Acquired IFNγ resistance impairs anti-tumor immunity and gives rise to T-cell-resistant melanoma lesions

Melanoma treatment has been revolutionized by antibody-based immunotherapies. IFNγ secretion by CD8+ T cells is critical for therapy efficacy having anti-proliferative and pro-apoptotic effects on tumour cells. Our study demonstrates a genetic evolution of IFNγ resistance in different melanoma patient models. Chromosomal alterations and subsequent inactivating mutations in genes of the IFNγ signalling cascade, most often JAK1 or JAK2, protect melanoma cells from anti-tumour IFNγ activity. JAK1/2 mutants further evolve into T-cell-resistant HLA class I-negative lesions with genes involved in antigen presentation silenced and no longer inducible by IFNγ. Allelic JAK1/2 losses predisposing to IFNγ resistance development are frequent in melanoma. Subclones harbouring inactivating mutations emerge under various immunotherapies but are also detectable in pre-treatment biopsies. Our data demonstrate that JAK1/2 deficiency protects melanoma from anti-tumour IFNγ activity and results in T-cell-resistant HLA class I-negative lesions. Screening for mechanisms of IFNγ resistance should be considered in therapeutic decision-making.

U nderstanding the mechanisms of T-cell inhibition by melanoma cells allowed for the development of new agents with considerable activity against metastatic disease including antibodies targeting the PD-L1/PD1 axis. PD-L1 expressed on melanoma cells binds its inhibitory PD1 receptor on cytotoxic CD8 þ T lymphocytes generating a checkpoint signal dampening the T cell's effector function 1 . Release from checkpoint blockade by treatment with anti-PD1 antibodies yields clinical benefit in a substantial proportion of melanoma patients, experiencing durable disease stabilization, tumour regression as well as complete remission [2][3][4] . Response to anti-PD1 therapy is strongly associated with the expression of its ligand on melanoma cells and the presence of CD8 þ T cells in the margin or center of metastatic lesions 5 . How T cells mediate disease stabilization or regression of bulky tumour masses remained unclear so far.
Upon activation by cognate HLA class I antigen complexes, T cells release cytolytic granules, containing perforins and granzymes, onto their target cells and secrete interferon (IFN)g acting on cells in the microenvironment 6 . Perforin/granzymemediated killing and induction of apoptosis by death receptor engagement have long been considered the major anti-tumour effector mechanisms of CD8 þ T cells. Accordingly, expression of cytolytic markers in pretreatment melanoma biopsies was found to be significantly associated with clinical benefit to antibodies targeting the T-cell checkpoint CTLA-4 (ref. 7). But evidence from different in vivo studies suggests that the anti-proliferative and pro-apoptotic activity of IFNg on melanoma cells contributes essentially to the efficacy of T-cell-mediated anti-tumour immunity.
IFNg binds to the heterodimeric IFNGR1/IFNGR2 receptor complex, leading to the activation of the receptor-associated kinases JAK1 and JAK2 that in turn phosphorylate STAT1. Phosphorylated STAT1 homodimers activate transcription of primary response genes including the transcriptional activator IRF1 that in turn coordinates the expression of secondary response genes 8 . Activation of the JAK1/2-STAT1-IRF1 signalling cascade in melanoma cells as well as other tumour cells can induce growth arrest and death via different pathways 9-12 . Recently, it was demonstrated that adoptively transferred tumour antigen-specific CD8 þ T cells infiltrating B16 melanoma lesions at low numbers arrested the growth of several times higher numbers of tumour cells in an IFNg-dependent manner 11 . Furthermore, T-cell-derived IFNg in combination with tumour-necrosis factor (TNF)a was found to be essential also for in vivo induction of tumour-cell senescence abrogating disease progression in a pancreatic tumour model 13,14 .
