Loss of function JAK1 mutations occur at high frequency in cancers with microsatellite instability and are suggestive of immune evasion

Immune evasion is a well-recognized hallmark of cancer and recent studies with immunotherapy agents have suggested that tumors with increased numbers of neoantigens elicit greater immune responses. We hypothesized that the immune system presents a common selective pressure on high mutation burden tumors and therefore immune evasion mutations would be enriched in high mutation burden tumors. The JAK family of kinases is required for the signaling of a host of immune modulators in tumor, stromal, and immune cells. Therefore, we analyzed alterations in this family for the hypothesized signature of an immune evasion mutation. Here, we searched a database of 61,704 unique solid tumors for alterations in the JAK family kinases (JAK1/2/3, TYK2). We used The Cancer Genome Atlas and Cancer Cell Line Encyclopedia data to confirm and extend our findings by analyzing gene expression patterns. Recurrent frameshift mutations in JAK1 were associated with high mutation burden and microsatellite instability. These mutations occurred in multiple tumor types including endometrial, colorectal, stomach, and prostate carcinomas. Analyzing gene expression signatures in endometrial and stomach adenocarcinomas revealed that tumors with a JAK1 frameshift exhibited reduced expression of interferon response signatures and multiple anti-tumor immune signatures. Importantly, endometrial cancer cell lines exhibited similar gene expression changes that were expected to be tumor cell intrinsic (e.g. interferon response) but not those expected to be tumor cell extrinsic (e.g. NK cells). From these data, we derive two primary conclusions: 1) JAK1 frameshifts are loss of function alterations that represent a potential pan-cancer adaptation to immune responses against tumors with microsatellite instability; 2) The mechanism by which JAK1 loss of function contributes to tumor immune evasion is likely associated with loss of the JAK1-mediated interferon response.

Introduction Immune evasion is one of the hallmarks of cancer because tumors subvert and even use immune responses to aid growth and invasion [1]. The immune system serves an important role in cancer surveillance and exerts considerable selective pressure on cancers that do develop. Immunosuppressed transplant patients have an increased rate of developing nonviral cancers [2] and the presence of tumor-infiltrating lymphocytes is a positive prognostic factor in several cancer types [3][4][5]. The presence of infiltrating lymphocytes has been further refined with two important observations. First, higher numbers of effector T cells tends to indicate a positive prognosis, suggesting that anti-tumor responses are antigen specific, while regulatory T cells indicate a negative prognosis [6][7][8]. Studies in mice have shown that T and B cell deficient animals have a higher incidence of tumors in a chemically induced model; furthermore, these tumors rarely engraft into wild-type hosts although wild-type to wild-type tumor transplants are successful, which further supports the antigen specificity of anti-tumor immune responses [9]. Second, tumors with a high-mutation phenotype resulting from mutation of POLD1/E or microsatellite instability (MSI) have both high numbers of neo-antigens and high numbers of tumor infiltrating lymphocytes, suggesting that tumors with more antigens elicit stronger immune responses [10,11]. Thus, immune evasion and immune editing are important parts of tumor growth and progression, especially for hyper-mutant tumors. One mechanism of immune evasion, checkpoint induction, has been targeted therapeutically with long term responses in a subset of patients [12]. Immune checkpoint inhibition has been successful in cancers with high levels of microsatellite instability (MSI-H) and is approved in diseases such as melanoma, non-small cell lung carcinoma, and metastatic bladder cancer that frequently have high tumor mutational burden [13][14][15]. Other cell intrinsic mechanisms of immune evasion include loss of proteins involved in antigen presentation, resistance to cell death, and receptor shedding of immune activating receptors [1,16,17]. Cell extrinsic mechanisms of immune evasion include secretion of immunosuppressive cytokines or recruitment of immunosuppressive cells such as regulatory T cells or myeloid derived suppressor cells [8,18,19]. In addition to subverting the immune response during tumor development, these mechanisms will also likely impact the success of checkpoint inhibitor therapy.
