Simultaneous Inhibition of LSD1 and TGFβ Enables Eradication of Poorly Immunogenic Tumors with Anti–PD-1 Treatment

The blockade of the immune inhibitory receptor PD-1 that mediates T-cell dysfunction has led to unprecedented, long-lasting responses in a wide variety of cancers. However, outcomes of large clinical trials have clearly shown that anti–PD-1–directed immune checkpoint blockade (ICB) therapy has not been effective in a majority of patients with cancer (1). Therefore, there are increasing interests and efforts in identifying potential targets for combination therapies to overcome tumor resistance to anti–PD-1 therapy. Genetic alterations frequently occur in cancer, some of which have been shown to suppress tumor immunogenicity and render unresponsiveness to PD-1 blockade (2). In addition to genetic alterations, epigenetic misregulation, another common feature of cancer, has emerged as a key regulator of tumor response to T-cell immunity and anti–PD-1 therapy. For instance, the MHC-I can be functionally impaired by transcriptional silencing mediated by the polycomb repressive complex 2 (PRC2) that catalyzes trimethylation of histone H3 lysine 27 (H3K27), which consequently attenuates tumor cell recognition by CD8⁺ T cells. Thus, inhibition of PRC2 has been reported to restore antitumor immunity in MHC-I–low tumor models (3, 4). Recent studies have also identified DNA methyltransferases (DNMT), LSD1 (H3K4me1/2 demethylase), and EZH2 (H3K27 methyltransferase) as suppressors of tumor responses to PD-1 blockade, and their individual inhibition can elicit tumor cell–intrinsic innate immune activation, thereby inflaming poorly immunogenic tumors with elevated T-cell infiltration and sensitizing tumors to anti–PD-1 therapy (5–9). Together, these studies point to a promising therapeutic approach that uses epigenetic drugs to sensitize refractory tumors to anti–PD-1 treatment.

Given that epigenetic regulators are broadly involved in controlling transcription programs, when they are targeted in tumor cells for the purpose of eliciting immune stimulatory effects in combination with anti–PD-1, tumor cell–derived immunoregulatory cytokines, ligands, or metabolites may also be altered in a way that exerts opposing effects and limits the combinatory effectiveness (10). For instance, TGFβ, which exerts potent suppression on immune surveillance, is secreted by tumor cells, in addition to multiple types of producers such as stromal cells, to enable immune evasion (11). Among a variety of immunosuppressive mechanisms, TGFβ in the tumor microenvironment (TME) is known to inhibit the cytotoxicity of CD8⁺ T cells (12), block T-cell infiltration (13, 14), and promote generation of regulatory T cells (Treg; ref. 15). In some cancers, the TGFβ pathway is epigenetically silenced (16, 17); therefore, when epigenetic regulators are targeted in combination with anti–PD-1 for treating such cancers, the TGFβ pathway might be reactivated and compromise the combinatory effect. Hence, a careful scrutinization of confounding effects of targeting epigenetic regulators may be warranted for developing effective combination therapies.

Our recent study has identified LSD1 as an inhibitor of tumor response to host immunity and PD-1 blockade, and its inhibition greatly sensitizes the refractory mouse B16/F10 melanoma to anti–PD-1 therapy (8). However, this combination did not result in a complete eradication of established tumors, similar to other combinations targeting epigenetic regulators together with ICB (6, 18, 19). In this study, we investigated the underlying mechanisms and identified TGFβ induction as a side effect of LSD1 ablation in tumor cells, which suppresses CD8⁺ T-cell cytotoxicity and blocks T-cell infiltration in response to PD-1 blockade. Accordingly, concurrent depletion of LSD1 and TGFβ renders these refractory tumors superiorly responsive to anti–PD-1 treatment, accompanied by tumor rejection. Importantly, this study also suggests that targeting LSD1 may sensitize certain otherwise refractory tumors to the emerging dual PD-1/TGFβ blockade therapy, which potentially extends the utility of this triple combination to the treatment of poorly immunogenic tumors.

We recently reported that LSD1 inhibition enhances tumor immunogenicity through activating the dsRNA–IFN pathway, resulting in an elevated T-cell infiltration in poorly immunogenic tumors, which sensitizes tumors to PD-1 blockade therapy (8). Although LSD1-null melanoma growth was greatly inhibited in response to anti–PD-1 treatment, tumors were not eradicated (8). Unexpectedly, further analyses showed that the mean fluorescence intensity (MFI) of the cytotoxic molecule granzyme B (GzmB) in individual CD8⁺GzmB⁺ tumor-infiltrating lymphocytes (TIL) was moderately decreased in LSD1-null tumors (Supplementary Fig. S1A), although the percentages of GzmB⁺ cells remained unchanged (8). We therefore hypothesized that certain immune suppressive molecules originating from tumor cells might be concurrently induced when LSD1 was inhibited, and such molecules could suppress the cytotoxicity of CD8⁺ TILs and compromise the immunostimulatory effect of targeting LSD1. To search and identify such immunosuppressive molecules, we reexamined the transcriptomic data obtained from ex vivo tumors (8). The gene ontology (GO) analysis of the upregulated genes caused by LSD1 loss, however, did not successfully identify any candidate pathways of interest as significant hits (Supplementary Fig. S1B; Supplementary Table S1). We then interrogated a list of individual genes with well-documented roles in suppressing T-cell immunity directly or indirectly through dendritic cells (DC; refs. 10, 20, 21), with two filtering criteria: (i) gene transcripts were reliably detected in LSD1-deficient B16/F10 tumor cells (logCPM > 1) and (ii) gene expression was significantly induced by LSD1 ablation compared with control (FC > 1.5 and FDR < 0.05). This analysis uncovered two hits: the Tgfb subfamily and Csf1 (encoding MCSF; Supplementary Fig. S1C–S1E; Supplementary Table S2).

