Chronic type II interferon promotes tumor growth through mitochondrial RNA-induced type I interferon and prostaglandin synthesis

Abstract

Interferons are pro-inflammatory cytokines that promote immune cell engagement to eliminate malignant cells. Paradoxically, chronic interferon signaling can also activate anti-inflammatory mechanisms that allow cancer cells to evade the immune system. Here, we sought to determine the cellular mechanisms underlying this switch from anti-tumorigenic to pro-tumorigenic interferon activity. We show that chronic type II interferon (IFN-II) uniquely induced tumor growth by activating a type I interferon (IFN-I) response mediated by release of double-stranded mitochondrial RNA (ds-mtRNA) into the cytoplasm. This IFN-I signal synergized with IFN-II to enhance tumor growth by increasing immunosuppressive prostaglandin E₂ (PGE₂) synthesis via increased cyclooxygenase 2 (COX2) expression. Elimination of PGE₂ synthesis in immunotherapy-resistant melanoma cells restored their responsiveness to anti-PD1 treatment, indicating that this covert mtRNA-IFN-prostaglandin pathway could be a therapeutic target to combat immunotherapy resistance.

Inflammation is a key contributing factor to cancer initiation and malignant progression, as well as the efficacy of cancer immunotherapies (1). The immune system initially defends the host against cancer through cytokine-mediated recruitment of effector immune cells for clearance of aberrant, pre-malignant cells. However, when subsets of these cells survive this protective immune response, they can adapt and become malignant despite sustained immune pressure (2). While this acquired capacity of tumor cells to evade immune control is a well-established hallmark of cancer (3), precisely how various tumor-intrinsic and tumor-extrinsic factors converge to promote immune evasion remains incompletely understood.

Interferons are potent immunosurveillance cytokines (4) classified into three types: type I, type II and type III. Each type binds its cognate receptor to drive the expression of 60–1,000 interferon-stimulated genes (ISGs) (5), the pattern of which is dependent on interferon type, dose and signal duration (5). Type I and type II interferons (IFN-I and IFN-II) are the primary orchestrators of immune responses in the tumor microenvironment (TME) (2, 4). While principally known for their anti-tumor effects, interferons also activate negative feedback signaling loops that limit inflammation-induced pathology (6). Exploitation of the anti-inflammatory aspects of interferon signaling allows cancer cells to survive and continue to grow (7), and accumulating evidence shows that prolonged interferon exposure drives a switch from a tumor-suppressive interferon response to a tumor-promoting one by mechanisms that remain poorly defined (8–10). Therefore, in this study, we sought to define the cellular mechanisms that drive the transition from tumor-repressive to tumor-promoting interferon signaling using the well-characterized mouse YUMM1.7 melanoma cell model (11) that we have used previously to address fundamental aspects of tumor growth and tumor immune responses (12, 13).

Chronic IFN-II exposure in vitro promotes tumor growth in vivo

We modeled the differential effects of acute versus chronic interferon signaling on cancer cells by treating YUMM1.7 cells with either IFN-I (IFNα or IFNβ) or IFN-II (IFNγ) for 24 hours (acute) or 28 days (chronic). Acute IFN-I or IFN-II stimulation did not affect cell growth, while chronic IFN-I and IFN-II both decreased melanoma cell proliferation in vitro (Figures 1A and S1A). To assess the effects of acute versus chronic interferon stimulation in vitro on tumor growth in vivo, we transferred acute and chronic IFN-I or IFN-II-treated cells into the flanks of C57BL/6 mice and monitored tumor progression. Despite having the highest expression of MHC-I and PD-L1 (PD-L1 being comparable to acute IFN-II-treated cells, Figures 1B and 1C), which should promote T cell-mediated killing, chronic IFN-II-treated cells generated larger tumors than acute IFN-I, acute IFN-II or chronic IFN-I-treated cells (Figure 1D). Additional experiments comparing acute and chronic IFN-II confirmed the specific effect of chronic IFN-II treatment on tumor growth (Figure 1E). Chronic IFN-II treatment not only increased tumorigenesis in a flank tumor model, but also increased metastasis to the liver after splenic injection (Figure S1D and S1E). A similar pro-tumorgenic effect of chronic IFN-II compared to acute IFN-II was also observed in B16-F10 and YUMMER1.7 melanoma cells (Figures S1B and S1C). Lastly, tumor immune-cell profiling revealed a decrease in monocytes, an increase in neutrophils (Figures 1F, 1G and S1F) and reduced cytokine production by CD8⁺ T cells (i.e. less effector function), all of which is consistent with an immunosuppressive TME (Figures 1H, 1I and S1G).

Fig. 1. Chronic IFN-II stimulation in vitro increases melanoma cell tumorigenicity in vivo.

