Opposing Functions of Interferon Coordinate Adaptive and Innate Immune Responses to Cancer Immune Checkpoint Blockade

•Blocking tumor IFNG signaling increases IFNG generated by exhausted T cells (TEX)•Higher immune vs. cancer ISGs disable inhibitory pathways, allows NK/ILC1s to mature•Tumors with adequate MHC-I and antigen are killed by TEX after checkpoint therapy•Tumors with low/absent MHC-I or poor antigens are killed by PD1+ TRAIL+ NK/ILC1s Interferon-gamma (IFNG) augments immune function yet promotes T cell exhaustion through PDL1. How these opposing effects are integrated to impact immune checkpoint blockade (ICB) is unclear. We show that while inhibiting tumor IFNG signaling decreases interferon-stimulated genes (ISGs) in cancer cells, it increases ISGs in immune cells by enhancing IFNG produced by exhausted T cells (TEX). In tumors with favorable antigenicity, these TEX mediate rejection. In tumors with neoantigen or MHC-I loss, TEX instead utilize IFNG to drive maturation of innate immune cells, including a PD1+TRAIL+ ILC1 population. By disabling an inhibitory circuit impacting PD1 and TRAIL, blocking tumor IFNG signaling promotes innate immune killing. Thus, interferon signaling in cancer cells and immune cells oppose each other to establish a regulatory relationship that limits both adaptive and innate immune killing. In melanoma and lung cancer patients, perturbation of this relationship is associated with ICB response independent of tumor mutational burden. Interferon-gamma (IFNG) augments immune function yet promotes T cell exhaustion through PDL1. How these opposing effects are integrated to impact immune checkpoint blockade (ICB) is unclear. We show that while inhibiting tumor IFNG signaling decreases interferon-stimulated genes (ISGs) in cancer cells, it increases ISGs in immune cells by enhancing IFNG produced by exhausted T cells (TEX). In tumors with favorable antigenicity, these TEX mediate rejection. In tumors with neoantigen or MHC-I loss, TEX instead utilize IFNG to drive maturation of innate immune cells, including a PD1+TRAIL+ ILC1 population. By disabling an inhibitory circuit impacting PD1 and TRAIL, blocking tumor IFNG signaling promotes innate immune killing. Thus, interferon signaling in cancer cells and immune cells oppose each other to establish a regulatory relationship that limits both adaptive and innate immune killing. In melanoma and lung cancer patients, perturbation of this relationship is associated with ICB response independent of tumor mutational burden. Immune checkpoint blockade (ICB) of the inhibitory receptors CTLA4 and PD1 can result in durable responses in multiple cancer types (Ribas and Wolchok, 2018Ribas A. Wolchok J.D. Cancer immunotherapy using checkpoint blockade.Science. 2018; 359: 1350-1355Crossref PubMed Scopus (2961) Google Scholar). Resistance and relapse are common and can be influenced by factors inherent to immune cells, cancer cells, or both (Patel and Minn, 2018Patel S.A. Minn A.J. Combination Cancer Therapy with Immune Checkpoint Blockade: Mechanisms and Strategies.Immunity. 2018; 48: 417-433Abstract Full Text Full Text PDF PubMed Scopus (319) Google Scholar). Important immune features include the status of T cell infiltration and the differentiation or activation state of T cells and innate immune cells. Features intrinsic to cancer cells that can impact ICB outcome include their repertoire of neoantigens, the ability to present antigens on major histocompatibility complex class one (MHC-I), and the expression of inhibitory receptor ligands. The clinical relevance of these immune and cancer cell factors is highlighted by common biomarkers for ICB response such as type I or II interferon (IFN) stimulated genes (ISGs) (Ayers et al., 2017Ayers M. Lunceford J. Nebozhyn M. Murphy E. Loboda A. Kaufman D.R. Albright A. Cheng J.D. Kang S.P. Shankaran V. et al.IFN-γ-related mRNA profile predicts clinical response to PD-1 blockade.J. Clin. Invest. 