The JAK-STAT family of proteins mediates signaling of the interferon (IFN), IL-6, and IL-2 families of cytokines [20]. JAK-STAT proteins are expressed by immune, stromal, and tumor cells, making them an important transducer of immune signals in nearly all cell types. JAK kinases are constitutively associated with cytokine receptors; cytokine binding results in activation of JAK kinases and phosphorylation of the receptor [21]. Phosphorylated receptors recruit STAT proteins, which are then themselves phosphorylated by the activated JAK kinases. Phosphorylated STAT proteins form homo-and hetero-dimers that translocate into the nucleus to function as transcription factors that regulate gene expression. Because the four JAK kinases (JAK1, JAK2, JAK3, and TYK2) mediate signaling for over 20 cytokines, genomic changes that alter their activity can have diverse effects [21]. For example, Jak1-deficient mice exhibit perinatal lethality and phenotypes as diverse as defective lymphopoeisis, lack of IFN response, and failure of cardiomyocytes and neuronal cells to respond to growth factors [22]. An immunodeficient patient with early-onset bladder cancer was found to have two different germline biallelic JAK1 hypomorph alterations [23]. In hematopoietic cancers, the activating mutation JAK2 V617F drives myeoloproliferative neoplasms and the effects of the JAK-STAT family are considered to be pro-tumorigenic [24]. In solid tumors, activation of STAT3 or STAT5 by mutation or cytokine signaling is pro-tumorigenic, while activation of STAT1 upregulates antigen presentation and has anti-tumorigenic effects [25]. However, the role of the JAK-STAT pathway in solid tumors has been ambiguous due to the low incidence of recurrent mutations in this pathway.
We hypothesized that immune surveillance presents a common selective pressure on high mutational burden tumors and, therefore, tumor-intrinsic immune evasion mechanisms would be enriched in tumors with high mutational burden. We investigated mutations within JAK family kinases suggestive of immune modulation in a collection of 61,704 unique solid tumors. Recurrent JAK1 frameshift mutations occurred at the highest frequency in endometrial cancers, but were also found in prostate, urinary, gastric, and colon cancers. JAK1 frameshifts occurred in MSI-H cancers with high mutational burdens. These mutations were observed in over a third of MSI-H cancers, which is in contrast to the low mutation rate of JAK-STAT family members in all solid tumors. JAK1 is a critical signaling kinase for the interferon, IL-6, and IL-2 pathways. Using The Cancer Genome Atlas (TCGA) data of MSI-H endometrial tumors, we found reduced expression of gene sets associated with interferon gamma, interferon alpha, IL-6, and IL-2 responses in JAK1 frameshift mutants compared to JAK1 wild-type tumors. Stomach adenocarcinomas with a JAK1 frameshift exhibited reductions in both interferon responses but reductions in IL-6 and IL-2 responses were not significant. Both endometrial and stomach tumors with JAK1 frameshifts also exhibited reductions in gene sets associated with immune surveillance such as the inflammatory response and antigen presentation. Thus, JAK1 frameshift mutations are a potential pan-cancer adaptation to the immune response that occurs against highly mutated MSI-H tumors. These alterations likely allow tumors to evade the immune response via a lack of antigen presentation and resistance to interferons.

Dataset collection and verification
The Foundation Medicine (FMI) dataset was a set of clinical samples sent to a CLIA (Clinical Laboratory Improvement Amendments)-certified, College of American Pathology (CAP)accredited, and New York State (NYS)-accredited laboratory (Foundation Medicine, Cambridge, MA) for next-generation sequencing [26]. Samples were subjected to an independent pathology review to verify the correct diagnosis. DNA was extracted and hybrid capture for all coding exons of 287 or 395 genes plus 19 or 31 introns frequently rearranged in cancer was performed; both versions of the test captured JAK1/2/3. Libraries were sequenced to a median depth >500X. Analysis for genomic alterations was performed as previously described [26]. Natural germline variants from the 1000 Genomes Project (dbSNP135) were removed. To maximize mutation-detection accuracy (sensitivity and specificity) in impure clinical specimens (!20% tumor nuclei), the test was previously optimized and validated to detect base substitutions at a !5% mutant allele frequency (MAF) with 99% sensitivity, indels (1-40bps) at !10% MAF with 98% sensitivity, focal homozygous deletions and amplifications (!8 copies) with > 95% sensitivity, and select gene fusions with >99% sensitivity [26,27]. Germline variants without clinical significance were further filtered by applying an internal algorithm to determine somatic/germline status [28].
Frameshift identification in The Cancer Genome Atlas data BAM files of Exomeseq for TCGA (http://cancergenome.nih.gov) cohort of uterine corpus endometrial cancer (UCEC), stomach adenocarcinoma (STAD) and colon adenocarcinoma (COAD) were downloaded from Cancer Genomics Hub (https://cghub.ucsc.edu/). GTFuse from Annai Systems is used to get sliced bam files and retrieve only reads of the interested regions such as JAK1 gene region in this case. JAK1 mutations for samples with matched normal were called using VarDict [29].

TCGA data analysis
Except for JAK1 frameshifts, we analyzed publically available TCGA data for Clinical (MSI and survival), RNASeqV2, and Somatic Mutations. RNA expression was measured using the "scaled estimate" column. Data sets were downloaded on June 1, 2016.
