Intrinsic Immunogenicity of Small Cell Lung Carcinoma Revealed by Its Cellular Plasticity

Small cell lung carcinoma (SCLC) is defined as a high-grade neuroendocrine carcinoma arising in the lung, and virtually all tumors display loss of the tumor suppressor genes RB1 and TP53 (1). Although most patients with SCLC initially respond to cytotoxic chemotherapy, resistance inevitably emerges, and the 5-year survival rate is dismal (2, 3). The addition of immune checkpoint blockade (ICB) to first-line chemotherapy has yielded relatively modest benefit, and durable responses are rare, despite SCLC's high tumor mutation burden (4, 5).

The pathologic diagnosis of SCLC is classically made by morphology and a limited panel of immunohistochemical stains. Morphologic features include small- to intermediate-sized cells with high nuclear-to-cytoplasmic ratio, nuclear molding, and a distinctive nuclear appearance (finely granular chromatin lacking prominent nucleoli), which is associated with the neuroendocrine phenotype. Tumors are overtly malignant, displaying a very high proliferation rate, apoptosis, and necrosis. Neuroendocrine differentiation can be demonstrated by immunohistochemical staining for synaptophysin, chromogranin A, and insulinoma-associated protein 1 (INSM1; refs. 6, 7).

Although SCLC has been largely approached diagnostically and clinically as a monolithic entity, recent transcriptomic profiling of SCLC cell lines and patient samples has indicated significant intertumoral and intratumoral heterogeneity. Most broadly, SCLC can be divided into neuroendocrine and nonneuroendocrine subtypes, the former showing classic neuroendocrine morphology, in vitro growth in nonadherent tumor clusters, and high expression of the transcriptional regulators ASCL1 and NEUROD1. Nonneuroendocrine SCLC comprises approximately 10% to 15% of all SCLCs; tends to show adherent in vitro growth with a mesenchymal, nonneuroendocrine morphology (8); downregulates ASCL1 and terminal neuroendocrine markers, such as chromogranin A and synaptophysin; and may upregulate transcription factors such as POU2F3, YAP1, or c-MYC (7, 9–12). In addition, loss of ASCL1 has also been associated with upregulation of AXL (11, 13, 14). Also, nonneuroendocrine SCLC cells have been shown to derepress specific endogenous retroviruses as well as STING, promoting intrinsic innate immune signaling (13). Furthermore, experiments in genetically engineered mouse models of SCLC and profiling of human SCLC have also shown the coexistence of neuroendocrine and nonneuroendocrine cells within tumors, as well as cellular plasticity sufficient to allow conversion between these subtypes (7, 12, 13, 15, 16).

The presentation of peptides on MHC I molecules to CD8⁺ T cells serves as a crucial mechanism of immunosurveillance of the intracellular compartment of all nucleated cells (17). Downregulation of MHC I expression and consequently decreased antigen presentation is a major mechanism of immune escape employed by intracellular pathogens, as well as by malignant cells. In tumor cells, alteration of the MHC I pathway can occur via genetic or epigenetic means and can affect any component of the MHC I antigen processing and presentation machinery (APM), from the immunoproteasome, to the import and loading of peptides on MHC I in the endoplasmic reticulum via the TAP complex, to the export of MHC I–peptide complexes to the plasma membrane (18, 19). Although tumors of different histologic subtypes have been shown to acquire genetic lesions of MHC I APM, usually as an adaptive response to immune system pressure in the setting of immunotherapy, SCLC has been known to be poorly immunogenic, displaying intrinsically low levels of MHC I (20–22), which has recently been suggested to be the consequence of epigenetic programming (23). However, the intersection of MHC I antigen presentation and SCLC heterogeneity has not been explored.

Here, we profile the cellular and immunologic heterogeneity of SCLC and discover that nonneuroendocrine SCLC derepresses MHC I antigen presentation, which is sufficient to evoke CD8⁺ T cell–mediated rejection. Patients harboring MHC I^hi SCLC show long-term durable responses to ICB, thus nominating MHC I as a novel biomarker of durable ICB response in this treatment-refractory tumor. Further, we demonstrate that the epigenetic regulator EZH2 can govern the transition to the nonneuroendocrine phenotype and that its transient inhibition can prime immunogenicity in vivo.

Although SCLC lacks effective MHC I antigen presentation (20–22), the spectrum of MHC I expression across SCLC and its relationship to tumor heterogeneity has not been characterized. We therefore profiled a large set of de novo primary SCLC patient samples (n = 102, Supplementary Table S1.1) using standard chromogenic IHC as well as multiplexed immunofluorescence (mIF). Most (H-score ≤30; n = 72, 71%) SCLCs displayed little to no MHC I (MHC I^lo/neg); however, 15% (H-score ≥140, n = 15) of cases showed uniformly high MHC I expression (MHC I^hi; Fig. 1A; Supplementary Table S1.2). Interestingly, the appearance of MHC I^hi tumors varied from that of classic neuroendocrine SCLC, exhibiting a more elongated morphology, variably prominent nucleoli, increased cytoplasm (decreased nuclear/cytoplasmic ratio), and decreased organoid architecture (Fig. 1B, Supplementary Fig. S1A). Careful reexamination of MHC I^lo/neg tumors also uncovered multiple cases with focal clusters of tumor cells that expressed high levels of MHC I (MHC I^foc) and showed morphology similar to that seen in MHC I^hi SCLC (Supplementary Fig. S1A and S1B).

Bulk RNA sequencing of a small subset of these MHC I^hi versus MHC I^lo/neg samples revealed that MHC I^hi SCLC upregulated a diverse array of MHC I antigen presentation machinery genes, including TAP1 and NLRC5, and also identified strong enrichment of genes associated with IFN signaling and epithelial-to-mesenchymal transition (EMT; Fig. 1C; Supplementary Table S1.3). In contrast to other previously identified markers of nonneuroendocrine SCLC such as POU2F3, YAP1, MYC, or vascular endothelial (VE)–cadherin (7, 12, 24), the MHC I^hi phenotype was more robustly associated with upregulation of the EMT marker AXL (Fig. 1D; Supplementary Fig. S1C; Supplementary Table S1.3), which we previously linked to an innate immune active state in SCLC (13). Using mIF, we confirmed that MHC I^hi SCLC tumors exhibited RB loss in all cases (n = 15/15) and showed INSM1 expression in most cases (n = 12/14, 1 case not quantified; Supplementary Fig. S1D; Supplementary Table S1.4), consistent with the given diagnoses of SCLC. Examination of protein expression of ASCL1 and chromogranin showed that MHC I^hi SCLC significantly downregulated both markers relative to MHC I^lo/neg tumors (Fig. 1E; Supplementary Fig. S1E; Supplementary Table S1.4). Taken together, these findings reveal that, although the MHC I^hi state retains some neuroendocrine features, it is significantly associated with downregulation of the key SCLC lineage oncogene, ASCL1, consistent with our morphologic findings that varied from prototypical SCLC.

As an increase in MHC I antigen presentation could affect the immune tumor microenvironment, we spatially characterized immune infiltration in MHC I^hi SCLC and MHC I^lo/neg SCLC using a multiplexed immunofluorescence assay. MHC I^hi SCLC was significantly more infiltrated by CD45⁺/PD-L1⁺ immune cells in intratumoral regions, as well as at the tumor/stromal edge, than MHC I^lo/neg SCLC. Furthermore, MHC I^hi SCLC displayed greater intratumoral infiltration of CD3⁺ T cells (Fig. 1F; Supplementary Fig. S1F), and CD8A was also identified as one of the top genes enriched in MHC I^hi tumors in our RNA sequencing (RNA-seq) analysis (rank 167 out of 16,189 genes; Supplementary Table S1.3), indicating enrichment for cytotoxic T cells. Although PD-L1 expression was statistically significantly increased in MHC I^hi SCLC cells, the absolute expression level was low compared with CD45⁺/PD-L1⁺ cells (Fig. 1F; Supplementary Fig. S1F), congruent with previous reports of low PD-L1 expression in SCLC tumor cells (25). Supporting these findings, we also observed similar increases in CD45⁺/PD-L1⁺ and CD3⁺ immune cell infiltration around tumor nests focally expressing high levels of MHC I in MHC I^foc tumors, which was not seen in the MHC I^lo areas of the same tumors (Supplementary Fig. S1G and S1H). Taken together, these findings suggest that MHC I^hi SCLC is associated with an immune microenvironment primed for an effector response, but one that displays an upregulation of the PD-1/PD-L1 immune checkpoint, primarily on infiltrating immune cells.

Because MHC I antigen presentation is crucial for the adaptive immune effector phase disinhibited by ICB, we hypothesized that patients bearing MHC I^hi SCLC may be uniquely responsive to such therapy. We gathered a cohort of patients with SCLC treated at the Dana-Farber Cancer Institute (DFCI) enriched for those with exceptional durable responses to ICB (n = 31; Supplementary Table S1.5–S1.6). Strikingly, when stratified by MHC I^lo/neg versus MHC I^hi expression as defined in our larger cohort, patients with MHC I^hi SCLC had dramatically more durable responses to ICB, which translated to a robust increase in overall survival, including when used in the post–first-line setting (Fig. 1G and H; Supplementary Fig. S2A and S2B). Analysis for potentially confounding variables found that MHC I expression was the only variable that predicted survival after treatment with ICB in our cohort containing exceptional responders (Supplementary Fig. S2C; Supplementary Table S1.5). We also analyzed a subset of patients with chemorefractory SCLC treated with nivolumab ± ipilimumab on the Checkmate 032 trial (n = 18). Although there was only one long-term responder, this patient's SCLC also exhibited high MHC I expression, and combined analysis of the DFCI/Checkmate 032 cohorts revealed a significantly higher rate of 12-month durable clinical benefit from ICB in MHC I^hi patients compared with MHC I^lo/neg patients (P < 0.001; Supplementary Fig. S2D). These data demonstrate that, although most human SCLCs are MHC I^lo/neg, a subset of tumors with nonneuroendocrine/mesenchymal features are MHC I^hi, which tracks with durable response to ICB and suggests intrinsic immunogenic capacity of these tumors.