Based on this knowledge, we postulated that melanoma cells from patients responding to immunotherapy should be sensitive to the anti-proliferative and pro-apoptotic effects of IFNg and that continuous cytokine exposure should select for the outgrowth of IFNg-resistant tumour subclones. Here we demonstrate that IFNg-resistant melanoma clones with inactivating JAK1/JAK2 mutations frequently evolve in patients receiving different types of immunotherapy. IFNg-resistant tumour cells are protected from cytokine-induced growth inhibition and apoptosis. Additionally, JAK1/JAK2-deficient lesions become T-cell-resistant by silencing HLA class I antigen presentation, which can no longer be restored by IFNg signalling. Our findings suggest sequential screening of tumour biopsies for genetic defects in the IFNg signalling cascade will aid therapeutic decision-making in patients with advanced melanoma.

Results
Acquired mutations in genes of the IFNc signalling pathway. Assuming that the growth-inhibitory and pro-apoptotic activity of T-cell-derived IFNg acts selectively on tumour cells, the evolution of genetic variants in melanoma with impaired cytokine signalling was explored. In a first step, we evaluated available exome data of 46 melanoma cell lines 15 , established from metastases of different patients in our laboratory, for aberrations in IFNGR1, IFNGR2, JAK1, JAK2, STAT1 and IRF1. Mutations in JAK1 (n ¼ 3), JAK2 (n ¼ 1) and STAT1 (n ¼ 1) were detected in 5 out of the 46 cell lines (Table 1). By Sanger sequencing we confirmed the mutations on freshly isolated DNA from the respective cell lines Ma-Mel-36, Ma-Mel-53, Ma-Mel-54a, Ma-Mel-85 and Ma-Mel-102. Independent of existing exome data, Sanger sequencing revealed a JAK1 mutation in a cell line from melanoma patient Ma-Mel-61 ( Table 1). The specific mutations present in the cell lines were also detected in situ in corresponding tumour tissue, with the exception of metastasis Ma-Mel-54a. As shown in Table 1, targeted sequencing revealed a homozygous status for the mutant allele in three of the six cell lines (Ma-Mel-54a, Ma-Mel-61g, Ma-Mel-102). To determine whether these mutations functionally impaired IFNg signalling, the cell lines were treated with recombinant IFNg for 48 h followed by protein expression analyses of pathway components and downstream targets.
Despite a STAT1 (c.947C4T) mutation frequency of approximately B100%, Ma-Mel-102 cells still showed a slight induction of pSTAT1 and IRF1 in the presence of IFNg ( Supplementary  Fig. 1a-c). However, the signals were weak and detectable only after long-term exposure of X-ray films, suggesting that the S316L exchange, located between the coiled-coil and DNA binding domains, strongly decreased STAT1 protein stability without necessarily leading to its complete inactivation. Parallel analyses on Ma-Mel-85 cells revealed a strong IFNg pathway activation compared with Ma-Mel-102 cells, accompanied with an elevated surface expression of CD54, HLA class I and PD-L1 ( Supplementary Fig. 1b-d). This is in line with a JAK1 (c.1548C4A) mutation frequency of only 50% ( Supplementary  Fig. 1a), suggesting that wild-type JAK1 was still active in Ma-Mel-85 cells. As expected, IFNg signalling was detected also in Ma-Mel-53 cells showing a JAK1 (c.2338G4A) mutation frequency of only 24% ( Supplementary Fig. 1a-c).    HLA-DR expression profiles. The IFNg-sensitive subpopulation Ma-Mel-36_sens strongly upregulated HLA-DR and PD-L1 surface expression in response to cytokine treatment, whereas the IFNg-resistant subpopulation Ma-Mel-36_res remained HLA-DR-negative and PD-L1-low under these conditions (Fig. 1b). By targeted sequencing we detected the JAK1 mutation (c.843C4A) encoding a truncated non-functional JAK1-Y281* variant in B60 and 100% of Ma-Mel-36_bulk and Ma-Mel-36_res cells, respectively, but not in Ma-Mel-36_sens cells (Fig. 1c). Accordingly, IFNg treatment resulted in pSTAT1 and IRF1 detection in lysates from Ma-Mel-36_bulk and Ma-Mel-36_sens cells but not from Ma-Mel-36_res cells (Fig. 1d).
