Epigenetic state determines the in vivo efficacy of STING agonist therapy

We have recently reported that STING expression is downregulated or lost in a notable subset of human melanoma cell lines mainly through promoter hypermethylation-driven silencing29,37, which can be restored through pharmacologic inhibition of DNA methylation (Fig. 2a). These findings led us to hypothesize that hypermethylation of the STING promoter also in mouse melanoma cell lines can mediate their coordinated transcriptional silencing and contribute to the impairment of the STING signaling function. To test this, we performed genome-wide DNA methylation profiling using the Infinium Mouse Methylation BeadChip microarray in B16-F10 and Yumm1.7 melanoma cell lines before and after their treatment with DNA methyltransferase inhibitor 5-aza-2’-deoxycytidine (5AZADC, decitabine). While we observed hypermethylation of STING promoter as indicated by β-values in both cell lines, their treatment with 5AZADC resulted in a decrease in STING promoter methylation levels (Fig. 2b, c). We also found reconstitution of STING protein expression in both B16-F10 and Yumm1.7 cell lines following their treatment with 5AZADC (Fig. 2d, e) further suggesting promoter hypermethylation-driven STING silencing and indicating their similarities in epigenetic regulation of STING to human melanoma cell lines. To further evaluate whether demethylation could restore functional activation of STING signaling, following 5AZADC treatment we next stimulated these cell lines with ADU-S100 and assessed STING-dependent downstream effects. We observed up to a 46-fold increase in induction of IFN-β (p < 0.0001) (Fig. 2f) and a notable increase in MHC class I surface expression in 5AZADC-pretreated B16-F10 and Yumm1.7 cells (Fig. 2g, h) compared to untreated controls following their exposure to ADU-S100. Although to a lesser extent, we also observed 5AZADC-mediated induction of IFN-β and upregulation of MHC class I in both B16-F10 and Yumm1.7 cell lines (Fig. 2f–h).

Consistent with our findings of 5AZADC-triggered IFN-β expression, multiple studies have shown that DNA demethylating agents among other epigenetic modulators can induce a type I IFN response in cancer cells through the de-repression of endogenous retroviruses (ERVs) and activation of the MDA5-MAVS pathway38,39,40. To determine whether the upregulation of MHC class I in response to treatment with 5AZADC occurs through type I IFN signaling, we blocked type I IFN receptor using an IFNAR blocking antibody in B16-F10 and Yumm1.7 cells. Indeed, in both cell lines blockade of IFNAR inhibited 5AZADC-induced upregulation of MHC class I (Supplementary Fig. 2a, b) as well as upregulation of LMP2 and TAP1, two key components of MHC class I antigen-processing pathway (Supplementary Fig. 2c, d).

To evaluate whether the profound effect of 5AZADC-pretreatment and ADU-S100 stimulation on IFN-β induction and MHC class I upregulation was driven by activation of STING signaling, we performed analogous experiments using B16-ISG and B16-ISG-STINGKO cells (Supplementary Fig. 3a). Similar to our observations in B16-F10 and Yumm1.7 cells, we found a 36-fold increase in induction of IFN-β (p < 0.0001) (Supplementary Fig. 3b) as well as a 4.4-fold increase in surface expression of MHC class I (p < 0.0001) (Supplementary Fig. 3c, d) in 5AZADC-pretreated B16-ISG cells in response to stimulation with the STING agonist, but not in 5AZADC-pretreated B16-ISG-STINGKO cells. Although we observed 5AZADC-induced upregulation of MHC class I in both B16-ISG and B16-ISG-STINGKO cell lines (Supplementary Fig. 3c, d), unlike 5AZADC-pretreated B16-ISG cells, further stimulation with the STING agonist did not increase surface expression of MHC class I in 5AZADC-pretreated B16-ISG-STINGKO cells, suggesting that DNA demethylation-induced increased IFN-β production and upregulation of MHC I in response to agonist stimulation in melanoma cells is STING mediated. Collectively, these data indicate that demethylation could enhance functional activation of STING signaling in mouse melanoma cell lines.