The observation of intratumoral heterogeneity in MHC I expression and neuroendocrine phenotype in MHC I^foc tumors (Supplementary Fig. S1A and S1B) suggested cell state plasticity and a possible epigenetic mechanism of regulation. Thus, to explore this further, we used the established in vitro human SCLC cell line NCI-H69 (H69) and an isogenic, adherent derivative (referred to hereafter as H69M) that contains an epigenetically distinct nonneuroendocrine subpopulation. These nonneuroendocrine cells are marked by expression of PD-L1 and AXL, and activate IFN signaling similar to patient MHC I^hi SCLC (Fig. 2A; refs. 13, 26). To systematically identify the loci associated with this epigenetic state change, we performed integrative unbiased global H3K27 chromatin immunoprecipitation sequencing (ChIP-seq) and RNA-seq analysis. Among the top genes upregulated in H69M enriched for both H3K27Me3 loss and H3K27Ac gain were TAP1, AXL, and YAP1 (Fig. 2A; Supplementary Table S2.1). Subset analysis of MHC I APM genes upregulated in H69M using the same method confirmed robust enrichment of mRNA expression and H3K27 acetylation (Ac) peaks in MHC I APM genes in H69M (Supplementary Fig. S3A and S3B), most notably at the TAP1 locus (Supplementary Fig. S3C). Preferential H3K27Ac gain and reciprocal H3K27Me3 loss at the TAP1 locus were at least as pronounced as the previously validated gene TMEM173 (STING; Supplementary Fig. S3D and S3E; Supplementary Table S2.2; ref. 13). Furthermore, we observed significant enrichment of this gene set, especially of TAP1 and AXL, in patient MHC I^hi SCLC tumors (Fig. 2B, Supplementary Fig. S3F). These findings confirm concordance between this in vitro model and patient samples, as well as further implicate TAP1 in this differential MHC I antigen presentation.

TAP1 forms part of a heterodimer that imports cytosolic peptides into the endoplasmic reticulum for loading onto MHC I, has been implicated in defective SCLC antigen presentation (20, 21), and was highly basally upregulated in bulk H69M cells (Supplementary Fig. S3G). We therefore examined whether recovery of MHC I expression in H69M could be related to this epigenetic derepression of TAP1. Flow cytometry revealed pronounced recovery of MHC I on H69M cells, specifically within the PD-L1^hi subpopulation that marks the nonneuroendocrine/mesenchymal cell state (Fig. 2C; ref. 13). Indeed, further dissection of PD-L1/MHC I^hi versus PD-L1/MHC I^lo cells confirmed that H69M PD-L1/MHC I^hi cells specifically upregulated TAP1 and TMEM173 together with EMT and IFN signatures, while downregulating neuroendocrine gene expression (ref. 27; Fig. 2D and E; Supplementary Table S3.1). Liquid chromatography–data independent analysis mass spectrometry (LC-DIAMS) of MHC I–associated immunopeptidomes of H69, bulk H69M, or purified H69M-PD-L1/MHC I^hi cells further demonstrated that H69 cells exhibited a profound defect in presentation of endogenously processed peptides, limited to signal peptides, pathognomonic for TAP1 downregulation (Fig. 2F; ref. 17). Bulk H69M cells exhibited partial rescue of endogenous peptide presentation, whereas H69M PD-L1/MHC I^hi cells showed robust recovery of surface peptide display (Fig. 2G). Indeed, H69M PD-L1/MHC I^hi cells exhibited near-complete recovery of the diversity of HLA-A2–restricted peptides, in contrast to H69 cells, whose peptide display was restricted solely to signal peptides (Fig. 2H; Supplementary Table S3.2). Thus, transition of SCLC to a nonneuroendocrine/mesenchymal state not only activates innate immune signaling pathways but also directly restores TAP1 expression and consequent antigen presentation, as revealed by direct physical detection of HLA-bound peptides.

We next sought to validate these results in additional SCLC cell lines stratified by adherence in cell culture (8), which tracks with downregulation of neuroendocrine markers and upregulation of AXL (Supplementary Fig. S4A and S4B; ref. 13). In contrast to neuroendocrine cells, MHC I expression was increased in multiple adherent, nonneuroendocrine SCLC cell lines and correlated with increased TAP1 mRNA expression (Fig. 3A). Furthermore, TAP1 overexpression in multiple neuroendocrine SCLC cell lines rescued MHC I expression (Supplementary Fig. S4C–S4H). Thus, endogenous MHC I antigen presentation is recovered in SCLC in its nonneuroendocrine state and is at least partly related to derepression of TAP1 epigenetic silencing.

Our findings of immunogenic plasticity in SCLC associated with response to ICB led us to explore pharmacologic approaches to convert SCLC cell state. Based on our ChIP-seq findings and its established role in neuroendocrine cell biology (13, 23, 28, 29), we next evaluated the role of the EZH2 H3K27 methyltransferase in neuroendocrine and nonneuroendocrine SCLC. Global H3K27Me3 levels were decreased in adherent cell lines and inversely related to TAP1 protein expression (Fig. 3B). Furthermore, analysis of SCLC Cancer Cell Line Encyclopedia data revealed anticorrelation between TAP1 and EZH2 expression, which tracked with cell line adherence as a surrogate for neuroendocrine status (Fig. 3C, Supplementary Fig. S5A). Short-term genetic and/or pharmacologic EZH2 inhibition (Supplementary Fig. S5B and S5C) in neuroendocrine SCLC slightly increased MHC I expression and sensitized cells to low-dose IFNγ (0.5 ng/mL; Supplementary Fig. S5D), consistent with a recent study implicating the polycomb repressor complex 2 (PRC2) complex in MHC I repression (23). Indeed, treatment of multiple neuroendocrine SCLC lines with a variety of PRC2 inhibitors for 6 days revealed that robust recovery of MHC I expression required concomitant type I or type II IFN exposure (Supplementary Fig. S5E–S5J). However, we also failed to observe AXL induction after this short-term EZH2 inhibition, even in H69 cells treated with GSK126 + IFNγ (Fig. 3D), revealing that such exposure might be insufficient to reset this broader epigenetic program.

Given the modest effect of short-term EZH2 inhibition alone on MHC I expression (Supplementary Fig. S5F), we next examined the downstream consequences of treating H69 cells with a 6-day GSK126 pulse and release, to determine whether inhibiting EZH2 could unmask cellular plasticity and intrinsic antigenicity over time. Interestingly, this approach, using multiple EZH2 inhibitors and neuroendocrine SCLC cell lines, facilitated recovery from cell cycle arrest and triggered adherence and expansion of a nonneuroendocrine subpopulation (referred to hereafter as H69EZ-G) that morphologically resembled H69M and upregulated MHC I expression (Supplementary Fig. S6A–S6D). Similar to patient MHC I^hi SCLC and nonneuroendocrine H69M, MHC I^hi H69EZ-G cells showed coexpression of AXL (Fig. 3E), highlighting the crucial role of an epigenetic cell state change in the basal upregulation of MHC I antigen presentation in nonneuroendocrine SCLC. Congruently, H69EZ-G cells basally upregulated TAP1 expression and exhibited persistent downregulation of H3K27Me3, specifically within the adherent MHC I^hi subpopulation (Fig. 3F; Supplementary Fig. S6E). RNA-seq of MHC I^hi H69EZ-G cells confirmed a similar phenotype as patient MHC I^hi SCLC and MHC I^hi H69M cells, with enrichment of IFN signaling, EMT, our previously described SPARCS ERV signature, and downregulation of neuroendocrine markers (ref. 13; Fig. 3G, Supplementary Table S4.1). H69EZ-G cells constitutively secreted multiple innate immune cytokines and chemokines comparable to H69M cells (Supplementary Fig. S6F), consistent with intrinsic activation of double-stranded RNA/double-stranded DNA sensors (13). In consonance with their STING derepression, both H69M and H69EZ-G cells preferentially responded to a cyclic dinucleotide STING agonist with dramatic upregulation of CXCL10, similar to other nonneuroendocrine SCLC models (Fig. 3H). CRISPR-mediated silencing of STING in H69M resulted in a marked downregulation of HLA-ABC expression (Fig. 3I), supporting a key role for nucleic acid sensing in enforcing surface antigen presentation in nonneuroendocrine SCLC. Taken together, these data suggest that EZH2 broadly regulates epigenetic SCLC plasticity, and its inhibition promotes a nonneuroendocrine cell state with intrinsic innate immune signaling, which leads to constitutive MHC I recovery (see schematic in Fig. 3J).