In line with the sequencing results, JAK1 protein expression was detected only in Ma-Mel-36_sens but not in Ma-Mel-36_res cells (Fig. 1e).
Assuming that JAK1 deficiency in these cells was due to a gene mutation and concurrent allelic loss, we performed single-nucleotide polymorphism (SNP) array analyses on DNA obtained from the three tumour cell populations and autologous As shown in Fig. 1a, metastasis Ma-Mel-36 developed after the patient had been treated with recombinant IFNa and a combination of dacarbacine/IFNa/interleukin (IL) 2, suggesting activated tumour-reactive T cells selectively enriched the IFNg-resistant cell subpopulation. Indeed, peripheral blood CD8 þ T cells from patient Ma-Mel-36 secreted IFNg in the presence of autologous melanoma cells, as determined by ELISpot assay (Fig. 1g). Pretreatment of tumour cells with IFNg slightly enhanced the activation of CD8 þ T cells by Ma-Mel-36_bulk and Ma-Mel-36_sens cells, whereas the T-cell-stimulatory capacity of Ma-Mel-36_res cells was not affected. Furthermore, impedancebased measurement of real-time proliferation in the xCELLigence system revealed a negative impact of IFNg on the expansion of Ma-Mel-36_bulk and Ma-Mel-36_sens cells, while Ma-Mel-36_res cells efficiently proliferated (Fig. 1h). This was measurable also in terms of cell numbers: a considerable reduction in Ma-Mel-36_bulk and in particular Ma-Mel-36_sens cells was noted in the presence of IFNg, due to an increase in apoptosis ( Fig. 1i and Supplementary Fig. 2d,e). In contrast, cell numbers and spontaneous apoptosis of IFNg-resistant Ma-Mel-36_res cells remained unaffected under these conditions ( Fig. 1i and Supplementary Fig. 2d,e).
As shown in Fig. 1e, Ma-Mel-36_sens cells responded to IFNa treatment whereas JAK1-deficient Ma-Mel-36_res cells were also resistant to type I IFN. Considering patient Ma-Mel-36 received IFNa-based therapies before the development of resistant lesions, we hypothesized that type I IFN signalling by affecting cell survival might have contributed to the enrichment of JAK1-deficient cells. However, in contrast to IFNg, IFNa treatment did not affect the survival of Ma-Mel-36_sens cells ( Supplementary Fig. 2f).
JAK2 deficiency blocks HLA class I upregulation by IFNc. By Sanger sequencing we found a JAK2 c.2876A4C exchange to be present in Ma-Mel-54a cells that, however, could not be detected in the corresponding tumour tissue (Table 1). To demonstrate that, in fact, the specific genetic alteration was acquired in the course of disease we sequenced DNA from a second cutaneous lesion (Ma-Mel-54b) of the patient, obtained one month after excision of metastasis Ma-Mel-54a (Fig. 2a). Indeed, tumour tissue Ma-Mel-54b and the corresponding cell line harboured the JAK2 mutation already present in Ma-Mel-54a cells (Fig. 2b). Both cell lines showed a JAK2 (c.2876A4C) mutation frequency of 100%, resulting in a Q959P exchange in the functionally important JAK2 JH1 kinase domain 16 . As shown in Fig. 2c, mutant JAK2-Q959P was no longer detectable by western blot. Accordingly, IFNg signalling was completely abrogated in both cell lines, which no longer showed CD54, HLA class I and PD-L1 upregulation in response to cytokine treatment (Fig. 2d,e).