2017; 127: 2930-2940Crossref PubMed Scopus (1756) Google Scholar, Harlin et al., 2009Harlin H. Meng Y. Peterson A.C. Zha Y. Tretiakova M. Slingluff C. McKee M. Gajewski T.F. Chemokine expression in melanoma metastases associated with CD8+ T-cell recruitment.Cancer Res. 2009; 69: 3077-3085Crossref PubMed Scopus (748) Google Scholar), tumor mutational burden (TMB) (Rizvi et al., 2015Rizvi N.A. Hellmann M.D. Snyder A. Kvistborg P. Makarov V. Havel J.J. Lee W. Yuan J. Wong P. Ho T.S. et al.Cancer immunology. Mutational landscape determines sensitivity to PD-1 blockade in non-small cell lung cancer.Science. 2015; 348: 124-128Crossref PubMed Scopus (5531) Google Scholar, Snyder et al., 2014Snyder A. Makarov V. Merghoub T. Yuan J. Zaretsky J.M. Desrichard A. Walsh L.A. Postow M.A. Wong P. Ho T.S. et al.Genetic basis for clinical response to CTLA-4 blockade in melanoma.N. Engl. J. Med. 2014; 371: 2189-2199Crossref PubMed Scopus (3028) Google Scholar), and expression of PDL1 (Taube et al., 2012Taube J.M. Anders R.A. Young G.D. Xu H. Sharma R. McMiller T.L. Chen S. Klein A.P. Pardoll D.M. Topalian S.L. et al.Colocalization of inflammatory response with B7-h1 expression in human melanocytic lesions supports an adaptive resistance mechanism of immune escape.Sci. Transl. Med. 2012; 4 (127ra37–127ra37)Crossref PubMed Scopus (1652) Google Scholar, Tumeh et al., 2014Tumeh P.C. Harview C.L. Yearley J.H. Shintaku I.P. Taylor E.J.M. Robert L. Chmielowski B. Spasic M. Henry G. Ciobanu V. et al.PD-1 blockade induces responses by inhibiting adaptive immune resistance.Nature. 2014; 515: 568-571Crossref PubMed Scopus (4365) Google Scholar). Both IFN-gamma (IFNG) and type I IFN (IFN-I) are among the known pathways that have critical roles in anti-tumor immunity. IFN enhances immune function by inducing expression of MHC-I (Dighe et al., 1994Dighe A.S. Richards E. Old L.J. Schreiber R.D. Enhanced in vivo growth and resistance to rejection of tumor cells expressing dominant negative IFN γ receptors.Immunity. 1994; 1: 447-456Abstract Full Text PDF PubMed Scopus (490) Google Scholar), which is constitutively expressed on many tissues including cancer cells, and by enabling dendritic cells (DCs) to cross prime T cells (Diamond et al., 2011Diamond M.S. Kinder M. Matsushita H. Mashayekhi M. Dunn G.P. Archambault J.M. Lee H. Arthur C.D. White J.M. Kalinke U. et al.Type I interferon is selectively required by dendritic cells for immune rejection of tumors.J. Exp. Med. 2011; 208: 1989-2003Crossref PubMed Scopus (715) Google Scholar, Fuertes et al., 2011Fuertes M.B. Kacha A.K. Kline J. Woo S.R. Host type I IFN signals are required for antitumor CD8+ T cell responses through CD8α+ dendritic cells.J Exp Med. 2011; 208: 2005-2016Crossref PubMed Scopus (786) Google Scholar). In this way, IFNs are important in the early phase of antigen recognition and the interaction between adaptive and innate immune cells. Accordingly, loss-of-function mutations and genomic alterations in the IFN signaling pathway have been associated with clinical ICB resistance and/or relapse (Gao et al., 2016Gao J. Shi L.Z. Zhao H. Chen J. Xiong L. He Q. Chen T. Roszik J. Bernatchez C. Woodman S.E. et al.Loss of IFN-γ Pathway Genes in Tumor Cells as a Mechanism of Resistance to Anti-CTLA-4 Therapy.Cell. 2016; 167: 397-404.e9Abstract Full Text Full Text PDF PubMed Scopus (732) Google Scholar, Shin et al., 2017Shin D.S. Zaretsky J.M. Escuin-Ordinas H. Garcia-Diaz A. Hu-Lieskovan S. Kalbasi A. Grasso C.S. Hugo W. Sandoval S. Torrejon D.Y. et al.Primary Resistance to PD-1 Blockade Mediated by JAK1/2 Mutations.Cancer Discov. 