We next investigated whether and how TGFβ subfamily members were regulated by LSD1 in tumor cells. As expected, LSD1 ablation in in vitro cultured B16/F10 (B16 in short) cells strongly induced the TGFβ1 transcript level and also moderately increased TGFβ2 and TGFβ3 (Fig. 1A). This upregulation was largely suppressed when LSD1 was reintroduced into the cells (Fig. 1A). We further quantified the secreted TGFβ1 by ELISA and showed that B16 cells, though normally secreting a low amount of TGFβ1 (22), became potent TGFβ1 producers upon LSD1 ablation (Fig. 1B). The ablation of CoREST (encoded by Rcor1), a key component of the LSD1/CoREST complex, also led to the upregulation of Tgfb1 as well as Serpine1, a well-characterized TGFβ downstream target (Fig. 1C). Given that LSD1 ablation strongly activated type 1 IFN pathway (8), we also tested the effect of autocrine IFNs on TGFβ expression and found that IFNβ depletion did not attenuate TGFβ induction (Supplementary Fig. S2A and S2B). Instead, it is likely that LSD1 directly regulates TGFβ upregulation, as we found it bound to the promoter region of the Tgfb1 gene, and its ablation increased active histone marks (Supplementary Fig. S2C). The induction of TGFβ subfamily members in response to LSD1 ablation occurred in multiple types of cancer cells of both mouse and human origins, including melanoma, lung carcinoma, and breast cancer (Supplementary Fig. S2D–S2L), suggesting that TGFβ expression is generally repressed by LSD1 across different tumor types.

We next asked whether TGFβ upregulation affected IFN activation in LSD1-null tumor cells. We deleted all three Tgfb genes in parental and LSD1-null B16 cells by CRISPR/Cas9 to obtain TGFβ triple knockout (TKO) and LSD1/TGFβ quadruple knockout (QKO) cells, and confirmed the removal of functional alleles of Tgfb genes by sequencing, transcript detection, and protein quantification (Supplementary Fig. S3A–S3C). Serpine1, a reporter gene for TGFβ pathway activation, was significantly induced in LSD1-deficient B16 cells, but this induction was by and large abolished in LSD1/TGFβ QKO cells (Supplementary Fig. S3D), providing an additional line of evidence for the loss of TGFβ pathway activity in these cells. We found that LSD1 ablation–stimulated IFN activation was largely unaffected by the elevated TGFβ expression, as IFNs and IFN stimulated genes (ISG) showed comparable expression between LSD1 knockout (KO) and LSD1/TGFβ QKO B16 cells (Fig. 1D and E). Taken together, these results demonstrate that LSD1 ablation in tumor cells induces TGFβ expression, which does not interfere with the simultaneous IFN activation.

We next investigated the in vivo biological significance of the induced TGFβ expression in LSD1-depleted tumors. To this end, we transplanted four groups of B16 tumors (Scramble, LSD1 KO, TGFβ TKO, and LSD1/TGFβ QKO) into syngeneic, immunocompetent mice for tumor growth. Although abrogation of all three TGFβs alone had no observable effect on tumor growth, abrogation of TGFβs in LSD1-null tumors showed a moderate but statistically significant effect on restraining tumor growth, evidenced by the further delayed growth of LSD1/TGFβ QKO tumors compared with LSD1 KO tumors and further extended survival (Fig. 1F and 1G). Thus, in wild-type (WT) B16 tumors with a low basal level of TGFβ expression (22; Fig. 1H), tumor cell–derived TGFβs seemed to have negligible effects on overall tumor growth. However, when LSD1 was ablated in tumor cells, TGFβ1 secretion was induced to a significant level (Fig. 1H), which exerted an immunosuppressive effect and counteracted the antitumor effect of LSD1 inhibition. Importantly, there were no discernable growth differences between LSD1/TGFβ QKO and scramble tumors in the immunodeficient T-cell receptor α (TCRα) KO mice (Fig. 1I), suggesting that the antitumor effect resulting from the abrogation of TGFβ induction in LSD1-null tumors relies on T-cell immunity. To further determine the roles of CD8⁺ versus CD4⁺ T cells in mediating the antitumor effect of LSD1/TGFβ ablation, we performed antibody-mediated depletion of either CD8⁺ or CD4⁺ T cells. We found that the growth difference between LSD1/TGFβ QKO and scramble tumors was largely diminished by the depletion of CD8⁺ but not CD4⁺ T cells (Fig. 1J), suggesting that it is mainly CD8⁺ T cells that mediate the antitumor effect of TGFβ removal in LSD1-null tumors.

In addition to evaluating the subcutaneous tumor growth model, we also evaluated the role of TGFβ induction in a lung metastasis model because TGFβs are well documented to promote tumor cell epithelial–mesenchymal transition (EMT), invasion, and metastasis (11). B16 cells were intravenously injected into mice to allow the circulation of tumor cells into lungs. The analysis of lung colonization by tumors showed that LSD1 loss inhibited B16 tumor metastasis only in WT but not TCRα KO mice (Supplementary Fig. S4A and S4B), suggesting that LSD1 ablation–elicited T-cell immunity is critical in controlling tumor metastasis. This observation is opposite to a previous report where LSD1 inhibition has been shown to increase breast cancer metastasis (23). The difference could be due to the use of xenograft (23) versus immunocompetent syngeneic models (this study) or due to different tumor types. The concurrent abrogation of TGFβs in LSD1-null tumors imparted further inhibitory effect on lung metastasis, evidenced by the observation that LSD1/TGFβ QKO tumors showed the least lung metastasis compared with either LSD1 KO or TGFβ TKO tumors (Supplementary Fig. S4A). Importantly, the cooperative effect between concurrent LSD1 and TGFβ ablation on controlling tumor lung metastasis also depended on T-cell immunity, because the absence of T cells completely abrogated this effect (Supplementary Fig. S4B). Of note, TGFβ ablation alone modestly suppressed tumor lung metastasis, and this inhibitory effect was observed only when T cells were present (Supplementary Fig. S4A and S4B). In summary, these findings suggest that tumor cell–derived TGFβs, unintentionally induced when LSD1 is removed in tumor cells, may compromise the immunostimulatory effect of LSD1 inhibition. Thus, blockade of TGFβs could potentiate the antitumor effect of LSD1 inhibition in certain tumors.