Assuming that JAK2 deficiency of Ma-Mel-54 cells was caused by the co-occurrence of a JAK2 gene mutation and allelic JAK2 loss, we performed SNP array analyses on DNA obtained from the two cell lines and autologous peripheral blood cells as a constitutive, normal control to detect aberrations of chromosome 9p to which the JAK2 gene maps at Chr.9p24.1. The same deletion on chromosome 9p, encompassing the region 9p24.3-p13.2 (Chr.9:203,861-37,578,327) was detected in Ma-Mel-54a and Ma-Mel-54b cells (Fig. 2f), demonstrating the common origin of JAK2 deficiency in both metastases. As shown in Fig. 2g, IFNg sensitivity of Ma-Mel-54a cells was restored upon transient JAK2 re-expression as indicated by the induction of signalling pathway components. Furthermore, Ma-Mel-54a-JAK2 transfectants proved to be sensitive towards the anti-proliferative activity of IFNg in contrast to non-transfected Ma-Mel-54a cells (Fig. 2h).
Interestingly, western blot analyses suggested a lack of HLA class I heavy chain expression in Ma-Mel-54a compared with Ma-Mel-54b cells (Fig. 2d). Indeed, Ma-Mel-54a cells only weakly expressed HLA class I surface molecules (Fig. 2e). By quantitative reverse transcription-PCR we demonstrated low level expression of specific messenger RNAs (mRNAs) involved in antigen presentation in Ma-Mel-54a cells (Fig. 2i)   HLA-B, HLA-C, TAP1, TAP2 and B2M genes as well as HLA class I surface molecules were strongly upregulated solely by IFNgtreated Ma-Mel-54a-JAK2 transfectants (Fig. 2j,k). Thus, JAK2 deficiency protected Ma-Mel-54a cells not only from anti-tumour IFNg activity but also conserved their HLA class I-low phenotype that in turn might have hampered effective T-cell recognition of the tumour cells. Due to a lack of cryopreserved autologous T cells from this patient, the functional significance of the HLA class I-low phenotype could not be investigated in more detail. By targeted sequencing on DNA from Ma-Mel-61g and blood cells as a constitutive normal control we detected a JAK1 c.1798G4T, JAK1-G600W mutation in 99% of the cells (Fig. 3c, Supplementary Fig. 3b), with the JAK1-G600W exchange affecting a conserved amino acid in the auto-inhibitory pseudokinase domain 17 . None of the IFNg-sensitive Ma-Mel-61 cell lines showed a JAK1 mutation (Fig. 3c, Supplementary  Fig. 3b). The JAK1-G600W mutant protein was still detectable by western blot (Fig. 3d), but was found to be completely inactive, as no upregulation of pSTAT1, STAT1 and IRF1 was detected in Ma-Mel-61g cells in response to IFNg treatment (Fig. 3e). Consistently, transfection of Ma-Mel-61g cells with an expression plasmid encoding wild-type JAK1, in contrast to JAK1-G600W, restored IFNg signalling ( Fig. 3f and Supplementary Fig. 3c). Again we detected autologous CD8 þ T cells secreting IFNg in the presence of Ma-Mel-61g cells in the patient's peripheral blood, whose activity might have selected for the outgrowth of IFNg-resistant cells (Fig. 3g). Accordingly, while proliferation of Ma-Mel-61g cells was not affected by IFNg, the cells became sensitive upon JAK1 re-expression (Fig. 3h). Analysing the genetic evolution of JAK1 deficiency, we found a deletion on chromosome 1p, encompassing the region 1p34.3-1p12 (Chr.1:40,061,699-118,932,325) including the JAK1 gene, to be present in all Ma-Mel-61 cell lines (Fig. 3i). This demonstrated that allelic JAK1 loss was an early event in the course of disease progression in this patient, predisposing to IFNg-resistance development, and that a subsequent inactivating JAK1 point mutation led to complete abrogation of type II IFN signalling in Ma-Mel-61g cells.