DNMT3A and DNMT3B are involved in melanoma STING silencing

The observation of promoter hypermethylation-mediated STING suppression in human37 and mouse melanoma cell lines (Fig. 2b, c) suggested a possible epigenetic mechanism driven by DNA methyltransferases (DNMTs) including DNMT1, DNMT3A, and DNMT3B. To further interrogate the role of DNMTs in melanoma-STING silencing, we transfected B16-F10 cells with small interfering RNA (siRNA) targeting DNMT1, DNMT3A, and DNMT3B. Depletion of DNMT3A and DNMT3B, but not DNMT1 resulted in the reconstitution of STING mRNA and protein expression in B16-F10 cells (Fig. 3a, b; Supplementary Fig 4a). To directly evaluate the functional activation of STING signaling in these cells, we next stimulated them with ADU-S100. Knockdown of DNMT3A and DNMT3B significantly increased STING-dependent induction of CXCL10 (up to 2.5-fold, p < 0.0001) and IFN-β (up to 3.6-fold, p < 0.0001) in B16-F10 cells compared with siControl in response to stimulation with the STING agonist (Fig. 3c, d). Consistent with these findings, reconstitution of STING expression following pharmacologic inhibition of DNA methylation with 5AZADC coincided with decreased expression of DNMT3A and DNMT3B in B16-F10 cells (Fig. 3e). Similar effects were observed in A375 and SK-MEL-28, two STINGlow human melanoma cell lines37 (Fig. 3f). Additionally, using a publicly available cBioPortal dataset41 we identified a negative, albeit not significant, correlation between STING and DNMT3A (Pearson’s r = −0.41), DNMT3B (Pearson’s r = −0.28) (Fig. 3g), as well as DNMT1 mRNA levels (Pearson’s r = −0.13) (Supplementary Fig. 4b) across metastatic melanoma samples, suggesting their involvement in melanoma-STING silencing.

Demethylation improves melanoma response to STING agonist therapy in STINGgt/gt mice

To examine whether demethylation-mediated reconstitution of STING and subsequent rescue of STING signaling in STINGlow melanoma cells could improve agonist-induced antitumor immune responses in vivo, we developed subcutaneous tumors of B16-F10 and Yumm1.7 in STINGgt/gt mice and treated them intratumorally with a combination of 5AZADC and ADU-S100 (Fig. 4a). While treatment with either 5AZADC or ADU-S100 alone was ineffective in delaying tumor growth, combined treatment of 5AZADC with ADU-S100 resulted in significantly enhanced inhibition of tumor growth in both B16-F10 (p < 0.05 on day 22) and Yumm1.7 (p < 0.0001 on day 21) models (Fig. 4b, c). Similarly, combination treatment with ADU-S100 and 5AZADC significantly delayed tumor growth in the aggressive model of B16-ISG (p < 0.0001 on day 21) (Fig. 4d); however, it failed to induce tumor control in the STING-deficient B16-ISG-STINGKO model (Fig. 4e) suggesting a critical role for STING in mediating antitumor responses to the combination therapy. Using immunoblot analysis, we further confirmed reconstitution of STING protein expression in B16-F10 and Yumm1.7 tumors following their in vivo treatment with 5AZADC (Fig. 4f, g). Consistent with these findings, we observed a marked increase in the transcription of Ifnb1 (p < 0.0001) and H2-k1 (p < 0.0001) genes downstream of STING signaling in tumors treated with the combination therapy compared to the agonist alone (Fig. 4h). Collectively, these results show that epigenetic reprogramming can restore melanoma-intrinsic STING signaling defects and therefore prime therapeutic responses to STING agonism in vivo.

Melanoma response to combination therapy with 5AZADC and ADU-S100 depends on CD8+ T cells