To understand how the TGFβ induction compromised the antitumor CD8⁺ T-cell immunity elicited by LSD1 ablation, we analyzed lymphocyte infiltration in the TME. As previously reported (8), LSD1 ablation strongly increased CD8⁺ T-cell infiltration, whereas the concurrent ablation of TGFβs in LSD1-null tumors resulted in no further increase (Fig. 2A). This was expected in some way, because IFN pathway activation, which is essential for LSD1 ablation–stimulated CD8⁺ T-cell infiltration (8), was comparable between LSD1 KO and LSD1/TGFβ QKO tumor cells (Fig. 1D). Other lymphocytes were also by and large numerically comparable between LSD1 KO and LSD1/TGFβ QKO tumors (Supplementary Fig. S5A–S5E), as well as in tumor-draining lymph nodes (TdLN) of tumor-bearing mice (Supplementary Fig. S6A–S6H). Of note, although TGFβ plays a well-documented role in promoting Treg cells (24), we did not detect an obvious alteration in Treg cell frequency when TGFβ levels were fluctuating in the TME (Supplementary Fig. S5D; Fig. 1H). This phenomenon is not unique to our tumor model, as previous studies also reported that TGFβ blockade by neutralizing antibodies did not change Treg cells in the TME (14). Thus, the secreted TGFβs by LSD1-null B16 tumors counteract the antitumor immunity in a way other than altering T-cell infiltration.

We next assessed the functional profile of CD8⁺ TILs, which the induced TGFβs might influence. The proliferation of CD8⁺ TILs was analyzed by measuring the expression of Ki-67, a cell proliferation marker, which appeared to be unaltered by either LSD1 ablation or concurrent TGFβ ablation (Fig. 2B). In addition, CD8⁺ TILs isolated from scramble, LSD1 KO, and LSD1/TGFβ QKO B16 tumors had comparable PD-1 expression (Fig. 2C), indicating similar levels of T-cell activation. In contrast, we found that the percentage of GzmB-expressing cells among CD8⁺ TILs was significantly increased in LSD1/TGFβ QKO tumors compared with those in scramble or LSD1 KO tumors (Fig. 2D). Importantly, the GzmB protein level (measured by MFI) in individual CD8⁺GzmB⁺ TILs, which was suppressed in LSD1 KO tumors, was restored when tumor cell–derived TGFβs were concurrently abrogated (Fig. 2E). As a control, TGFβ ablation alone in B16 tumor cells had no overt effect on CD8⁺ T-cell infiltration, proliferation, or GzmB expression (Supplementary Fig. S7A–S7D), in line with the unaltered tumor growth of TGFβ TKO tumors (Fig. 1F).

To further characterize CD8⁺ TILs in the TME of genetically modified B16 tumors, we used RNA sequencing (RNA-seq) to interrogate their transcriptional phenotype. LSD1 ablation in B16 tumor cells appeared to have a minor effect on the gene expression profile of CD8⁺ TILs, and we found only 25 and 10 genes downregulated and upregulated, respectively, in CD8⁺ TILs from LSD1 KO tumors compared with those from scramble tumors (Supplementary Fig. S8A; Fig. 2F). When Tgfb genes were deleted on top of LSD1 KO in B16 tumors, more genes were significantly upregulated in CD8⁺ TILs (119 genes for QKO vs. scramble or 99 genes for QKO vs. LSD1 KO alone; Fig. 2F). As a control, TGFβ ablation alone only elevated the expression of 20 genes (Fig. 2F). These data suggest that ablation of LSD1 and TGFβ in B16 tumors cooperatively induces gene expression in CD8⁺ TILs, which was supported by the result obtained from comparing the expression of a list of genes among all samples (Fig. 2G). The GO enrichment analysis revealed that the upregulated genes (QKO vs. KO) were significantly enriched in GO terms related to innate immune response and granzyme activity (Fig. 2H), and a list of representative genes are shown in Supplementary Fig. S8B. Consistently, the analysis of a panel of T-cell cytotoxicity-related molecules showed that the transcripts of several granzyme genes, the expression of which appeared to be suppressed in LSD1 KO tumors, were elevated by the concurrent ablation of LSD1 and TGFβ (Fig. 2I). The expression of FASL, IFNg, and TNF by CD8⁺ TILs, however, did not appear to be much affected by LSD1 or/and TGFβ depletion in tumors (Fig. 2I–2K). These data thus demonstrate that granzyme expression by CD8⁺ TILs is selectively and transcriptionally suppressed in LSD1 KO tumors as a result of TGFβ induction.

Additionally, we analyzed the abundance and phenotype of myeloid cells in the TME, which could also be affected by TGFβ induction. We found that the abundances of DCs and macrophages were comparable between LSD1 KO and LSD1/TGFβ QKO tumors, whereas the Gr1^hi myeloid-derived suppressor cells (MDSC) showed a moderately higher abundance in LSD1/TGFβ QKO than LSD1 KO tumors (Supplementary Fig. S9A–S9E). Notably, the modest suppression of CD86 expression in two subsets of DCs and the upregulation of arginase expression in MDSCs detected in LSD1-null tumors can be restored by simultaneous TGFβ ablation (Supplementary Fig. S9F–S9I), implicating an immunosuppressive effect of TGFβ induction through targeting innate immune cells. Collectively, these results suggest that TGFβ induction in LSD1-null tumors compromises the antitumor effect of CD8⁺ TILs through suppressing their cytotoxicity, which might involve a direct effect of TGFβ on T cells and/or an indirect effect mediated by innate immune cells.

