TGFβ1-Mediated SMAD3 Enhances PD-1 Expression on Antigen-Specific T Cells in Cancer

Programmed death-1 (PD-1; encoded by PDCD1) is a coinhibitory receptor induced on T cells by antigenic stimulation (1). PD-1 expression on functional memory CD8⁺ T cells declines upon the resolution of inflammation and the clearance of antigen during acute infections (2). Conversely, PD-1 expression is maintained on exhausted T cells in chronic infections. In cancer, the PD-1 pathway is highly engaged within the tumor microenvironment (TME), with tumor and immune system cells expressing high levels of the PD-1 ligands PD-L1 (also known as B7-H1) and PD-L2 (also known as B7-DC), and tumor-infiltrating CD4⁺ and CD8⁺ T cells expressing high levels of PD-1 (3, 4). Blockade of PD-1 has been effective in prolonging patient survival in melanoma, renal cell carcinoma, non–small cell lung cancers, Hodgkin lymphoma, and many other cancer types (5–8). Similarly, chronic infection with the hepatitis C virus (HCV), hepatitis B virus (HBV), or human immunodeficiency virus (HIV) sustains high levels of PD-1 on viral-specific CD8⁺ T cells (9–11).

Binding of PD-1 on T cells to its ligands, PD-L1 and PD-L2, can inhibit T-cell effector function (12). Pathogen- or tumor-driven inflammation can induce PD-L1 and PD-L2 expression. For example, PD-L1 is highly expressed on many human tumors (4, 13), and its expression is highly colocalized with infiltrating CD8⁺ T cells in patients with melanoma (14). Similarly, patients with chronic liver disease from HCV and HBV infection also show increased levels of PD-L1 on hepatocytes and Kupffer cells in the liver (15). Elevated PD-L1 and PD-L2 expression may enhance engagement of PD-1 on T cells and pathogen evasion of host immune responses (4, 16–19). The levels of PD-1 on tumor-infiltrating lymphocyte (TIL) subsets in many cancers are much higher than those seen on normally activated or memory T cells in peripheral blood or in corresponding normal tissue (20). This induction of receptor, together with ligand upregulation, is likely responsible for the profound inhibition of effector antitumor T-cell activity in the TME. Although IFNγ, a T-cell effector cytokine, is known to enhance PD-L1 expression on tumor cells (13), and some cytokines have been shown previously to affect T-cell expression of PD-1 (21, 22), the molecular mechanisms that permit expression of PD-1 on human T cells at very high levels have not been fully elucidated. This is critical to our understanding of PD-1 inhibition of T-cell control of tumors or chronic viral infections and modulation of that pathway through immunotherapy.

As part of a cytokine screen to identify those that regulate PD-1 induction on T cells, we found that TGFβ1 modified antigen-driven PD-1 induction to the greatest extent. TGFβ1 is a regulatory cytokine that suppresses immune function in cancers and in chronic viral infections (23–26). The SMAD transcription factors transduce signals from TGFβ superfamily ligands that regulate cell proliferation, differentiation, and death through the activation of receptor serine/threonine kinases. High serum levels of TGFβ are associated with poor prognosis in cancer (27, 28), and TME-derived TGFβ can suppress antitumor T-cell responses (29, 30). Accordingly, the blockade of TGFβ1 signaling on T cells has been effective in restoring T-cell effector functions (31, 32). The known suppressive mechanisms of TGFβ1 include SMAD2/3-dependent inhibition of effector cytokine production by CD8⁺ T cells in cancer (33) and development of CD4⁺ regulatory T cells (Treg) that suppress neighboring effector cells through both contact-independent and contact-dependent mechanisms (34, 35).

Here, we report a novel molecular mechanism of immunosuppression in which TGFβ enhances antigen-driven PD-1 gene transcription selectively through SMAD3, resulting in enhanced surface expression of PD-1 protein. Utilizing mice with T cells conditionally deleted of Smad2 or Smad3, we found that TGFβ1-enhanced PD-1 expression is abrogated in Smad3-deficent T cells. In contrast, Smad2-deficient T cells expressed PD-1 at levels comparable with wild-type (WT) mice. This suggests that enhanced PD-1 expression on T cells is predominantly regulated by SMAD3. The effect of SMAD3 was specific to PD-1, as the expression of other inhibitory receptors was not decreased by Smad3 deficiency. Mice with Smad3-deficient T cells more effectively controlled tumors in association with loss of the subset of antigen-specific TILs displaying the highest levels of PD-1 and increased TIL and draining lymph node (DLN) cytokine production. PD-1 blockade did not provide further antitumor activity beyond that produced by T cell–specific Smad3 knockout, demonstrating that PD-1 induction by the TGFβ1/SMAD3 axis is critical in suppressing antitumor T-cell function. Thus, our findings suggest that TME-derived TGFβ1 augments PD-1 expression on TILs, suppressing CD8⁺ T cells that engage tumor antigens and enhancing tumor immune resistance.