Cancer cell line encyclopedia (CCLE) RNAseq data CCLE RNAseq data is downloaded from Omicsoft Corporation (http://www.omicsoft.com/). The FPKM gene expression value provided by Omicsoft is converted to transcripts per million (TPM) using the following formula [30]: Gene expression and gene set enrichment analysis (GSEA) Tumors samples from TCGA or cell lines from the CCLE [31] were divided into JAK1 frameshift and wild-type groups. Gene expression was compared using Student's t as the distance metric (difference in mean between the two groups divided by standard deviation). Genes were then rank ordered and analyzed using the GseaPreRank tool in GSEA 2.2.2 [32,33]. We analyzed gene sets in Hallmarks version 5.1 using the weighted enrichment statistic and a custom set of immune genes using the weighted p2 enrichment statistic since the immune gene sets were small. Immune gene sets are provided (S1 Appendix). Note the MHCI gene set is referred to as antigen presentation in the text. The IFN Gamma response in Figs 3 and 5 is the sum of z-scores for the top 25 genes ranked by the Student's t for that disease.

Statistics and visualization
Statistics were calculated using RStudio 0.99 [34], R 3.2.1 [35]; Scipy version 0.15.1, Anaconda version 2.3.0 [36], in Python 2.7.10; or GraphPad Prism. Data were visualized using the cBioportal lollipop plot generator [37,38], R standard plotting functions and package NMF [39] to display heatmaps, or Matplotlib version 1.4.3 [40] in Python. Not all samples in TCGA have all of the required information (i.e. MSI, RNA-Seq); therefore, TCGA cohorts were adjusted to include the maximum number of samples for a figure panel based on the data required to perform the analysis and the data available for each sample.

Ethics approval and consent to participate
This study, including the consent procedure, was reviewed and approved by the Western Institutional Review Board (WIRB). For samples within the Foundation Medicine dataset, written patient consent was obtained at the time of testing. Patients were not consented for release of raw sequencing data.

Availability of data and material
Patients were not consented for release of sequence information by Foundation Medicine. A subset of curated Foundation Medicine mutation data has been released through the National Cancer Institute's Genomics Data Commons: https://portal.gdc.cancer.gov Program: FM The results published here are in part based upon data generated by TCGA Research Network: http://cancergenome.nih.gov/ [41].
Premature terminating mutations (frameshift and nonsense) in JAK1 were enriched in endometrial cancers (8.2%), prostate cancers (2.7%), urinary cancers (1.2%), and several gastrointestinal tract cancers (gastric, small intestine, and colorectal; <1%, Fig 1D). In contrast, early terminating mutations in JAK2 and JAK3 were not disease specific and occurred in fewer than 1% of all disease groups. In total we observed 135 frameshift mutations at the K860/P861 hotspots, with 50.4% (n = 68) occurring in endometrial cancer ( Fig 1E). We also observed 90 frameshift mutations at P430/L431, with 45.6% (n = 41) in endometrial cancer. Prostate cancer showed a slightly increased incidence of P430/L431 frameshifts compared to K860/P861 (12 vs. 10, Fig 1E). The P429/P430 proline pair is conserved across JAK family members; we looked for frameshifts at this position and found none in solid tumors ( Fig 1A-1C red boxes). A smaller analysis of 3,666 tumors for alterations in TYK2 identified only two frameshifts in this gene (data not shown). Thus, frameshifts in the JAK family appeared to be JAK1 specific and were recurrent across several solid tumor types.
Loss of function mutations such as frameshifts generally occur in tumor suppressor genes, in which a second event is required to inactivate the remaining wild-type allele. Therefore, we analyzed the mutant allele frequency of JAK1 early terminating mutations and JAK1 ploidy to determine the cancer cell fraction in tumor samples. This analysis was limited to samples with greater than 40% tumor purity to increase accuracy. Of samples with a mutation resulting in premature termination of JAK1, 38.1% (51/134) had a homozygous mutation in all copies of JAK1, thus inactivating the protein in these samples ( Fig 1F). Another 19.4% (26/134) of samples had two truncating mutations in JAK1 and neither was present in all copies. Unfortunately, we cannot assess whether these mutations are on the same or different alleles. At least one copy of JAK1 remained functional in 31.3% (42/134) of tumors that had a clonal JAK1  11.2% (15/134) of JAK1 frameshift samples had a subclonal mutation defined as less than 0.75 copies (Fig 1F). In total, 58% of tumors with JAK1 alterations likely contained inactivation of both JAK1 alleles and an additional 31% had lost one copy. To further assess if the JAK1 frameshift mutations were loss of function events, we evaluated expression of JAK1 mRNA in endometrial tumors from TCGA and endometrial cell lines from the CCLE. JAK1 mRNA expression was significantly downregulated in both endometrial tumor samples and endometrial cell lines with a JAK1 frameshift compared to JAK1 wildtype endometrial samples (Fig 1G and 1H, tumor: P < 10 −18 , cell line: P < 10 −4 , Mann-Whitney U test). We concluded that multiple mutations in JAK1 and reduced expression were indicative of loss of function mutations commonly found in tumor suppressor genes.