(A) Cell proliferation rate following acute or chronic IFN-I or IFN-II treatment, captured using the IncuCyte Live-Cell Analysis System with measurements representing the average percentage confluence per well at each timepoint (n=16 wells per treatment condition). Callout box shows confluence measurements for each group at 120 hours, the point at which control cells reached 100% confluence. (B-C) Bar graphs (top) and respective histograms (bottom) of MHC-I (B) and PD-L1 (C) surface expression (shown as mean fluorescence intensity on the day of cell implantation in vivo) after acute or chronic IFNβ or IFNγ treatment. (D) Tumor growth curves of YUMM1.7 cells implanted into the flanks of mice following acute or chronic IFNβ or IFNγ treatment (n=5 mice per condition). Lines represent the average volume per group and error bars represent SEM. Volumes at endpoint were used to determine statistical significance. (E) Combined final weights of tumors generated after implantation of acute or chronic IFNγ-treated cells in vivo. Histogram shows mean ± SEM of biological replicates (n=22 mice per condition). (F) Percentage of tumor-infiltrating monocytes, depicted as a percentage of the CD45⁺/CD11b⁺/Ly6G^- population ± SEM (n=4 mice per condition). (G) Percentage of tumor-infiltrating neutrophils, depicted as percent of CD45⁺ population ± SEM (n=4 mice per condition). (H) Percentage of CD8⁺ T cells that are double IFNγ⁺ and TNFα⁺ ± SEM (n=4 mice per condition). (I) Percentage of CD8⁺ T cells that are double PD1⁺ and IFNγ⁺ (n=4 mice per condition).

Immune profiling in E-H is representative of three separate replicates. Statistical significance was determined using one-way ANOVA with Dunnett’s multiple comparisons test for A and Welch’s t-test for D-I (ns = p>0.05, * = p≤0.05, ** = p≤0.01, *** = p≤0.001, **** = p≤0.0001).

Chronic IFN-II induces a tumor cell-intrinsic IFN-I response that drives enhanced tumor growth in vivo

To better understand the tumor cell-intrinsic drivers of the increased tumorigenicity uniquely resulting from chronic IFN-II treatment, we analyzed global gene expression changes in response to acute or chronic IFN-I and IFN-II treatment by RNA-seq. As expected, we found that differentially expressed genes (DEGs) induced by each of the interferons clustered based on type (IFN-I vs. IFN-II) for both acute and chronic treatment, and that chronic treatment differed from acute for respective IFN types, with the chronic IFN-II cluster separating furthest from the other treatment clusters (Figures S2A and S2B). Furthermore, IFN-II showed the largest increase in DEGs from acute to chronic stimulation, indicating a unique transcriptional response (Figure S2C). We also performed ATAC-seq on cells exposed to acute IFN-I and IFN-II, as well as chronic IFN-II, which revealed a dramatically different chromatin landscape between acute and chronic IFN-II signaling, with chronic IFN-II showing the largest number of peaks defining both gained and lost chromatin accessibility (Figure 2A), consistent with the larger number of DEGs observed. Deconvoluting specific transcriptional differences between these IFN-induced conditions using standard gene set enrichment analysis (GSEA) proved difficult due to the high degree of overlap between pre-defined interferon or inflammatory response pathway sets (Figure S2D), as well as the dominance of ISGs in each of our conditions and our own overlap between interferon types and duration. To address this, we developed a customized approach to assess nuanced differences in ISG subsets (and other genes) that changed in response to interferon type and duration, with the goal of better defining genes involved in the tumor-promoting, chronic IFN-II response we elucidated. This involved first defining interferon type-specific ‘classifier’ genes, for acute and chronic stimulation separately (i.e. genes that are differentially upregulated between interferon types, and between treatment and control). Expression patterns of interferon classifier genes defined this way plotted across all treatments reveals a unique expression signature in response to chronic IFN-II treatment (Figure S2E). Next, genes that were differentially expressed between interferon types were clustered based on changes in expression as a function of interferon treatment duration (Figure S2F). This revealed two signature gene clusters that changed based on interferon type during the transition from acute to chronic signaling (Figure S2F). Cluster A is a set of genes primarily associated with an IFN-I response after acute stimulation, but that changed to be primarily associated with an IFN-II response upon chronic signaling. Cluster B represents the inverse. To probe this IFN ‘type-switch’ in ISG expression further, we defined a more stringent list of interferon type-specific genes based only on acute interferon treatment. Doing this, we found that Cluster A stood out as a set of genes regulated by IFN-I in an acute setting, but was then upregulated by IFN-II in a chronic setting (Figure 2B). However, Cluster B did not withstand this more stringent criterion, revealing that IFN-II-specific genes are not induced strongly by chronic IFN-I treatment (Figure 2C). Based on this analysis, Irf7 is a signature example of an IFN-I-specific gene and Irf1 is a signature example of an IFN-II-specific gene, distinctions which have been recently validated across several cell lines (14) and will be used diagnostically in experiments going forward.

Fig. 2. Chronic IFN-II exposure induces IFN-I production and signaling in melanoma cells that enhances their tumorigenicity.

(A) Venn diagrams showing gained and lost accessibility sites between indicated treatments determined by ATAC-seq. (B) Heatmap showing type I IFN-specific differentially expressed gene (DEG) list, defined by acute IFN expression pattern. Type I genes must have a minimum log₂FC greater than 2 compared to control and be more than twice as highly induced by IFN-I than IFN-II ((Type I lfc- Type II lfc) > 2). (C) Type II IFN-specific gene list, defined by acute IFN expression pattern. To be considered a Type II gene, it must have a minimum log₂FC greater than 2 compared to control and be more than twice as highly induced by IFN-II than IFN-I ((Type II lfc- Type I lfc) > 2). (D) Percentage overlap of acute or chronic IFN-II STAT1 CUT&RUN peaks with acute IFN-I peaks. IFN-I represented by IFNβ. (E) Percentage overlap of acute or chronic IFN-II STAT2 CUT&RUN peaks with acute IFN-I peaks. IFN-I represented by IFNβ. (F) mRNA levels of Ifnb1 in control, acute and chronic IFN-II-treated cells determined by RT-qPCR. Dots represent biological replicates (n=6) and bars represent mean ± SEM. (G) Level of IFNβ protein in control, acute and chronic IFN-II-treated cell supernatants determined by ELISA. Dots represent biological replicates (n=3) and lines represent mean ± SEM. (H) Tumor growth curves of acute and chronic IFN-II-treated control, STAT2, and IFNAR KO cells implanted into the flanks of mice (n=4 mice per group). Lines represent the average volume per group and error bars represent SEM. (I) Final tumor weights of acute and chronic IFN-II-treated control, STAT2 and IFNAR KO tumors. Histograms represent mean ± SEM. (J) Images of resultant tumors in each group; all images are from the same photograph and therefore scaled equivalently.