To assess the effects of cytokines known to alter T-cell development, function, and/or proliferation on PD-1 expression, we isolated CD3⁺ T cells from healthy donor peripheral blood mononuclear cells (PBMC) and activated them with αCD3/αCD28–conjugated beads in the presence of one of 16 cytokines across a range of concentrations (Supplementary Fig. S1A, data shown at 500 ng/mL). The cells were labeled with carboxyfluorescein succinimidyl ester (CFSE) to monitor cellular proliferation, and PD-1 expression was measured (Fig. 1A, representative plots). αCD3/αCD28 induces higher PD-1 expression compared with resting CD8⁺ and CD4⁺ T cells, confirming T-cell receptor (TCR) and costimulation-dependent PD-1 expression (Fig. 1A, left and middle graphs). Although most of the cytokines tested had no effect or only a modest effect on PD-1 expression upon T-cell activation, major enhancement of PD-1 expression was observed with TGFβ1 (Fig. 1A, middle and right graphs; and Supplementary Fig. S1A). In conjunction with T-cell stimulation, IL2, IL6, IL12, and TNFα induced only modest enhancement of αCD3/αCD28–induced PD-1 expression (Supplementary Fig. S1A). Because increased TGFβ1 production is a hallmark of most TME, we chose to further explore its role in PD-1 expression. The coculture of T cells with TGFβ1 further enhanced PD-1 expression on both CD8⁺ (Fig. 1B, left; open symbols) and CD4⁺ (Fig. 1B, right; open symbols) T cells versus αCD3/αCD28 alone (Fig. 1B, closed symbols) on all generations (Fig. 1B). TGFβ1 did not have any effects on cellular proliferation as measured by CFSE dilution (Fig. 1B, black vs. open bar graph), suggesting that enhanced PD-1 expression is not simply due to altered cellular proliferation.

Although human memory T-cell populations, such as CMV- and EBV-specific T cells, express intermediate levels of PD-1, naïve T cells do not express PD-1 (36). To test whether TGFβ1-mediated enhancement of PD-1 expression depends on the basal level of PD-1 expression, we isolated naïve T cells (phenotype CCR7⁺ CD45RA⁺) and memory T cells (phenotype CCR7⁺ CD45RA⁻ or CCR7⁻ CD45RA⁺) from healthy donors. The cells were activated with αCD3/αCD28–conjugated beads with or without TGFβ1. Although TGFβ1 increased PD-1 expression on αCD3/βCD28–stimulated naïve and memory CD4⁺ and CD8⁺ T cells (Fig. 1C, representative plots), the effect was more pronounced on naïve T cells than on memory T cells for both CD4 and CD8 subsets (Fig. 1D, dark and light gray bars). In the absence of αCD3/αCD28, TGFβ1 does not affect the basal levels of PD-1 expression on either naïve or memory T-cell subsets. This suggests that TGFβ1 enhancement of PD-1 expression is dependent on T-cell activation. Furthermore, we found that TGFβ1 increased PD-1 surface expression in a concentration-dependent manner (Fig. 1E, left). In contrast, TGFβ1 did not affect the expression of the T-cell activation marker HLA-DR on T cells (Fig. 1E, right), demonstrating that the TGFβ1 effect on PD-1 expression does not simply reflect a general effect on T-cell activation–induced antigens. Changes in intracellular and surface levels of PD-1 were positively and directly correlated (Supplementary Fig. S1B). Finally, enhanced surface expression of PD-1 was preceded by increased transcription of PD-1, shown as kinetic changes of PDCD1 mRNA levels across different time points (Fig. 1F).

Next, we investigated whether blockade of TGFβ1 signaling can abrogate TGFβ-dependent PD-1 enhancement. TGFβ1 binds TGFβ receptor I (TGFβRI) and TGFβRII and acts through SMAD-dependent and SMAD-independent mechanisms (37). Upon binding of the high-affinity TGFβRII by TGFβ1, TGFβRI and TGFβRII heterodimerize and TGFβRI, a serine/threonine kinase, phosphorylates SMAD2/3. To address the role of TGFβ1 receptor signaling, the cells were activated in the presence of TGFβ1 with varying concentrations of either an antibody that blocks the activity of TGFβ1 but not TGFβ2 or TGFβ3 [neutralizing antibody (nAb); Fig. 2A] or a TGFβRI kinase inhibitor (SB431542; Fig. 2B). Both TGFβ1 nAb and SB431542 decreased TGFβ1-dependent PD-1 expression in a dose-dependent manner, although SB431542 was more effective than TGFβ1 nAb. SB431542-mediated TGFβR signaling inhibition was shown by the diminished phosphorylation levels of SMAD2 (Fig. 2C). Analogous to the effects on surface expression, SB431542 blocked the TGFβ1-dependent increase of PDCD1 mRNA levels (Fig. 2D). Given the critical role of nuclear factor of activated T cell (NFATc1) during TCR-dependent PD-1 induction (38), we tested whether TGFβ1-dependent PD-1 expression requires NFATc1 by treating cells with cyclosporine A, a calcineurin inhibitor that exerts its immunosuppressive effects by keeping the transcription factor NFATc1 inactive. We found that cyclosporine A completely abrogated not only TCR-dependent but also TGFβ1-enhanced PD-1 expression (Supplementary Fig. S1C), suggesting that TGFβ1 requires TCR-induced NFATc1 activity to enhance the PD-1 expression. On the basis of the critical role of TGFβR kinase activity, we next assessed downstream molecules in the TGFβR signaling cascade.