To determine whether SCLC is intrinsically immunogenic in its nonneuroendocrine state, we used a CRISPR-derived Rb1;Trp53;Rbl2/p130 (RPP) SCLC genetically engineered mouse model (GEMM) generated in a C57BL/6 background (30, 31). SCLC lines derived from established tumors gave rise to both neuroendocrine suspension tumor cells (referred to hereafter as RPP cells) and nonneuroendocrine adherent tumor cells (referred to hereafter as RPP-A cells) with inactivation of the same tumor suppressor genes, where RPP-A cells exhibited a phenotype highly concordant with our RNA-seq data from human cell lines and patient samples (Supplementary Fig. S7A–S7D; Supplementary Table S4.2). RPP cells were largely devoid of MHC I expression (H2-K^b and H2-D^b) and endogenously processed MHC I–restricted peptide display, which was recovered in RPP-A cells, over independent cell line derivations from independent mice (Fig. 4A; Supplementary Fig. S8A–S8C). Similar to H69 cells, short-term treatment of RPP cells with an EZH2 inhibitor promoted partial recovery of MHC I in response to a low dose of IFNγ (Supplementary Fig. S8D).

In support of differential immunogenicity, expression of exogenous ovalbumin (OVA; ref. 32) resulted in basal endogenous presentation of the H2-K^b–restricted ovalbumin antigen SIINFEKL in RPP-A.OVA cells, but not in RPP.OVA cells (Supplementary Fig. S8E and S8F), which tracked with killing by cognate OT-I T cells (Fig. 4B). RPP.OVA cells were rendered sensitive by high-dose IFNγ stimulation (Supplementary Fig. S8G), confirming their immunogenic capacity with sufficient extrinsic stimulation.

We next directly interrogated the intrinsic immunogenicity of neuroendocrine or nonneuroendocrine SCLC in vivo by subcutaneous inoculation of RPP or RPP-A cells into syngeneic BL6 mice. Strikingly, RPP-A cells formed nascent tumors that were spontaneously rejected over the long term and heavily infiltrated with immune cells, in contrast to RPP cells, which effectively evaded an immune response and formed large tumors over time (Fig. 4C; Supplementary Fig. S9A and S9B). RPP-A cells generated tumors in immunodeficient NOD/SCID gamma (NSG) mice (Supplementary Fig. S9C), confirming their intrinsic tumorigenic capacity and the requirement for adaptive immunity to mediate their rejection. Furthermore, mixing studies combining RPP-A cells with RPP cells at a 1:4 ratio revealed that RPP-A cells in a mixed population can still recruit T cells in vitro and promote rejection in immunocompetent mice (Supplementary Fig. S9D–S9G), suggesting that immune responses evoked by a minority nonneuroendocrine population can be sufficient to target neuroendocrine SCLC cells in a heterogenous tumor.

To characterize the immune response against nonneuroendocrine SCLC cells in greater depth, we performed paired single-cell RNA and T-cell receptor (TCR) sequencing of CD45⁺ immune cells 14 days following RPP or RPP-A tumor cell implantation. Single-cell data clustering and gene signature analysis revealed that RPP-A tumors were more heavily infiltrated with CD8⁺ T cells (Fig. 4D; Supplementary Figs. S10A–S10C, S11A–S11D; Supplementary Table S5.1–S5.2), whereas RPP tumors were uniquely enriched with multiple immune-suppressive M2 macrophage markers (ref. 33; Supplementary Figs. S10A–S10C, S11E; Supplementary Table S5.1–S5.2). Interestingly, TCR clonotype analysis uniquely identified an immunodominant TCR clone (Trav14D-3-DV8, Trbv29) comprising 11% of all TCRs within the RPP-A CD8 T-cell population and two additional expanded clones also expressing Trbv29 (Fig. 4E; Supplementary Table S5.3). Trav14D-3-DV8, Trbv29 TCR clonotypes were not identified in RPP tumors, which did not harbor any expanded T-cell clones (Fig. 4E; Supplementary Table S5.4). Moreover, differential gene expression analysis revealed high-level granzyme, NKG2 family members (KLRE1 and KLRC1), and Ly6c2 expression (Supplementary Fig. S11F; Supplementary Table S5.3) specific to these Trbv29⁺ CD8 T cells, suggestive of highly sensitive cytotoxic effectors. Furthermore, although RPP cells exhibited quantitatively greater numbers of CD4 T cells, only RPP-A tumors showed relative expansion of CD4 T-cell clones with transcriptional features of developing effector memory (Supplementary Fig. S11G), consistent with their ability to support a diversity of T cell–mediated immune responses against RPP-A–associated antigens.

We next cloned the immunodominant Trav14D-3-DV8, Trbv29 CD8 TCR into the TCR-null line BW5147.3 (Supplementary Fig. S12A and S12B), to assess in vitro recognition of RPP versus RPP-A cells. Trav14D-3-DV8, Trbv29–expressing T cells were activated only by RPP-A cells derived from two independent SCLC GEMMs (Fig. 4F; Supplementary Fig. S12C). High-dose IFNγ stimulation led to an increase in TCR responsiveness to RPP-A cells but not RPP cells, suggesting unique derepression of the antigenic determinant of the TCR in RPP-A cells (Fig. 4F). Taken together, these data confirm that nonneuroendocrine SCLC is intrinsically immunogenic and capable of evoking T cell–mediated rejection.

Finally, we sought to determine whether the EZH2 inhibitor–induced cell state change in neuroendocrine SCLC could facilitate immune rejection in this syngeneic mouse model. Since EZH2 inhibition can also affect T cells in vivo (34), we sought to experimentally isolate our analysis of tumor-intrinsic immunogenicity by examining the effects of EZH2 inhibition on RPP cells ex vivo prior to implantation. Similar to our studies in human SCLC cells, pulse treatment with GSK126 followed by subsequent release derived an adherent cell line from the murine RPP cell line. This cell line (RPP-EZ) displayed a morphologic and immunologic phenotype intermediate between RPP and RPP-A, showing adherence but not fully elongated/fibroblastic morphology, with a moderate increase in basal MHC I expression and recognition by the immunodominant TCR in the presence of IFN (Fig. 4G; Supplementary Fig. S12D). In consonance with these findings, RPP-EZ tumors displayed significant growth impairment relative to RPP tumors but did not show durable long-term responses (Fig. 4H). Of note, we did not observe in vitro proliferation changes or differences in early tumor growth between RPP and RPP-EZ, which, if anything, displayed faster engraftment. Because STING was restored and active in the nonneuroendocrine state, and our in vitro data suggested that IFN priming may be necessary to unmask robust immunogenicity, we next tested the effects of the STING agonist ADU-S100 in our RPP-EZ and RPP syngeneic models. Immune profiling demonstrated a significant increase in immune cell infiltration, including CD4⁺ and CD8⁺ T cells, in both RPP and RPP-EZ tumors exposed to STING agonist (Supplementary Fig. S12E). When intratumoral ADU-S100 injection was performed in two independent experiments, long-term durable responses were observed preferentially in RPP-EZ tumors (P < 0.01; Fig. 4H; Supplementary Fig. S12F). Indeed, this combination yielded the most significant improvement in overall survival (Fig. 4I), suggesting that EZH2 inhibition both restores SCLC antigenicity and can prime therapeutic response to STING agonism.

Once considered a uniform diagnosis, SCLC subtyping is rapidly evolving to identify therapeutic vulnerabilities. Here we have uncovered a nonneuroendocrine population of SCLC with intrinsic immunogenicity that enriches for durable response to immune checkpoint blockade. Mechanistic dissection of the epigenetic features that govern this immunologic plasticity reveal TAP1 and STING as direct regulators of MHC I antigen presentation, as well as a rationale to combine EZH2 inhibitors with STING agonists in a syngeneic mouse model. In addition to confirming that EZH2 regulates MHC I expression in SCLC (23, 35), here we more broadly demonstrate that transition to a nonneuroendocrine cell state in SCLC leads to robust presentation of MHC I–restricted antigens coupled with intrinsic innate immune signaling, sufficient to cause antigen-specific T cell–mediated rejection in a syngeneic system. Notably, the immunodominant T-cell clone we identified recognizes a putative rejection antigen solely presented on nonneuroendocrine cells, suggesting the intriguing possibility that epigenetic modulation may uncover immunopeptidome repertoire plasticity sufficient to reveal immunogenic antigens. Indeed, we previously identified derepression of unique endogenous retroviruses in this context (13), suggesting the possibility that peptides derived from specific retroelements, in addition to more canonical mutated neoantigens or tumor-associated antigens, could be the source of such immunogenicity. That MHC I^hi SCLC displays increased immune infiltration by both CD3⁺/CD8⁺ T cells and CD45⁺/PD-L1⁺ immune cells suggests that the immune microenvironment of such tumors may be poised for an antitumor response, but these tumors may avoid elimination via immune checkpoints such as PD-L1, the presence of immunosuppressive leukocyte populations, and/or the concurrent production of immune-suppressive cytokines.

