in the level of immunosuppressive adenosine. Subsequently, this may alter the suppressive TME and promote an antitumor immune response (Nishikawa and Koyama, 2021). Therefore, our current findings provide a mechanistic basis for the therapeutic applications of blocking CD39 activity, specifically by targeting the immunosuppressive function of intratumoral Treg cells, particularly T-bet–expressing pTreg cells, which represent the dominant Treg cell population in the TME.

C57BL/6J mice, CD45.1 congenic mice, and TCRbd^−/− mice were bred in specific pathogen–free facilities at Tsinghua University. Foxp3^YFP-cre and Foxp3^GFP mice were obtained from Alexander Rudensky at Memorial Sloan Kettering Cancer Center. Foxp3^DTR mice were generated as previously described (Kim et al., 2007) by crossing Foxp3^YFP-cre mice with Rosa26^DTR mice obtained from Yan Shi at Tsinghua University. Ifng^icreRosa26^YFP fate mapping mice were generated by crossing Ifng^icre mice with Rosa26^YFP mice (Srinivas et al., 2001) obtained from Xuyu Zhou at the University of Chinese Academy of Sciences. OT-II Ifng^YFP reporter mice were generated by crossing OT-II mice (Barnden et al., 1998) with Ifng^YFP reporter mice (also known as Great mice) (Reinhardt et al., 2009), which were both purchased from The Jackson Laboratory. CD45.1/CD45.2 (double-positive) OT-II Ifng^YFP reporter mice were then generated by crossing OT-II Ifng^YFP reporter mice with CD45.1 congenic mice. OT-II Cd4^CreTgfbr2^fl/fl mice were generated by crossing OT-II mice with Tgfbr2^fl/fl mice (Chytil et al., 2002) provided by Xiao Yang at Academy of Military Medical Sciences China and then crossing with Cd4^Cre mice purchased from TACONIC. OT-II Cd4^CreTgfbr2^fl/fl mice were then crossed with CD45.1 congenic mice to obtain CD45.1/CD45.2 (double-positive) OT-II Cd4^CreTgfbr2^fl/fl mice. Il17f^CreRosa26^tdTomato mice were generated by crossing Il17f^Cre mice with Rosa26^tdTomato (Ai9) mice (Madisen et al., 2010) provided by YanGang Sun at the Institute of Neuroscience, Chinese Academy of Sciences. Foxp3^YFP-creTbx21^fl/fl mice were generated by crossing Tbx21^fl/fl mice (provided by Zhongyun Dong at Tsinghua University) with Foxp3^YFP-cre mice, as previously described (Di Giovangiulio et al., 2019). Foxp3^YFP-creEntpd1^fl/fl mice were generated by crossing Entpd1^fl/fl mice (purchased from GemPharmatech) with Foxp3^YFP-cre mice. Animals were bred and kept in specific pathogen–free facilities at the Experimental Animal Center of Tsinghua University. All experimental studies were approved and carried out in accordance with regulations implemented by the governmental and institutional guidelines for animal welfare.

E.G7 mouse lymphoma and B16-F10-OVA mouse melanoma cell lines were purchased from ATCC and Biocytogen, respectively. Hepa1-6 mouse hepatoma cell line was a kind gift from Dr. Haiyan Liu (Soochow University, Suzhou, China). E.G7 cells were grown in RPMI 1640 medium supplemented with 10% (vol/vol) heat-inactivated fetal bovine serum (FBS), 2 mM L-glutamine, 100 U/ml penicillin, and 100 µg/ml streptomycin. B16-F10-OVA and Hepa1-6 cells were grown in Dulbecco’s Modified Eagle Medium supplemented with 10% (vol/vol) heat-inactivated FBS, 2 mM L-glutamine, 100 U/ml penicillin, and 100 µg/ml streptomycin. Cell cultures were maintained at 37°C with 5% CO₂.

1.5 × 10⁶ E.G7, 0.7 × 10⁶ B16-F10-OVA, or 7 × 10⁶ Hepa1-6 tumor cells were resuspended in 200 µl PBS. The suspension of tumor cells was inoculated subcutaneously into the left flanks of the experimental mice 6–10 wk old. Tumor volumes were measured using a caliper every 2 or 3 days and then calculated by using the formula below: tumor volume = 0.5 × length × width². The tumor-bearing mice were sacrificed on days 18–23 after tumor inoculation or when the tumor diameters reached 20 mm. Distal LNs, TDLNs, SPLs, and tumors were isolated and processed for phenotypic analyses.