Our data suggest that human PD-1 expression is under direct transcriptional control by TGFβ1, so we hypothesized that TGFβ1 modulates human PDCD1 promoter activity. We identified putative SMAD-binding elements (SBE), one distal to (SBE-D) and the other proximal to (SBE-P) the PDCD1 transcription start site (Fig. 3A). To test this hypothesis, Jurkat T cells were transfected with a luciferase vector containing the 1.9 kb–long PDCD1 promoter region, and luciferase activity was measured after different treatments. αCD3/αCD28 activation induced PDCD1 promoter activity (Fig. 3B, gray bars), and mutation in the NFATc1-binding site abrogated such induction (Supplementary Fig. S1D and S1E). Because Jurkat T cells express minimal levels of the TGFβ receptors, the T cells were cotransfected with TGFβRI and TGFβRII plasmids (Supplementary Fig. S1F and S1G). The addition of TGFβ1 to αCD3/αCD28 enhanced NFATc1-dependent PDCD1 promoter activity (Fig. 3B, WT; and Supplementary Fig. S1F). The introduction of site-directed mutations in SBEs (shown in bold letters in Fig. 3A, named SBE-D and SBE-P) significantly diminished TGFβ1-driven PDCD1 promoter activity, and the introduction of both mutations (SBE-D/P) further decreased the effect (Fig. 3B). To further validate our luciferase-based reporter system, we utilized the chromatin immunoprecipitation (ChIP) assay to verify SMAD3 binding to the human PDCD1 promoter. Although αCD3/αCD28 did not induce SMAD3 binding, the addition of TGFβ1 significantly enhanced SMAD3 binding to the human PDCD1 promoter (Fig. 3C, right graph). This binding was specifically due to TGFβ1 receptor signaling, as it was abrogated by treatment with the TGFβRI kinase inhibitor SB431542 (Fig. 3C, right graph). There was no effect of TGFβ1 on binding of SMAD3 to the GAPDH promoter (Fig. 3C, left graph). We also confirmed that NFATc1 binds to the human PDCD1 promoter following αCD3/αCD28 stimulation and found that this binding was in fact enhanced by TGFβ1 (Supplementary Fig. S1H).

TGFβR1 has serine/threonine kinase activity that phosphorylates SMAD2 and SMAD3 (39). SMAD2 and SMAD3 bifurcate the signaling pathway by forming heterodimers with SMAD4 (considered a co-SMAD; refs. 40, 41). Thus, we further investigated whether SMAD2 or SMAD3 is the dominant regulator of PDCD1 promoter activity by using siRNA (Supplementary Fig. S2A and S2B). We found that knockdown of SMAD3 expression (but not SMAD2) abrogated TGFβ1 enhancement of PDCD1 promoter activity (Fig. 3D). In addition, we tested whether a specific inhibitor of SMAD3, SIS3, that inhibits phosphorylation of SMAD3 but not SMAD2, affects PDCD1 promoter activity similarly to knockdown of SMAD3 (42). SIS3 inhibited TGFβ1 enhancement of PDCD1 promoter activity (Fig. 3E, white bars) without significantly altering NFATc1-dependent PDCD1 promoter activity (Fig. 3E, gray bars). Thus, our data collectively showed that SMAD3 is a key mediator of TGFβ1-enhanced PDCD1 promoter activity and increased PDCD1 transcription levels.

To investigate the role of SMAD3 on PD-1 T-cell surface expression, we treated human CD3⁺ cells with SIS3 and found that SIS3 treatment decreased PD-1 surface expression on both human CD4⁺ (Fig. 4A, left) and CD8⁺ (Fig. 4A, right) T cells in a dose-dependent manner. Next, we investigated whether SMAD3 deficiency can abrogate TGFβ1-dependent PD-1 expression on murine T cells. CD4⁺ T cells were isolated from WT, Smad2 f/f; Cd4-cre (Smad2 cKO), and Smad3 f/f; Cd4-Cre (Smad3 cKO) mice and activated with αCD3/αCD28 with or without TGFβ1 (Fig. 4b). Cre-mediated gene knockout of Smad2 and Smad3 in CD4⁺ T cells was confirmed by Western blot analysis (Supplementary Fig. S2C). Consistent with our human T-cell findings, TGFβ1 minimally increased PD-1 expression on Smad3 cKO CD4⁺ T cells compared with WT CD4⁺ T cells, as shown in both the representative histogram (Fig. 4B, left) and the mean fluorescence intensity (MFI) of PD-1 (Fig. 4B, right). In contrast, Smad2 cKO CD4⁺ T cells maintained high PD-1 expression in response to TGFβ1 (Fig. 4B). Similarly, when WT, Smad2 cKO, and Smad3 cKO OT-I [ovalbumin (Ova)-specific CD8⁺ T cells] cells were activated with type I Ova in the presence of TGFβ1, Smad3 cKO OT-I showed decreased PD-1 expression (Fig. 4C). In contrast, the expression of lymphocyte activation gene3 (LAG3), another inhibitory receptor, decreased on Smad3 cKO OT-I and OT-II (Ova-specific CD4⁺ T cells) cells activated in the presence of TGFβ1, suggesting that TGFβ1 has differential effects on inhibitory receptors (Supplementary Fig. S2D). Taken together, the in vitro results in both human and murine T cells support the notion that TGFβ1 enhancement of PD-1 transcription is dependent selectively on SMAD3.