Statistical significance was determined using one-way ANOVA with Dunnett’s multiple comparisons test for F-H and one-way ANOVA with Sidak’s multiple comparisons test for I (ns = p>0.05, * = p≤0.05, ** = p≤0.01, *** = p≤0.001, **** = p≤0.0001).

Since STAT1 and STAT2 are the main transcriptional regulators of IFN-I and IFN-II signaling, we performed CUT&RUN for STAT1 and STAT2 under conditions of acute and chronic IFN-I (IFNβ) or IFN-II (IFNγ) stimulation. We found that the overlap between IFN-I and IFN-II differential peaks increased for both STAT1 and STAT2 in response to chronic IFN-II (Figures 3D and 3E). At “Type I” ISGs (Figure 2B), STAT1 and STAT2 peaks were present in acute IFN-I, but not acute IFN-II conditions (Figure S3A). In contrast, STAT1 and STAT2 were bound to “Type I” ISGs in both chronic IFN-I and chronic IFN-II (Figure S3A), consistent with the increased “Type I” gene expression pattern observed from RNA-seq in chronic IFN-II (Figure 2B). At “Type II” ISGs (Figure 2C), we observed STAT1 binding in acute IFN-II, but not acute IFN-I (Figure S3B). Additionally, both STAT1 and STAT2 show increased binding to the majority of “Type II” ISGs in chronic IFN-II, but not chronic IFN-I stimulation (Figure S3B), which is also consistent with amplified expression of “Type II” ISGs observed from RNA-seq (Figure 2C). Together, these data show that STAT1, and particularly STAT2, gained binding at both ISG types under chronic IFN-II stimulation. Representative tracks showing STAT1 and STAT2 peaks overlayed with ATAC peaks (Figure S3C) show that STAT2 binding increases at both “IFN-I” and “IFN-II” genes in the chronic IFN-II condition. All the data above indicate that IFN-I signaling is induced after chronic IFN-II stimulation, which was confirmed by increased expression of IFNβ, both at the RNA and protein level (Figures 2F and 2G). Therefore, we conclude that YUMM1.7 melanoma cells that experience chronic IFN-II unleash intrinsic IFN-I signaling, but not vice-versa. To extend these findings into a more relevant human model, A375 cells (human melanoma) were subjected to the same chronic IFN regimen as described for YUMM1.7 cells, and a similar IFN-I signature was induced after chronic IFN-II treatment (Figures S4A and S4B).

Fig. 3. Chronic IFN-II-mediated IFN-I induction occurs via cytosolic release of mitochondrial RNA.

(A) Ifnb1, Irf7 and Irf1 mRNA expression in SCR, STING KO, MAVS KO, or STING/MAVS double KO (DKO) YUMM1.7 cells after chronic IFN-II treatment, measured by RT-qPCR. Each dot represents a biological replicate, bars represent mean ± SEM (n=10 control, n=7 STING and MAVS KO, n=3 DKO). (B) Representative images taken by automated spinning disk confocal microscopy showing increased MAVS clustering following chronic IFN-II treatment in YUMM1.7 cells. TOMM20 (mitochondria, magenta), MAVS (cyan). Scale bars, 10μm, 2μm (inset). Quantification (right) was performed on >400 cells for each treatment group across 3 independent experiments. Each dot represents one cell. (C) Representative images of dsRNA staining with J2 antibody for quantification of cytosolic dsRNA in control vs. chronic IFN-II-treated YUMM1.7 cells. Top two rows were treated with vehicle control and bottom two rows were treated with IMT1B for 96 hrs prior to fixing. dsRNA (cyan), TOMM20 (mitochondria, magenta). Scale bars, 10μm, 2μm (inset). Quantification of cytosolic dsRNA in vehicle vs. IMT1B-treated groups (right). Each dot represents mean fluorescence intensity (MFI) of one cell (n=15–21 cells per condition), performed across 3 independent experiments. (D) Representative graph of Ifnb1, Irf7 and Irf1 mRNA expression following 96hrs IMT1B treatment in chronic IFN-II-treated YUMM1.7 cells determined by RT-qPCR. Each dot represents one technical replicate, bars represent mean ± SD. Experiment was performed in triplicate. (E) Representative graphs depicting levels of mitochondrial transcripts in control vs. chronic IFN-II-treated YUMM1.7 cells in whole cell (left) and cytosolic (right) fractions, determined by RT-qPCR. Each dot represents one technical replicate, bars represent mean ± SD (n=3). Cell fractionation and subsequent RT-qPCR was performed in triplicate. (F) Representative images of YUMM1.7 control (left) and chronic IFN-II-treated (right) cells stably expressing a MitoTag GFP construct (cyan) and stained by RNA-FISH targeting the mitochondrial transcript Rnr2 (red), counterstained with Hoescht (yellow, nucleus). Top row was treated with vehicle, middle row was treated with VBIT-4 for 96hrs, and bottom row was treated with IMT1B for 96hrs. Scale bars, 10μm, 2μm (inset). Quantification of extramitochondrial mtRNA Rnr2 transcript foci is shown on the far right (n=32–44 cells per condition, imaged across a minimum of 4 independent experiments).