2017; 7: 188-201Crossref PubMed Scopus (766) Google Scholar, Zaretsky et al., 2016Zaretsky J.M. Garcia-Diaz A. Shin D.S. Escuin-Ordinas H. Hugo W. Hu-Lieskovan S. Torrejon D.Y. Abril-Rodriguez G. Sandoval S. Barthly L. et al.Mutations Associated with Acquired Resistance to PD-1 Blockade in Melanoma.N. Engl. J. Med. 2016; 375: 819-829Crossref PubMed Scopus (1918) Google Scholar), and unbiased genetic screens have identified this same pathway as being important for immunotherapy response in certain mouse models (Manguso et al., 2017Manguso R.T. Pope H.W. Zimmer M.D. Brown F.D. Yates K.B. Miller B.C. Collins N.B. Bi K. LaFleur M.W. Juneja V.R. et al.In vivo CRISPR screening identifies Ptpn2 as a cancer immunotherapy target.Nature. 2017; 547: 413-418Crossref PubMed Scopus (568) Google Scholar, Mezzadra et al., 2017Mezzadra R. Sun C. Jae L.T. Gomez-Eerland R. de Vries E. Wu W. Logtenberg M.E.W. Slagter M. Rozeman E.A. Hofland I. et al.Identification of CMTM6 and CMTM4 as PD-L1 protein regulators.Nature. 2017; 549: 106-110Crossref PubMed Scopus (372) Google Scholar). In contrast, some patients have tumors with mutations in the IFN pathway that nonetheless respond to ICB (Hellmann et al., 2018Hellmann M.D. Nathanson T. Rizvi H. Creelan B.C. Sanchez-Vega F. Ahuja A. Ni A. Novik J.B. Mangarin L.M.B. Abu-Akeel M. et al.Genomic Features of Response to Combination Immunotherapy in Patients with Advanced Non-Small-Cell Lung Cancer.Cancer Cell. 2018; 33: 843-852.e4Abstract Full Text Full Text PDF PubMed Scopus (609) Google Scholar, Sade-Feldman et al., 2017Sade-Feldman M. Jiao Y.J. Chen J.H. Rooney M.S. Barzily-Rokni M. Eliane J.-P. Bjorgaard S.L. Hammond M.R. Vitzthum H. Blackmon S.M. et al.Resistance to checkpoint blockade therapy through inactivation of antigen presentation.Nat. Commun. 2017; 8: 1136Crossref PubMed Scopus (500) Google Scholar) or have high serum levels of IFNG that associates with ICB progression (Huang et al., 2017Huang A.C. Postow M.A. Orlowski R.J. Mick R. Bengsch B. Manne S. Xu W. Harmon S. Giles J.R. Wenz B. et al.T-cell invigoration to tumour burden ratio associated with anti-PD-1 response.Nature. 2017; 545: 60-65Crossref PubMed Scopus (953) Google Scholar). These apparently “paradoxical” observations may represent feedback inhibition properties of IFN signaling (Snell et al., 2017Snell L.M. McGaha T.L. Brooks D.G. Type I Interferon in Chronic Virus Infection and Cancer.Trends Immunol. 2017; 38: 542-557Abstract Full Text Full Text PDF PubMed Scopus (234) Google Scholar). In the context of chronic pathogen infection, persistent IFN signaling and ISGs dampen immune responses to prevent immune-mediated pathology while allowing for a host-pathogen stalemate (Cheng et al., 2017Cheng L. Ma J. Li J. Li D. Li G. Li F. Zhang Q. Yu H. Yasui F. Ye C. et al.Blocking type I interferon signaling enhances T cell recovery and reduces HIV-1 reservoirs.J. Clin. Invest. 2017; 127: 269-279Crossref PubMed Scopus (124) Google Scholar, Teijaro et al., 2013Teijaro J.R. Ng C. Lee A.M. Sullivan B.M. Sheehan K.C.F. Welch M. Schreiber R.D. de la Torre J.C. Oldstone M.B.A. Persistent LCMV infection is controlled by blockade of type I interferon signaling.Science. 2013; 340: 207-211Crossref PubMed Scopus (538) Google Scholar, Wilson et al., 2013Wilson E.B. Yamada D.H. Elsaesser H. Herskovitz J. Deng J. Cheng G. Aronow B.J. Karp C.L. Brooks D.G. Blockade of chronic type I interferon signaling to control persistent LCMV infection.Science. 