PD-1 is highly expressed on TILs, and its high expression is associated with decreased effector function in advanced-stage human cancer (3, 43–46). Given that TME-derived TGFβ1 can suppress antitumor immunity (30, 31), we hypothesized that SMAD3 contributes to high TIL PD-1 expression. To investigate whether TGFβ1 regulates PD-1 expression through SMAD3 in vivo, WT, Smad2 cKO, and Smad3 cKO mice were challenged with B16 melanoma, and PD-1 expression was assessed on TILs. SMAD2 and SMAD3 are known suppressors of T-cell function (33), and growth of B16 melanoma in Smad2 and Smad3 cKO mice was indeed significantly delayed (Fig. 4D). Although both Smad2 cKO and Smad3 cKO mice had comparably decreased volumes of B16 melanoma versus WT, the PD-1^hi subset population was significantly lower on Smad3 cKO CD8⁺ TILs, but not on Smad2 cKO CD8⁺ (Fig. 4E). In contrast, LAG3 expression was significantly enhanced on Smad2 cKO CD8⁺ TILs but unaffected by Smad3 cKO (Supplementary Fig. S2E). PD-1 expression on CD4⁺ TILs was not significantly different among WT, Smad2, and Smad3 cKO groups, although the level of PD-1 expression was much lower on CD4 T cells than on CD8 T cells in all three mouse groups (Supplementary Fig. S2F). The lack of a significant difference in PD-1 expression on CD4⁺ T cells in vivo could be due to the minimal fraction of antigen-specific CD4⁺ effector T cells in the TME. Alternatively, because PD-1 is expressed by Tregs (47, 48) that constitute the majority of CD4⁺ TILs in B16 melanoma, most CD4⁺ TILs may not be specific for tumor antigens. Supporting this notion, we found that the majority of CD4⁺ PD-1⁺ TILs express FOXP3 (Supplementary Fig. S3A). However, in the presence of TCR signaling in vitro, FOXP3⁺ and FOXP3⁻ CD4⁺ T cells are capable of enhancing PD-1 expression to the same extent in response to TGFβ1 (Supplementary Fig. S3B), and the FOXP3⁻CD4⁺ TILs did not express lower levels of PD-1 in the Smad3 cKO than in the WT mice (Supplementary Fig. S3C).

To test whether TGFβ1-induced SMAD3 activation upregulated PD-1 on tumor antigen–specific T cells, we utilized B16 melanoma cells stably expressing Ova as a model tumor antigen (B16-Ova). CD45.1 WT mice were challenged with B16-Ova and CD45.2 OT-I cells from WT, and Smad3 cKO OT-I mice were adoptively transferred into the tumor-bearing mice after tumor cells became palpable. The tumor growth was monitored, and lymphocytes infiltrating into tumors were harvested after 5 days. We found that transferred Smad3 cKO OT-I cells limit tumor growth more effectively than WT OT-I cells do (Fig. 5A). Consistent with our in vitro data, neither Smad3 cKO TILs nor Smad2 cKO TILs showed significantly increased cellular proliferation by CFSE (Fig. 5B and C) when gated on CD45.2⁺ (antigen experienced) donor cells. When a proliferated subset (i.e., CFSE-negative subset) was further gated, Smad3 cKO TILs showed significantly fewer of the PD-1^hi–expressing T cells highly characteristic of WT TILs (Fig. 5D, top). This effect of SMAD3 was specific to PD-1, as there was no reduction in the LAG3^hi subset in Smad3 cKO TILs (Fig. 5D, bottom). PD-1^hi–expressing cells did not decrease in Smad2 cKO OT-I (Fig. 5E, top), further supporting that SMAD3 is a critical mediator of PD-1 expression in the TME. In contrast, Smad2 cKO TILs maintained high levels of PD-1 (Fig. 5E, top), which is consistent with our observation in polyclonal CD8⁺ T cells (Fig. 4E). Similar to TILs, LAG3 expression on OT-I cells was not significantly affected in Smad3 cKO mice. Conversely, PD-1 and LAG3 expression on T cells in DLNs and in non-DLNs (NDLN) was comparable between WT and Smad3 cKO OT-I (Supplementary Fig. S4A and S4B) or Smad2 cKO OT-I (Supplementary Fig. S4C and S4D), suggesting that the effect of TGFβ1 is specific to the TME. To confirm the specificity to the TME, WT and DNTGFβRII Tg⁺ mice were challenged with B16 melanoma, and tumor volume was measured. In association with enhanced antitumor immune responses in DNTGFβRII Tg⁺ mice (Supplementary Fig. S5A), decreased PD-1 expression was also observed on CD8⁺ TILs, but not on T cells from DLNs (Supplementary Fig. S5B). As in Smad2 and Smad3 cKO mice, the DNTGFβRII Tg⁺ mice did not show significant decreases in LAG3 expression on TIL or DLN T cells (Supplementary Fig. S5C).

We next assessed the effect of Smad2 and Smad3 cKO on T-cell function. The number of TILs obtained in the B16 model is smaller than in some other tumor models and isolation of TILs from the tumors harvested from the Smad2 and Smad3 cKO mice particularly challenging. To confirm our findings with B16 melanoma and to permit functional analysis of TILs and T cells from the DLNs, we used a cancer model with more abundant TILs, the MC38 colon cancer model. WT, Smad2 cKO, and Smad3 cKO mice were challenged with MC38 colon cancer, and PD-1 expression was assessed. As with the B16 melanoma (Fig. 4D), tumor growth in Smad2 and Smad3 cKO mice was significantly delayed (Fig. 6A). In contrast to CD8⁺ TILs, the PD-1^hi subset population was absent in CD4⁺ TILs in all groups, and differences could not be assessed. As in B16 melanoma, the MFI of PD-1 on CD8⁺ TILs was significantly lower in the Smad3 cKO than in the WT or Smad2 cKO mice (Fig. 6B). We performed intracellular cytokine staining (ICS) to examine CD8⁺ T-cell production of IFNγ, TNFα, IL2, FOXP3, and granzyme B in WT and Smad3 cKO mice, but had insufficient TILs to perform ICS analysis for the Smad2 cKO group. We saw no significant differences between the Smad3 group and the WT group in FOXP3 or granzyme B levels when examining TILs or DLNs. However, in the Smad3 cKO group compared with the WT group, there was significantly higher production of TNFα from CD8⁺ TILs and of IFNγ and IL2 in the DLNs (Fig. 6C). Thus, decreased PD-1 expression secondary to the loss of Smad3 signaling in TILs and T cells in the DLNs is associated with increased production of multiple cytokines and functionality.