E.G7, B16-F10-OVA, and Hepa1-6 tumors were digested with 1 mg/ml collagenase D supplemented with 10 U/ml DNase I for 40 min at 37°C. The cell suspension was centrifuged with a discontinuous Percoll gradient (GE Healthcare). Lymphocyte fractions were isolated from the interface for further analysis.

For isolation of IEL, the small intestine or colon tissues were digested with RPMI 1640 medium containing 1 mM dithiothreitol, supplemented with 100 U/ml penicillin, 100 µg/ml streptomycin, 5 mM EDTA, and 20 mM HEPES at 37°C for 30 min. The cell suspension was centrifuged with a discontinuous Percoll gradient (GE Healthcare). Lymphocyte fractions were isolated from the interface for further analysis. For isolation of LPL, the small pieces of small intestine or colon tissues were subsequently digested with RPMI 1640 medium containing 0.5 mg/ml Collagenase D, 1 mg/ml Dispase, 40 µg/ml DNase I, supplemented with 100 U/ml penicillin, 100 µg/ml streptomycin, and 20 mM HEPES at 37°C for 30 min. The cell suspension was centrifuged with a discontinuous Percoll gradient (GE Healthcare). Lymphocyte fractions were isolated from the interface for further analysis.

CD4⁺ T cells were isolated from the SPL and LNs of CD45.1/CD45.2 (double-positive) Foxp3^GFP reporter mice by using Mouse CD4 (L3T4) MicroBeads (Miltenyi Biotec) according to the manufacturer’s protocol. Treg cells (CD4⁺GFP⁺) were sorted from the isolated CD4⁺ T cells by using BD FACSAria Cell Sorter. CD90.2⁺ cells were isolated from the SPL and LNs of CD45.2 Foxp3^GFP reporter mice by using Dynabeads FlowComp Mouse Pan T (CD90.2) Kit (Invitrogen) according to the manufacturer’s protocol. Conventional CD4⁺ T cells and CD8⁺ T cells were then sorted from the isolated CD90.2⁺ cells by using BD FACSAria Cell Sorter. The sorted cells were mixed at the following 1:10:8 ratios of Treg cells (0.1 × 10⁶):conventional CD4⁺ T cells (1.0 × 10⁶):CD8⁺ T cells (0.8 × 10⁶) and then intravenously co-transferred into TCRbd^−/− mice. After 3 days, E.G7 tumor cells were inoculated into the left flanks of the TCRbd^−/− mice. The mice were sacrificed on day 21 after tumor inoculation. LNs, SPLs, and tumors were processed for flow cytometry.

CD4⁺ T cells were isolated from the SPL and LNs of CD45.1/CD45.2 (double-positive) Foxp3^GFP reporter mice by using Mouse CD4 (L3T4) MicroBeads (Miltenyi Biotec) according to the manufacturer’s protocol. Treg cells (CD4⁺ GFP⁺) were sorted from the isolated CD4⁺ T cells by using BD FACSAria Cell Sorter. CD90.2⁺ cells were isolated from the SPL and LNs of CD45.2 Foxp3^DTR mice by using Dynabeads FlowComp Mouse Pan T (CD90.2) Kit (Invitrogen) according to the manufacturer’s protocol. Conventional CD4⁺ T cells and CD8⁺ T cells were then sorted from the isolated CD90.2⁺ cells by using BD FACSAria Cell Sorter. The sorted cells were mixed at the following 1:10:8 ratios of Treg cells (0.1 × 10⁶):conventional CD4⁺ T cells (1.0 × 10⁶):CD8⁺ T cells (0.8 × 10⁶) and intravenously co-transferred into TCRbd^−/− mice. After 3 days, E.G7 tumor cells were inoculated into the left flanks of the TCRbd^−/− mice. On days 9, 11, 13, 15, and 17 after tumor inoculation, the tumor-bearing TCRbd^−/− mice were either intraperitoneally injected with DT (Calbiochem) (30 µg/kg) diluted in 200 µl PBS as previously described (Mayer et al., 2014) (designated DT-treated group) or intraperitoneally injected with vehicle control PBS (designated control group). The mice were sacrificed on day 19 after tumor inoculation. LNs, SPLs, and tumors were processed for flow cytometry. Different congenic markers of CD45 carried by the donor cells were utilized for distinguishing tTreg cells (CD45.1⁺) and pTreg cells (CD45.1⁻).