Statistical significance was determined using one-way ANOVA with Dunnett’s multiple comparisons test for A, unpaired Welch’s t-test for B, two-way ANOVA with Sidak’s multiple comparisons test for C-E, and one-way ANOVA with Sidak’s multiple comparisons test for F (ns = p>0.05, * = p≤0.05, ** = p≤0.01, *** = p≤0.001, **** = p≤0.0001).

To determine if the IFN-I signaling that emanates from chronic IFN-II exposure contributes to the enhanced tumor growth we observed (Figures 1D and 1E), we knocked out 1) the interferon alpha receptor (IFNAR), which eliminates recognition of autocrine or paracrine IFN-I signals (Figure S5A, S5B and S15A), and 2) STAT2, which eliminates the formation of the transcription factor ISGF3 needed for induction of IFN-I ISGs (Figure S5A and S5C). These knockout cell lines were then subjected to the chronic IFN-II treatment regimen and subsequently implanted into C57BL/6 mice. Knock-out of either IFNAR or STAT2 reversed the enhanced in vivo growth of chronic IFN-II-treated cells (Figures 2H, 2I and 2J). RT-qPCR analysis of Irf7 and Irf1 in STAT2 and IFNAR knockouts showed that the IFN-I signal was eliminated and IFN-II signaling remained intact, respectively (Figure S5D and S5E). Moreover, B2m (proxy for MHC-I) and Pdl1 expression was comparable between the chronic IFN-II treated control cells and the STAT2 or IFNAR knockout cells, showing that antigen presentation and immune checkpoint signaling are not overtly affected (Figure S5F and S5G). These results demostrate that the IFN-I signal instigated by chronic IFN-II stimulation is required for the increased tumorigenicity of chronic IFN-II treated melanoma cells.

Mitochondrial RNA release drives IFN-I signaling in response to chronic IFN-II

We next endeavored to determine the source of the IFN-I signal observed in chronic IFN-II-treated cells. Induction of a sterile IFN-I response is most commonly driven by mislocalized cellular nucleic acids, with STING and MAVS serving as central signaling proteins for cytosolic DNA- and RNA-sensing pathways, respectively. Therefore, we generated STING, MAVS, and STING/MAVS double knockout (DKO) cell lines and treated them chronically with IFN-II to determine whether IFN-I induction was driven by cytosolic DNA, RNA, or both (Figures S6A, S6B and S6C). In all three lines, the IFN-II response remained intact as measured by Irf1 mRNA levels (Figure 3A), however, knockout of MAVS blunted IFN-I signaling as measured by Ifnb1 and Irf7 mRNA levels, while knockout of STING had no effect alone and no additive effect in combination with MAVS in the DKO (Figure 3A). With MAVS now implicated genetically, we next tested directly whether MAVS activation was occurring following chronic IFN-II treatment using confocal microscopy to visualize and quantify clustering of MAVS, an commonly used indicator of MAVS activation (15–17). We observed a significant increase in MAVS clustering in IFN-II-treated cells, confirming activation of MAVS (Figure 3B). That these puncta represent bona fide MAVS clusters was confirmed by their absence in chronic IFN-II-treated MAVS KO cells (Figure S7A), and by their increased presence after acute Poly(I:C) treatment, albeit to a lesser degree than chronic IFN-II-treated cells (Figure S7A). In addition, we observed both mitochondrial and non-mitochondrial localization of MAVS clusters and quantified this. Both types were increased in Poly(I:C) and chronic IFN-II-treated cells, but to a much larger degree in the latter (Figure S7A). This pattern may represent a relevant manifestation of chronic cytosolic RNA signaling, as non-mitochondrial MAVS puncta have been previously associated with signaling from peroxisomes (18), as well as trafficking through the endolysosomal pathway to temper signaling output (19), suggesting that chronic IFN-II induces sustained MAVS activation that the cells are trying to regulate. Although MAVS RNA signaling is clearly a major component of the IFN-I signal we observe in chronic IFN-II-treated cells, it was not accounting for the entire ISG response (Figure 3A), suggesting other innate RNA sensors are involved. Consistent with this, via KO of upstream RNA sensors, we also found a significant contribution from TLR3 and PKR, with minimal or variable contribution from MDA5, RIG-I and OAS1 (Figure S6I). Together, these findings established RNA sensing as the dominant mechanism underlying IFN-I induction following chronic IFN-II treatment. We therefore confirmed the presence of immunostimulatory double-stranded RNA (dsRNA) in the cytosol of chronic IFN-II-treated cells using the J2 antibody (Figure 3C).