TGFβ1 is known to inhibit CD8⁺ T-cell effector function through SMAD2/3 (33) and many different mechanisms (37). Our data provide evidence that the enhancement of PD-1 expression represents a newly defined mechanism through which TGFβ1/SMAD3 suppresses T-cell function. To address how significant the impact of SMAD3-mediated PD-1 upregulation is on tumor evasion of T-cell responses, we treated Smad3 cKO mice bearing B16 melanoma with an anti–PD-1 blocking antibody (αPD-1) previously shown to have therapeutic efficacy in WT mice bearing B16 melanoma (49, 50). If the effect of Smad3 cKO on tumor growth is mediated through a mechanism other than PD-1 or if the effects of Smad3 cKO on PD-1 expression are not sufficient to negate that mechanism of tumor evasion, treatment with αPD-1 would confer additional therapeutic benefits in Smad3 cKO mice. To assess this, WT and Smad3 cKO mice were challenged with B16 melanoma cells and were given either isotype-matched IgG or αPD-1. We found that the tumor volume was decreased with αPD-1 compared with IgG-treated WT mice (Fig. 7, WT), attesting to the general role of the PD-1 pathway in immune resistance in this tumor model. In contrast, αPD-1 had no effect on tumor growth in Smad3 cKO mice (Fig. 7, Smad3).

We show here that TGFβ1, signaling selectively through SMAD3, significantly upregulates PD-1 in the context of TCR engagement. We show that this mechanism is important in generating a PD-1^hi population of T cells in the TME, where TGFβ1 expression is commonly very high. Thus, upregulation of both PD-1 ligands and the PD-1 receptor itself contributes to PD-1 pathway–mediated tumor immune resistance.

PD-1 expression can be differentially regulated by the environmental context in which a T cell encounters antigen. Upon activation, NFATc1 transiently induces PD-1 expression on T cells (38). Once PD-1 expression is induced, it is sustained in chronic infections or tolerogenic environments (2), but a high level of PD-1 expression is not achieved when antigen is encountered in an inflammatory environment, such as in Listeria monocytogenes infection (51). Further supporting the notion that the level of PD-1 expression is context dependent, there has been emerging evidence that cytokines can regulate NFATc1-induced PD-1 expression. IFNα promotes PD-1 expression on murine T cells through STAT1-mediated transcriptional regulation of Pdcd1 gene expression (21, 22). IL6 also increases PD-1 expression through a STAT3-dependent mechanism in murine CD8⁺ T cells in vitro (52), and we found similar regulation in human CD4⁺ and CD8⁺ T cells (Supplementary Fig. S1A). IL12 has differential effects on PD-1 in vivo and in vitro. IL12-conditioned tumor-specific memory CD8⁺ T cells have lower PD-1 expression in vivo with stronger antitumor immune responses (21). In contrast, we have found that IL12 increases PD-1 expression on human CD4⁺ and CD8⁺ T cells in vitro, consistent with others' findings on murine CD8⁺ T cells (52). Thus, although our data agree with the literature that IL6 and IL12 modulate PD-1 expression, TGFβ1 has the greatest effect on PD-1 expression, which has not been shown previously. The effects of some cytokines could be greater in vivo than we observed in vitro. However, our in vivo data demonstrating that the loss of TGFβ1 signaling has a profound impact on high-level PD-1 expression upon TCR engagement while signaling from other cytokines remains intact suggest that TGFβ1 signaling through SMAD3 is the major regulator of T-cell expression of PD-1 and function.

The data on the effect of other cytokines on PD-1 expression also collectively show that the regulatory mechanisms of PD-1 expression are highly conserved between human and mouse. This is further supported by high-sequence homology between human and murine Pdcd1 proximal promoter regions, including the NFATc1-binding site (52). We demonstrate that SMAD3-dependent PD-1 regulation is also conserved in showing that SMAD3 has the greatest effects on PD-1 expression on both human and murine T cells. NFATc1 was previously shown to be critical for PD-1 induction in mice (38), and mutation of antigens such that the TCR is no longer engaged results in a decline in human PD-1 expression in chronic infection with HCV or HIV (10, 53). Supporting previous findings that PD-1 expression depends on TCR engagement, we find that antigenic stimulation is required for TGFβ1 to enhance PD-1 expression and that NFATc1 binding to the human PDCD1 promoter following αCD3/αCD28 is enhanced by TGFβ1. Furthermore, TGFβ1 enhances PD-1 expression on both CD4⁺ and CD8⁺ T cells regardless of their naïve or memory status, although its effect was more pronounced on naïve T cells than on memory T cells.