To determine whether T cells were the primary targets of tumor cell–derived TGFβs in LSD1-null tumors, we specifically blocked the TGFβ pathway in T cells using CD4-dnTGFBRII transgenic mice, which express a dominant-negative TGFβ receptor (dnTβRII) on αβ T cells that attenuates T-cell response to TGFβ signals (25). We speculated that if LSD1 ablation–induced TGFβs act on T cells, the growth difference between LSD1 KO and LSD1/TGFβ QKO tumors observed in WT mice should be diminished when they were implanted in the dnTβRII transgenic mice. We found when LSD1 KO tumors were implanted in dnTβRII transgenic mice, their growth was significantly suppressed to a level even slower than that of LSD1/TGFβ QKO tumors (Fig. 3A). Importantly, the expression of GzmB by CD8⁺ TILs in LSD1 KO tumors was increased by the presence of dnTβRII to a similar level as those in LSD1/TGFβ QKO tumors, whereas GzmB expression by CD8⁺ TILs in scramble and LSD1/TGFβ QKO tumors remained largely unchanged (Fig. 3B and 3C). Of note, scramble and LSD1/TGFβ QKO tumors with low levels of TGFβ expression also responded to the blockade of TGFβ receptor signaling in T cells by exhibiting reduced tumor growth (Fig. 3A), which was possibly due to the significant reduction of Treg cells (Fig. 3D and 3E), considering the critical role of the TGFβ pathway in Treg induction and maintenance (24). Thus, these results suggest that TGFβ induction in response to LSD1 loss in tumor cells acts on αβ T cells and ultimately suppresses the cytotoxicity of CD8⁺ TILs, although the detailed underlying mechanism remains to be further explored.

TGFβ subfamily proteins are known to play a tumor cell growth–inhibitory role by suppressing cell-cycle progression, promoting differentiation and inducing apoptosis (11). Therefore, the slower growth of LSD1 KO tumors compared with LSD1/TGFβ QKO tumors in dnTβRII transgenic mice, where TGFβs' effect on T cells is blocked, could be due to a growth-inhibitory effect of TGFβs on B16 tumor cells. Indeed, we found that the B16 cell growth inhibition as a result of LSD1 KO was partially rescued by the concurrent ablation of Tgfb genes (Supplementary Fig. S10A), and the addition of exogenous TGFβ1 to the cultured LSD1/TGFβ QKO cells reinstalled the growth inhibition (Supplementary Fig. S10B). To identify the mechanism by which TGFβ upregulation inhibited LSD1 KO B16 cell growth, we analyzed cell-cycle progression and apoptosis. We detected a modest but statistically significant G1 arrest in LSD1 KO cells, which was rescued by the concurrent TGFβ deletion (Supplementary Fig. S10C and S10D). In addition, annexin V staining was elevated in LSD1 KO but not LSD1/TGFβ QKO cells (Supplementary Fig. S10E and S10F). These results demonstrate that TGFβ upregulation caused by LSD1 ablation inhibits in vitro B16 cell growth at least in part through suppressing cell-cycle progression and promoting apoptosis.

To further confirm the growth-inhibitory effect of TGFβs on LSD1 KO B16 tumor cells in vivo, we ectopically expressed a TGFβ dominant-negative receptor (dnTGFβRII) on the surface of Lsd1 KO B16 cells (Fig. 3F), which did not interfere with TGFβ expression but rendered B16 cells unresponsive to TGFβ signals (Fig. 3G). When implanted in WT mice, LSD1 KO tumors expressing dnTGFβRII grew significantly faster than those responsive to TGFβ signals (Fig. 3H). Thus, TGFβs induced by LSD1 ablation in B16 tumors play at least two opposing roles: primarily, paracrine TGFβs play a protumor role by acting on T cells and suppressing CD8⁺ T-cell cytotoxicity, which compromises the antitumor effect of LSD1 loss; secondarily, autocrine TGFβs play an antitumor role by acting on tumor cells and inhibiting cell survival/proliferation.

Recent studies have identified TGFβ as a critical inhibitor of tumor response to anti–PD-1 therapy mainly through expelling T-cell infiltration (13, 14). In the poorly immunogenic B16 tumor model, neither TGFβ deletion alone nor their deletion on top of LSD1 loss altered T-cell infiltration in the TME (Fig. 2A; Supplementary Fig. S7A); instead, a critical role of TGFβ lies in suppressing CD8⁺ T-cell cytotoxicity, which might be responsible for the incomplete response of LSD1-null tumors to PD-1 blockade. To test this possibility, we examined the response of LSD1/TGFβ QKO tumors to anti–PD-1 treatment and found that those tumors showed profound growth control in response to PD-1 blockade and, excitingly, approximately 44% of QKO tumors were rejected by PD-1 blockade without recurrence up to 60 days (Fig. 4A and 4B). This is in stark contrast to our previous report that LSD1-null B16 tumors, though reduced, could not be completely rejected by anti–PD-1 treatment (8).

To determine whether mice that rejected LSD1/TGFβ QKO tumors upon anti–PD-1 treatment had developed immunity, surviving mice were rechallenged with WT B16 tumors on the left hind flank 60 days after clearance of the primary tumors. Although naïve, age-matched control mice all grew tumors and reached the endpoint within 30 days, 5 out of 6 mice cured by LSD1/TGFβ ablation plus anti–PD-1 treatment rejected secondary B16 tumors (Supplementary Fig. S11A and S11B). Mice that survived B16 rechallenge were subsequently challenged with the irrelevant MC38 colorectal adenocarcinoma on the right front flank 60 days after secondary tumor clearance. In contrast to the rechallenge of B16 tumors, MC38 tumors grew similarly in those mice compared with their counterparts in the naïve, age-matched mice (Supplementary Fig. S11C). These results suggest that tumor-bearing mice cured by LSD1/TGFβ ablation plus anti–PD-1 treatment may have developed immunologic memory.

Next, we sought to substitute genetic perturbation of TGFβ with systemic delivery of TGFβ-blocking antibodies. Although anti–PD-1 already suppressed LSD1-null tumor growth, the inclusion of anti-TGFβ in the combination exerted additional inhibitory effect on tumor growth, leading to the rejection of 2 out of 9 tumors (Fig. 4C). Notably, although a single treatment with anti–PD-1 or anti-TGFβ had no overt effect on scramble B16 tumors, their combination gave rise to a synergistic effect on restraining tumor growth (Fig. 4C). To explore the generalizability of our study, we performed a similar tumor growth experiment in a D4m.3A melanoma model and found that, most impressively, the combination of anti-TGFβ and anti–PD-1 drastically increased the rejection rate of LSD1-null tumors compared with anti–PD-1 alone (80% vs. 20%), whereas this combination had no obvious effect on scramble tumor growth (Fig. 4D and 4E). These results demonstrate an inhibitory role of TGFβs that compromises a full response of LSD1-null tumors to anti–PD-1 treatment, highlighting the importance of blocking the TGFβ pathway when LSD1 is targeted in this combination immunotherapy.