To evaluate the therapeutic potentials of selective depletion of pTreg cells, DT was administered at the later phase of tumor growth. The tumor-bearing TCRbd^−/− mice were either intraperitoneally injected with DT (Calbiochem) (30 µg/kg) diluted in 200 µl PBS as previously described (Mayer et al., 2014) (designated DT-treated group) or intraperitoneally injected with vehicle control PBS (designated control group) on days 15, 17, 19, and 21 after tumor inoculation. The mice were sacrificed on day 22 after tumor inoculation. LNs, SPLs, and tumors were processed for flow cytometry.

1.5 × 10⁶ E.G7 tumor cells were subcutaneously injected into CD45.1 Foxp3^YFP-cre mice. 7 × 10⁶ Hepa1-6 tumor cells were subcutaneously injected into CD45.2 Foxp3^YFP-cre mice. The E.G7 and Hepa1-6 tumor-bearing Foxp3^YFP-cre mice were sacrificed on day 21 and day 23, respectively. Tumors were subjected to TIL isolation as described above. CD4⁺ T cells were isolated by using Mouse CD4 (L3T4) MicroBeads (Miltenyi Biotec) according to the manufacturer’s protocol. Intratumoral Treg cells (CD4⁺ YFP⁺) were sorted from the isolated CD4⁺ T cells by using BD FACSAria Cell Sorter. Treg cells sorted from LNs of tumor-free mice were included as controls. Total RNA was extracted with TRIzol reagent (Life Technologies) according to the manufacturer’s instructions. RNA integrity was determined by Agilent 2100 Bioanalyzer. RNA sequencing library was constructed by BGI Genomics. BGISEQ-500 was used to obtain sequence reads. The clean reads were mapped to the Mus muculus genome. Gene expression was represented by Reads Per Kilobases per Million Reads. DEGs were identified by false discovery rate–adjusted P value <0.05. GSEA was performed by GSEA software.

Genes of nucleotide metabolism were retrieved from the GeneCards gene database. RNA sequencing data of intratumoral Treg cells in human breast cancer, HCC, CRC, and NSCLC were retrieved from previous reports (De Simone et al., 2016; Plitas et al., 2016; Zheng et al., 2017). Gene Ontology (GO) and pathway enrichment analysis were performed on the publicly available RNA sequencing results of intratumoral Treg cells in human breast cancer patients (Plitas et al., 2016). The Venn diagram of the DEG was constructed by using a web tool of Bioinformatics & Evolutionary Genomics (http://bioinformatics.psb.ugent.be/webtools/Venn/). Publicly available single-cell RNA sequencing data for T cells in human cancers were analyzed and visualized by using the ScRNA-seq Data Portal for T cells in Pan-Cancer (http://cancer-pku.cn:3838/PanC_T/).

CD4⁺ T cells were isolated from the spleen and lymph nodes of CD45.1 Foxp3^YFP-cre mice by using Mouse CD4 (L3T4) MicroBeads (Miltenyi Biotec) according to the manufacturer’s protocol. Treg cells (CD4⁺YFP⁺) were sorted from the isolated CD4⁺ T cells by using BD FACSAria Cell Sorter. CD90.2⁺ cells were isolated from the SPL and LNs of CD45.2 Foxp3^YFP-cre mice by using Dynabeads FlowComp Mouse Pan T (CD90.2) Kit (Invitrogen) according to the manufacturer’s protocol. Conventional CD4⁺ T cells and CD8⁺ T cells were then sorted from the isolated CD90.2⁺ cells by using BD FACSAria Cell Sorter. The sorted cells were mixed at the following 1:10:8 ratios of Treg cells (0.1 × 10⁶):conventional CD4⁺ T cells (1.0 × 10⁶):CD8⁺ T cells (0.8 × 10⁶) and then intravenously co-transferred into TCRbd^−/− mice. After 3 days, Hepa1-6 tumor cells were inoculated into the left flanks of the TCRbd^−/− mice. The mice were sacrificed on day 18 after tumor inoculation. Tumors were subjected to TIL isolation as described above. CD4⁺ T cells were isolated by using Mouse CD4 (L3T4) MicroBeads (Miltenyi Biotec) according to the manufacturer’s protocol. Intratumoral Treg cells (CD4⁺YFP⁺) were sorted from the isolated CD4⁺ T cells by using BD FACSAria Cell Sorter. Treg cells sorted from LNs of tumor-free Foxp3^YFP-cre mice were included as controls. Total RNA was extracted with TRIzol reagent (Life Technologies) according to the manufacturer’s instructions. RNA integrity was determined by Agilent 2100 Bioanalyzer. RNA sequencing library was constructed by BGI Genomics. BGISEQ-500 was used to obtain sequence reads. The clean paired-end reads were aligned to the Mus musculus reference genome (GRCm38) using STAR v2.7.11a (Dobin et al., 2013) via RSEM v1.3.3 (Li and Dewey, 2011) with default parameters. The reads count matrix and fragments per kilobase of exon model per million mapped fragments matrix were quantified using RSEM, with Gencode vM25 (Frankish et al., 2021) for gene annotation. Differential expression analysis was performed by DESeq2 (Love et al., 2014), and ClusterProfiler (Yu et al., 2012) was used for GSEA. DEGs in Hepa1-6 intratumoral pTreg cells versus Hepa1-6 intratumoral tTreg cells were identified by a fold-change >2 and adjusted P value <0.05. DEGs in Hepa1-6 intratumoral pTreg or Hepa1-6 intratumoral tTreg cells versus LN-derived Treg cells were identified by a fold-change >4 and adjusted P value <0.01.