Given the important role of CXCL10 in regulating CXCR3+ T cell trafficking into tumors, we next tested whether demethylation-mediated rescue of STING signaling in STINGlow B16-F10 melanoma models could restore melanoma-intrinsic CXCL10 induction. Quantitative reverse transcription PCR analysis indicated a 2.6-fold increase (p < 0.0001) in CXCL10 gene expression in B16-F10 tumors in STINGgt/gt mice following the combination treatment with 5AZADC and ADU-S100 (Fig. 5a). This increased CXCL10 induction correlated with higher frequency of intratumoral CXCR3+ CD8+ T cells (Fig. 5b), demonstrating that reversal of epigenetic silencing of STING using a demethylating agent can promote tumor infiltration of CD8+ T cells. To directly assess whether the observed antitumor effects were attributed to the generation of functional tumor antigen-specific CD8+ T cells, we next performed antibody-mediated depletion of either CD8+ or CD4+ T cells in B16-F10-bearing STINGgt/gt mice receiving the combination therapy. Successful depletions were confirmed by flow cytometry analysis of splenocytes (Supplementary Fig. 5). While depletion of CD4+ T cells did not alter the tumor control, depletion of CD8+ T cells completely abrogated the antitumor activity of the combination therapy (p < 0.001 on day 22) (Fig. 5c). These data indicate that CD8+ T cells are critical for the therapeutic efficacy of the combination therapy with 5AZADC and ADU-S100.

CD8+ TILs in 5AZADC and ADU-S100 combination-treated mice indicate less exhausted phenotype

To characterize how combination therapy with 5AZADC and ADU-S100 can impact the tumor microenvironment, we next assessed the abundance and the phenotype of CD8+ TILs. Flow cytometry analysis indicated that combination treatment led to a significant increase in the frequency of tumor infiltrating CD8+ T cells (p < 0.05) (Fig. 6a, d). These CD8+ TIL also expressed lower levels of inhibitory receptors LAG-3 and PD-1 relative to those in the control treated tumors suggesting that combination therapy with 5AZADC and ADU-S100 can induce a functional CD8+ T cell population with an activated rather than an exhausted state (Fig. 6b, c, e–h). These findings, together with the observation that in the absence of CD8+ T cells combination therapy did not result in a therapeutic benefit (Fig. 5c), underscore the importance of CD8+ T cells in inducing antitumor responses in the setting of demethylation-mediated tumor cell-intrinsic STING reactivation.

Combination therapy with 5AZADC and ADU-S100 promotes activation and effector function of CD8+ T cells

To assess whether intratumoral treatment with the combination of ADU-S100 and 5AZADC induces systemic immune effects, we examined the frequency and differentiation status of T cells in the spleens of tumor-bearing mice. Within the splenic T cell population, combination treatment with 5AZADC and ADU-S100 led to an increase in the percentage of CD8+ T cells (Fig. 7a; Supplementary Fig. 6a). We also observed higher frequency of activated CD69+ CD44+ CD8+ splenic T cells in tumor-bearing mice treated with the combination of 5AZADC and ADU-S100 compared to those treated with either single agent (Fig. 7b, c; Supplementary Fig. 6b). Furthermore, the combination treatment significantly increased the central memory phenotype of CD8+ T cells (TCM, CD44+ CD62L+; p < 0.01) in the spleens of tumor-bearing mice (Fig. 7d, e; Supplementary Fig. 6c). Similarly, CD8+ T cells in tumors treated with the combination of ADU-S100 and 5AZADC indicated higher frequency of TCM (Supplementary Fig. 7a, b). In addition, the activation markers CD69+ and CD44+ were expressed at higher frequency on CD8+ TIL from tumors treated with the combination therapy than those treated with ADU-S100 or 5AZADC alone (p < 0.01) (Supplementary Fig. 8a–c). We next performed intracellular cytokine staining on splenocytes from tumor-bearing mice to determine whether the combination treatment could also impact the functional state of CD8+ T cells. Splenic CD8+ T cells in tumor-bearing mice treated with the combination of 5AZADC and ADU-S100 expressed higher levels of IFN-γ and TNF-α relative to the mice treated with ADU-S100 alone (Fig. 7f, g; Supplementary Fig. 6d, e) indicating that combination therapy stimulates a stronger systemic antitumor T cell response. Overall, these results demonstrate that demethylation-mediated reactivation of STING in tumor cells can promote expansion, activation, and effector function of CD8+ T cells.