Deficiencies in the mismatch repair pathway lead to frequent frameshift mutations particularly at mono and di-nucleotide microsatellite repeats, otherwise known as microsatellite instability (MSI). Since frameshifts in JAK1 occurred at 7-mer and 8-mer mononucleotide repeats, we tested the association between MSI status and the presence of a JAK1 frameshift alteration in a subset of samples for which MSI status was available. Diagnostically, MSI is scored as microsatellite stable (MSS), MSI-low (MSI-L), or MSI-high (MSI-H); however, for our initial analysis we excluded MSI-L samples. JAK1 frameshifts were significantly enriched in MSI-H endometrial cancers (38%; 50/131) versus microsatellite stable (MSS) endometrial cancers (<1%; 3/657), (Fig 2A, P = 1.78 x 10 −39 , Fisher's exact test). MSI-H prostate and urinary cancers also showed high rates of JAK1 frameshifts: 54% (13/24) and 30% (3/10), respectively (Fig  2A). In gastrointestinal cancers, JAK1 frameshifts were observed in 6% (9/158) and 15% (4/27) of MSI-H colorectal and gastric cancers, respectively (Fig 2A). MSI is rare in other tumor types such as lung cancer; however, 17% (29/173) of other tumors that were MSI-H also had a JAK1 frameshift. The co-occurrence of JAK1 frameshifts and MSI-H was statistically significant in all diseases (Fig 2A, P < 10 −4 for all diseases, Fisher's exact test). MSI-H tumors exhibit increased mutation rates; therefore, we also tested the association of JAK1 frameshifts with tumor mutational burden (TMB). In endometrial cancers, JAK1 frameshifts were associated with a significantly higher TMB (median 23.1 mutations per Mb) compared to samples without a JAK1 frameshift (median 4.5 mutations per Mb; Fig 2B, P = 5.0 x 10 −39 , Mann-Whitney U test), which is consistent with the increased TMB of MSI-H tumors [42]. Significantly, increased TMB in JAK1 frameshift samples was also observed in prostate, colorectal, and gastric cancers (Fig 2C-2E, P < 0.001, all groups, Mann-Whitney U test). Thus, JAK1 frameshifts are a likely pan-cancer alteration that occur primarily in MSI-H cancers with high mutational burden.

Confirmation of JAK1 frameshifts in TCGA data
We sought to confirm our results in a cohort of tumors from TCGA and independently called JAK1 frameshifts in these samples (see Methods). FMI disease types were realigned to match TCGA categories. In endometrial adenocarcinoma (UCEC), 3 ; in the FMI data set we did not observe any JAK1 frameshifts in MSS CRC samples. Finally, we could not confirm the association of JAK1 frameshifts and MSI in prostate cancer since only 3 JAK1 frameshift mutations were detected out of 498 samples tested. Furthermore, MSI status was not provided for TCGA prostate cancer samples. Comparison of MSI-H tumors between the FMI and TCGA cohorts showed similar incidence of JAK1 frameshifts in UCEC and STAD, while the incidence of JAK1 frameshifts in COAD was significantly greater in TCGA ( Fig 2G, COAD P = 0.0144, Fisher's exact test). We also analyzed the association between TMB and JAK1 frameshift alterations. In the TCGA cohort, we defined coding mutations as non-silent, non-splice site mutations across the genome to be consistent with our measure of TMB in the FMI cohort. In UCEC, TMB was significantly higher in samples with a JAK1 frameshift (Fig 2H; P = 1.8 x 10 −6 , Mann-Whitney U test). STAD, and to a lesser extent COAD, also exhibited higher TMBs in samples with a JAK1 frameshift (Fig 2I and 2J; STAD: P = 2.0 x 10 −9 , COAD: P = 0.02). Thus, we confirmed with an independent cohort from TCGA that JAK1 frameshifts are found primarily in MSI-H tumors that have a high TMB with a similar incidence to the FMI cohort.