7 × 10⁶ Hepa1-6 tumor cells were subcutaneously injected into Foxp3^YFP-cre mice and Foxp3^YFP-creTbx21^fl/fl mice. The Hepa1-6 tumor-bearing mice were sacrificed on day 23. Tumors were subjected to TIL isolation as described above. CD4⁺ T cells were isolated by using Mouse CD4 (L3T4) MicroBeads (Miltenyi Biotec) according to the manufacturer’s protocol. Intratumoral Treg cells (CD4⁺ YFP⁺) were sorted from the isolated CD4⁺ T cells by using BD FACSAria Cell Sorter. Conventional CD4⁺ T cells (CD4⁺YFP⁻) and Treg cells (CD4⁺YFP⁺) sorted from Hepa1-6 tumors and tumor-free LN of Foxp3^YFP-cre mice, respectively, were included for comparison. Cells were pooled from multiple mice per sample.

BD Rhapsody 8-lanes cartridges (664887; BD Biosciences) were loaded with 80,000 cells to capture 50,000–60,000 cells per sample. Single cells were isolated with the BD Rhapsody HT Xpress Single-Cell Analysis System using the BD RhapsodyTM Enhanced Cartridge Reagent Kit (633773; BD Biosciences) and BD Rhapsody cDNA Kit (633733; BD Biosciences) according to the manufacturer’s recommendations (BD Biosciences). cDNA libraries were prepared using the Whole Transcriptome Analysis Amplification Kit (633801; BD Biosciences) following the BD Rhapsody System mRNA Whole Transcriptome Analysis Preparation Protocol (BD Biosciences). All the libraries were sequenced in a paired-end mode (PE150) on DNBseq-T7 (GenePlus).

Single-cell sequencing data preprocessing was performed using the BD Rhapsody Sequence Analysis Pipeline (version 2.1; BD Biosciences Inc.) with GRCm39 as the mouse reference genome. Then the filtered Cell-by-Feature data tables of the four samples were conducted in Seurat (version 5.0.2) R package and subjected to the following quality control (QC): First, selecting the cells with detected genes no less than 200 and genes expressed in at least three cells. Then, low-quality cells with <500 unique molecular identifiers (UMIs) and 300 expressed genes or >7,500 expressed genes were filtered out, as well as >20% mitochondrial gene expression. In addition, the cells with a ratio of log₁₀ (nUMI) and log₁₀ (nGene) <0.8 and the potential doublets which were calculated by DoubletFinder (version 2.0.4) with 7.5% doublet formation rate were also removed.

After QC, the filtered count data were log normalized and scaled. Then, about 2,000 genes with the highest variance were selected. The integration pipeline was applied by Seurat “harmony” function (version 1.2.0), and the first 30 principal components (PCs) were used for nonlinear dimensional reduction analysis. The four samples were clustered by setting up resolution to 0.2 and visualized by Uniform Manifold Approximation and Projection (UMAP).

DEGs between the two groups were obtained from Seurat “FindMarkers” function. The marker genes of each cluster were calculated by “FindallMarkers” function with the default setting.

GO analyses were performed by “enrichGO” function of clusterProfiler (version 4.10.1) R package. Heatmap of log normalized expression data was generated by ComplexHeatmap (Version 2.18.0) R package. Volcano plots were generated by ggplot2 (version 3.5.0) and ggrepel (version 0.9.5) R package.

The loom file was obtained by running velocyto (version 0.17.17) package. Then, RNA velocity was performed with a scvelo (version 0.2.4) python toolkit by setting model to “stochastic.”