Evolution of T-cell-resistant lesions. Similar to Ma-Mel-61g, IFNg signalling was abrogated in Ma-Mel-61h cells due to the homozygous JAK1 c.1798G4T, JAK1-G600W mutation (Fig. 4a,b). In contrast to Ma-Mel-61g, however, Ma-Mel-61h cells demonstrated a stable HLA class I-negative phenotype. Lack of HLA heavy chains and HLA class I surface expression was observed by western blot and flow cytometry, respectively (Fig. 4b,c). Staining of Ma-Mel-61g and Ma-Mel-61h tissue sections revealed the presence of HLA class I-negative tumour cells in both lesions ( Fig. 4d and Supplementary Fig. 3d). Quantification of mRNAs involved in antigen presentation indicated a complete lack in the expression of HLA-B, HLA-C, TAP1 and TAP2 in Ma-Mel-61h cells compared with the control cells Ma-Mel-61b (Fig. 4e). Transient JAK1 re-expression and subsequent IFNg treatment induced de novo HLA-B, HLA-C, TAP1 and TAP2 mRNA expression demonstrating a reversible silencing of antigen presentation in Ma-Mel-61h cells (Fig. 4e). This could not be observed for Ma-Mel-61h cells expressing JAK1-G600W, confirming functional inactivity of the mutant protein. Consistently, enhanced expression of specific mRNAs could also be measured for IFNg-treated Ma-Mel-61g-JAK1 but not for Ma-Mel-61g-JAK1-G600W transfectants (Fig. 4e). As shown in Fig. 4f, the subpopulation of Ma-Mel-61h-JAK1 transfectants demonstrated de novo HLA class I surface expression after IFNg-treatment, resulting in detection of the previously ignored tumour cells by autologous CD8 þ T cells (Fig. 4g). Overall, these data demonstrated that JAK1 deficiency in Ma-Mel-61h cells was followed by silencing of antigen presentation, generating a T-cell-resistant melanoma phenotype. Similar results were obtained for JAK2-deficient Ma-Mel-54a cells (Fig. 2g,i,k), suggesting the broader significance of our findings.
Genetic alterations defined in different melanoma data sets. The detection of IFNg-resistant melanoma metastases in our patient cohort led us to assess the presence of alterations in type II IFN signalling pathway components in independent sample homozygous deletions, that in most cases were mutually exclusive ( Fig. 5a; Supplementary Data 1, Supplementary Fig. 4). Interestingly, a remarkable fraction of the biopsies (44%, 16   affecting genes of the IFNg signalling cascade were more frequent in metastatic samples compared with primary tumours (Fig. 5b) and were detectable in metastases from patients receiving neo-adjuvant IFNa and from patients without treatment (Supplementary Data 1). Tumour biopsies with mutations did not show elevated expression of IFNG or CD8A mRNA in comparison to biopsies without mutations, suggesting a comparable activity of T cells in both types of lesions ( Supplementary Fig. 5). Support of our finding of recurrent genetic alterations in type II IFN signalling pathway genes was also obtained from additional published data sets. A mutation frequency of 22% (11 of 49), based on SNV/Indels and homozygous deletions, was detected in melanoma cell lines studied by the Cancer Cell Line Encyclopedia (CCLE) 19 . Considering only SNV/Indels, the frequency in melanoma tissue samples was 7% (20 of 287) for the TCGA melanoma collection (Fig. 5a) 18 , 3% (3 of 91) for melanoma biopsies studied by Krauthammer et al. 20 (Fig. 5c), 6% (3 of 49) for melanoma cell lines from CCLE 19 (Fig. 5d) and 10% (12 of 121) for the cell lines analysed by Hodis et al. 15 (Fig. 5e). Overall these analyses found mutations to be present in a considerable fraction of melanoma cells.