Endometrial tumors with JAK1 frameshifts exhibit decreased expression of IFN response and immune cell associated genes To study the functional effects of JAK1 frameshifts on tumor phenotype, we compared mRNA expression profiles of JAK1 frameshift and JAK1 wild-type samples from MSI-H UCEC samples in the TCGA dataset. We used gene set enrichment analysis (GSEA) to analyze pathways that were altered in samples with JAK1 frameshift mutations. Gene expression changes were rank ordered using Student's t as the signal to noise metric (S2 Appendix, tab UCEC). In UCEC, IRF9 was the most significantly downregulated gene in samples with JAK1 frameshifts; this protein forms a complex with STAT1 and STAT2 to mediate IFN induced transcription. JAK1 was the 7 th most significantly down regulated gene. In the Hallmark collection of gene sets, ten gene sets were significantly decreased in JAK1 frameshift samples (Table 1, P < 0.05 after family wise error rate (FWER) correction). All GSEA results are shown in an additional file (S3 Appendix). JAK1 mediates signaling in four of these gene sets including IFNγ, IFNα, IL-6, and IL-2 pathways ( Table 1). Five of these gene sets were related to immune responses including inflammatory response, TNFA signaling, allograft rejection, complement, and coagulation (Table 1). Four gene sets were significantly increased in the JAK1 frameshift samples that were related to growth and division including E2F targets, G2M checkpoint, fatty acid metabolism, and mitotic spindle (Table 1). Since five significantly altered gene sets were related to immune responses, we developed a custom set of gene signatures for GSEA derived from a published set of differentially expressed genes in immune cells [43] and manually curated an antigen presentation by MHC class I specific gene set (referred to as antigen presentation). These gene sets were small (range 16-61 genes) and GSEA was run with a high penalty. As a result, enrichment at the front of the rank list had to be higher relative to the hallmark gene sets in order to achieve significance. In this analysis, seven immune gene sets showed significantly decreased expression in JAK1 frameshift samples while none showed significantly increased expression ( Table 1). Several of the reduced gene sets have been associated with anti-tumor immunity including antigen presentation, neutrophils, M1 macrophages, and activated NK cells (Table 1). Thus, in MSI-H UCEC, frameshift mutations in JAK1 resulted in reduced IFN and IL-6 responses as well as reduced M1 macrophages and activated NK cell signatures.
Since GSEA measures average expression, we also examined the expression of mRNAs in the IFNγ response gene set for individual samples. This plot showed similar down regulation of these genes for most JAK1 frameshift tumors (Fig 3A, marker column 1: green JAK1 frameshift positive). However, a subset of samples with JAK1 frameshifts exhibited an IFNγ response signature (Fig 3A, right). One possible explanation for this observation is that the JAK1 loss of function alteration in these tumors was subclonal or affected only one copy, which was a frequent occurrence in the FMI dataset (Fig 1F). We quantified the IFNγ response signature in each sample (see Methods) and observed a bimodal distribution for samples with a JAK1 frameshift (Fig 3B, red bars) but a unimodal distribution for samples without a JAK1 frameshift (Fig 3B, grey bars). We investigated the correlation of the IFNγ response signature with JAK1 expression since expression is reduced in cell lines and tumor samples with a JAK1 frameshift (Fig 1G and 1H). We observed a significant correlation between the expression of JAK1 and IFNγ response in samples with or without a JAK1 frameshift (Fig 3C, JAK1  . From these data, we conclude that homozygous loss of JAK1 leads to loss of the IFNγ response gene set, consistent with a role for JAK1 as a tumor suppressor in this context. We next assessed whether overlap of gene sets with the 200-gene IFNγ response set resulted in false enrichment of similar gene sets. The overlap of the IFNγ response gene set with hallmark gene sets that were altered in both tumors and cell lines was highly variable, with 73/97 genes for IFNα, 37/200 for allograft rejection and inflammatory response, and 5/200 for EMT. To control for overlap with the IFNγ response gene set, we removed overlapping genes and ran the GSEA analysis again. All hallmark gene sets with a FWER P < 0.001 in Table 1 remained significant, while the P value for KRAS, IL2 signaling, and coagulation increased above 0.05. There was also overlap of the IFNγ response gene set with immune gene sets: 9/16 for