On day 0, C57BL/6J mice were inoculated with 1.5 × 10⁶ E.G7 tumor cells. On day 3, CD4⁺ T cells were isolated from the SPL and LNs of CD45.1/CD45.2 (double-positive) OT-II Ifng^YFP reporter mice by using Mouse CD4 (L3T4) MicroBeads (Miltenyi Biotec) according to the manufacturer’s protocol. Naive CD4⁺ T cells (CD4⁺CD25⁻ CD44⁻CD62L^high) were sorted from the isolated CD4⁺ T cells by using BD FACSAria Cell Sorter. The sorted naive CD4⁺ T cells were seeded in a 48-well plate precoated with 2 µg/ml anti-CD3 and 2 µg/ml anti-CD28 (BioXcell). The cells were cultured in RPMI 1640 medium supplemented with 10% (vol/vol) heat-inactivated FBS, 2 mM L-glutamine, 100 U/ml penicillin, 100 µg/ml streptomycin, and 0.1% β-mercaptoethanol, and differentiated into Th1 cells in the presence of IL-12 (15 ng/ml), IL-2 (25 U/ml) and anti-IL-4 (10 µg/ml). On day 7 (the fourth day of culture), CD4⁺YFP⁺ cells (IFN-γ–producing cells) were sorted and transferred intravenously into E.G7-bearing B6 mice (0.8 × 10⁶ cells per mouse) which were on day 7 after tumor inoculation. The tumor-bearing mice were sacrificed on day 21. TILs were isolated for phenotypic analyses by flow cytometry.

On day 0, CD90.2⁺ cells were isolated from the SPL and LNs of CD45.2 Foxp3^GFP reporter mice by using Dynabeads FlowComp Mouse Pan T (CD90.2) Kit (Invitrogen) according to the manufacturer’s protocol. Treg cells (CD4⁺GFP⁺) and CD8⁺ T cells were sorted from the isolated CD90.2⁺ cells by using BD FACSAria Cell Sorter. CD4⁺ T cells were isolated from the SPL and LNs of CD45.1/CD45.2 (double-positive) OT-II Cd4^CreTgfbr2^fl/fl mice (designated KO group) or CD45.1/CD45.2 (double-positive) OT-II Tgfbr2^fl/fl mice (designated WT group) by using Mouse CD4 (L3T4) MicroBeads (Miltenyi Biotec) according to the manufacturer’s protocol. Naive CD4⁺ T cells (CD4⁺CD25⁻CD44⁻CD62L^high) were sorted from the isolated CD4⁺ T cells by using BD FACSAria Cell Sorter. The sorted cells were mixed at the following 1:10:8 ratios of Treg cells (0.1 × 10⁶):naive conventional CD4⁺ T cells (1.0 × 10⁶):CD8⁺ T cells (0.8 × 10⁶) and intravenously co-transferred into TCRbd^−/− mice. On day 3, B16-OVA tumor cells were inoculated into the left flanks of the TCRbd^−/− mice. The mice were sacrificed on day 16 for TIL phenotypic analyses by flow cytometry.

CD4⁺ T cells were isolated from the SPL and LNs of CD45.1 Foxp3^YFP-cre mice by using Mouse CD4 (L3T4) MicroBeads (Miltenyi Biotec) according to the manufacturer’s protocol. Treg cells (CD4⁺YFP⁺) were sorted from the isolated CD4⁺ T cells by using BD FACSAria Cell Sorter. CD90.2⁺ cells were isolated from the SPL and LNs of CD45.1/CD45.2 (double-positive) Foxp3^YFP-creTbx21^fl/fl mice (designated KO group) or CD45.1/CD45.2 (double-positive) Foxp3^YFP-cre mice (designated WT group) by using Dynabeads FlowComp Mouse Pan T (CD90.2) Kit (Invitrogen) according to the manufacturer’s protocol. Conventional CD4⁺ T cells and CD8⁺ T cells were then sorted from the isolated CD90.2⁺ cells by using BD FACSAria Cell Sorter. The sorted cells were mixed at the following 1:10:8 ratios of Treg cells (0.1 × 10⁶):conventional CD4⁺ T cells (1.0 × 10⁶):CD8⁺ T cells (0.8 × 10⁶) and intravenously co-transferred into TCRbd^−/− mice. After 3 days, E.G7 tumor cells were inoculated into the left flanks of the TCRbd^−/− mice. The mice were sacrificed on day 21 after tumor inoculation for TIL phenotypic analyses by flow cytometry.