Our studies in the three melanoma patient models identified allelic JAK1 and JAK2 losses as initial genetic alteration predisposing to IFNg-resistance development. This led us to screen available SNP microarray data from 59 'in-house' melanoma cell lines for loss of heterozygosity (LOH) in the JAK1 and the JAK2 locus 21 . As shown in Fig. 5f, LOH for JAK1 was detected in 25% (15 of 59) and for JAK2 in 76% (45 of 59) of the cell lines suggesting a high risk of resistance development in the course of an effective anti-tumour T-cell response.
Based on the above sequencing results and functional data, we asked for the impact of alterations in IFNg signalling genes on the control of disease progression. When assessed in the largest available TCGA melanoma cohort with survival data (479 samples) these alterations were found to have a statistically significant negative impact on patient survival (Fig. 6a). On the other hand, elevated mRNA levels for STAT1 as well as its downstream target IRF1, indicating active IFNg signalling, were strongly associated with improved overall survival (Fig. 6b). The direct correlation between IFNG and CD8A mRNA expression pointed to CD8 þ T cells as the major cytokine source ( Supplementary Fig. 6a,b).
Mutations emerge before checkpoint blocking therapy. As resistance to IFNg signalling could impact therapy efficacy, we screened 59 formalin-fixed, paraffin-embedded (FFPE) tumour samples from patients receiving anti-PD1 therapy for corresponding mutations by targeted sequencing of DNA isolated from macrodissected tumour cells and autologous control blood cells ( Table 2). In 19% of the biopsies (11 out of 59) mutations were identified, affecting IRF1 (n ¼ 1), JAK1 (n ¼ 5) and JAK2 (n ¼ 5), some of them clearly inactivating as the stop codon mutation in sample 15-12774 (IRF1 W195*), and the frame shift mutations in specimens D5923-13 (JAK2-D710fs) and 18298-15 (JAK2-T376fs). Furthermore, we determined the mutation frequency in pre-treatment biopsies from patients having received anti-CTLA-4 therapy, in this case evaluating existing exome sequencing data 7 . Mutations in type II IFN signalling pathway genes were identified in 9% of samples (10 out of 110) ( Table 2).
Overall, these sequencing results demonstrate that genetic alterations in type II IFN signalling pathway components are present in a considerable number of melanomas. Although we could not detect significant differences in responses to anti-PD1 or anti-CTLA-4 treatment between patients with or without mutations (Supplementary Table 1), it is important to note that two of the three patients with clearly inactivating mutations (15-12774: IRF1 W195*; 18298-15: JAK2 D710fs) showed progressive disease under anti-PD1 treatment, while the third patient (18298-15: JAK2 T376fs) showed a partial response, suggesting that these mutations might have contributed to therapy resistance. Considerably larger cohorts of anti-PD1 and anti-CTLA-4 treated patients will be required to allow conclusive statistical analysis to be performed in future studies.

Discussion
Remarkable response rates in treatment of metastatic melanoma have been reported for adoptive T-cell transfer 22,23 and therapy with immune checkpoint-blocking antibodies, including anti-PD1 monotherapy 2,3 as well as anti-PD1 and anti-CTLA-4 combination therapy 4,24 . Long-term data from anti-CTLA-4 therapy suggest that a number of patients will show a durable complete response and may even be healed of metastatic disease 25 . Despite the considerable therapeutic potential, not all patients benefit equally well from immunotherapy. Primary as well as acquired therapy resistance is a major concern and the identification of resistance mechanisms is crucial for advancing treatment of melanoma and other malignancies.