antigen presentation, 13/48 for M1 macrophages, 8/59 for activated NK cells, and 3/60 for neutrophils. Analysis of these gene sets after removing genes that overlap with IFNγ response resulted in M1 macrophages and activated NK cells having a FWER P > 0.05, and the antigen presentation gene set became too small to test. However, IFNγ is a known activator of NK cells and is involved in M1 macrophage polarization; therefore, the reduction in enrichment for these gene sets was expected and we concluded that overlap with the IFNγ response gene set was not a confounder of these results. Gene expression analysis was performed on resection specimens composed of tumor, stromal, and immune components. To establish which gene sets were altered in tumor cells versus stromal or immune components, we performed the same GSEA analysis on 29 endometrial cell lines characterized in the CCLE (HEC1A was excluded because of its relation to HEC1B). Ten of these cell lines (34.5%) had a JAK1 frameshift. Consistent with the TCGA UCEC tumor samples, JAK1 and IRF9 were the first and third most significantly downregulated genes, respectively (S2 Appendix, tab CCLE). Eight hallmark gene sets were significantly downregulated in JAK1 frameshift samples and five of those were reduced in endometrial tumor samples including IFNα, IFNγ, inflammatory response, allograft rejection, and coagulation ( Table 2, S3 Appendix). The trend of upregulated versus downregulated gene sets was also consistent between UCEC and CCLE even if significance was not reached (Fig 4). We also analyzed the immune gene signatures in the CCLE. The majority of immune gene signatures were downregulated in JAK1 frameshift samples from the CCLE; however, only two gene sets in the CCLE comparison were statistically significant (Fig 4, Table 2, M0 and M2 macrophages). Neither of these gene sets were significantly altered in UCEC tumor samples (Fig 4). Thus, we conclude that changes in the hallmark gene sets were likely reflective of tumor cell intrinsic changes in gene expression while changes in the immune gene sets were likely tumor cell extrinsic.  Stomach adenocarcinomas with JAK1 frameshifts exhibit decreased expression of IFN associated genes We also analyzed the effect of JAK1 frameshifts on gene expression in MSI-H STAD. GSEA analysis of MSI-H STAD revealed four significantly down regulated pathways in STAD with a JAK1 frameshift including IFNα response, IFNγ response, mitotic spindle, and G2M checkpoint (Table 3). Significantly upregulated pathways include the epithelial mesenchymal transition (EMT), cholesterol homeostasis, angiogenesis, and coagulation (Table 3). Analyzing immune signatures showed that antigen presentation and memory B cells were significantly  down regulated (Table 3). Gene expression changes were also clear at the sample level with the majority of STAD with a JAK1 exhibiting reduced expression of genes in the IFNγ response gene set (Fig 5A and 5B). The IFNγ response correlated with JAK1 expression in samples with and without JAK1 frameshifts (Fig 5C; JAK1 frameshift Pearson's r = 0.59, P = 0.01; JAK1 nonframeshift Pearson's r = 0.44, P = 0.0003). As in UCEC, JAK1 expression and JAK1 MAF exhibited an inverse correlation although it did not reach statistical significance (Fig 5D, Pearson's r = -0.327, P = 0.19). The IFNγ response also had a negative correlation with JAK1 MAF in JAK1 frameshift samples that trended towards significance (Fig 5E, Pearson's r = 0.-42, P = 0.078). These findings suggest that in STAD, like UCEC, loss of IFN responsiveness requires complete loss of JAK1.
To begin to analyze broader effects of JAK1 frameshifts, we compared GSEA results of STAD and UCEC. The core effect of IFN unresponsiveness shown by decreases in the IFNα response, IFNγ response, and antigen presentation gene sets were conserved between STAD and UCEC (Fig 6). Immune gene sets were also both generally reduced in STAD and UCEC; however, only two reached significance in STAD. Several gene sets exhibited tumor type specificity. Gene sets associated with proliferation including the mitotic spindle, G2M checkpoint, and E2F targets were increased in JAK1 frameshift UCEC but decreased in JAK1 frameshift STAD (Fig 6). In contrast, gene sets reminiscent of a recently described immunotherapy resistant signature (IPRES) such as angiogenesis and EMT gene sets were higher in JAK1 frameshift STAD but lower in JAK1 frameshift UCEC (Fig 6) [44]. We conclude that JAK1 frameshifts result in reduced interferon responses in UCEC and STAD; however, subsequent phenotypic alterations such as proliferation and mesenchymal phenotype are likely cell type specific.