2013; 340: 202-207Crossref PubMed Scopus (512) Google Scholar). In cancer, this dichotomous function of IFN is exploited through chronic signaling by tumor cells that can promote resistance to ICB (Benci et al., 2016Benci J.L. Xu B. Qiu Y. Wu T.J. Dada H. Twyman-Saint Victor C. Cucolo L. Lee D.S.M. Pauken K.E. Huang A.C. et al.Tumor Interferon Signaling Regulates a Multigenic Resistance Program to Immune Checkpoint Blockade.Cell. 2016; 167: 1540-1554.e12Abstract Full Text Full Text PDF PubMed Scopus (593) Google Scholar). IFN-driven resistance can be inhibited by genetic ablation of the IFNG receptor (IFNGR) and/or IFN-I receptor (IFNAR) in cancer cells, resulting in a decrease in PDL1, other inhibitory ligands, and the GzmB antagonist SERPINB9 (Jiang et al., 2018Jiang P. Gu S. Pan D. Fu J. Sahu A. Hu X. Li Z. Traugh N. Bu X. Li B. et al.Signatures of T cell dysfunction and exclusion predict cancer immunotherapy response.Nat. Med. 2018; 24: 1550-1558Crossref PubMed Scopus (1465) Google Scholar). Expansion of exhausted T cells (TEX) can then ensue to restore ICB response through unknown mechanisms. Together, these observations highlight the importance of understanding how the opposing functions of IFN signaling impact cancer immunotherapy. Loss of the beta-2 microglobulin (B2M) subunit of MHC-I appears to be a common resistance mechanism to ICB (Sade-Feldman et al., 2017Sade-Feldman M. Jiao Y.J. Chen J.H. Rooney M.S. Barzily-Rokni M. Eliane J.-P. Bjorgaard S.L. Hammond M.R. Vitzthum H. Blackmon S.M. et al.Resistance to checkpoint blockade therapy through inactivation of antigen presentation.Nat. Commun. 2017; 8: 1136Crossref PubMed Scopus (500) Google Scholar). However, diminished expression or loss of B2M can also occur in patients who respond to ICB (Rizvi et al., 2018Rizvi H. Sanchez-Vega F. La K. Chatila W. Jonsson P. Halpenny D. Plodkowski A. Long N. Sauter J.L. Rekhtman N. et al.Molecular Determinants of Response to Anti-Programmed Cell Death (PD)-1 and Anti-Programmed Death-Ligand 1 (PD-L1) Blockade in Patients With Non-Small-Cell Lung Cancer Profiled With Targeted Next-Generation Sequencing.J. Clin. Oncol. 2018; 36: 633-641Crossref PubMed Scopus (823) Google Scholar, Rodig et al., 2018Rodig S.J. Gusenleitner D. Jackson D.G. Gjini E. Giobbie-Hurder A. Jin C. Chang H. Lovitch S.B. Horak C. Weber J.S. et al.MHC proteins confer differential sensitivity to CTLA-4 and PD-1 blockade in untreated metastatic melanoma.Sci Transl Med. 2018; 10 (eaar3342)Crossref PubMed Scopus (283) Google Scholar), suggesting that innate immune cells might contribute to ICB response in some cases. Indeed, conventional NK cells and innate lymphoid cells (ILCs) are capable of destroying cancers through either perforin-mediated cytotoxicity or TNF-family death receptors such as TRAIL (Spits et al., 2016Spits H. Bernink J.H. Lanier L. NK cells and type 1 innate lymphoid cells: partners in host defense.Nat. Immunol. 2016; 17: 758-764Crossref PubMed Scopus (288) Google Scholar). NK/ILC effector function is regulated through cellular maturation, combinations of activating and inhibitory receptors, and possibly immune checkpoint receptors like PD1, TIM3, and TIGIT (Gao et al., 2017Gao Y. Souza-Fonseca-Guimaraes F. Bald T. Ng S.S. Young A. Ngiow S.F. Rautela J. Straube J. Waddell N. Blake S.J. et al.Tumor immunoevasion by the conversion of effector NK cells into type 1 innate lymphoid cells.Nat. Immunol. 2017; 18: 1004-1015Crossref PubMed Scopus (375) Google Scholar, Zhang et al., 2018Zhang Q. Bi J. Zheng X. Chen Y. Wang H. Wu W. Wang Z. Wu Q. Peng H. Wei H. et al.Blockade of the checkpoint receptor TIGIT prevents NK cell exhaustion and elicits potent anti-tumor immunity.Nat. Immunol. 