Our finding that exceptional durable responders to ICB exhibit this nonneuroendocrine, MHC I^hi phenotype and that patients with MHC I^hi SCLC more generally show preferential clinical benefit to ICB agrees with recent independent characterization of an inflamed SCLC subtype (36), underscoring the intrinsic immunogenicity of this SCLC population and suggesting that immune cell PD-L1 expression may be an important immunosuppressive mediator in this context. Furthermore, a biomarker to stratify ICB responders from nonresponders in SCLC is sorely needed. Although the incorporation of ICB has improved response rates to first-line treatment in SCLC, overall survival remains limited, and new biomarkers could help select patients who are expected to progress in order to facilitate enrollment in clinical trials. In contrast to non–small cell lung cancer, tumor cell PD-L1 levels in SCLC are uniformly low and fail to predict ICB response (37, 38). Although tumor mutation burden (TMB) tracks to some degree with clinical benefit to ICB in SCLC (39), these cancers possess a high number of mutations overall given their close association with cigarette smoking. Furthermore, TMB is a continuous rather than discrete variable, and it lacks precision to explain differential antigenicity. In contrast, tumor cell MHC I is the ultimate target of CD8 T cells that are reinvigorated by ICB, and hence it is not surprising that MHC I status separates the patients with SCLC who are able to achieve durable response. Although larger cohorts of patients will be necessary to validate these results, MHC I, as well as other features of this nonneuroendocrine state, could serve as straightforward and much-needed biomarkers for ICB response in SCLC.

By integrating epigenetic and expression profiling, as well as protein expression across our HLA⁺ cell line models and patient SCLC, we provide an initial description of the MHC I^hi SCLC phenotype relative to the current paradigm in the literature. Broadly, we find that MHC I^hi SCLC does not express the neuroendocrine markers ASCL1 or NEUROD1, or other more recently proposed markers of a nonneuroendocrine state, such c-MYC or YAP1. Instead, MHC I^hi SCLC is enriched for expression of AXL, a state marker for EMT, which appears—like MHC I APM—to be regulated epigenetically. Corroborating our findings, a recently described inflamed SCLC subtype also demonstrates a similar nonneuroendocrine transcriptional signature marked by expression of AXL and upregulation of immune genes, including those involved with MHC I APM (36). In addition, this finding has strong parallels to the melanoma literature in which the MITF lineage–specific oncogene is also downregulated in an AXL-positive innate immune active state (40). These data thus identify not only AXL upregulation as an important feature of the nonneuroendocrine state but also a novel paradigm for innate immune regulation that has parallels across lineage oncogene–driven tumors.

Because the MHC I^hi phenotype results in unique responsiveness to STING agonists, including upregulation of the T-cell chemoattractant CXCL10, these populations may especially evoke T-cell infiltration upon STING pathway stimulation (41, 42). Indeed, we provide initial evidence in a murine model that ex vivo EZH2 inhibition can convert neuroendocrine SCLC to an antigenic cell state that is responsive to STING agonism in vivo. Future preclinical therapeutic studies could focus on combining EZH2 inhibitors (or other epigenetic inhibitors that alter the SCLC cell state) with STING agonists to overcome the intrinsic immunotherapy resistance of the SCLC neuroendocrine state (43). This will require pharmacodynamic/pharmacokinetic optimization of clinical-grade EZH2 inhibitors and STING agonists that are currently in development. Although work remains to elucidate the cellular and immunologic interplay of neuroendocrine and nonneuroendocrine SCLC subpopulations, identification of nonneuroendocrine subpopulations in human SCLC via cell state or immunologic profiling may also prove useful in stratifying patients for additional types of immunotherapy, including cancer vaccines. Indeed, as these subpopulations are likely to have escaped immune editing (44), they are likely to possess an immunopeptidome enriched for (neo)antigens, resulting in particular vulnerability to immune attack.

In total, 102 cases of formalin-fixed, paraffin-embedded (FFPE) primary resections of SCLC were obtained from various commercial tissue vendors (BioChain Institute; Discovery Life Science; Indivumed GmbH; The MT Group). All cases had been reviewed by a board-certified pathologist at New England Pathology Associates on hematoxylin and eosin (H&E) slides to ensure tumor content. FFPE sections at 4 μm were prepared and stored at 4°C until use. All cases were deemed histologically compatible with SCLC after expert pathologic review (N.R. Mahadevan, L.M. Sholl). SCLC diagnosis in MHC I^hi/foc cases was supported with tumor cell–specific RB inactivation in all cases.

For clinical correlation studies, DFCI/Brigham and Women's Hospital (BWH) patients with SCLC from 2014 to 2019 were selected who had received ICB (n = 31) via electronic medical record (EMR) review. This cohort was enriched for patients with exceptional responses (stable disease/partial response for greater than 1 year). MHC I IHC was performed as below, and scoring used the same cutoffs as detailed in the text and below. Clinical course information, including radiography and pathology, was abstracted from the EMR and pathology information systems (Powerpath). Targeted next-generation sequencing (NGS) data were abstracted from institutional targeted NGS (Oncopanel; ref. 45) results. All patients provided written informed consent for this research under DFCI protocol 02–180. SCLC human tumor samples were collected and analyzed according to Dana-Farber/Harvard Cancer Center institutional review board–approved protocols. These studies were conducted according to the Declaration of Helsinki and approved by Dana-Farber and BWH institutional review boards. An additional cohort of n = 18 FFPE samples was obtained from the Checkmate 032 trial of nivolumab ± ipilimumab in recurrent SCLC (30). Demographic information was used to ensure no overlap with the DFCI/BWH cohort. After MHC I IHC as described below, survival and response analyses were performed.

Staining of MHC I (HLA-A, HLA-B, and HLA-C, 1:6,000 dilution, clone EMR8–5; Abcam, cat. ab70328) for all cases (n = 102) and chromogranin (1:5,000 dilution, clone LK2H10; Thermo Fisher, cat. MS-324-PABX), as well as AXL (1:4,000 dilution, clone EPR19880; Abcam, cat. ab219651) for the MHC I^hi/foc (n = 17) cases and a randomly selected subset of MHC I^lo/neg (n = 40) cases, was performed with a BOND RX fully automated stainer (Leica Biosystems). FFPE tissue sections were baked for 2 to 3 hours at 40°C and then loaded into the BOND RX. Slides were deparaffinized (BOND DeWax Solution; Leica Biosystems) and rehydrated through a series of washes of graded ethanol to deionized water. Antigen retrieval (BOND Epitope Retrieval Solution 1 or 2; Leica Biosystems) was performed at pH 6.0 (chromogranin) or 9.0 (MHC I and AXL) for 30 minutes at 98°C. Slides were then stained with primary antibodies for 40 minutes. Next, rabbit anti-mouse IgG (Post Primary, BOND Polymer Refine Detection Kit; Leica Biosystems) was applied for 10 minutes, then subsequently detected by anti-rabbit polymeric horseradish peroxidase (Poly-HRP, BOND Polymer Refine Detection Kit; Leica Biosystems). Finally, slides were counterstained with hematoxylin, dehydrated in graded alcohol and xylene, and mounted with EcoMount mounting medium (Biocare Medical). Pathologists S. Rodig and N.R. Mahadevan assigned an H-score (range, 0 to 300) to each sample by multiplying the percentage of tumor cells positive for MHC I staining (0% to 100%) by the average intensity of staining (0 to 3+) of the tumor cells. The cutoff for MHC I^hi cases was calculated to include extreme outliers of a positively skewed distribution: median + 2.5 interquartile ranges. The same method of assigning an H-score was used by pathologist N.R. Mahadevan to score chromogranin and AXL staining.

Multiplexed immunofluorescence staining was performed with the BOND RX fully automated stainer (Leica Biosystems). FFPE tissue sections were baked for 3 hours at 60°C and then loaded into the BOND RX. Slides were deparaffinized (BOND DeWax Solution; Leica Biosystems) and rehydrated through a series of washes of graded ethanol to deionized water. Antigen retrieval (BOND Epitope Retrieval Solution 1; Leica Biosystems) was performed at pH 6.0 for 30 (RB) or 10 (all other target antigens) minutes at 98°C. Slides were then stained with primary antibodies with an incubation time of 40 minutes. As an additional step for mouse antibodies, rabbit anti-mouse IgG (Post Primary, BOND Polymer Refine Detection Kit; Leica Biosystems) was applied for 10 minutes. The slides were then incubated with anti-rabbit Poly-HRP (BOND Polymer Refine Detection Kit; Leica Biosystems) for 10 minutes. Next, the slides were incubated with Opal Fluorophore Reagents (Akoya Biosciences) for 5 minutes to visualize signal of the antibody complexes. This process was repeated for all antibodies included in a panel. The target antigens, antibody clones, and dilutions for panels 1 and 2 are listed in Supplementary Table S6.1. Slides were then manually counterstained with 4′,6-diamidino-2-phenylindole (DAPI; NucBlue Fixed Cell ReadyProbes Reagent; Invitrogen), washed with deionized water, air dried, and mounted with ProLong Diamond Anti-fade Mountant (Invitrogen).