Demethylation can further improve STING agonist efficacy in mice with intact STING activity

To further investigate the functional importance of demethylation-induced STING reactivation in tumor cells in enhancing antitumor responses to STING agonist therapy in a host with intact STING activity, we injected C57BL/6 mice with B16-F10 or Yumm1.7 tumor cells and treated them with a combination of ADU-S100 and 5AZADC (Fig. 8a). In this setting, where STING is functional in the host antigen presenting cells, ADU-S100 alone resulted in tumor control compared to 5AZADC or vehicle treated controls; however, combination therapy was more effective in inhibiting the tumor growth in both B16-F10 and Yumm1.7 models (Fig. 8b, c). Although there was no detectable difference in the percentage of CD8+ TIL (Fig. 8d), tumors treated with a combination of ADU-S100 and 5AZADC were more enriched with effector cytokine expressing CD8+ T cells than those treated with ADU-S100 alone (Fig. 8e, f). These data show that restoration of STING activity in tumor cells can augment antitumor responses to STING agonist therapy even in STING-intact hosts and indicate that both tumor and host-derived STING, together, can play a critical role in promoting antitumor immunity.

Discussion

Direct activation of the host STING pathway through intratumoral injections of synthetic cyclic dinucleotides has shown to be an attractive therapeutic strategy in multiple preclinical cancer models22,25,42. However, clinical trials in cancer patients currently testing STING activating drugs as a monotherapy or in combination with immune checkpoint inhibitors have so far provided little, if any, evidence of objective responses26,27,28.

This discrepancy in outcomes to STING agonist therapies raises the possibility that targeting the STING pathway could be functionally relevant not only in antigen presenting cells of the host, which has been the focus of prior preclinical studies, but also in tumor cells with well documented epigenetically driven dysfunctional STING activity across several human cancer types, including melanoma29,35,43,44. Despite this common suppression, the ability to target and restore STING signaling in tumor cells using epigenetic modulators35,37 provides a rationale to explore whether epigenetic reprogramming of tumor cell-intrinsic STING activity can improve the therapeutic efficacy of STING agonists. In this study, we investigated this rationale in vivo using multiple approaches with murine melanoma models in both STING-deficient and STING-sufficient hosts.

Consistent with our previous findings in human melanoma37, we found promoter hypermethylation-driven STING silencing in two distinct murine melanoma cell lines. Both in vitro and in vivo, pharmacologic inhibition of DNA methylation with 5AZADC resulted in reconstitution of STING protein expression indicating the involvement of common epigenetic mechanisms in regulating STING expression in these experimental models. We further showed that demethylation-mediated reversal of STING silencing not only greatly increased agonist-induced production of IFN-β by tumor cells in in vitro experiments, but also within the tumors in STING-deficient mice. In host dendritic cells, STING-mediated IFN-β induction has been well characterized and shown to be essential for their activation and the cross-priming of cytotoxic T cells22,23. Similarly, initiation of an adaptive immune response to radiation therapy has been linked to STING-mediated IFN-β induction by dendritic cells45. Nevertheless, compelling clinical and experimental evidence support the importance of IFN-β production by tumor cells themselves in mediating antitumor responses to radiation therapy in combination with immune checkpoint inhibitors46,47,48. In agreement with these studies, we found that demethylation-mediated restoration of STING-dependent IFN-β production in tumors resulted in greater tumor control. IFN-β induction in tumor cells can exert direct effects both on tumor cells as well as APCs16,47. Indeed, we have previously shown that in human melanoma cell lines STING-mediated IFN-β induction can augment their surface expression of MHC class I and therefore render them better targets for cytotoxic CD8+ T cell recognition and killing29,37. The importance of IFN-β production by cancer cells in the recruitment and activation of Batf3-dependent dendritic cells has been also demonstrated in different settings47,49. Although mechanistically it remains unclear, it is conceivable that demethylation-mediated restoration of STING-dependent IFN-β production in tumors can take a more active and direct role in driving the maturation and activation of tumor-infiltrating dendritic cells.