2018; 19: 723-732Crossref PubMed Scopus (536) Google Scholar). Recent evidence indicates that type one ILCs (ILC1s) can participate in anti-tumor immunity or cancer immune surveillance. This includes ILC1-like populations (Dadi et al., 2016Dadi S. Chhangawala S. Whitlock B.M. Franklin R.A. Luo C.T. Oh S.A. Toure A. Pritykin Y. Huse M. Leslie C.S. Li M.O. Cancer Immunosurveillance by Tissue-Resident Innate Lymphoid Cells and Innate-like T Cells.Cell. 2016; 164: 365-377Abstract Full Text Full Text PDF PubMed Scopus (214) Google Scholar) and intratumoral ILC1s that are generally poorly cytotoxic (Cortez et al., 2017Cortez V.S. Ulland T.K. Cervantes-Barragan L. Bando J.K. Robinette M.L. Wang Q. White A.J. Gilfillan S. Cella M. Colonna M. SMAD4 impedes the conversion of NK cells into ILC1-like cells by curtailing non-canonical TGF-β signaling.Nat. Immunol. 2017; 18: 995-1003Crossref PubMed Scopus (200) Google Scholar, Gao et al., 2017Gao Y. Souza-Fonseca-Guimaraes F. Bald T. Ng S.S. Young A. Ngiow S.F. Rautela J. Straube J. Waddell N. Blake S.J. et al.Tumor immunoevasion by the conversion of effector NK cells into type 1 innate lymphoid cells.Nat. Immunol. 2017; 18: 1004-1015Crossref PubMed Scopus (375) Google Scholar). Although the ability of NK/ILC1s to eradicate tumors with diminished MHC-I and/or a poor neoantigens is of significant interest, how to mobilize these innate immune cells to facilitate tumor response is unclear. A large proportion of human cancers differentially express a subset of ISGs that can predict resistance to radiation and chemotherapy (Weichselbaum et al., 2008Weichselbaum R.R. Ishwaran H. Yoon T. Nuyten D.S.A. Baker S.W. Khodarev N. Su A.W. Shaikh A.Y. Roach P. Kreike B. et al.An interferon-related gene signature for DNA damage resistance is a predictive marker for chemotherapy and radiation for breast cancer.Proc. Natl. Acad. Sci. USA. 2008; 105: 18490-18495Crossref PubMed Scopus (390) Google Scholar). Coincidentally, this ISG resistance signature (ISG.RS) is also associated with resistance to ICB, as demonstrated by elevated expression in murine tumors from Res 499 melanoma cells (Figure 1A), which were derived from an ICB-resistant B16-F10 tumor (Twyman-Saint Victor et al., 2015Twyman-Saint Victor C. Rech A.J. Maity A. Rengan R. Pauken K.E. Stelekati E. Benci J.L. Xu B. Dada H. Odorizzi P.M. et al.Radiation and dual checkpoint blockade activate non-redundant immune mechanisms in cancer.Nature. 2015; 520: 373-377Crossref PubMed Scopus (1619) Google Scholar). In contrast, ISGs can also predict clinical ICB response, especially ISGs typically associated with IFNG signaling (Ayers et al., 2017Ayers M. Lunceford J. Nebozhyn M. Murphy E. Loboda A. Kaufman D.R. Albright A. Cheng J.D. Kang S.P. Shankaran V. et al.IFN-γ-related mRNA profile predicts clinical response to PD-1 blockade.J. Clin. Invest. 2017; 127: 2930-2940Crossref PubMed Scopus (1756) Google Scholar). To begin reconciling these seemingly disparate observations, we examined the ISG.RS and genes from the IFNG hallmark gene set (IFNG.GS) by dividing them into two non-overlapping subsets (Figure 1B and Table S1) and creating a metagene (the average scaled expression of all genes in the set). The expression of these ISG metagenes was then examined across different cellular populations in human melanomas using previously published single-cell RNA-seq data (Tirosh et al., 2016Tirosh I. Izar B. Prakadan S.M. Wadsworth 2nd, M.H. Treacy D. Trombetta J.J. Rotem A. Rodman C. Lian C. Murphy G. et al.Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq.Science. 