A minimum of four representative images per sample were obtained using the Mantra Quantitative Pathology Workstation (Akoya Biosciences) at 20× magnification. Following image acquisition, each field of view was spectrally unmixed and analyzed using inForm 2.4 Image Analysis Software (Akoya Biosciences). Images were manually categorized into tumor and stromal regions using inForm's tissue classification tool. All nucleated cells within a field of view were then identified via nuclear DAPI staining. Using a supervised machine-learning algorithm within the program, each cell was assigned a phenotype according to biomarker expression within its defined nuclear and membrane compartments. For cases of sufficient image quality stained with panel 1, an H-score for ASCL1 and INSM1 was generated by setting intensity thresholds for 0 to 3 binning (from 0 for “no signal” to 3 for “strong signal”) under pathologist supervision. The percentage of cells in each intensity bin was multiplied by the value of that bin (i.e., 0 to 3), and the combined scores automatically produced an H-score in the inForm software. For each case, an average H-score was calculated based on 4 to 10 representative fields of view. Cells phenotypically determined to be nonmalignant were excluded from analysis. To quantify immune infiltrate in cases stained with panel 2, a minimum threshold of positive staining was assigned for each image, categorizing cells as negative or positive for a particular biomarker. To calculate cell density of a biomarker (per megapixel), the percentage of positive cells was multiplied by the total number of cells present in the tissue category, divided by the number of pixels in the tissue category, then divided again by the conversion factor of 1,000,000 (pixels per megapixel). For each case, an average H-score was calculated based on 5 to 12 representative fields of view. When quantifying biomarker expression in malignant cells, cells phenotypically determined to be nonmalignant were excluded from analysis.

Use of human SCLC cell lines NCI-H69, H69M (parental, control, and STING KO) NCI-H82, NCI-H2081, and NCI-H196 was continued from our prior work (6). The human SCLC cell line CORL47 (Sigma) was kindly provided by Dr. M. Oser. The murine lymphoma cell line BW5147.3 was obtained from the laboratory of E. Reinherz. COLO668 and 293T cells were obtained from ATCC.

H69, H69M, H69EZ-G, H82, H2081, and CORL47 were cultured in RPMI-1640 (Thermo Fisher Scientific, #11875–119) containing 10% FBS (cat. 100–106; Gemini Bio-products), 1× penicillin/streptomycin (cat. 15140122; Thermo Fisher Scientific), and prophylactic Plasmocin (2.5 mg/mL; cat. ant-mpp; Invivogen). H196 and 293T were maintained in DMEM (cat. 11965–118; Thermo Fisher Scientific) containing 10% FBS and 1× penicillin/streptomycin. RPP and RPP-A cell lines were cultured in RPMI (cat. 11875–119; Thermo Fisher Scientific), containing 10% FBS (cat. 100–106, Gemini Bio-products), 1× penicillin/streptomycin (cat. 400–109; Gemini Bio-Products), and HITES [1× insulin-transferrin-selenium (cat. I3146; Sigma-Aldrich), 10 nmol/L β-estradiol (cat. E2257; Sigma-Aldrich), and 10 nmol/L hydrocortisone (cat. 10687010; Thermo Fisher Scientific)]. BW5147.3 was maintained in DMEM (cat. 11965–118) containing 10% FBS, 1× penicillin/streptomycin (cat. 15070063), 400 μg/mL G418 (cat. 10131035; Thermo Fisher Scientific), and 400 μg/mL hygromycin (cat. H0135; Sigma-Aldrich). All cell lines were routinely monitored for Mycoplasma using the Takara PCR Mycoplasma Detection Set (cat. 6601; Kusatsu) and were negative.

Total cellular RNA was extracted using the RNeasy Mini Kit (Qiagen) according to manufacturer's instructions. After extraction, 500 ng to 1 μg total RNA was used to generate cDNA with the SuperScript III First-Strand Synthesis SuperMix for qRT-PCR kit, which includes both oligo-dT and random primers (Thermo Fisher Scientific). qRT-PCR of the indicated genes was performed using SYBR green PCR Master Mix (Applied Biosystems) and the Applied Biosystems 7300 Fast real-time PCR system and software. Target gene expression was normalized to 36B4 and analyzed using the –ΔΔCt relative quantification method. The sequences of primers used are listed in Supplementary Table S6.2.

Protein was isolated from cell lines and measured by BCA (Pierce Biotechnology). Protein extracts were subjected to polyacrylamide gel electrophoresis using the 4% to 12% NuPAGE gel system (Invitrogen), transferred to polyvinylidene difluoride (Millipore) membranes, and immunoblotted using antibodies (Cell Signaling Technologies) that specifically recognize TAP1 (cat. 12341), EZH2 (clone D2C9, cat. 5246S), H3K27Me3 (clone C36B11, cat. 9733), STING (clone D2P2F, cat. 13647S), and β-actin (cat. 4970). Secondary antibodies were from LICOR Biosciences: IRDye 800CW Goat 495 anti-Mouse IgG (H + L; cat. 926–32210) and IRDye 800CW Goat anti-Rabbit IgG (H + L; cat. 926–22 32211). LICOR blocking buffer (cat. 927–40000) was used to dilute primary and secondary antibodies, with the exception of the anti-TAP1 antibody, which was diluted in HIKARI Signal Enhancer Solution 1 (cat. NU00101; Nacalai USA). Imaging of blots and quantitation of bands were performed using the LICOR Odyssey system.

Proteome Profiler Human Cytokine Array Kit (cat. ARY005B), human and mouse CXCL10 ELISA (cat. SIP100 and DY466), and mouse IL2 ELISA (cat. M2000) were performed according to manufacturer's (R&D Systems) instructions. For cytokine array, conditioned media (CM) from SCLC cells at basal conditions were collected after 48 hours. For CXCL10 ELISA, CM were collected after 24 hours. For IL2 ELISA, CM from SCLC-T cell cocultures were collected after 72 hours.

Single mouse tumor cell suspensions were generated using the mouse Tumor Disassociation Kit (cat. 130-096-730; Miltenyi Biotec) per the manufacturer's protocol using the gentleMACS Octo Disassociator with heaters using the 37C_m_TDK1 protocol. After red blood cell lysis and filtration, cell suspensions were subjected to flow cytometry. For extracellular staining, after live/dead staining with Zombie NIR Fixable Viability Kit (cat. 423106; Biolegend) per the manufacturer's instructions, single-cell suspensions were stained with fluorophore-conjugated primary antibodies (see Supplementary Table S6.3) in PBS containing 2% FBS at 2 μg/mL. For experiments with intracellular staining, after live/dead and extracellular staining, cells were fixed in fresh 4% paraformaldehyde and permeabilized in 100% ice-cold methanol for 30 minutes. After washing, cells were incubated with fluorophore-conjugated anti-H3K27Me3 (cat. 12158, clone C36B11; Cell Signaling Technologies) at 1:100 dilution. After washing, cells were resuspended in PBS containing 2% FBS and analyzed on a LSRFortessa flow cytometer (Becton Dickinson). Levels were compared with isotype control antibodies. Cell cycle analysis was performed using the BD Cycletest Plus Kit (cat. 340242; BD Biosciences) per the manufacturer's protocol. The data analyses were performed with FlowJo software (TreeStar).

Pan-HLA immunoprecipitation (IP) was performed as previously described (46). Briefly, adherent cells were released nonenzymatically to preserve surface antigen expression by incubation at 37°C for 30 minutes in nonenzymatic cell dissociation medium [PBS supplemented with EDTA (5 mmol/L), EGTA (5 mmol/L), and FCS (1%)]. Following centrifugation, the cell pellet (0.5–1 × 10⁶ cells) was lysed (Triton X-100, 1.2%; octyl-β-D-glucopyranoside, 1.77%; Tris-HCl, 20 mmol/L; EDTA, 1 mmol/L; NaCl, 100 mmol/L; complete protease inhibitor cocktail, ×1; Roche) and incubated on ice for 20 minutes. Lysed cells were centrifuged for 10 minutes at 21,000 × g, and the supernatant was centrifuged for 1 hour at 105,000 × g (Airfuge; Beckman-Coulter) and 4°C to pellet cellular debris. The lysate was then added to the antimonomorphic HLA class I antibody (clone W6/32; Biolegend) preincubated Protein G Sepharose beads (GE Healthcare) and rotated for 3 hours at 4°C. The lysate–bead suspension was then centrifuged at 900 × g for 3 minutes and the supernatants retained for confirmation of depletion of HLA class I. The beads were then washed six times in wash buffer (0.5 mL/wash; octyl-β-D-glucopyranoside, 1.77%; Tris-HCl, 20 mmol/L; EDTA, 1 mmol/L; NaCl; 100 mmol/L) followed by two washes in salt buffer (Tris-HCl, 20 mmol/L; EDTA, 1 mmol/L; NaCl; 100 mmol/L). Residual liquid was removed using fine pipette tips and the bead samples with bound HLA-peptide stored at −80°C until processed for MS analysis.

All Poisson detection MS methodology and liquid chromatography (LC) data independent acquisition (DIA) MS methodologies were performed as described previously (46). In brief, mass spectra were collected on a quadrupole-TOF (Sciex 6600+) instrument in a DIA format. The m/z region 400 to 680 was split into 11 minimally overlapping windows of variable width designed to transmit equal ion fluxes with MHC I immune peptidomes. MS data were collected in a series of a single full-range MS spectrum followed with 11 MS/MS spectra for each transmitted window. The MS/MS spectra were compared with reference patterns obtained from synthetic peptides using an algorithm based on the theory of sampling a Poisson process. High LC/MS sensitivity was promoted using electrospray ionization with 20-μm ID alkane-modified polystyrene-divinylbenzene monolithic columns (fabricated in-house) at flow rates of roughly 10 nL/min. Elution positions of the synthetic peptides relative to shared endogenous immune peptides using the same column configuration were also determined, and this provides a restrictive map for the elution positions of the epitope candidates in the tumor DIA MS data.