To further clarify the mechanistic roles of LSD1 ablation, TGFβ blockade, and PD-1 blockade in their cooperative effect on tumor growth control, we analyzed the immune phenotype of B16 tumors in response to single or different combinatory treatments. We found that LSD1 loss played a major part in elevating CD8⁺ T-cell infiltration, as did the combination of PD-1 and TGFβ blockade, but undesirably compromised the cytotoxicity (GzmB MFI) of CD8⁺ TILs, even when PD-1–blocking antibodies were applied (Fig. 4F–4H). In line with the genetic data (Supplementary Fig. S7A, S7C, and S7D; Fig. 2E), TGFβ blockade alone had no overt effect on either T-cell infiltration or CD8⁺ T-cell cytotoxicity in scramble B16 tumors, but helped restore the cytotoxicity (GzmB MFI) of CD8⁺ TILs in LSD1-null tumors (Fig. 4F and 4H). PD-1 blockade alone significantly increased the cytotoxicity (GzmB⁺ percentage) of CD8⁺ TILs, but it had a limited effect on T-cell infiltration in our experimental settings (Fig. 4F and 4G). Importantly, the triple combination of LSD1 ablation, PD-1 blockade, and TGFβ blockade cooperatively elevated T-cell infiltration while unleashing CD8⁺ T-cell cytotoxicity (Fig. 4H,Figure 4), leading to the improvement of tumor rejection. With respect to T-cell infiltration, Lsd1 ablation likely worked through inflaming poorly immunogenic tumors for better T-cell recruitment, since it did not affect the proliferative response of CD8⁺ TILs to dual PD-1/TGFβ blockade (Fig. 4I). As a control, the frequencies of Treg cells were found to be mostly comparable between different groups of treatment (Fig. 4J). TGFβ inhibition by genetic depletion seemed to be more effective than blocking antibodies on unleashing T-cell cytotoxicity when applied together with the PD-1–blocking antibodies (Fig. 4K and 4L). In summary, LSD1 ablation, TGFβ blockade, and PD-1 blockade cooperatively potentiate both T-cell infiltration and cytotoxicity that enables the eradication of certain poorly immunogenic tumors.

The combination of epigenetic therapy and immunotherapy represents an emerging approach for cancer treatment (26). Our present study shows that, when the histone demethylase LSD1 is targeted for promoting T-cell infiltration in the poorly immunogenic B16 and D4m.3A tumors, the immunosuppressive TGFβs are concurrently induced to a significant level that suppresses CD8⁺ T-cell cytotoxicity and also impedes the combinatory effect of LSD1 ablation and PD-1 blockade through restraining both CD8⁺ T-cell cytotoxicity and infiltration. Importantly, LSD1 depletion appears to enable and significantly improve the response of certain poorly immunogenic tumors to dual PD-1/TGFβ blockade, which is currently being explored as a new combination treatment strategy (27). Our findings thus raise a promising strategy that combines LSD1 inhibition with the blockade of TGFβ and PD-1 for treating certain poorly immunogenic or “cold” tumors.

The regulatory role of LSD1 in TGFβ expression was first uncovered a decade ago, when LSD1 downregulation, which was reported in breast carcinomas, led to the upregulation of TGFβ and promoted breast cancer metastasis in xenograft tumor models (23). However, the biological significance of this initial report was not fully investigated. In fact, LSD1 is commonly upregulated in various cancer types and considered to mostly play a tumor-promoting role through multiple mechanisms including maintaining cancer stem cell renewal, sustaining tumor cell proliferation, and suppressing antitumor immunity (28), which makes LSD1 a promising target for cancer treatment. Our study identifying a suppressive effect of LSD1 on TGFβ expression is consistent with this previous report, and we have further extended this regulation to multiple types of tumor cells. More importantly, our study using syngeneic tumor models has investigated an unexplored role of this regulation, in which LSD1 inhibition-induced TGFβs suppress CD8⁺ T-cell cytotoxicity in the TME and consequently counteract the antitumor effect of LSD1 inhibition–stimulated CD8⁺ T-cell infiltration. The TGFβ signaling pathway in αβ T cells was involved in mediating the immune-suppressive effect of tumor cell–derived TGFβs. Although a direct inhibitory effect of TGFβs on CD8⁺ T cells has been previously reported (12), we could not rule out the possibility that CD4⁺ T cells might also be involved in mediating the suppression of CD8⁺ T-cell cytotoxicity in response to the elevated TGFβs. Meanwhile, we also observed an antitumor effect of LSD1 inhibition–induced TGFβs accomplished through inhibiting tumor cell survival/proliferation (Supplementary Fig. S10). This is not surprising, because TGFβ is a pleiotropic cytokine known to play both antitumor and protumor roles depending on tumor types, tumor stages, and genetic and epigenetic properties, as well as the TME (11, 29). Under the circumstance of targeting LSD1 for controlling B16 tumor growth, the protumor effect of the induced TGFβs outweighed its antitumor effect (Fig. 1F), highlighting the importance of blocking TGFβ in conjunction with LSD1 inhibition.