2016; 352: 189-196Crossref PubMed Scopus (2033) Google Scholar). This revealed that the IFNG.GS is predominantly expressed by intratumoral immune cells such as T cells, NK cells, and macrophages (Figures 1B and S1A). In contrast, the ISG.RS is predominantly expressed in cancer cells, albeit with variable expression.Figure S1ISGs and Immune Cell Populations Expressed in Human Melanoma, Related to Figure 1Show full caption(A) Boxplots of ISG metagene expression in immune cell populations determined by single-cell RNA-seq of human melanoma. The width of the boxplots for the indicated cell types are proportional to the population size (of note, immune and cancer cells were sorted before sequencing). P values for comparisons between cancer cells and each immune population are p < 0.001. (B) Gene set enrichment analysis of cancer and resistance-associated ISGs (ISG.RS) and IFNG-related ISGs (IFNG.GS) in melanoma patients treated with anti-PD1. Shown are enrichment plots along with the normalized enrichment score (NES) and p value. The leading edge for the ISG.RS is labeled. (C) Relative frequencies of immune populations in the melanoma tumors inferred by CIBERSORT. For immune cell types with both resting and activated populations, the difference between activated and resting was used. (D) Multivariable random forest model for probability of response for melanoma patients treated with anti-PD1. Shown are the adjusted effects of model variables on the probability of response (left plots, yellow boundaries indicate one standard error) and average variable importance scores with standard deviations (right plot). Predictor values are metagene expression values for ISG.RS and IFNG.GS or log10 frequency for TMB. Variable importance score represents the increase in classification error rate when the variable is perturbed. The classification error rate for the model is 36%. (E) Random forest model with variable selection based on minimal depth was performed on bootstrapped samples. Variables include inferred frequencies of various immune populations (based on CIBERSORT), the ratio of IFNG.GS to ISG.RS (dISG), TMB, and other control variables. Shown are the relative frequencies that each variable was selected based on minimal depth after resampling versus the average variable importance score (VIMP) (with standard deviations). The inset shows the distribution of the number of variables in each bootstrapped model. Similar results were also obtained with lasso and logistic regression.View Large Image Figure ViewerDownload Hi-res image Download (PPT) (A) Boxplots of ISG metagene expression in immune cell populations determined by single-cell RNA-seq of human melanoma. The width of the boxplots for the indicated cell types are proportional to the population size (of note, immune and cancer cells were sorted before sequencing). P values for comparisons between cancer cells and each immune population are p < 0.001. (B) Gene set enrichment analysis of cancer and resistance-associated ISGs (ISG.RS) and IFNG-related ISGs (IFNG.GS) in melanoma patients treated with anti-PD1. Shown are enrichment plots along with the normalized enrichment score (NES) and p value. The leading edge for the ISG.RS is labeled. (C) Relative frequencies of immune populations in the melanoma tumors inferred by CIBERSORT. For immune cell types with both resting and activated populations, the difference between activated and resting was used. (D) Multivariable random forest model for probability of response for melanoma patients treated with anti-PD1. Shown are the adjusted effects of model variables on the probability of response (left plots, yellow boundaries indicate one standard error) and average variable importance scores with standard deviations (right plot). Predictor values are metagene expression values for ISG.RS and IFNG.GS or log10 frequency for TMB. Variable importance score represents the increase in classification error rate when the variable is perturbed. The classification