We next sought to determine the cellular origin of this immunostimulatory dsRNA, as identifying its source could provide mechanistic insight into how chronic IFN-II treatment initiates IFN-I signaling. Under sterile conditions, cytosolic immunostimulatory RNA can originate either from the nucleus or mitochondria. We initially hypothesized that mitochondrial RNA (mtRNA) was a contributor since bi-directional transcription of the circular mtDNA is a well-established source of completely base-paired dsRNA that activates IFN-I signaling when released into the cytosol under conditions of cellular stress (20–23). Therefore, we inhibited mitochondrial transcription in cells that had already undergone chronic IFN-II treatment using IMT1B, a specifc inhibitor of the dedicated mtRNA polymerase, POLRMT (24) (Figure S6D). IMT1B treatment did not affect the IFN-II response (Figure 3D), but was sufficient to reduce the IFN-I signal (Figure 3D), IFNβ production (Figure S6E), MAVS clustering (Figure S7B) and the amount of dsRNA in the cytoplasm (Figure 3C). Further confirming the role of mtRNA as a source of the IFN-I induction in chronic IFN-II-treated. cells, we found an increase in mtRNA in purified cytoplasmic fractions by RT-qPCR (Figure 3E, S6G and S6H). Interestingly, no increase in cytosolic mtDNA, which is also highly immunostimulatory (20, 25), was observed (Figure S6F), consistent with STING KO having no effect. We also performed single-molecule mtRNA-FISH using probes specific for two different mtDNA-encoded transcripts, mt-Nd4 and mt-Rnr2, and quantified the number of mtRNA foci outside of mitochondria by confocal microscopy. Both probes revealed an increase in extramitochondrial foci in chronic IFN-II-treated cells, confirming mtRNA is released following chronic IFN-II treatment (Figures 3F, S7D).

While there are multiple possible mechanisms of mtRNA release, it has been reported that mtRNA can be released through VDAC pores to initiate IFNβ production (26). Therefore, we tested whether inhibition of VDAC-mediated mitochondrial nucleic acid release by VBIT-4 treatment affected extramitochondrial mtRNA foci and found that cytosolic mtRNA foci are decreased with VBIT-4 treatment in chronic IFN-II treated cells (Figures 3F and S7D). VBIT-4 treatment also decreased MAVS clustering as quantified by confocal microscopy and IFN-I signaling as quantified by RT-qPCR (Figures S7B and S7C). Furthermore, chronic IFN-II-induced IFN-I signaling was also observed in human melanoma (A375) cells (Figures S4A and S4B), where it was also MAVS-dependent (Figures S8A-D) and accompanied by elevated cytosolic dsRNA that was reversed by IMT1B treatment (Figure S8E). These results establish that chronic IFN-II stimulation drives the release of ds-mtRNA which activates MAVS-dependent IFN-I signaling in both mouse and human melanoma cells.

Chronic IFN-II stimulation upregulates COX2-dependent prostaglandin synthesis

Having established that chronic IFN-II treatment induces the release of ds-mtRNA into the cytosol, we next asked whether this phenotype was accompanied by alterations in mitochondrial oxidative phosphorylation (OXPHOS), as this is a known precursor to mitochondrial nucleic acid release (20–23). In chronic IFN-II treated YUMM1.7 cells, we observed decreased OXPHOS and a shift toward increased glycolysis, as evidenced by decreased oxygen consumption by the mitochondrial electron transport chain (ETC), decreased citrate abundance and increased lactate abundance (Figure S9A-S9D). This reduction in ETC activity was accompanied by increased mitochondrial membrane potential (i.e. hyperpolarization) and increased mitochondrial and cytosolic ROS (Figure S9E-S9J). Together, these data suggest that mtRNA release is associated with altered mitochondrial OXPHOS following chronic IFN-II exposure, however, the precise relationship between these phenotypes remains to be defined.

Given we were observing significant mitochondrial metabolic changes, we reasoned that chronic IFN-II signaling may more broadly reprogram cellular metabolism. To systematically assess this, we probed our RNA-seq dataset for changes in expression of metabolic genes that were unique to chronic IFN-II treatment. We assigned scores to each treatment condition for every KEGG metabolic pathway (Figure S10A) and plotted the difference between IFN-II and IFN-I expression scores for acute versus chronic conditions, yielding quadrants indicative of IFN type-predominance of each pathway (Figure 4A). Three polyunsaturated fatty acid (PUFA) metabolism pathways (highlighted in purple in Figure 4A) stood out as more strongly induced by IFN-I in an acute setting, but by IFN-II in a chronic setting, with genes representing arachidonic acid and ether lipid metabolism being the strongest contributors to this signature (Figure 4B). Therefore, we next directly profiled fatty acid (FA) metabolic parameters in acute and chronic IFN-II-treated cells. Total saponified FAs were increased in response to both acute and chronic IFN-II treatment (Figure S10B), and chronic IFN-II-treated cells proliferated much more slowly than control cells when cultured in lipid-depleted FBS (Figure S10C), suggesting increased reliance on exogenous FA uptake. Fatty acid oxidation rate is low following chronic IFN-II treatment (Figure S10D), consistent with decreased OXPHOS and indicative that the FA uptake occurring in these cells is not for the purpose of ATP production through β-oxidation. Thus, we hypothesized that chronic IFN-II-treated cells are utilizing fatty acids for another, non-bioenergetic purpose.

Fig. 4. IFN-I and IFN-II synergize to upregulate prostaglandin synthesis.