Discussion
In this study, we showed that loss of function frameshift mutations in JAK1 may have a role in immune evasion. These frameshifts appear in tumors that are MSI-H and have a high TMB, which are therefore likely to elicit immune responses. We observed a similar incidence of these frameshifts in two independent patient cohorts after normalizing by MSI-H status. We additionally showed alterations in gene expression patterns using TCGA and CCLE data that are suggestive of direct effects such as tumor cell-intrinsic reductions in IFN response genes and indirect effects through alterations to inflammation and macrophage polarization. From these data, we derived two primary conclusions: 1) JAK1 frameshifts are loss of function alterations that represent a potential pan-cancer adaptation to immune responses against MSI tumors. We observed these frameshifts in cancers where MSI is both common (i.e. endometrial cancer) and rare (i.e. prostate cancer). Furthermore, the frameshift mutations themselves are likely a result of mismatch repair deficiency which causes MSI. MSI-H tumors have a high TMB that causes an immune response that can be mitigated by loss of function mutations in JAK1. 2) The mechanism by which JAK1 loss of function allows immune evasion is likely loss of JAK1dependent interferon responses. JAK1 is required for interferon induced antigen presentation and growth inhibition [21,22]. Of the three pathways that utilize JAK1, only IFN response was significantly down regulated in both endometrial and stomach cancers. Endometrial cancer cell lines with JAK1 frameshifts have been shown to be resistant to the effects of IFNγ, including a lack of growth arrest and a failure to up regulate antigen processing and presentation [45]. However, some tumor samples with JAK1 frameshift mutations in the TCGA dataset exhibited high expression of IFNγ response genes. Using JAK1 expression and MAF, we showed that these samples likely have heterozygous or subclonal JAK1 frameshift mutations with complete loss of JAK1 required to abrogate signaling. About 40% of samples in the FMI dataset had a heterozygous deficiency in JAK1. Another explanation for IFN signaling in these samples is JAK1 independent NF-κB pathway signaling, which has been demonstrated for IFNα treatment [46]. However, most of the genes in the IFN response gene sets have a STAT1 consensus transcription factor binding site (data not shown) [47], while the JAK1 independent NF-κB pathway would be expected to activate NF-κB responsive genes. In total, we conclude that JAK1 frameshift mutations are a potential pan-cancer adaptation to anti-tumor immunity and that these alterations likely result in a lack of antigen presentation and resistance to interferons.
Based on this data, we speculate that JAK1 frameshifts will also affect treatment with checkpoint inhibitors. Current companion diagnostics for checkpoint inhibitors utilize PD-L1 immunohistochemistry to identify patients likely to respond to these agents in some tumor types. Since PD-L1 (CD274) is an IFNγ responsive gene, tumors with a JAK1 frameshift will likely have reduced expression of PD-L1 and may therefore have a negative result on the companion diagnostic assay. In STAD, high expression of CD274 in MSI-H JAK1 frameshift STAD was rare compared to MSI-H JAK1 wild-type STAD (data not shown). Therapeutically, loss of JAK1 or MHCI pathway genes was shown to be associated with adaptive resistance to single agent pembrolizumab in a recent study of four relapsed melanoma patients [48]. In these patients, heterozygous point mutations in JAK1 or MHCI pathway components in the primary sample exhibited loss of heterozygosity in the relapsed sample. Loss of B2M was also observed previously in melanoma patients treated with ex vivo expanded CD8 + T cells and IL-2. Using TCGA data, we found that 71% (60/85) of STAD, 80% (36/45) of COAD, and 31% (52/169) of UCEC had a frameshift in the MHCI pathway; however, the occurrence of these mutations was independent of JAK1 frameshifts and did not affect our analysis. While the response rate to checkpoint inhibitors has been higher in MSI-H cancers compared to MSS cancers of the same disease type, the determinants of response within MSI-H cancers remain to be described [13]. We speculate that tumors with complete loss of function of JAK1 will exhibit primary resistance to checkpoint inhibitors. Tumors with heterozygous or subclonal frameshifts in JAK1 will relapse after acquiring another JAK1 alteration or undergoing loss of heterozygosity at the JAK1 locus. Additional data, including paired genomics and immunotherapy treatment outcomes, will be needed to validate these conclusions; nevertheless, our analysis strongly suggests loss of function JAK1 alterations represents a tumor-intrinsic mechanism of immune evasion and possibly renders MSI-H tumors resistant to checkpoint inhibitors.