error rate for the model is 36%. (E) Random forest model with variable selection based on minimal depth was performed on bootstrapped samples. Variables include inferred frequencies of various immune populations (based on CIBERSORT), the ratio of IFNG.GS to ISG.RS (dISG), TMB, and other control variables. Shown are the relative frequencies that each variable was selected based on minimal depth after resampling versus the average variable importance score (VIMP) (with standard deviations). The inset shows the distribution of the number of variables in each bootstrapped model. Similar results were also obtained with lasso and logistic regression. To understand the potential consequences of these differences in IFNG.GS and ISG.RS expression patterns, we analyzed bulk RNA-seq data combined from two cohorts of melanoma patients treated with anti-PD1 (Figure 1C) (Hugo et al., 2016Hugo W. Zaretsky J.M. Sun L. Song C. Moreno B.H. Hu-Lieskovan S. Berent-Maoz B. Pang J. Chmielowski B. Cherry G. et al.Genomic and Transcriptomic Features of Response to Anti-PD-1 Therapy in Metastatic Melanoma.Cell. 2016; 165: 35-44Abstract Full Text Full Text PDF PubMed Scopus (1739) Google Scholar, Riaz et al., 2017Riaz N. Havel J.J. Makarov V. Desrichard A. Urba W.J. Sims J.S. Hodi F.S. Martín-Algarra S. Mandal R. Sharfman W.H. et al.Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab.Cell. 2017; 171: 934-949.e16Abstract Full Text Full Text PDF PubMed Scopus (972) Google Scholar). As expected, the majority of genes in the IFNG.GS are depressed in the majority of tumors from non-responders to anti-PD1 (Figure S1B). However, like ICB-resistant murine Res 499 tumors, most ISG.RS genes are enriched in tumors from non-responders (Figure S1B). Consistent with the importance of CD8 T cells in response, tumors with high IFNG.GS but low ISG.RS also have the greatest proportion of CD8 T cells (Figure 1D, top right quadrant) as inferred by CIBERSORT (Newman et al., 2015Newman A.M. Liu C.L. Green M.R. Gentles A.J. Feng W. Xu Y. Hoang C.D. Diehn M. Alizadeh A.A. Robust enumeration of cell subsets from tissue expression profiles.Nat. Methods. 2015; 12: 453-457Crossref PubMed Scopus (5003) Google Scholar) (Figure S1C). The higher frequencies of CD8 T cells are accompanied by increased number of activated NK cells (Figure 1D, orange regression line), which also has been associated with clinical ICB response (Riaz et al., 2017Riaz N. Havel J.J. Makarov V. Desrichard A. Urba W.J. Sims J.S. Hodi F.S. Martín-Algarra S. Mandal R. Sharfman W.H. et al.Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab.Cell. 2017; 171: 934-949.e16Abstract Full Text Full Text PDF PubMed Scopus (972) Google Scholar). To understand how these immune and interferon-related variables independently contribute to ICB response, we utilized a multivariable logistic regression model. This revealed that while higher IFNG.GS increases the odds ratio for response, ISG.RS independently decreases the likelihood (Figure 1E). The significance of both of these variables are independent of tumor mutational burden (TMB) status, which expectedly correlates with response. In contrast, neither the abundance of CD8 T cells nor NK cells are significant in the model. A random forest model, which does not assume linearity and incorporates interaction effects, revealed that ISG.RS exhibits a higher importance score than either IFNG.GS or TMB (Figure S1D). In total, these data suggest that while expression of IFNG.GS by immune cells is associated with CD8 T cell abundance, accumulation of activated NK cells, and ICB response, all of these effects are opposed by high levels of ISG.RS in cancer cells. Although the IFNG.GS and ISG.RS predict opposite