(A) Comparison of KEGG metabolic pathway expression across YUMM1.7 treatment groups. For each pathway, a group score was calculated and assigned values according to its relative Type I vs. Type II expression in acute (y-axis) and chronic (x-axis) treatments. Terms with a positive median score in chronic Type II exposure, a p-value < 0.1, and a (chronic - acute) score > 0.2 were identified as noteworthy and annotated (red and purple). Noteworthy terms following the pattern of acute Type I and chronic Type II induction (bottom right quadrant) are denoted in purple. (B) Genes that cluster as most highly expressed in chronic IFNγ across the three pathways identified in (A). (C) Quantification of secreted eicosanoid species from acute and chronic IFN-I or IFN-II-treated cells, determined by LC-MS. Data normalized to cell number. (D-E) Western blot analysis of COX2 expression in SCR, STAT2 and IFNAR knockout cells that underwent chronic IFNγ treatment (n=3). (F) Tumor weights following injection of acute or chronic IFNγ-treated YUMM1.7 cells with or without aspirin administration. (G) Combined tumor weights of three replicates of tumor implantation of acute or chronic IFNγ-treated cells with or without dual knockout of COX1/COX2 following IFNγ treatment. Plot shows mean ± SEM of biological replicates (dots represent individual mice). (H) PGE₂ concentration in cultures of control, acute IFNγ chronic IFNγ or chronic IFNγ-treated cells with subsequent dual knockout of COX1/COX2, measured by ELISA. Bars depict mean ± SEM (n=3). (I) Percentage of tumor-infiltrating monocytes, depicted as a percentage of the CD45⁺/CD11b⁺/Ly6G^- population ± SEM (n=5 mice per condition). (J) Percentage of tumor-infiltrating neutrophils, depicted as percent of CD45⁺ population ± SEM (n=5 mice per condition). (K) Percentage of CD8⁺ T cells that are double PD1⁺ and IFNγ⁺ ± SEM (n=5 mice per condition). (L) Fraction of CD8⁺ T cell infiltration, depicted as percentage of CD8⁺ T cells as a portion of the CD45⁺ population ± SEM (n=5 mice per condition).

Statistical significance was determined using one-way ANOVA with Sidak’s multiple comparisons test for F-L (ns = p>0.05, * = p≤0.05, ** = p≤0.01, *** = p≤0.001, **** = p≤0.0001).

Consistent with these findings, PUFA levels were elevated following chronic IFN-II treatment (Figure S10B and S10G), which in combination with high expression of PUFA metabolizing enzymes (Figure 4A, 4B and S10A), suggested engagement of downstream lipid signaling pathways. Among these, phospholipase A2 (PLA2) stood out as a commonly upregulated enzyme across PUFA metabolism pathways (Figure 4B), linking PUFA availability to downstream eicosanoid production through esterification of arachidonic acid (27) (Figure S10E). A comprehensive eicosanoid lipidomic analysis of supernatants from melanoma cells treated with acute or chronic IFN-I or IFN-II revealed increased abundance of nearly all eicosanoid species, most prominantly in chronic IFN-II-treated samples, with prostaglandins representing the dominant class both basally and after IFN stimulation (Figures 4C and S10H). Amplification of COX2, which is the inducible isoform of cyclooxygenase responsible for ~90% of the PGE₂ production in Braf^V600E mutant melanoma cells such as YUMM1.7 (28), was also confirmed at the protein level after chronic IFN-II treatment (Figure S10F).

We next measured Ptgs2 (gene encoding COX2) expression in chronic IFN-II-treated cells in which innate immune IFN-I signaling was disrupted to determine whether these pathways were linked. Paralleling what we observed for the induction of IFN-I (Figures 3A and S6I), chronic IFN-II-induced upregulation of Ptgs2 mRNA was blunted in MAVS, DKO, TLR3 and PKR KO YUMM1.7 cells, but not in STING KO cells (Figures S11A and S11C), showing that the innate immune RNA-sensing pathways positively regulates its expression. Similarly, PTGS2 induction was reduced in MAVS and DKO A375 human melanoma cells (Figure S11B). Further solidifying the connection between IFN-II-induced IFN-I and Ptgs2/COX2 upregulation, chronic IFN-II-dependent amplification of COX2 protein was also blunted in both STAT2 and IFNAR KO cells (Figures 4D and 4E). These data led us to hypothesize that IFN-I and IFN-II cooperatively amplify Ptgs2/COX2 expression. To test this directly, we treated cells acutely (24 hours) with IFN-I, IFN-II or both and found that while each cytokine modestly induced COX2 expression, combined treatment resulted in a synergistic increase (Figure S11D). This effect was dependent on STAT1 and STAT2, albeit to different degrees, but not on STAT3 (Figures S11E-S11F). Finally, chronic IFN-II stimulation induced Ifnb1, Irf7 and Ptgs2 expression across multiple tumor models, including YUMM1.7, A375 (human melanoma), B16-F10 (mouse melanoma), and FC1199 (mouse pancreatic ductal adenocarcinoma (PDAC)), showing that the induced IFN-I signaling and subsequent COX2 regulation is not idiosyncratic to YUMM1.7 mouse melanoma cells (Figures S12A-S12L). All of the above results, coupled to the fact that prostaglandins are well-known for their immune-suppressive and pro-tumorigenic effects (29), led us to next test the role of prostaglandins in the enhanced tumorigenic phenotype imparted by chronic IFN-II on melanoma cells.