The development of cancer therapies by targeting the TGFβ pathway has faced drawbacks, partly because of the antitumor role of the TGFβ pathway in certain tumor contexts, which is supported by long-standing evidence (11), as well as safety issues observed in previous clinical trials (30). The fact that the TGFβ pathway plays a significant part in suppressing antitumor immunity points to a therapeutic approach combining TGFβ pathway inhibition and immunotherapy that has gained renewed attention since ICB was demonstrated to be successful. The combinatorial effect of TGFβ pathway inhibition and PD-1/PD-L1 blockade varies across a range of reported syngeneic tumor models. In immunogenic but “immune-excluded” tumor models including EMT6 and MC38, where TGFβ signaling in fibroblasts is thought to be a key barrier for T-cell penetration into tumor parenchyma, cotargeting PD-1/PD-L1 and TGFβ has shown superior tumor eradication (14, 31). However, in poorly immunogenic tumor models such as KPC1, TC1, and 4T1, there is a lack of persuasive combinatorial effect of anti–PD-1/PD-L1 and LY364947 (a TGFBRI inhibitor; ref. 22) as well as between anti–PD-1 and anti-TGFβ in the absence of vaccination (32) or radiation therapy (33), presumably due to low mutational burdens or the absence of T-cell recognition. Indeed, in our study, the poorly immunogenic B16/F10 melanoma showed a moderate response and D4m.3A melanoma showed no response to anti-TGFβ/anti–PD-1 treatment (Fig. 4C and 4D). The lack of a robust response could be overcome by including LSD1 inhibition in the combination (Fig. 4C and 4D) because this triple combination can potentiate both T-cell infiltration and cytotoxicity.

Our study centers on identifying immunosuppressive mediators that are unintentionally induced by LSD1 inhibition, which then counteract the immunostimulatory effect of LSD1 inhibition and impede the synergistic effect of combining LSD1 inhibition and PD-1 blockade in tumor control. The findings from this study illustrate a duality of epigenetic perturbations in the combination with ICB therapy, highlighting the importance of thorough examinations of transcriptional programs and biological processes regulated by epigenetic regulators under investigation for the purpose of provoking immune responses in combination immunotherapy. Therefore, our findings may be relevant when small molecules targeting other epigenetic regulators are used in immunotherapy. In summary, our findings suggest a triple combination strategy that combines LSD1 inhibition, TGFβ blockade, and PD-1 blockade for treating certain ICB-refractory tumors. Extensive future studies are warranted to discover biomarkers for tumors that respond to this combination therapy in order to further develop this new cancer treatment strategy.

All cancer cell lines were cultured in DMEM growth medium (Gibco, cat#1195) supplemented with 10% heat-inactivated FBS (Gemini Bio, cat#900–108) and 1% penicillin/streptomycin (Gibco, cat#15140122) in a 5% CO₂ incubator at 37°C. B16/F10, D4m.3A, and MC38 cell lines were from Drs. Glenn Dranoff (Novartis), David Fisher (MGH), and Arlene Sharpe (HMS), respectively. These cell lines were confirmed to be Mycoplasma-free by the Universal Mycoplasma Detection Kit (ATCC 30–1012K) according to the manufacturer's instructions. T-cell stimulation were performed in R10 medium (RPMI-1640 supplemented with 10% FBS, 1% penicillin/streptomycin, 12 mmol/L HEPES, and 50 μmol/L β-mercaptoethanol).

The guide RNA (gRNA) oligos targeting mouse Tgfb1, Tgfb2, Tgfb3, and Rcor1 (encoding CoREST; sequences listed in Supplementary Table S3) were annealed and cloned into a lenti-CRISPR-v2-Puromycin⁺ vector (Addgene, cat#52961), respectively. Lentivirus carrying lenti-CRISPR plasmid was prepared by cotransfecting HEK293T cells with four helper plasmids (pHDM-VSV-G, pHDM-tat1b, pHDM-HgPM2, and pRC-CMVRaII), followed by viral supernatant collection after 72 hours. To delete three Tgfb genes, WT or Lsd1 KO (clone g5-4) B16 cells were transduced with a mixture of lenti-CRISPR virus carrying respective gRNA with the addition of 8 μg/mL polybrene (Sigma-Aldrich, cat#H9268) and selected with 1 μg/mL puromycin for 2 days. Cells were then transferred into puromycin-free fresh medium and seeded at low density to allow colony formation from single cells. Colonies were then picked and expanded for knockout validation of all three Tgfb genes by sequencing of target genomic regions or ELISA.

For rescue assay, Lsd1 KO B16 cells were transduced with lentiviral pHAGE-CMV-Flag-HA-LSD1 and selected with puromycin for 2 weeks to obtain a stable cell line. To establish stable cell lines ectopically expressing dominant-negative TGFβ receptor (dnTGFβRII), cDNAs were amplified from pCMV5-HA-TBRII (delta Cyt; Addgene, cat#14051) and cloned into MSCV-PIG (Puro IRES GFP) retroviral vector (Addgene, cat#18751). Retrovirus carrying MSCV-PIG was used to transduce Lsd1 KO B16 cells, followed by puromycin selection for 2 weeks to obtain a stable cell line.

For total RNA extraction, supernatant was removed and cells were lysed by directly applying TRIzol (Life Technologies, cat#15596018) onto cells. RNA extraction was performed according to the manufacturer's instructions. The extracted RNA was reverse-transcribed into cDNA using the PrimeScript RT Reagent Kit (TaKaRa, cat#RR037B). Briefly, RNA samples were first mixed with random 6-mers and oligo dT primers and denatured at 70°C for 5 minutes and cooled down on ice, followed by the addition of buffer and reverse transcriptase. The reaction was then incubated at 37°C for 30 minutes for reverse transcription and terminated at 85°C for 15 seconds. The obtained cDNA samples were diluted in H₂O and used for real-time quantitative PCR (qPCR). SYBR green (Life Technologies, cat#A25743) and gene-specific primers (listed in Supplementary Table S3) were used for PCR amplification and detection on a QuantStudio 3 real-time PCR system (Applied Biosystems). The qPCR data were normalized to GAPDH and presented as fold changes of gene expression in the test sample compared with the control.