Foxp3^YFP-creEntpd1^fl/fl mice (designated KO group) and Foxp3^YFP-cre mice (designated WT group) were inoculated with 7 × 10⁶ Hepa1-6 tumor cells. The tumor-bearing mice were sacrificed on day 18 for TIL isolation, followed by CD4⁺ T cells isolation by using Mouse CD4 (L3T4) MicroBeads (Miltenyi Biotec) according to the manufacturer’s protocol. NRP1⁻ Treg cells (NRP1⁻CD4⁺YFP⁺) were sorted from the isolated intratumoral CD4⁺ T cells by using BD FACSAria Cell Sorter. CD90.2⁺ cells were isolated from the SPL and LNs of C57BL/6J mice by using Dynabeads FlowComp Mouse Pan T (CD90.2) Kit (Invitrogen) according to the manufacturer’s protocol. Naive CD8⁺ T cells (CD8⁺CD25⁻CD44⁻CD62L^high) were sorted from the isolated CD90.2⁺ cells by using BD FACSAria Cell Sorter. The sorted naive CD8⁺ T cells were stained with 5 µM CellTrace Violet (CTV) (Invitrogen) for 20 min at room temperature. The CTV-labeled naive CD8⁺ T cells were seeded in 1 × 10⁴ cells per well in a 96-well U-bottom plate precoated with 2 µg/ml anti-CD3 and 2 µg/ml anti-CD28 (BioXcell). The CTV-labeled naive CD8⁺ T cells were cultured either in the presence of KO intratumoral NRP1⁻ Treg cells, or WT intratumoral NRP1⁻ Treg cells at 1:1 ratio of CD8⁺ T cells:Treg cells, or in the absence of Treg cells. The cells were cultured in RPMI 1640 medium supplemented with 10% (vol/vol) heat-inactivated FBS, 2 mM L-glutamine, 100 U/ml penicillin, 100 µg/ml streptomycin, and 0.1% β-mercaptoethanol. Unstained and unstimulated CD8⁺ T cells were included as controls. After 72 h of co-culture, the cells were harvested and processed for phenotypic analyses. DI of CD8⁺ T cells were evaluated by FlowJo software. Percentages of suppression were determined by using the following formula: $Suppression %=100−DITreg/DIwithout Treg × 100$ as previously described (McMurchy and Levings, 2012).

0.7 × 10⁶ B16-F10-OVA tumor cells were subcutaneously inoculated into the left flanks of Foxp3^YFP-cre (WT) mice and Foxp3^YFP-creTbx21^fl/fl (KO) mice or Foxp3^YFP-creEntpd1^fl/fl (KO) mice. 75 µg anti-PD-1 (clone J43; Bioxcell) antibody was intraperitoneally injected every 3 days from day 9 onwards. Tumor growth was monitored every 2–3 days. Survival probability of each group of mice was evaluated.

The study was conducted in accordance with the guidelines of ethical regulation for human experimentation (The Institutional Review Board of Tsinghua University). All the experimental protocols were approved by the Medical Ethics Committee of Tsinghua University. Tumor tissues, paratumor tissues, and matched blood samples of HCC patients were collected from China–Japan Friendship Hospital, Beijing, China. The collected specimens were processed and subjected to Ficoll–Hypaque (GE Healthcare) gradient centrifugation. Lymphocyte fractions were isolated from the interface and stained with antibodies for flow cytometry analysis. The samples were analyzed following safety regulations.

Tissues were fixed in 4% paraformadehyde for 24 h and then embedded in paraffin wax. Paraffin-embedded tissue sections were cut at a thickness of 5 μm by using a Reica RM2235 Microtome and followed by H&E staining according to standard protocols. Images were taken by using a Nikon Eclipse 90i motorized upright microscope.