Our proliferation assays showed that TGFβ1-mediated PD-1 enhancement is independent of cellular proliferation. In contrast to others' findings for which SMADs were proposed to play a role in TGFβ1 suppression of T-cell proliferation (54), we found that TGFβRI-dependent signaling, as demonstrated by phosphorylation of SMAD2 (Fig. 2C), did not result in suppression of T-cell proliferation. TGFβ1-mediated suppression of proliferation can be overcome by CD28-mediated costimulation, and it is possible that αCD3/αCD28 used in our in vitro culture system masked inhibition (55). However, neither Smad2 cKO OT-I nor Smad3 cKO OT-I showed significantly altered proliferation in vivo (Fig. 5B and C). Nevertheless, we observed that isolated Smad3 cKO CD4⁺ T cells have increased IL2 expression compared with WT littermates when activated with αCD3/αCD28 (Supplementary Fig. S5D), consistent with previous reports (56).

Others have suggested a potential association between TGFβ1 signaling and high PD-1 expression, but the direct causal relationship, the molecular mechanism, and biological implications of this association have not ever been characterized (32, 57, 58). TGFβ1 signaling consists of SMAD- and non-SMAD–dependent pathways, and SMAD-dependent gene regulation (SMAD2 and SMAD3) has been well characterized (59, 60). Some genes are preferentially and exclusively regulated by SMAD2 or SMAD3 as with ID1 and MYC (61, 62). On the other hand, SMAD2 and SMAD3 can redundantly regulate the expression of many genes that are under control of TGFβ1 (63). Our luciferase assay and in vitro data suggest that PD-1 regulation is predominantly under the control of Smad3. Although our in vitro data support a minor role for SMAD2 in TGFβ1-dependent PD-1 enhancement, our in vivo data clearly demonstrated no enhancement of PD-1 expression through SMAD2, with Smad2 cKO mice showing a small increase rather than decrease in PD-1 expression. The in vivo data could reflect enhanced SMAD3 activity in compensation for SMAD2 deletion in T cells.

Our in vivo studies mainly focused on CD8⁺ T cells because the PD-1 expression difference was greater in CD8⁺ T cells than in CD4⁺ T cells in vivo and the levels of PD-1 on CD4⁺ T cells much lower than on CD8⁺ T cells. Supporting this notion, TGFβ1-suppression of antitumor immunity in vivo appears to be dependent on CD8⁺ T cells but not on CD4⁺ T cells in a murine mouse model (31). This discrepancy could be due to cellular intrinsic difference between the CD4⁺ and CD8⁺ subsets of T cells (64). Our observation that stimulated OT-I and OT-II T cells respond similarly to TGFβ1 in vitro (Supplementary Fig. S2D) but CD4 and CD8 TILs in vivo do not may be explained by an absence of antigenic recognition by CD4⁺ TILs in vivo due to CD4⁺ T cells being primarily Tregs (Supplementary Fig. S3B and S3C). Alternatively, mechanisms of PD-1 regulation unique to CD4⁺ T cells may exist in vivo given that the FOXP3⁻ CD4⁺ T cells in the Smad3 cKO mice did not express lower levels of PD-1 than WT (Supplementary Fig. S3D).

Interestingly, the effect of TGFβ1/SMAD3 on PD-1 expression of CD8⁺ T cells was specifically on TILs, but not on those originating from tumor DLNs. We did not find the percentage of PD-1^hi T cells isolated from the tumor DLNs in WT mice to be significantly different from that of DNTGFβRII Tg⁺ and Smad3 cKO mice (Supplementary Figs. S4A and S4B and S5B). This may be due to the fact that the PD-1^hi CD8⁺ T-cell population prominent in the TME was absent in DLNs in WT mice and suggests that TGFβ1 levels could be much lower outside of the TME. Supporting this notion, others found that TGFβ1 expression is higher in human head and neck squamous cell carcinoma tissue than in adjacent mucosal tissue (65). Nevertheless, it is yet to be determined whether the dominant source of TGFβ1 in the TME is derived from tumor or T cells. In spite of extensive evidence that TGFβ1 suppresses antitumor immunity, tumor-specific deletion of TGFβ1 did not enhance antitumor immune responses (33). In contrast, others have reported that the deletion of T cell–derived TGFβ1 was sufficient to prevent tumor growth (32).

Although Smad3 cKO mice did not mount immune responses as potent as those of DNTGFβRII Tg⁺ mice, B16 melanoma growth in Smad3 cKO mice appeared comparable with that in Pdcd1 KO mice or anti–PD-1 antibody–treated WT mice (Fig. 7). Although adoptive transfer of naïve antigen-specific T cells is known to confer minimal antitumor effects, transferred Smad3 cKO OT-I effectively controlled tumor growth (Fig. 5A). Collectively, these results provide direct evidence that PD-1–mediated antitumor immunity depends in part on Smad3 activation and that Smad3-driven PD-1 upregulation is relevant to tumor immune evasion. Our data clearly show that αPD-1 treatment decreases B16 tumor growth in WT mice (49, 50) but not in Smad3 cKO mice.

In summary, our data demonstrate a novel immunosuppressive function of TGFβ1 in regulating high-level PD-1 expression on T cells encountering cognate antigen. In addition to other suppressive roles for TGFβ1, TGFβ1 enriched in the TME may induce high levels of PD-1 on T cells as they encounter antigens on the tumor surface, reducing T-cell effector function and limiting the antitumor T-cell response. In addition, our data provide mechanistic understanding of the regulation of high-level PD-1 expression. Although it is well known that T cells against intact antigen in the setting of chronic viral infections, such as HCV and HIV, or malignancy express very high levels of PD-1, it is not known how those high levels are induced. This study elucidates a mechanism through which the highest levels of PD-1 are induced. Indeed, high TGFβ1 serum levels are associated with worse disease outcome in HCV infection (66), and TGFβ1 expression is high in the TME of advanced stages of cancer, which may further limit the efficacy of T cells against disease in those settings (67–69). Given the potential for autoimmunity with PD-1 therapy, it is worth investigating whether inhibitors of SMAD3 used in combination with other immunotherapeutic agents activate T cells expressing the highest levels of PD-1 rather than all T cells bearing PD-1. As the PD-1^hi subset of TILs may in fact contain the highest proportion of true tumor-specific cells, these may be the most important target population for SMAD3 blockade.