The ELISA assay was performed with a Mouse TGFβ1 DuoSet ELISA Kit (R&D Systems, cat#DY1679–05) according to the manufacturer's instructions. For measuring TGFβ1 secreted by cultured cells, cells were cultured for 24 hours and changed to FBS-free medium for an additional 24 hours, followed by supernatant collection and clearance by centrifugation. For measuring TGFβ1 protein level in implanted tumors, tumor masses were excised and snap-frozen in liquid nitrogen. Tumor samples were then homogenized in protein extraction buffer (100 mmol/L Tris pH 7.4, 150 mmol/L NaCl, 1 mmol/L EGTA, 1 mmol/L EDTA, 1% Triton X-100, 0.5% sodium deoxycholate and protease inhibitor cocktail) and incubated at 4°C on a rotator for 30 minutes, followed by centrifugation to clear the lysate. Protein concentration in the lysate was quantified by Bio-Rad protein assay (Bio-Rad, cat #5000006), to which TGFβ1 concentration was normalized.

B16 cells in culture at 80% confluence were dissociated into single-cell suspension with trypsin, washed, and resuspended into fresh medium. The numbers of viable cells were counted and diluted appropriately for seeding on 12-well plates (200 cells per well). Cells were allowed to grow for 6 days, with an addition of fresh medium on day 3 without removing old medium. After that, the medium was removed and cells were stained with crystal violet solution (0.5% w/v crystal violet powder, 80% v/v H₂O and 20% v/v methanol).

B16 cells were harvested by trypsinization and fixed with 70% ethanol overnight. After fixation, cells were resuspended in 1X PBS and permeabilized with 0.1% Triton X-100. Cells were then treated with RNase A (0.8 mg/mL, Thermo Fisher Scientific, cat#RN0531), stained with propidium iodide (Thermo Fisher Scientific, cat#P3566), and analyzed on a BD LSRII. The cell-cycle profiles were analyzed to determine the fraction of cells in G1, S, and G₂–M phase using FlowJo.

Female WT C57BL/6 mice aged 6 to 10 weeks were purchased from The Jackson Laboratory and allowed to acclimate to housing conditions at the Boston Children's Hospital Animal Facility for 1 week. B6.129S2-Tcratm1Mom/J (TCRα KO) mice were originally purchased from The Jackson Laboratory (stock #002116) and bred at the Boston Children's Hospital Animal Facility. For experiments with B6.Cg-Tg(Cd4-TGFBR2)16Flv/J mice, those mice originally purchased from The Jackson Laboratory (stock #005551) were bred at the Boston Children's Hospital Animal Facility, and both female and male mice were used. Mice were anesthetized with isoflurane, shaved at right hind flank, and injected subcutaneously with 250,000 or 500,000 B16/F10 or D4m.3A tumor cells. Tumors were measured with a caliper every 2 to 3 days once palpable (long diameter and short diameter). Tumor volumes were calculated using the volume formula for an ellipsoid: 1/2 × D × d² where D is the longer diameter and d is the shorter diameter. Mice were sacrificed when tumors reached 2,000 mm³ or upon ulceration/bleeding. All experimental mice were housed in specific pathogen–free conditions, and all animal procedures were performed in accordance with animal care guidelines and with the prior approval by the Boston Children's Hospital Institutional Animal Care and Use Committee.

For antibody treatments, mice were administered 100 μg antibody via intraperitoneal injection initiated based on a set day or set tumor volume and continued every 2 to 3 days as indicated. The following antibodies from BioXCell were used: anti–PD-1 (clone 29F.1A12), anti-TGFβ (clone 1-D11), or rat IgG2a isotype control (clone 2A3). Prior to treatments, mice were randomized such that treatment groups had similar average tumor volumes prior to treatment initiation.

For CD4⁺ or CD8⁺ T-cell depletion, anti-CD4 depleting antibodies (clone GK1.5, BioXcell), anti-CD8α depleting antibodies (clone YTS169.4, BioXcell), or isotype control (clone LTF-2, BioXcell) were intraperitoneally injected into mice the day before tumor implantation at a dose of 200 μg, followed by injection of 100 μg every 3 days.

B16/F10 cells (200,000) with indicated genetic modifications were transferred intravenously into WT or TCRα KO female mice via tail- vein injection. Lungs were removed 14 days postinjection and fixed overnight in Fekete's solution. Visible metastases were counted in a blinded fashion.

Tumors were excised on day 14 post implantation and cut into 2-mm-sized pieces in type I collagenase (Worthington Biochemical Corporation, cat# LS004194) and DNase I (Sigma-Aldrich, cat#10104159001). Samples were then incubated at 37°C for 30 minutes and passed through a 70-μm cell strainer. To enrich leukocytes, samples were spun through a Percoll (GE Healthcare Life Sciences, cat# 17–0891–01) gradient. Leukocytes were collected from the interface of the 40% and 70% Percoll gradient, stained, and analyzed for fluorescent markers. For cytokine staining, enriched leukocytes were stimulated in R10 medium with PMA (50 ng/mL, Sigma-Aldrich, cat#P1585) and ionomycin (500 ng/mL, Sigma-Aldrich, cat#I0634) in the presence of GolgiPlug (BD Biosciences, cat#555029), followed by cellular surface staining and intracellular staining. The eBioscience Foxp3 Fixation/Permeabilization kit (cat# 00–5523–00) was used for intracellular staining. All antibodies were purchased from BioLegend, Thermo Fisher Scientific, R&D Systems or BD Biosciences: CD45.2 BV421 (clone 104), CD11b BV605 (clone M1/70), CD3e BV510 (clone 145–2C11), CD4 APC (clone RM4–5), CD8b APC-Cy7 (clone YTS156.7.7), CD8a BV605 (clone 53–6.7), Foxp3 PE (clone FJK-16s), Granzyme-B FITC (clone GB11), Ki-67 PerCP-Cy5.5 (clone B56), CD44 FITC (clone IM7), CD11c PE (clone N418), CD49b AF700 (clone DX5), Gr-1 APC-Cy7 (clone RB6–8C5), TNFα FITC (clone MP6-XT22), IFNγ PE (clone XMG1.2), PD-1 PE-Cy7 (clone 29F.1A12), CD86 AF700 (clone GL-1), CD103 FITC (clone 2E7), F4/80 PE-Cy7 (clone BM8), Arginase PE (clone A1exF5), CD64 FITC (clone X54–5/7.1), and human TGFBRII PE (clone 25508).