Prostaglandin production drives chronic IFN-II-induced tumor growth

To evaluate the functional contribution of prostaglandins to the pro-tumorigenic effect of chronic IFN-II treatment, we first tested whether systemic administration of aspirin, a COX1/2 inhibitor would reverse the enhanced tumor growth. Mice were injected with either acute or chronic IFN-II-treated melanoma cells, then half of each cohort was administered aspirin in their drinking water (Figure S13A). While aspirin had minimal effect on tumor growth in the mice implanted with acute IFN-II-treated cells, it completely reversed the enhanced tumor growth of chronic IFN-II-treated cells (Figure 4F). Although consistent with tumor-derived prostaglandins being responsible for the enhanced tumorigenesis, aspirin is a systemic intervention that could affect many cells. Thus, to investigate the role of tumor-derived prostaglandins, we performed a dual knockout of COX1 and COX2 (COX1/2 KO) in YUMM1.7 melanoma cells that had already undergone chronic IFN-II treatment before implanting into mice (Figures S15B-S15E). Similar to aspirin, the dual KO of COX1 and COX2 had minimal effect on tumor growth of acute IFN-II-treated cells, but completely reversed the enhanced tumor growth of the chronic IFN-II-treated cells (Figures 4G and S13B-S13D). As expected, double KO of COX1 and COX2 eliminated PGE₂ production in chronic IFN-II cells (Figure 4H). Moreover, dual COX1/2 KO also reversed the immunosuppressive phenotype observed in tumors from chronic IFN-II-treated cells as evidenced by increased monocyte and decreased neutrophil infiltration, as well as restored CD8⁺ T cell cytokine production in chronic IFN-II COX1/2 KO tumors compared to chronic IFN-II control tumors (Figures 4I–4K). This occurred without any overt effect of COX1/2 knockout on IFN-II signaling, B2m or Pdl1 expression (Figures S13E-S13G). In addition, higher CD8⁺ T cell infiltration was observed (Figure 4L). Lastly, Irf7 expression remained intact in the COX1/2 knockout cells (Figure S13H), confirming that IFN-I signaling was unaffected and hence demonstrating that prostaglandin production is downstream of IFN-I induction.

Acquired resistance to immunotherapy is associated with increased Ptgs2/COX2 expression

While the YUMM1.7 cell line used for most of this study is a useful tool for studying cancer biology, it is minimally immunogenic, as evidenced by similar growth rates in immunodeficient and immunocompetent mice (11), and thus not an ideal model for studying chronic tumor interferon responses in vivo. The YUMMER1.7 cell line (generated by UV irradiation of YUMM1.7) is more immunogenic and responds to immune checkpoint therapies (30). Additionally, YUMMER1.7 cells treated with chronic IFN-II in vitro produced significantly higher levels of PGE₂ (Figure S14A), showing that the IFN-prostaglandin pathway is engaged in this cancer model. Therefore, we employed the YUMMER1.7 tumor model with an immunotherapy regimen to better address how chronic IFN-II and upregulated prostaglandin synthesis is involved in tumorigenesis and immunotherapy responses in vivo (Figure 5A). We implanted YUMMER1.7 cells into the flanks of C57BL/6 mice and administered anti-PD1 immunotherapy. Tracking tumor growth kinetics of each mouse allowed us to divide the resultant tumors into three categories: responders, partial responders, and non-responders (Figure S14B). In the first round of tumor implantation and immunotherapy treatment, anti-PD1 responders had the lowest expression of Ptgs2 (COX2) (Figure 5B). Next, we generated a YUMMER1.7 cell line from a tumor which initially responded to anti-PD1 immunotherapy and re-implanted these cells into another set of mice to determine if they would develop resistance to anti-PD1 therapy. Indeed, a subset of mice failed to respond to immunotherapy after re-implantation of the responsive cell line (Figure S14C) and these had significantly higher expression of Ptgs2 (COX2) (Figure 5C). Finally, to ensure more consistent resistance to anti-PD1 immunotherapy, we took a non-responding tumor cell line and subjected it to a third round of anti-PD1 immunotherapy, and again saw an increase in Ptgs2 expression (Figure 5D). Finally, to directly test the role of COX2-PGE₂ signaling in this acquired resistance, we generated COX1/2 knockout lines from the immunotherapy-resistant cells (Figures S15H and S15I) and re-implanted them alongside scramble control immunotherapy-resistant counterparts. Strikingly, loss of COX1/2 largely restored sensitivity to immunotherapy, with nine out of ten mice exhibiting complete and sustained tumor regression (Figures 5E and 5F). Similarly, knockout of COX1/2 in the YUMMER1.7 parental line (Figures S15F and S15G) signicantly increased anti-PD1 response, with all mice now showing complete remission with anti-PD1 immunotherapy (Figure S14E). These results establish COX2-PGE2 signaling as a critical downstream effector of chronic IFN-II-driven tumor progression and immunotherapy resistance.

Fig. 5. Increased tumor-cell-intrinsic prostaglandin synthesis mediates melanoma immunotherapy resistance.