Single-cell suspensions of LN, SPL, TDLN, or TILs were subjected to antibody staining for surface markers, cytokines, intranuclear proteins, and transcription factors. For intracellular cytokine staining, the cell suspension was stimulated with Phorbol 12-myristate 13-acetate (50 ng/ml; Sigma-Aldrich) and ionomycin (500 ng/ml; Sigma-Aldrich) in the presence of Brefeldin A (Golgiplug; BD Bioscience) in 96-well plates for 4 h in 5% CO₂ at 37°C. For surface marker staining, the cell suspension was incubated with Fc receptor-blocking anti-CD16/32 (Biolegend) for 15 min, then stained with Fixable Viability Dye eF506 (eBioscience) and surface markers for 40 min. After surface marker staining, the cells were fixed and permeabilized by using FOXP3 Transcription Factor Staining Buffer Set (eBioscience) according to the manufacturer’s instructions. For intracellular staining by fluorochrome-conjugated anti-mouse antibody, the staining was performed for 1 h. Anti-tdTomato staining was performed by using unconjugated primary anti-RFP (Rockland) antibody, followed by staining with secondary antibody donkey anti-rabbit IgG (Clone Poly4064; Biolegend) for 30 min. All the staining steps were carried out at 4°C. After staining, the cells were acquired by a flow cytometer (LSRFortessa; BD) and analyzed by using FlowJo software. The following fluorochrome-conjugated anti-mouse antibodies were used for staining: anti-CD45 (Clone 30-f11; eBioscience), anti-CD45.2 (Clone 104; Biolegend), CD45.1 (Clone A20; Invitrogen), anti-TCRβ (Clone H57-594; BD Biosciences), anti-CD3 (Clone 145-2C11; BD Biosciences), anti-CD4 (Clone RM4-5; Biosciences), anti-CD8 (Clone 53–6.7; eBioscience), anti-FOXP3 (Clone FJK-16s; eBioscience), anti-GFP (Invitrogen), anti-T-bet (Clone 4B10; Biolegend), anti-CD25 (Clone PC61.5; eBioscience), anti-NRP1 (Clone 3DS304M; eBioscience), anti-ICOS (Clone 7E.17G9; eBioscience), anti-CD39 (Clone 24DMS1; eBioscience), anti-CD73 (Clone eBioTY/11.8; eBioscience), anti-PD-1 (Clone J43; eBioscience), anti-GITR (Clone DTA-1; BD Bioscience), anti-CTLA-4 (Clone UC10-4B9; eBioscience), anti-TIM-3 (Clone RMT3-23; Biolegend), anti-Ly108 (Clone 13G3-19D; eBioscience), anti-TIGIT (Clone GIGD7; eBioscience), anti-OX40 (Clone OX86; Biolegend), anti-TCF1 (Clone C63D9; CST), anti-CD44 (Clone IM7; BD Biosciences), anti-CD62L (Clone MEL-14; BD Biosciences) and anti-Ki67 (BD Biosciences), anti-IFN-γ (Clone XMG1.2; BD Biosciences), anti-TNF (Clone MP6-XT22; BD Biosciences), anti-Granzyme B (Clone QA16A02; Biolegend), anti-CD107a (Clone 1D4B; Biolegend), anti-CD11b (Clone M1/70; eBioscience), anti-Ly6C (Clone HK1.4; eBioscience), anti-Ly6G (Clone 1A8; BD Biosciences), anti-Siglec-F (Clone 1RNM44N; eBioscience), anti-CD19 (Clone 1D3; BD Biosciences), anti-CD11c (Clone HL-3; BD Biosciences), anti-NK1.1 (Clone PK136; eBiosciences); anti-OVA Tetramer-SIINFEKL (MBL); anti-CX3CR1 (Clone SA011F11; Biolegend); anti-A2AR (Clone 7F6-G5-A2; Novus Biologicals); anti-RORγt (Clone Q31-378; BD Biosciences); anti-GATA3 (Clone L50-823; BD Biosciences); anti-BCL6 (Clone K112-91; BD Biosciences); anti-IL-4 (Clone 11B11; BD Biosciences); anti-IL-13 (Clone eBio13A; eBioscience); anti-IL-17A (Clone eBio17B7; eBioscience); and anti-IRF4 (Clone 3E4; eBioscience). The following fluorochrome-conjugated anti-human antibodies were used for staining: anti-CD3 (Clone OKT3; Biolegend); anti-CD4 (Clone L200; BD Biosciences); anti-FOXP3 (Clone 236A/E7; BD Biosciences); anti-CD39 (Clone A1; Biolegend); and anti-T-bet (Clone 4B10; eBioscience). All reagents and resources used in the study are listed in Table S1.

All experiments were repeated at least once independently showing similar results. The Student’s t test or Mann–Whitney U test was used to evaluate the significance differences unless otherwise specified. Two-way analysis of variance (ANOVA) was used to determine the significance of differences in tumor volumes. The Kaplan–Meier method was used to evaluate the survival probability. The number of replicates and statistical tests used in the study were specified in the figure legends. All statistical tests are two-sided. Differences were considered statistically significant at P value <0.05. *P < 0.05, **P < 0.01, ***P < 0.001; NS, not significant.

All animal studies were approved by the Institutional Animal Care and Use Committee of Tsinghua University. All the experimental protocols for human experimentation were approved by the Medical Ethics Committee of Tsinghua University.