For cell line comparison, RNA was isolated directly from duplicate or triplicate samples with the Qiagen RNeasy isolation kit (cat. 8028) using on-column DNAse digestion. RNA libraries were prepared from 250 ng total RNA using the Illumina Exome Capture Kit per the manufacturer's instructions. RNA-seq was performed per the standard protocols at the Dana-Farber Molecular Biology Core Facilities with the Illumina NextSeq 500 instrument. Data quality controls and replicate correlation were evaluated using the VIPER software package (47).

For patient samples, 30 blocks, comprising all MHC I^hi cases (n = 15) and MHC I^lo/neg cases (n = 15), were submitted for paired whole-genome and transcriptome sequencing. Samples with less than 70% overall tumor content were scribed by a pathologist to enrich for tumor. Up to 10 consecutive FFPE 5-μm sections were macrodissected or whole slide scraped and pooled into a single tube for downstream DNA and RNA isolation. DNA and RNA were coextracted using the Qiagen AllPrep Kit (cat. 80204). DNA and RNA were qualified and quantified using Nanodrop and Qbit to determine concentration and Agilent Bioanalyzer to determine quality. RNA samples with less than 100 ng were excluded from downstream analysis. RNA-seq libraries were prepared using the TruSeq Illumina Stranded Total Gold with RiboZero using 100 ng total RNA. Libraries were quantified and qualified using KAPA library quantification kits (cat. 07960255001) and Bioanalyzer, and all samples with a less than 2 nmol/L library concentration were excluded from sequencing. Libraries were sequenced on the Novaseq to a read depth of 80 million paired-end reads with read lengths of 100 bp. These data are deposited in the Gene Expression Omnibus (GEO) database under accession GSE168295.

Differential gene expression analysis for both cell line and patient sample bulk RNA-seq data was performed using the R package, DESeq2 (48). Gene Set Enrichment Analysis (GSEA) of bulk RNA-seq data was performed using the generated lists of differentially expressed genes with R package ClusterProfiler (49) for the Hallmark Gene Sets from the MSigDB collection (50), Neuroendocrine marker genes (27), and SPARCS gene signature set (13). The GSEA P values were adjusted with the Benjamini–Hochberg method, and the adjusted P value threshold was set to 0.05. Visualization of the results and downstream analyses were performed using R software (r-project.org).

ChIP-seq was performed as described previously (51). Briefly, cells were cross-linked with 1% formaldehyde for 10 minutes, washed in 5 mg/mL BSA before being resuspended in lysis buffer (50 mmol/L Tris-HCl, pH 8.1; 10 mmol/L EDTA; 1% SDS; 1× protease inhibitor cocktail), and sonicated with the Covaris M220 sonicator to obtain chromatin fragment lengths of 100 to 1,000 bp. Fragmented chromatin was diluted in IP buffer (20 mmol/L Tris-HCl, pH 8.1; 150 mmol/L NaCl; 2 mmol/L EDTA; 1% Triton X-100) and incubated with Protein G magnetic beads (Dynabeads; Thermo Fisher Scientific) preincubated with anti-H3K27Ac (ab4729; Abcam) or anti-H3K27Me3 (clone C36B1; Cell Signaling Technologies) antibodies. Immunoprecipitates were washed six times with wash buffer (50 mmol/L HEPES, pH 7.6; 0.5 M LiCl; 1 mmol/L EDTA; 0.7% Na deoxycholate; 1% IGEPAL CA-630) and twice with TE buffer. Immunoprecipitated (or no IP input) DNA was treated with RNase A and Proteinase K on the beads and recovered in 1% SDS and 0.1 mol/L NaHCO₃. Up to 10 ng DNA was used for the library construction using the NEBNext Ultra II DNA Library Prep Kit (cat. E7645; New England Biolabs). Sequencing was performed on the NextSeq500 (Illumina) for 38 nucleotides each from paired ends according to the manufacturer's instructions.

Sequenced reads were aligned to human reference genome hg19 by bowtie2 with default settings with -k 1 parameter. After removing duplicate reads with samtools, ChIP-enriched regions (peaks) were identified by magnetic-activated cell sorting (MACS; ref. 52). ChIP signals for each sample were visualized on an integrative genome viewer genome browser (53) using wiggle files with a 10-bp resolution for H3K27Ac modification and 1-kb resolution for H3K27Me3 modification generated by MACS with tag shift that was rescaled to normalize to a total number of uniquely alignable sequences by the WigMath function of the Java-Genomic Toolkit. These data are deposited in the GEO database under accession GSE168195.

The H69M upregulated genes (1,326) were used to identify associated H3K27Ac peaks found in H69M or H69 cells, resulting in 4,937 peaks. H3K27Me3 signal and H3K27Ac signal normalized by the MAnorm algorithm were used to draw a scatterplot for deltas and log fold changes (logFC), respectively, and plotted. The thresholds of 3.0 logFC for H3K27Ac and –3 delta for H3K27Me3 were used to identify 78 epigenetically differential genes, which were further filtered with differentially expressed genes in H69M versus H69 at the threshold of the –4.0 DEseq2 score described above to identify 72 epigenetically regulated genes. Similarly, to determine whether MHC I antigen-presenting machinery was enriched, 823 genes from reactome ID R-HSA-983169 (MHC I–mediated antigen processing and presentation) was used to run single-sample GSEA (ssGSEA) against rank-ordered differentially expressed genes between H69M versus H69. The 91 genes overlapping between 1,326 H69M-upregulated genes and 823 MHC I antigen presentation and processing genes were used to identify associated H3K27Ac peaks found in H69M or H69 cells, resulting in 319 peaks. H3K27Me3 signal and H3K27Ac signal normalized by the MAnorm algorithm were used to draw the scatter plot for deltas and log fold changes, respectively. A heat map was drawn using H3K27Me3 signal and H3K27Ac signal around the 319 peaks and clustered by hierarchical clustering using deepTools.

The full-length human TAP1 sequence was obtained from the Harvard PlasmID database and cloned into the pLX_304 vector using the Gateway system (Thermo Fisher Scientific). A pLX_304 vector-bearing nanoLuc (nLuc) was used as a control. NCI-H82 and CORL47 cells were transfected by nucleofection (AmaxaTM 4D-Nucleofector X Unit) according to the manufacturer's instructions (Lonza) and maintained under blasticidin (10 μg/mL) selection.

Oligonucleotide coding for guide RNAs that target the EZH2 gene was chosen from the Avana library and the Brunello library (54). A nontargeting single-guide RNA (sgRNA) from the Gecko library v2 was used as a control (ctrl) sgRNA (55). LentiGuide vectors were cloned as previously described (55, 56). sgRNA target sequences are listed in Supplementary Table S6.4. 293T cells were transduced using X-treme Gene 9 (Roche) according to the manufacturer's instructions. The 72-hour supernatant of transduced 293T cells was collected through a 0.45-μm filter and added to the target NCI-H82.Cas9 cells (gift from M. Oser) in the presence of polybrene (8 μg/mL). Cells and virus were centrifuged for 30 minutes at 2,000 rpm at 37°C for 2 hours. On day 5, puromycin and blasticidin were added to select infected cells.

Recombinant human IFNγ (cat. 285-IF) and IFNβ (cat. 8499-IF), as well as murine IFNγ (cat. 485-MI) and IFNβ (cat. 8234-MB-010) cytokines, were purchased from R&D Systems and reconstituted in 1% BSA. For STING agonism experiments, cells were treated with ADU-S100 (50 μmol/L, cat. CT-ADUS100; ChemieTek) for 24 hours. For EZH2 inhibition experiments, cells were treated with the indicated concentrations of EZH2 inhibitor (GSK126, cat. 15415, Cayman Chemical Company; CPI-1205, cat. S8353, Selleck Chemicals; EPZ-6438 was obtained from the laboratory of Jun Qi) or DMSO for 6 days. EZH2 inhibitor was replenished on day 3 with all cells carried each time. After the EZH2 inhibitor treatment period, equal numbers of control or drug-treated cells were exposed to the indicated concentrations of IFNγ or IFNβ for 24 hours before harvesting of RNA or flow cytometry. For H69EZ-G and RPP-EZ derivation, GSK126 (5 μmol/L)-treated or control cells were thoroughly washed twice in PBS and all cells recultured in cell culture medium without drug. Cultures were routinely passaged as necessary and examined daily for the emergence of adherent populations.

Multiplex assays were performed using the bead-based immunoassay approach Bio-Plex Pro Human Cytokine 40-plex Assay (cat. 171AK99MR2) on a Bio-plex 200 system (cat. 171000201; Bio-Rad Laboratories) and the Human Cytokine/Chemokine Magnetic Bead Panel (cat. HCYTMAG-60K-PX30) on a Luminex MAGPIX system (Merck Millipore). Conditioned media concentration levels (pg/mL) of each protein were derived from five-parameter curve-fitting models. Fold changes relative to the corresponding control were calculated and plotted as log2FC. Lower and upper limits of quantitation were imputed from standard curves for cytokines above or below detection.