All animals were housed and handled in compliance with Johns Hopkins Animal Care and Use policy. C57BL/6 DNTGFβRII Tg⁺ and C57BL/6 Cd4-Cre transgenic mice were purchased from The Jackson Laboratory. CD45.1 congenic mice were purchased from the NCI at Frederick, MD. Smad2 flox/flox (fl/fl) and Smad3 fl/fl mice were generated by S.-J. Lee's Laboratory at Johns Hopkins School of Medicine (Baltimore, MD) and backcrossed to C57BL/6 at least six generations. OT-I and OT-II mice were generous gifts from Drs. Charles Drake and Hyam Levitsky at Johns Hopkins School of Medicine. Age-matched female mice were utilized in all in vivo experiments.

Human PBMCs were isolated from leukopheresis by Ficoll–Hypaque density gradient. Isolated human PBMCs were subjected for CD3⁺ T-cell isolation using the Pan T-cell Isolation Kit (Miltenyi Biotec) as instructed in the manual. The isolated cells were activated for 72 hours with αCD3/αCD28 Dynabeads (Invitrogen) at a cell-to-bead ratio of 1:1 in RPMI + 10% FBS (supplemented with HEPES buffer, penicillin/streptomycin, and l-glutamine). Murine CD4⁺ and CD8⁺ T cells were isolated from the spleen and lymph nodes using the Negative Selection Kit (Invitrogen) and were activated with plate-coated αCD3 (10 μg/mL) and soluble αCD28 (2 μg/mL) or with cognate Ova peptides in the presence of irradiated splenocytes for 72 hours.

Jurkat T cells (clone E6-1) were purchased from the ATCC and were kept as a frozen stock. Jurkat T cells (1.5 × 10⁷) were transfected with 10 μg pGL-3 Firefly Luciferase Vector (Promega) and 1 μg of pRL-TK Vector (Promega) by electroporation using Nucleofector II (Amaxa/Lonza). The cells were rested in a 6-well plate overnight and activated with plate-coated αCD3 (10 μg/mL) and soluble αCD28 (5 μg/mL) with or without rhTGF-β1 (50 ng/mL). After 24 hours, the cells were harvested and lysed followed by luminescence measurement using Dual-Luciferase Assay (Promega). Where indicated, the cells were cotransfected with empty vector (pSG-V5), TGF-βRI-His (Addgene plasmid #19161), and TGF-βRII (Addgene plasmid #11766). For siRNA-mediated knockdown, the cells were cotransfected with 1.5 μmol/L of siRNA for SMAD2 (Santa Cruz Biotechnology, SC-38374) and SMAD3 (Santa Cruz Biotechnology, SC-38376).

Human recombinant IL1α, IL2, IL4, IL6, IL10, IL12, IL13, IL15, IL17, IL18, IL21, IL23, INFα, IFNγ, TGFβ1, and TNFα were purchased from Peprotech and used at 5, 50, and 500 ng/mL. The data shown in Supplementary Fig. S1A are using 500 ng/mL only because the relative effects of cytokines were not different at other doses. Primary human T cells were treated with neutralizing TGFβ1 antibody (Abcam, 2Ar2) and with small-molecule inhibitors SB431542, cyclosporine A (Sigma-Aldrich), and SIS3 (Calbiochem) for 1 hour prior to activation at the indicated concentration range.

After indicated time of culture, human T cells were harvested and centrifuged at 400 × g (or 1,500 rpm) for 5 minutes. The cells were washed in FACS buffer (1× PBS + 2% FBS) and stained with Aqua Viability Dye (Invitrogen) as instructed in the manual. After wash, the cells were stained with PD-1 PE (BioLegend), CD8 PerCP, CD4 Pacific Blue, CD3 FITC (eBioscience), and HLA-DR APC (eBioscience) or qDot605 (Invitrogen). The similar protocol was used for murine T cells and PD-1 PE, CD4 or CD8 PerCP, FITC (eBioscience), LAG3 APC or PacBlue, CD4 BV605, CD8 BV570, CD3 AF700, PD-1 PE Cy7, and CD44 AF700 (BioLegend) were used for flow cytometry. For intracellular staining, the cells were treated with FOXP3/Transcription Factor Staining Buffet Set (eBioscience) and stained with FOXP3 FITC, TNFα PE (eBioscience), IL2 PE-CF594, or IFNγ APC (BD Biosciences).

Total RNA was extracted from primary T cells under the indicated conditions using RNeasy Plus Kit (Qiagen). Extracted RNA (100 ng) was reverse transcribed using SuperScript III First-Strand Synthesis System (Invitrogen). Generated cDNA was subjected for real-time PCR assay. PDCD1 primer sequences are forward 5′-CACTGAGGCCTGAGGATGG-3′; reverse 5′-AGGGTCTGCAGAACACTGGT-3′. All target genes were normalized to 18s rRNA or 28s rRNA as described previously.