Similar to our observations in human melanoma cell lines, CXCL10 was upregulated in mouse melanoma tumors following demethylation-mediated reactivation of STING signaling. Together with CXCL9, another CXCR3-binding chemokine, CXCL10 is a dominant chemokine for the recruitment of tumor-specific CD8+ T cells into the tumors31,32. Also, as a member of the 12-chemokine gene signature classifier33, its expression has been correlated with the presence of tumor-associated tertiary lymphoid structures50,51,52 which are rapidly emerging as a powerful prognostic marker in melanoma and a number of other cancer types53,54,55. In our studies, tumor-localized CXCL10 induction was correlated with higher frequency of CD8+ T cells within tumors. These data suggest that STING-functional tumors can initiate a CXCL10/CXCR3 axis and subsequently guide trafficking and entry of effector T cells into the tumor sites. Additionally, CXCL10 has been shown to promote effector T cell generation and function56,57. In line with these data, we found a CD8+ T cell-dependent mechanism in mediating antitumor effects in response to the combination therapy. This finding also coincides with our observation of STING-dependent MHC class I upregulation in human and mouse melanoma cells following demethylation-mediated restoration of STING signaling which can enhance both their antigenicity and susceptibility to lysis by CD8+ T cells29,37. Although it was proposed that STING agonist can induce T cell death through the activation of proapoptotic transcriptional programs58, recent studies have indicated that this effect depends on the type and doses of STING agonists59,60. Unlike the mouse STING ligand DMXAA that was shown to trigger T cell death, stimulation of T cells with cyclic dinucleotides including cGAMP did not cause T cell death. Also, two recent reports have shown that T cell-intrinsic STING signaling could control differentiation and effector functions of CD8+ and CD4+ T cells60,61. Although our initial in vivo studies were performed in STINGgt/gt mice with T cells lacking STING and therefore insensitive to STING agonist, our later experiments in STING-sufficient C57/BL6 mice did not indicate any unfavorable effect on the frequency or function of T cells of the mice treated with either ADU-S100 alone, or when combined with 5AZADC.

Furthermore, consistent with mobilization and expansion of CD8+ T cells within the tumors, we found an increase in the frequency of splenic CD8+ T cells in tumor-bearing mice treated with intratumoral injections of ADU-S100 and 5AZADC. These splenic CD8+ T cells mainly displayed an effector/memory phenotype as indicated by higher expression of antigen-experience and activation markers including CD44 and CD69, as well as higher frequency of IFN-γ effector cytokine producing cells. Based on these observations, it is likely that local tumor cell-induced type I IFN response driven by the rescue and activation of STING signaling can promote a systemic antitumor CD8+ T cell immunity. This notion is further supported by two recent publications describing a novel role for cancer cells and their type I IFN production in priming CD8+ T cell responses through MHC class I dressing of dendritic cells49,62. Although additional experiments will be needed to fully understand the factors contributing to the systemic activation that we observed, regardless of the exact mechanism, our data indicate that targeting tumor-cell intrinsic STING signaling through intratumoral combination therapy is capable of priming antitumor CD8+ T cell responses.

As restoration and reactivation of STING pathway in tumor cells can reinstate their antigenicity as well as T cell tumor trafficking, a key question for future work will be if this strategy could improve the efficacy of T cell-based immunotherapies in patients with melanoma particularly those involving adoptive cell transfer of autologous TIL. Similarly, it will be important in the future to determine how targeting tumor cell-intrinsic STING activity in solid tumors beyond melanoma can affect immunotherapy outcomes. Restoration of tumor cell-intrinsic STING activity through epigenetic remodeling may be also beneficial for CAR T-cell therapies by modulating the tumor microenvironment and providing a favorable milieu for CAR T-cell trafficking and persistence in solid tumors63.

In summary, our data demonstrate that epigenetic silencing of STING in melanoma cells is not only a marker of tumor-intrinsic immune evasion mechanism but can indeed confer resistance to STING agonist therapy. Accordingly, we have shown that a rational combination of a clinically available DNA methylation inhibitor with a STING agonist can lead to robust antitumor responses in the setting of STINGlow murine models of melanoma. Therefore, identification and pharmacologic restoration of tumor cell-intrinsic STING signaling defects through epigenetic reprograming might be critical for the successful use of STING agonist therapies in the clinic. Although additional work will be necessary to further identify optimal dosing levels and therapeutic schedules of each component, insights from our study provide a framework to design proper clinical treatment modalities with appropriate patient selection in melanoma and perhaps other solid tumors.