RPP and RPP-A cell lines were transduced to stably express OVA antigen using the pLVX–lucOS–IRES–Neo lentiviral vector (32). CD8⁺ T cells were isolated from 8- to 12-week-old C57BL/6-Tg (TcraTcrb)1100Mjb/J OT-I mice (stock 003831; Jackson Laboratory) using magnetic separation and LS columns per the manufacturer's protocol (cat. 130-049-401; Miltenyi Biotec). OT-I T cells were activated with Dynabeads Mouse T-Activator CD3/CD28 beads (cat. 11456D; Thermo Fisher Scientific) for 24 hours before coculture. RPP.OVA and RPP-A.OVA cells (target) were seeded in 96-well plates at a density of 10,000 cells per well and cocultured with activated OT-I CD8⁺ T cells (effector) at effector-to-target (E/T) ratios of 0, 1, 5, and 10 for 48 hours. Tumor cell survival was measured by luminescence as previously described (32). Percent survival was calculated by dividing luminescence of each E/T ratio by luminescence of appropriate tumor cell condition without T cells, and percent specific tumor cell lysis was calculated as 100 – percent survival.

The RPP SCLC mouse cell lines were derived from SCLC tumors that were generated in LSL-Cas9 BL6 mice that were intratracheally injected with adeno-associated virus that encode Cre–recombinase and sgRNAs targeting Rb1, Trp53, and Rbl2 (RPP; refs. 30, 31). Once tumors developed, independently infected mice (631 and 703) were euthanized with CO₂, and their tumors were quickly extracted, washed in ice-cold PBS, and minced several times using an ethanol-sterilized razor blade. Then, 3 mL collagenase/hyaluronidase (cat. 07912; Stemcell Technologies) diluted 1:10 in RPMI media containing 10% FBS, penicillin and streptomycin, and HITES and 1 mL dispase (cat. 354235; Corning) were added to the tumor and incubated at 37°C for 20 to 40 minutes (until most of the tumor cells were in suspension). The cells were then collected, centrifuged at 201 × g for 5 minutes, resuspended in RPMI HITES media, filtered through a 70-μm cell strainer (cat. 352350; Becton-Dickinson), centrifuged again at 201 × g for 5 minutes, resuspended in fresh RPMI HITES media, and placed in ultra-low-adherence tissue culture dishes (cat. 3471; Corning). Media were subsequently replaced every 3 days. Histopathology of the tumor from which the cell lines were derived showed SCLC. Immunoblot analysis was used to confirm expression of the neuroendocrine transcription factor ASCL1 and Cas9. RPP-A derivatives of the parental cell line were established by plating cells on tissue culture plates at 200,000 cells/mL for 4 days and the supernatant containing suspension cells and replacing with fresh complete media every 3 to 4 days until the RPP-A cell lines were established. CRISPR amplicon sequencing of the 631RPP and 631RPP-A cell lines was performed as described previously (30). Unless otherwise indicated, RPP and RPP-A designations refer to tumor cells derived from the 631 mouse [631RPP and 631RPP-A (30)].

RPP, RPP-A, or RPP-EZ cells (8 × 10⁶) in 1:1 PBS/Matrigel (Corning #354234) were subcutaneously injected into the flank of 8-week-old syngeneic C57BL/6 or NSG mice (Charles River Laboratories) anesthetized with isoflurane. Tumor volume was determined from caliper measurements of tumor length (L) and width (W) according to the formula (L × W)²/2. Both tumor size and body weight were measured three times per week. RPP-EZ tumors were injected with vehicle or ADU-S100 treatments (first treatment, 50 μg; second treatment, 35 μg) on day 19 of 23 for the larger cohort (n = 7) and on day 21 of 24 for the smaller cohort (n = 3 mice per group). Mice were euthanized at preset or humane endpoints and tumors collected and processed with gentleMACS C tubes (Miltenyi 130-093-237) per the manufacturer's protocol prior to flow cytometry. All experiments were conducted in accordance with a DFCI Institutional Animal Care and Use–approved protocol.

Viable CD45⁺ (20,000) and CD3⁺ (30,000) cells were isolated via FACS from pooled RPP or RPP-A tumors, washed, assessed for viability, and loaded onto a 10X Chromium instrument (10X Genomics) per the manufacturer's instructions. Single-cell RNA and TCR libraries were generated using the Chromium Single Cell 5′ Library Construction (cat. PN-1000020) and Chromium Single Cell V(D)J Enrichment Kit, Mouse T Cell kits (cat. PN-1000071; 10X Genomics) per user guide. Quality control of the completed libraries was performed using the Bioanalyzer High Sensitivity DNA Kit (cat. 5067–4626; Agilent) and then sequenced using the Illumina NextSeq 500 platform.

Single-cell RNA-seq data were aligned to the mouse genome (mm10). After initial filtering with default parameters, the feature matrix generated by Cell Ranger v 3.0.2 was used to perform downstream analysis using R toolkit Seurat (v.3.1.3; ref. 57). Cells expressing fewer than 100 or more than 7,000 genes and cells with more than 20% of unique molecular identifiers mapping to mitochondrial genes were removed from consideration. In addition, genes that were expressed in fewer than six cells were filtered out. The filtered matrix was log-normalized with scaling factor 10,000, centered genewise, and then underwent dimensionality reduction using principal component analysis (PCA) on the highly varying genes. Clustering was performed on the chosen principal components using the shared nearest neighbor algorithm as implemented in Seurat, with top 50 dimensions. The clusters were visualized using the UMAP algorithm. Marker genes for each cluster were identified with Seurat's function FindMarkers (Wilcoxon rank-sm test without thresholds). To identify dominant cell type in each cluster, we used the enrichment analysis using the AUCell R package (58), using cell type–specific previously published signatures (59). TCR chain assembly was performed using the Cell Ranger pipeline, with default parameters. Cells with low confidence or nonproductive TCR were filtered out. Cells with at least one α or β chain were used in integrative analysis with the transcriptome data. A clonotype was assigned to a specific cell type (cluster) if most of the cells with this clonotype belonged to the corresponding cluster. To integrate RPP and RPP-A data sets, functions FindIntegrationAnchors and IntegrateData were applied using top 20 dimensions of the transcriptome data. Approaches similar to those described above were used to perform clustering and identify cell types in the integrated data set, and function AverageExpression was used to obtain average expression of the genes within a cell type for RPP and RPP-A populations.

TCR cloning into Tcra;Tcrb-null BW5147.3 cells was adapted from Holst and colleagues (60) and as described in Brazin and colleagues (61). Briefly, the Trbv29-Trbc1-P2A-Trav14D-3-DV8-Trac1 minigene derived from TCR sequencing data was commercially synthesized (Integrated DNA Technologies) using optimized constant regions, and Gateway attb1 and attb2 sites were added via PCR cloning. The minigene was cloned into the Lenti plx_307 backbone, and 293T cells were transduced using X-treme Gene 9 (Roche) according to the manufacturer's instructions. On day 2, the supernatant of transduced 293T cells was collected through a 0.45-μm filter and added to the target cells in the presence of polybrene (8 μg/mL). Cells and virus were centrifuged at 2,000 rpm at 37°C for 2 hours. On day 5, puromycin (1 μg/mL) was added to select infected cells. After expansion, selected cells were sorted to >98% purity using an anti-mouse pan-TCRβ antibody (cat. 109211, clone H57-597; Biolegend) expression (BD FACSMelody).

Tumor cell–T-cell cocultures were performed at a stimulator/responder (S/R) ratio of 1:1. Tumor cells (100,000) were seeded in a 24-well plate and stimulated or not with IFNγ (100 ng/mL) for 24 hours. Tumor cells were washed once in PBS and replated in cell culture medium. After washing, 100,000 parental BW5147.3 or transgenic Trav14D-3-DV8, Trbv29 transgenic T cells were added to tumor cells, and cocultures were incubated for 72 hours, after which cell-free supernatant was harvested for ELISA.

Mouse cell–derived tumor spheroids were generated from RPP and RPP-A cells by seeding 1 × 10⁶ cells into a 6-well ultra-low attachment dish (cat. 3471; Corning) and incubating at 37°C, 5% CO₂ for 24 hours. Spheroids were stained with Cell Tracker Red (cat. C34552; Thermo Fisher Scientific) to allow identification using a fluorescence microscope. Collagen hydrogel was formulated to a concentration of 2.5 mg/mL using type I rat tail collagen (cat. 354236; Corning) following the addition of 10× PBS with phenol red with pH adjusted using NaOH to the 7.0 to 7.5 range, which was confirmed using PANPEHA Whatman paper (Sigma-Aldrich) as previously described (62, 63). The spheroids were then pelleted and resuspended in 250 μL collagen hydrogel, and 10 μL collagen–spheroid suspension was pipetted into the central gel region of each device of the microfluidic chip (cat. DAX-1; AIM Biotech) to generate a three-dimensional tumor microenvironment model. The devices were incubated for 40 minutes at 37°C in humidity chambers to allow ample collagen polymerization. After polymerization, fluorescently stained T cells (eBioscience cell proliferation dye eFluor 450, cat. 65–0842) were resuspended in RPMI supplemented with 10% FBS, 1% penicillin–streptomycin, and 2-mercaptoethanol. Then, 200 μL T-cell suspension, containing 5 × 10⁴ cells, was loaded into the fluidic channel on one side of the device. Next, 200 μL RPMI-supplemented media without cells was added to the media channel of the other side of the microfluidic device.