(A) Schematic representation of the generation of a YUMMER1.7 anti-PD1 immunotherapy resistant cell line. Parental YUMMER1.7 cells underwent three rounds of in vivo passaging, transitioning from a responding into a non-responding tumor cell line. Ptgs2 expression was assessed in each resultant tumor cell line following rounds 1–3 of in vivo growth. (B) Ptgs2 mRNA expression in round 1 tumor cell lines (growth curves shown in Figure S14B), determined by RT-qPCR. Bars depict mean ± SEM of biological replicates. (C) Ptgs2 mRNA expression in round 2 tumor cell lines (growth curves shown in Figure S14C), determined by RT-qPCR. Bars depict mean ± SEM of biological replicates. (D) Ptgs2 mRNA expression in round 3 tumor cell lines (growth curves shown in Figure S14D), determined by RT-qPCR. Bars depict mean ± SEM of biological replicates. (E) Tumor growth curves of anti-PD1-resistant YUMMER1.7 tumor cells isolated from the final round of immunotherapy following re-implantation of either control or COX1/2 knockout lines and subsequent anti-PD1 treatment. Lines depict averages of n=10 mice per group. (F) Individual tumor growth curves of either control or COX1/2 knockout YUMMER1.7 immunotherapy resistant cell lines.

Statistical significance was determined using one-way ANOVA with Sidak’s multiple comparisons test for B-D (ns = p>0.05, * = p≤0.05, ** = p≤0.01, *** = p≤0.001, **** = p≤0.0001).

Conclusions

This study provides new mechanistic insight into the important issue of how chronic IFN-II signaling switches the activity of this critical cytokine from ant-tumorigenic to pro-tumorigenic, a transition that suppresses anti-tumor immunity and immunotherapy responses. Specifically, we show that chronic IFN-II exposure instigates release of dsRNA from mitochondria that activates MAVS-dependent IFN-I signaling both in mouse and human melanoma cells. These IFN-II and IFN-I signals then synergize to upregulate COX2-dependent PGE₂ synthesis, an immunosuppressive pathway implicated in cancer progression and chemo-, immuno-, and targeted therapy resistance (29, 31–38). While PGE₂-mediated immune suppression and TME remodeling has been characterized (38–44), our results directly link chronic IFN-II signaling in tumor cells to activation of this pathway. Chronic IFN-II also resulted in lower mitochondrial OXPHOS (oxygen consumption rate), increased mitochondrial membrane potential and elevated ROS. While we did not establish a causal link between these mitochondrial phenotypes and ds-mtRNA release, these and other forms of mitochondrial stress are known to instigate mtDNA and/or mtRNA release (20, 45, 46), suggesting that chronic IFN-II-driven mitochondrial remodeling facilitates ds-mtRNA release and subsequent IFN-I induction. Consistent with this, ds-mtRNA release was inhibited by VBIT-4 (Figures 3F and S7D), which directly or indirectly inhibits VDAC-dependent mitochondrial nucleic acid release (26, 47, 48). From our results, we conclude that IFN-I induction downstream of chronic IFN-II exposure arises from a network of partially redundant RNA-sensing mechanisms, involving both MAVS-dependent and MAVS-independent components. Moreover, MAVS itself has diverse functions beyond innate immune signaling, including roles in apoptosis (49), autophagy and mitochondrial dynamics (50, 51), and has been reported to exert both pro- and anti-tumorigenic activities in a context- and tumor-type-dependent manner (52, 53). Consequently, the precise contribution of MAVS and mtRNA-driven signaling to tumor progression will require additional in vivo approaches that systematically dissect this signaling complexity. Furthermore, while we implicate ds-mtRNA as a major contributor, our study does not discount the possibility of nuclear RNA species also being involved in the IFN-I activation. Despite these limitations, this study establishes a previously unrecognized mechanism by which chronic IFN-II signaling generates a broad RNA-dependent IFN-I response that is required for its pro-tumorigenic effects.

IFN-I signaling is known to regulate the balance between immune-mediated tumor elimination or tumor escape (9), and chronic IFN-II exposure initiates epigenetic modifications that preferentially enhance IFN-I-related gene expression and hinder CD8⁺ T cell function and immunotherapy responses (10). Consistent with this, two recent clinical trials have shown that inhibition of JAK kinases that are central to IFN signaling enhances the clinical efficacy of anti-PD-1 immunotherapy (54, 55). However, specific ISGs profoundly and differentially impact patient responses to immunotherapy (8–10, 41), emphasizing the need for a more refined understanding of context-dependent IFN signaling and the functional roles of specific ISGs in terms of cancer growth and therapeutic intervention. In this regard, our results illuminate a mechaninstic basis for how a covert IFN-I signal is liberated in response to chronic IFN-II, marking a pivotal point at which IFN-II signaling shifts toward promoting tumor progression, with enhanced COX2-dependent PGE₂ production explaining the potent pro-tumorigenic effects of this signaling cascade. Importantly, this signaling axis was conserved across both mouse and human melanoma cells, supporting its potential translational relevance. Therefore, this study points to new potential therapeutic avenues such as inhibiting mtRNA release, targeting downstream consequences on lipid metabolism, or fine-tuning how and when to inhibit IFN-I signaling or COX2 activity to halt cancer progression and/or enhance immunotherapy responses.

Supplementary Material

Materials and Methods

Figs. S1 to S15

Tables S1 to S14

References 56–73

Data and material availability:

Data for YUMM1.7 RNA-seq, ATAC-seq and CUT&RUN are available at the Gene Expression Omnibus (GEO) under accession codes GSE307166, GSE307167 and GSE307168 and for A375 RNA-seq under accession code GSE330381.