clinical outcomes, their expression is positively correlated, consistent with IFN controlling both metagenes (Figure 1F). An explanation for this apparent “paradox” lies in the relative expression of each metagene. When expression of the ISG.RS exceeds the IFNG.GS, resistance is favored (Figure 1F, left plot, red circles below diagonal). In contrast, most responses occur when IFNG.GS is similar to or greater than ISG.RS (Figure 1F, blue circles). Based on these findings, we combined the two metagenes into a ratio of IFNG.GS over ISG.RS (or, the difference of these two metagenes in log transformed space). By logistic regression, this composite variable (dISG) is strongly associated with response and is independent of TMB (Figure 1F, right plot and inset). Specifically, the probability of response is low when either the ratio or TMB is low but increases when either increase. Furthermore, random forest machine learning and bootstrapping revealed that the ISG ratio has the highest robustness and average variable importance compared to TMB and multiple immune features (Figure S1E). In total, the single-cell and bulk RNA-seq analysis suggests that distinct ISGs differentially expressed by cancer and immune cells can oppose each other to influence CD8 T cell infiltrate and NK activation and can be combined into a ratio that predicts ICB response independent of TMB (Figure 1G). Motivated by these findings, we sought to understand the mechanistic underpinnings inferred by these statistical relationships. If the probability of ICB response is influenced by the ratio of IFNG-related ISGs expressed by immune cells over inhibitory ISGs expressed by cancer cells, one way to enhance the ratio in favor of response is to prevent IFN signaling in cancer cells. We first confirmed whether the ISG.RS, which is elevated in ICB-resistant Res 499 tumors, is regulated by IFN signaling in cancer cells (hereafter referred to as tumor IFN signaling). Indeed, CRISPR knockout of IFNGR and/or IFNAR significantly diminishes ISG.RS levels (Figure 1H). However, loss of tumor IFN signaling can render cancers less responsive to immunotherapy due to compromised MHC-I and antigen processing (Manguso et al., 2017Manguso R.T. Pope H.W. Zimmer M.D. Brown F.D. Yates K.B. Miller B.C. Collins N.B. Bi K. LaFleur M.W. Juneja V.R. et al.In vivo CRISPR screening identifies Ptpn2 as a cancer immunotherapy target.Nature. 2017; 547: 413-418Crossref PubMed Scopus (568) Google Scholar, Zaretsky et al., 2016Zaretsky J.M. Garcia-Diaz A. Shin D.S. Escuin-Ordinas H. Hugo W. Hu-Lieskovan S. Torrejon D.Y. Abril-Rodriguez G. Sandoval S. Barthly L. et al.Mutations Associated with Acquired Resistance to PD-1 Blockade in Melanoma.N. Engl. J. Med. 2016; 375: 819-829Crossref PubMed Scopus (1918) Google Scholar), suggesting that the impact from ablating tumor IFN signaling might be context dependent. In light of this, we surmised two situations whereby the benefit of inhibiting IFN-driven resistance could outweigh the potential negative impact on MHC-I. The first is when constitutive MHC-I is high, minimizing effects that loss of IFN-inducible MHC-I has on CTL-mediated killing. A second situation is when tumors have depleted or poor neoantigens. Here, diminished CTL recognition presumably makes MHC-I status less consequential for T cell-mediating killing, but interference with IFN-driven resistance might improve killing by NK or other innate lymphoid cells. We first characterized various mouse tumor models for differences in MHC-I expression, TMB, and predicted neoantigen status (Figure 2A). Of these, CT26 colorectal cancer has the highest TMB (Figure 2B) and maintains high MHC-I in the absence of IFNG signaling (Figures 2C and 2D)