JAK1 frameshift alterations have been described mainly in the context of endometrial cancer [49]. The underlying biochemistry of these alterations is well explained, with frameshifts showing a loss of function phenotype for IFNγ signaling [45]. This study was consistent with knockout studies in cell lines and mice that show JAK1 is a critical mediator of signaling for the IFN, IL-6, and IL-2 families of cytokines [22]. We extended on these studies by analyzing a database of over 60,000 solid tumors to show that these mutations occur across multiple cancer types and are associated with MSI-H and high TMB. Thus, the previous association of these alterations with endometrial cancer is not unexpected since MSI is frequent in endometrial cancer and JAK1 frameshifts occur in 40-50% of MSI-H endometrial cancers [42]. The rare occurrence of JAK1 frameshifts in prostate or urinary cancer can be explained in part by the rarity of MSI in these tumor types. This positive association with MSI strengthens the conclusion that JAK1 frameshifts are likely immune evasion mutations, and may represent a mechanism through which MSI-H tumors can circumvent immune pressures. Previous reports have shown frequent JAK1 frameshifts in endometrial cancer but none were observed in CRC, another tumor type with an established MSI-H phenotype [49]. We observed a difference in prevalence between these two diseases but did detect CRC samples with JAK1 frameshifts. Analyzing TCGA and CCLE expression data with GSEA, we confirmed indirectly the underlying biochemistry in human tumors samples by showing that expression of response genes for IFN and IL-6 families of cytokines are reduced in tumor samples and cell lines with JAK1 frameshifts. Beyond biochemical effects, reductions in gene sets associated with anti-tumor immunity were also observed in human cancers. Thus, these mutations are not limited to endometrial cancer and the direct effects of IFN; they also occur in MSI-high tumors and result in both unresponsiveness to IFN and a blunted tumor immune response.
Gene expression analysis in this study was performed on tumor samples collected by the TCGA consortium and are thus composed of tumor, immune, and stromal cells. While we adapted previous methods designed to deconvolute expression patterns between cell types in these types of specimens [43,50], the analysis is still being performed on a mixture of cells. Therefore, gene expression changes we attributed to tumor cell intrinsic processes may instead be tumor microenvironment-relevant. To control for this potential confounder, we performed GSEA on endometrial cell lines, without immune or stromal components, from the CCLE. This analysis suggested that the effects of JAK1 frameshifts on IFN signaling were likely tumor cell intrinsic whereas decreased expression of activated immune cell genes was likely cell extrinsic. Interestingly, the inflammatory response and allograft rejection gene sets were significantly different in both the cell line and tumor analyses. The changes to these gene sets may simply reflect overlap, known or unknown, with IFN response genes. The cell line data also could not control for gene sets associated with growth and division since cell lines divide continuously and are grown in media with external growth factors. In endometrial tumor samples, we observed increased expression of gene sets associated with cell growth and division in tumors with JAK1 frameshifts and it is tempting to conclude that unresponsiveness to the growth arrest induced by interferon allows these tumors to grow faster than JAK1 wild-type tumors. However, given that we were analyzing a mixture of cells, it was unclear which cell types were dividing and we thus have refrained from making conclusions about cell growth and division processes. Despite these limitations, we observed the expected phenotype of IFN unresponsiveness in tumors with a loss of function JAK1 mutation. The immune alterations we observed were centered on the type or activation state of the immune response. In endometrial cancer, we observed differences in antigen presentation, M1 macrophage differentiation, and activated NK cells, but not resting NK cells or other macrophage types. We speculate these changes in expression of UCEC samples to be the result of tumor-cell extrinsic changes of macrophage polarization away from an anti-tumor M1 phenotype to a more neutral M0 or pro-tumor M2 phenotype. These data are consistent with observations by us and others that MSI-H endometrial cancers had lymphocytic infiltrates regardless of JAK1 alteration status (data not shown and [51]). In addition, STAD exhibited increased expression of gene sets similar to the IPRES immunotherapy resistance signature [44]. Based on these data, we speculate that JAK1 frameshift alterations and the resultant unresponsiveness to IFN lead to reduced anti-tumor immune responses.
In this study, we found no evidence for prognostic value for JAK1 frameshifts in TCGA data and the effect of these mutations on outcome will likely require the incorporation of additional clinical variables. A separate study of MSI-H endometrial cancers also showed no prognostic value of JAK1 frameshift mutations in patients receiving standard clinical care [51]. MSI-H tumors have a better prognosis than MSS tumors in colorectal cancer [16]. In endometrial cancer, MSI-H tumors have an intermediate prognosis, with POLE mutated cancers having the best prognosis and serous-like cancers having the worst [42]. Furthermore, mutation of the B2M gene in MSI-H colon cancers is an independent positive prognostic factor. B2M mutant colon cancers formed large primary tumors but rarely metastasized [16]. Clinically, many MSI-H cancers will likely be cured by surgery and adjuvant chemotherapy. However, a proportion will progress, and in this context, JAK1 frameshifts may play a role in immune evasion. This may also be true for other gastrointestinal cancers where JAK1 and other MHCI pathway components may be contributing to this phenotype. JAK1 frameshifts may also play a role in evasion to checkpoint inhibitors.