The data presented in this work signify that under the selective pressure of an effective T-cell response tumour clones evolve that are considerably less susceptible or even resistant to T-cell effector mechanisms. As such the direct cytotoxic effects of CD8 þ T cells mediated by release of cytolytic granules or death receptor engagement are an essential but most likely insufficient part of the overall anti-tumour response that also depends on the secretion of IFNg. By induction of growth arrest and cell death IFNg has a broader impact on tumour cells and their   [9][10][11]13,14 . Accordingly, analyses on TCGA melanomas revealed a strong association between patient survival and elevated mRNA levels of STAT1 and its downstream target IRF1, indicating IFNg-dependent pathway activation. Furthermore, the strong correlation between IFNG and CD8A mRNA in melanomas, both known as favourable prognostic markers [26][27][28][29] , argues for CD8 þ T cells as a major IFNg source. However, a significant contribution by other lymphocytes such as CD4 þ T cells of the Th1 phenotype or natural killer cells cannot be excluded 13,30 . ARTICLE Assuming that IFNg exerts a strong selective pressure on tumour cells we screened an 'in-house' collection of short-term cultured melanoma cell lines for mutations in genes of the IFNg signalling pathway and detected JAK1, JAK2 and STAT1 alterations in cells and corresponding tumour tissue from 6 out of 47 patients. In two heterozygous JAK1 mutants IFNg signalling was still active but was strongly impaired and no longer detectable in homozygous STAT1 and JAK1/2 mutants, respectively. JAK1/2-deficient tumour cells emerged in disease stage IV metastases under/after immunotherapy. The different treatments, including IFNa, IL2 and combinations thereof, might have induced or boosted the effector functions of tumour-reactive CD8 þ T cells, favouring mutant outgrowth. Indeed, we demonstrated that JAK1/2 loss protected melanoma cells from anti-proliferative and pro-apoptotic IFNg activity. Since JAK1 is a component also of the type I IFN signalling pathway, an additional selective pressure of IFNa on tumour cells cannot be excluded 8,31 . In contrast to the JAK1/2-deficient melanoma cells, STAT1 mutant cells were established from a treatment-naive stage III lymph node metastasis suggesting that in this case spontaneous anti-tumour T-cell responses enriched these cells. In addition to melanoma, inactivating mutations in genes related to IFNg signalling, in particular JAK1, have recently been described for microsatellite instable endometrial and colorectal cancers, arguing for a contribution to disease progression also in other malignancies [32][33][34][35][36][37] .
In our patient models, JAK1 deficiency originated from an initial chromosome 1p aberration causing mono-allelic JAK1 loss in melanoma cells and a subsequent mutation inactivating the remaining JAK1 allele. Losses of the short arm of chromosome 1 are not uncommon in cutaneous and uveal melanoma with larger deletions occurring in around 10% of cutaneous melanomas 38,39 . More focal deletions as well as copy number neutral losses of heterozygosity may occur, that we detected at a high frequency in a cohort of 59 melanoma cell lines 21 , showing also a very high frequency of allelic JAK2 losses (76%). All of these alterations in addition to gene mutations would predispose to JAK1/JAK2 inactivation and IFNg-resistance in tumours if put under selective pressure by the immune system.
Interestingly, our data demonstrate that IFNg-resistant JAK1/2-deficient melanoma cells progress to a 'higher level' of immunotherapy resistance. We provide evidence for the first time that IFNg-resistant HLA class I-positive metastases can evolve into HLA class I-negative lesions thereby gaining complete CD8 þ T-cell resistance. The HLA class I-negative phenotype is caused by a coordinated silencing of genes involved in antigen presentation (HLA-B, HLA-C, TAP1, TAP2, B2M). Downregulation of this set of genes has previously been reported for melanoma and other tumour entities. The underlying molecular silencing mechanisms remain unclear but are most likely of epigenetic nature [40][41][42] . IFNg is well known for its role in upregulating antigen processing and presentation thereby augmenting the detection and elimination of malignant cells by tumour antigen-specific CD8 þ T cells 43,44 . However, in case of JAK1/2 deficiency IFNg-induced restoration of antigen presentation in tumour cells is abrogated. Phenotypically HLA class I-negative JAK1/2-deficient metastases share features with tumours lacking HLA class I surface expression due to inactivating B2M mutations as described by us and others 33,[45][46][47][48] . HLA class I-negative metastases will be resistant towards any type of immunotherapy that is dependent on the activity of HLA class I-restricted tumour antigen-specific CD8 þ T cells, including adoptive cell therapy and checkpoint modulators. However, in contrast to B2M mutants, melanoma cells of the regulatory HLA class I-negative phenotype can regain HLA class I expression to adapt to specific environmental conditions such as metastatic sites (for example, lung, liver) enriched for natural killer cells that are specialized in recognition and killing of HLA class I-negative malignant cells 49 .