After 72 to 96 hours of incubation (37°C, 5% CO₂), T-cell migration was measured. Whole devices were imaged using a Nikon Eclipse 80i fluorescence microscope equipped with z-stack-capable stage (Prior) and an Andor Zyla camera (Andor). Image capture and analysis were performed using the NIS-Elements AR software package. Whole-device images were created by stitching in multiple captures at 4× magnification with 5% overlap. Nine z-stacks were taken 10 μm apart. Using the NIS software, an extended depth of focus image was generated, merging the multiple z-stacks into a single image. A binary layer was added to the image to exclude pixels below background intensity. A rectangular region of interest (ROI) was drawn encompassing the gel area, and the binary area (μm²) within the ROI was measured using the software's measurement tool. Quantification of cell migration/infiltration into the three-dimensional tumor microenvironment was performed by measuring the total cell area of cell tracker dye in the entire gel region as binary area.

All graphs depict mean ± SEM unless otherwise indicated. Tests for differences between two groups were performed using two-tailed unpaired Student t test or two-tailed Mann–Whitney test as specified in the figure legends. One-way ANOVA and Bonferroni correction was performed where applicable. Log-rank test was used for patient and mouse survival analyses, and the χ² test was used to compare observed and expected proportions with MHC^hi in our clinical cohort of patients experiencing durable clinical benefit from ICB (defined as 12-month progression-free survival). P values were considered significant if less than 0.05. Asterisks used to indicate significance correspond with *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. GraphPad Prism 9 (GraphPad Software) was used for statistical analysis of experiments, data processing, and presentation.

E.H. Knelson reports grants from Takeda outside the submitted work. P.C. Gokhale reports grants from Elstar Therapeutics, grants from Epizyme Inc, and grants from Daiichi Sankyo outside the submitted work. C.P. Paweletz reports personal fees from Dropworks and personal fees from BioRad outside the submitted work; Scientific Advisory Board—Dropworks and Scientific Advisory Board—XSphera Biosciences; research funding from Daiichi Sankyo, Bicycle Therapeutics, Transcenta, Bicara Therapeutics, AstraZeneca, Intellia Therapeutics, and Takeda. M.M. Awad reports personal fees from Merck, personal fees from ArcherDx, personal fees from Mirati, personal fees from Gritstone, personal fees from NextCure, personal fees from EMD Serono, personal fees from NovaRx, personal fees from Novartis, grants and personal fees from Bristol Myers Squibb, grants and personal fees from Lilly, grants and personal fees from AstraZeneca, grants from Genentech, personal fees from Maverick, personal fees from Nektar, personal fees from Syndax, and personal fees from Blueprint Medicine during the conduct of the study. E.M. Van Allen reports personal fees from Tango Therapeutics, personal fees from Genome Medical, personal fees from Invitae, personal fees from Monte Rosa Therapeutics, personal fees from Manifold Bio, personal fees from Illumina, grants from Novartis, grants from BMS, personal fees from Enara Bio, and personal fees from Janssen outside the submitted work; in addition, E.M. Van Allen has a patent for institutional patents filed on chromatin mutations and immunotherapy response, and methods for clinical interpretation pending. X. Wang reports being an employee and stock owner of Bristol-Myers Squibb. B. Chen reports personal fees from Bristol-Myers Squibb outside the submitted work. L.M. Sholl reports grants from Genentech, personal fees from AstraZeneca, personal fees from EMD Serono, and personal fees from Foghorn Therapeutics outside the submitted work. P.A. Jänne reports grants and personal fees from AstraZeneca, grants and personal fees from Boehringer Ingelheim, personal fees from Pfizer, personal fees from Roche/Genentech, personal fees from Chugai, personal fees from Ignyta, personal fees from LOXO Oncology, grants and personal fees from Eli Lilly, personal fees from SFJ Pharmaceuticals, personal fees from Voronoi, grants and personal fees from Daiiichi Sankyo, personal fees from Biocartis, personal fees from Novartis, personal fees from Sanofi Oncology, grants and personal fees from Takeda Oncology, personal fees from Mirati Therapeutics, personal fees from Transcenta, personal fees from Silicon Therapeutics, personal fees from Syndax, personal fees from Nuvalent, personal fees from Bayer, personal fees from Esai, grants from PUMA, grants from Astellas, and grants from Revolution Medicine outside the submitted work; in addition, P.A. Jänne is an inventor of a DFCI- owned patent on EGFR mutations issued and licensed to Lab Corp. E. Gjini reports other support from Bristol-Myers Squibb during the conduct of the study and currently is an employee of Bristol-Myers Squibb. R. Edwards reports other support from Bristol-Myers Squibb outside the submitted work, is an employee of Bristol-Myers Squibb and receives salary and owns stock. S. Rodig reports grants from Bristol-Myers Squibb during the conduct of the study; grants from Merck, grants from Affimed Inc, and grants from KITE/Gilead outside the submitted work; and is on the Scientific Advisory Board for Immunitas, Inc. M.G. Oser reports grants from Takeda Pharmaceutical Company, grants from Eli Lilly and Company, personal fees from Eli Lilly and Company, and personal fees from Cedilla Therapeutics outside the submitted work. D.A. Barbie reports personal fees from Qiagen and grants from BMS during the conduct of the study; personal fees and other from Xsphera, grants from Gilead, grants from Novartis, grants from Lilly/Loxo Oncology, and personal fees from Tango Therapeutics outside the submitted work. No disclosures were reported by the other authors.

N.R. Mahadevan: Conceptualization, formal analysis, supervision, investigation, methodology, writing–original draft, project administration, writing–review and editing. E.H. Knelson: Conceptualization, formal analysis, investigation, writing–review and editing. J.O. Wolff: Formal analysis, validation, investigation, methodology. A. Vajdi: Data curation, formal analysis, investigation, methodology. M. Saigi: Investigation. M. Campisi: Investigation, methodology. D. Hong: Investigation. T.C. Thai: Investigation. B. Piel: Investigation. S. Han: Investigation. B.B. Reinhold: Formal analysis, investigation, methodology. J.S. Duke-Cohan: Investigation. M.J. Poitras: Investigation. L.J. Taus: Investigation. P.H. Lizotte: Investigation, methodology. A. Portell: Investigation. V. Quadros: Investigation. A.D. Santucci: Investigation. T. Murayama: Investigation. I. Canadas: Conceptualization, investigation. S. Kitajima: Methodology. A. Akitsu: Methodology. M. Fridrikh: Investigation. H. Watanabe: Formal analysis, investigation. B. Reardon: Investigation. P.C. Gokhale: Project administration. C.P. Paweletz: Project administration. M.M. Awad: Project administration. E.M. Van Allen: Supervision. A. Lako: Project administration. X. Wang: Resources. B. Chen: Resources. F. Hong: Formal analysis. L.M. Sholl: Formal analysis, investigation. M.Y. Tolstorukov: Supervision. K. Pfaff: Supervision. P.A. Jänne: Supervision. E. Gjini: Supervision, investigation. R. Edwards: Resources. S. Rodig: Resources, formal analysis, supervision, methodology. E.L. Reinherz: Conceptualization, supervision, writing–review and editing. M.G. Oser: Conceptualization, resources, supervision, writing–review and editing. D.A. Barbie: Conceptualization, resources, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

We thank Jun Qi for reagents, as well as Marina Vivero and Maurizio Zanetti for insightful discussions on this work. Additionally, we thank the BMS Translational Pathology Histology Core Laboratory for preparation of H&E and unstained slides and the Dana-Farber/Harvard Cancer Center in Boston, MA, for the use of the Specialized Histopathology Core, which provided histology and immunohistochemistry service. This work was supported by NIH R01CA190294 (D.A. Barbie), the Parker Institute for Cancer Immunotherapy (D.A. Barbie, E.L. Reinherz), Dunkin Donuts Award (D.A. Barbie, E.L. Reinherz), Schaubert Family Funds (D.A. Barbie), Gross-Loh Research Fellowship (E.H. Knelson), and BMS II-ON Funding (D.A. Barbie, S. Rodig). Additional funding was provided by NIH T32HL7627-35 and NIH T32CA251062 (N.R. Mahadevan), NCI-T32 CA009172 (E.H. Knelson), and NCI/NIH K08CA222657 (M.G. Oser); Damon Runyon Cancer Research Foundation Clinical Investigator Award (M.G. Oser); Kaplan Family Funds (M.G. Oser); MIT-POLITO grant BIOMODE; Compagnia di San Paolo, under the joint “Doctorate of Bioengineering and Medical-Surgical Sciences” of University of Turin and Politecnico di Torino (M. Campisi); and Fundación CRIS contra el cáncer Grant (M. Saigi). Dana-Farber/Harvard Cancer Center is supported in part by an NCI Cancer Center Support Grant NIH 5 P30 CA06516. Further, we thank the Sequencing Facility of the Department of Oncological Sciences at Icahn School of Medicine at Mount Sinai (ISMMS) for sequencing assistance. This work was supported in part through Tisch Cancer Institute at ISMMS and the computational resources and staff expertise provided by Scientific Computing at ISMMS. Research reported in this article was supported by the Office of Research Infrastructure of the National Institutes of Health under award numbers S10OD018522 and S10OD026880. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. We thank Anika Adeni, Kesi Michael, and Arrien Bertram for assistance in obtaining patient specimens. We are grateful to Benjamin Ferland, Thomas Thayer, and Teri Bowman for technical assistance and Jennifer Largaespada and Hannah Cassatly for superior administrative assistance. We are grateful to Neal Lindeman, Lecia Sequist, Jim DeCaprio, Camille Kotton, Anna Farago, and Jacob Sands for invaluable mentoring throughout this project.

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.