B16 cells in culture were fixed in 1% formaldehyde (Thermo Fisher Scientific, cat#28908) for 10 minutes at room temperature on a shaker, followed by quenching in 125 mmol/L glycine. Next, cells were washed twice with ice-cold 1X PBS and lysed in sonication buffer (50 mmol/L HEPES pH7.9, 140 mmol/L NaCl, 1 mmol/L EDTA, 1% Triton X-100, 0.1% sodium deoxycholate, 0.2% SDS) supplemented with protease inhibitor, then subjected to sonication to obtain DNA fragments mostly between 300 and 800 bp. Subsequent procedures were conducted by following the Epigenesys protocol (https://www.epigenesys.eu/en/). The following antibodies were used for immunoprecipitation: anti-LSD1 (Abcam, cat#ab17721), anti-H3K4me1 (Abcam, cat#ab8895), anti-H3K4me2 (EMD Millipore, cat#07–030), anti-H3K27ac (Active Motif, cat#39034), and rabbit normal IgG (Cell Signaling Technology, cat#2729).

CD8⁺ TILs were isolated from implanted B16 tumors on day 14 as described above. CD8⁺ TILs from two individual tumors were combined as a biological replicate. CD8⁺ TILs were stained with antibodies against CD45.2 (clone 104), TCRβ (clone H57–587), and CD8 (clone YTS156.7.7), and 7-AAD Viability Staining Solution (BioLegend, cat# 420404) was used to exclude dead cells. The CD45.2⁺TCRβ⁺CD8⁺ cells were then sorted on a BD Aria. Approximately 30,000 to 70,000 sorted cells for each biological replicate were directly lysed in 1 mL TRIzol Reagent (Life Technologies, cat#15596018). After incubation for 5 minutes, 0.2 mL chloroform was added and mixed by inverting the tubes several times. Samples were incubated for 2 to 3 minutes and later centrifuged for 15 minutes at 12,000 g at 4°C. The upper aqueous phase containing the RNA was collected and mixed with equal volume of 70% ethanol, which was then loaded into a spin column from a RNeasy Micro Kit (Qiagen, cat#74004) and subjected to RNA isolation according to the instruction manual. The on-column DNase digestion was conducted to eliminate DNA contamination.

Purified total RNA was quantified by Qubit (Invitrogen) and used for polyA⁺ RNA isolation with a Nebnext Poly(A) mRNA Magnetic Isolation Module (New England Biolabs, cat#E7490S) according to the manufacturer's instructions. The polyA⁺ enriched RNA was then used to generate a directional RNA library with a NEBNext Ultra II Directional RNA Library Prep Kit for Illumina (New England Biolabs, cat#E7760L) and NEBNext Multiplex Oligos for Illumina (New England Biolabs, cat#E7335L) according to the manufacturer's instructions. Library concentrations and quality were assessed on a Bioanalyzer and by qPCR. The library was sequenced at Novaseq (Illumina) to generate reads from paired ends (2 × 150 bp). The raw data were deposited at the Gene Expression Omnibus (GEO) under the subseries entry GSE161569.

The software hisat2 (version 2.1.0) was used to generate genome indices for mouse reference genome (GRCm38/mm10) and extract splice sites with gencode vM24 annotation. Next, the high-quality paired-end RNA-seq reads were aligned to mouse reference genome, and the consequence of alignment served as the input for featureCounts to quantify raw read counts for protein-coding genes. We used R package DESeq2 (version 1.38.0) to identify differentially expressed genes (DEG) between different groups. The raw read count per gene served as the input for DESeq2. Because three samples were collected in each condition, they were treated as biological replicates to improve the reliability of DEG identification. Statistical tests for differential expression were based on a model using the negative binomial distribution. The reported statistical significances were corrected for multiple testing using the Benjamini–Hochberg procedure with a false discovery rate less than 0.05. In addition, to be called DEGs we required the fold change to be greater than 1.5. The upregulated genes in the QKO condition versus the KO condition were queried to Gene Ontology Consortium for GO enrichment assessment, with specification of biological process.

Statistical analyses were performed using GraphPad Prism 8 software, and statistical significance was determined by P < 0.05. An unpaired Student t test was used for comparisons between two groups and a two-way ANOVA was used for multiple comparisons of tumor growth. For comparing mouse survival curves, a Log-rank (Mantel–Cox) test was used.

W. Sheng has a patent for “Methods of Treating Cancer Using LSD1 Inhibitors and/or TGF-Beta Inhibitors in Combination with Immunotherapy” pending. Y. Shi reports grants from NCI and grants from Ludwig Institute for Cancer Research during the conduct of the study; other from Constellation Pharmaceuticals, Inc; other from Athelas Therapeutics, Inc; other from Imago Biosciences; and other from Active Motif, Inc outside the submitted work. In addition, Y. Shi has a patent for “Methods of Treating Cancer Using LSD1 Inhibitors and/or TGF-Beta Inhibitors in Combination with Immunotherapy” pending. Y. Shi is a cofounder and equity holder of Constellation Pharmaceuticals, Inc., Athelas Therapeutics, Inc., and K36 Therapeutics, Inc. and holds equity of Imago Biosciences. Y. Shi is also a consultant for Active Motif, Inc. No disclosures were reported by the other authors.

W. Sheng: Conceptualization, data curation, formal analysis, investigation, methodology, writing–original draft. Y. Liu: Data curation, formal analysis, investigation, methodology, writing–original draft. D. Chakraborty: Investigation, writing–review and editing. B. Debo: Investigation, writing–review and editing. Y. Shi: Conceptualization, supervision, funding acquisition, writing–original draft.

We thank all the Shi lab members for discussion and suggestions. We thank Arlene Sharpe, David Fisher, and Glenn Dranoff for sharing cell lines. We thank the Department of Immunology's Flow Cytometry Facility (HMS) for assistance on flow cytometric analysis. This work was supported by funds from the NIH (R35 CA210104), Boston Children's Hospital, and Ludwig Institute for Cancer Research. Y. Shi is an American Cancer Society Research Professor.

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.