Human PDCD1 promoter (1.9 kb) was cloned from the genomic DNA of isolated CD3⁺ T cells and the sequence was confirmed. The amplified clones were ligated to SacI/XhoI-digested pGL3-Basic Vector (Promega) using the In-Fusion Cloning Kit (Clontech). Site-directed mutagenesis was carried out using following primers for NFAT, SBE-D, and SBE-P sites using QuikChange Lightning Kit (Agilent Technologies). Primer sequences were as follows: forward: 5′-GATGCTCTTTTTGGACTGTTTCGG-3′, reverse 5′-CCGAAACAGTCCAAAAAGAGCATC-3′ (NFAT); forward: 5′-ACCTTAGCTGGATGGCAGCA-3′, reverse 5′-TGCTGCCATCCAGCTAAGGT-3′ (SBE-D); forward: 5′-CGCGCCTCGCATCCATCATCTT-3′, reverse: 5′-AAGATGATGGATGCGAGGCGCG-3′ (SBE-P).

ChIP assay was performed according to the manufacturer's guidance (Invitrogen MAGnify ChIP system). Briefly, isolated CD3⁺ T cells were activated with αCD3/αCD28–conjugated beads for 24 hours and fixed with 2% formaldehyde. Sonicated DNA was immunoprecipitated with αSMAD3 (Cell Signaling Technology), and αNFATc1 (Santa Cruz Biotechnology). The immunoprecipitated chromatin was analyzed on Roche LightCycler 480 by SYBR Green using the following primers for PDCD1 promoter. PDCD1: forward 5′-CCTCACATCTCTGAGACCCG-3′, reverse 5′-CCGAAGCGAGGCTAGAAACC-3′; GAPDH: 5′-TACTAGCGGTTTTACGGGCG-3′, 5′-TCGAACAGGAGGAGCAGAGAGCGA-3′.

Human or murine T cells were activated as indicated and harvested and lysed in RIPA Buffer (Cell Signaling Technology). Protein extract concentrations were measured using the BCA Protein Assay Kit (Thermo Scientific) and followed by heating under reducing conditions. The equal amounts of extracts were loaded/run on NuPAGE Precast gels (Invitrogen), and transferred membranes were blotted with the following antibodies: pSMAD2, total SMAD2, total SMAD3 (Cell Signaling Technology), and β-actin (Sigma).

B16 melanoma cell lines were purchased from the ATCC and kept as frozen stock in 2014. B16 melanoma cell lines (1 × 10⁵) were injected on a flank in 100 μL volume. Tumor volumes were measured every other day using a caliper and assessed using the formula 1/2 (length × width²). CD45.1 host mice were injected with 1 × 10⁵ B16-Ova melanoma cell lines on a flank. WT OT-I or cKO OT-I (8 × 10⁶) were labeled with the CellTrace CFSE Cell Proliferation Kit (Life Technologies) and were adoptively transferred into tumor-bearing mice by retro-orbital injection on day 12. The tumors were harvested on day 5 after the adoptive transfer, and lymphocytes were purified using Percoll (GE Healthcare) gradient. In a blocking experiment, 5 × 10⁵ B16 melanoma cells were injected on a flank, and 100 μg Armenian hamster IgG isotype control (Rockland) or anti–PD-1 antibody (G4) were injected intraperitoneally twice a week from day 0. The B16 melanoma cell lines have tested negative for Mycoplasma but have not been authenticated by the laboratory.

MC38 colon adenocarcinoma cell lines were purchased from the ATCC and kept as frozen stock. B16 melanoma cell lines (4 × 10⁵) were injected on a flank in 100 μL volume. Tumor volumes were measured using a caliper and assessed using the formula ½ (length × width²). The tumors were harvested on day 23, and lymphocytes were purified using Percoll (GE Healthcare) gradient. For intracellular cytokine staining, lymphocytes were stimulated with phorbol 12-myristate 13-acetate (PMA, 50 ng/mL) and ionomycin (500 ng/mL) in the presence of Brefeldin A and Monensin (eBioscience) for 4 hours prior. After stimulation, cells were permeabilized and stained for intracellular cytokines. MC38 lines have tested negative for Mycoplasma but have not been authenticated by the laboratory.

No potential conflicts of interest were disclosed.

Conception and design: B.V. Park, F. Pan, A.L. Cox

Development of methodology: F. Pan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B.V. Park, Z.T. Freeman, A. Ghasemzadeh, M.A. Chattergoon, A. Rutebemberwa, T.V. Huynh, S.M. Sebald, S.-J. Lee

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B.V. Park, Z.T. Freeman, A. Ghasemzadeh, M.A. Chattergoon, F. Pan

Writing, review, and/or revision of the manuscript: B.V. Park, A. Rutebemberwa, F. Pan, D.M. Pardoll, A.L. Cox

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Steigner, M.E. Winter, S.M. Sebald

Study supervision: F. Pan, A.L. Cox

We would like to acknowledge Ada Tam and Richard L. Blosser from the Cancer Research Flow Cytometry Core and Tricia L. Nilles from the School of Public Health Flow Cytometry Core at Johns Hopkins University for their technical help. We are grateful to Juan Fu and Young Kim for providing anti–PD-1 antibody reagents and Xingmei Wu for maintaining and genotyping mouse colonies. Also, we want to thank the Jeff Wrana and Joan Massagué Laboratory for their generous donation of plasmids to Addgene.

This work was supported by grants U19 AI088791, R01AR060636, RO1AI099300, RO1AI089830, and P30CA006973 from the Bloomberg-Kimmel Institute for Cancer Immunotherapy and the National Institutes of Health.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.