Fig. S1 illustrates the therapeutic effects of pTreg cell depletion in tumor-bearing mice, including tumor growth data and flow cytometry analysis of TILs following pTreg cell depletion. Fig. S2 depicts the role of T-bet in pTreg cell function within tumor immunity, showing tumor growth, Treg cell frequencies, and changes in CD8⁺ T cell responses as well as Th1 cell populations in Tbx21 conditional knockout mice. Fig. S3 explores the conversion of Th1 cells into pTreg cells in tumors in response to TGF-β signaling, presenting tumor growth data and flow cytometry analysis. Fig. S4 demonstrates the role of T-bet in supporting Treg cell suppressive functions and preventing Th2/Th17 responses within the TME, with gene expression data and flow cytometry results. Fig. S5 shows selective CD39 expression on pTreg cells across various tissues, as well as A2AR expression on tumor-infiltrating CD8⁺ T cells in relation to exhaustion markers. Table S1 lists the reagents and resources used in the study.

The bulk RNA sequencing data underlying Fig. 3, A and B; and Fig. 8, A–G are openly available in the GEO database at GSE240961. The bulk RNA sequencing data underlying Fig. 4, A–K; and Fig. 8 H are openly available in the GEO database at GSE283943. The single-cell RNA sequencing data underlying Fig. 7 and Fig. S4 are openly available in the GEO database at GSE285225. All other relevant data are available from the corresponding author upon reasonable request. Correspondence and requests for materials should be addressed to C. Dong ([email protected]).

We would like to acknowledge Prof. Alexander Rudensky (Memorial Sloan Kettering Cancer Center, New York, NY, USA) for providing Foxp3^YFP-cre mice, Prof. Yan Shi (Tsinghua University, Beijing, China) for providing Rosa26^DTR mice, Prof. Xuyu Zhou (University of Chinese Academy of Sciences, Beijing, China) for providing Rosa26^YFP reporter mice, Prof. Xiao Yang (Academy of Military Medical Sciences China, Beijing, China) for providing Tgfbr2^fl/fl mice, Dr. YanGang Sun (Institute of Neuroscience, Chinese Academy of Sciences, Shanghai, China) for providing Rosa26^tdTomato (Ai9) mice, and Dr. Zhongjun Dong (Tsinghua University, Beijing, China) for providing Tbx21^fl/fl mice. We express our gratitude to Dr. Haiyan Liu (Soochow University, Suzhou, China) for providing Hepa1-6 mouse hepatoma cell line. We gratefully acknowledge Tsinghua University Institute for Immunology and School of Medicine for infrastructure support. Special thanks go to Dr. Longyan Wu (Core Facility of Institute for Immunology Tsinghua University, Beijing, China) for conducting single-cell capture and library preparation; and Xiuyu Zhou (Core Facility of Institute for Immunology Tsinghua University, Beijing, China) and Rui Sun (Experimental Animal Center of Tsinghua University, Beijing, China) for technical support. We are deeply grateful to our lab members (Tsinghua University, Beijing, China), special thanks go to Dr. Xiaohu Wang, Weiwei Fu, Xiao Ding, and Jing Li for their intellectual communication; and Yu Feng, Lin Sun, Ling Jin, Qinli Sun, Jian Gong, Dongli Cai, Yuling Li, and Qiuyan Lan for technical support and advice.

S.-N. Tan was supported by the China Postdoctoral Science Foundation (Grant number: 2018M641388) and a postdoctoral fellowship from Center for Life Sciences at Tsinghua-Peking University. This study was also supported in part by grants from the Natural Science Foundation of China (31991173 and 31991170 to C. Dong), Innovative research team of high-level local universities in Shanghai (SHSMU-ZLCX20211600 to C. Dong), Zhejiang Provincial Natural Science Foundation of China (LD25C120002 to C. Dong), National Key Research and Development Program of China (2021YFC2302403 to L. Ni), and Tsinghua University-Xiamen Chang Gung Hospital Joint Research Center for Anaphylactic Disease.

Author contributions: S.-N. Tan: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Validation, Visualization, Writing - original draft, J. Hao: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Resources, Software, Validation, Visualization, Writing - review & editing, J. Ge: Formal analysis, Methodology, Software, Validation, Visualization, Y. Yang: Investigation, Methodology, L. Liu: Resources, J. Huang: Resources, M. Lin: Formal analysis, Methodology, Software, Visualization, X. Zhao: Data curation, Software, Visualization, G. Wang: Formal analysis, Methodology, Software, Visualization, Z. Yang: Investigation, Methodology, Resources, L. Ni: Project administration, Writing - review & editing, C. Dong: Conceptualization, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Writing - original draft, Writing - review & editing.