Recently, resistance to anti-PD1 and anti-CTLA-4 therapy has been associated with sustained IFNg signalling upregulating ligands for multiple inhibitory receptors on T cells, as well as IFNg resistance protecting from cytokine-induced cell cycle arrest/apoptosis [33][34][35]50 . It will be of importance to determine how far the alterations we detected in pretreatment biopsies will undergo positive selection in tumours recurring upon anti-PD1 therapy or whether mutations will evolve de novo as recently described 33 . Of equal importance will be the identification of novel IFNg-resistance mechanisms. Epigenetic factors as well as altered expression of negative IFNg pathway regulators in tumour cells or microenvironmental influences could have an additional relevant role in conferring resistance or reduced sensitivity to IFNg 51,52 . Furthermore, the resistance mechanisms could go beyond IFNg and apply to other cytokines such as TNFa 13,14 . In this regards, the combined action of IFNg and TNFa has been demonstrated to destroy tumour cells and their stroma thereby essentially contributing to the eradication of established mouse tumours 53 .

Methods
Patients samples. Peripheral blood samples and tumour tissues were collected after written informed patient consent with institutional review board approval. Melanoma cell lines were established from excised metastatic lesions. Cell lines were confirmed to be mycoplasma-free in monthly intervals and authenticated by genetic profiling on genomic DNA at the Institute for Forensic Medicine (University Hospital Essen) using the AmpFLSTR-Profiler Plus kit (Applied Biosystems). Melanoma cells were cultured in RPMI1640 or DMEM medium with L-glutamine (Gibco/Life technologies) and 10% fetal calf serum. Cells were seeded and rested overnight followed by addition of IFNg (500 U ml À 1 , Boehringer Ingelheim) or IFNa2b (1,000 U ml À 1 , Essex Pharma) and incubation for indicated periods.
Immunohistochemistry. Serial cryostat tissue sections were stained with antibodies specific for HLA-DR,-DP,-DQ, kindly provided by S.  Kaplan-Meier survival plots and log-rank tests in R (R Development Core Team; http://www.R-project.org). Coexpression plots were obtained for TCGA SKCM samples using cBioPortal 57 . Mutation calls, including SNV and Indels, from 110 patients before anti-CTLA-4 antibody therapy were reported previously 7 . Additional mutation data for melanoma tissues and cell lines with SNV/Indels 15,20 as well as for melanoma cell lines including also homozygous deletions, as part of the CCLE project 19 were assessed using cBioPortal 57 . SNP array data from 59 melanoma cell lines analysed with the 250 k StyI SNP array of the Affymetrix GeneChipV 500 K array set (Affymetrix, Santa Clara, CA), GEO accession number GSE17534 (ref. 21) were assessed for LOH using Affymetrix genotyping console software. In 44 cases, SNP data from corresponding germline DNA were available for paired analysis. In the remaining cases, the SNP data from tumour samples were compared with baseline values obtained from combined analysis of the SNP data from the 44 available germline cases.
Plasmid generation and transfection. Wild-type JAK1 was amplified using Phusion High-Fidelity DNA Polymerase (NEB) and the following primers: JAK1-SPAfo: 5 0 -ATCGTCCTCGAGATGCAGTATCTAAATATAAAA-3 0 and JAK1-SPAre: 5 0 -ATTGCTCATATGTTTTAAAAGTGCTTCAAATCC-3 0 . After restriction digest with XhoI and NdeI, the PCR product was ligated into the PMZ3F vector (kindly provided by the laboratory of Jack Greenblatt, University of Toronto) 60 . Protein expression was verified by immunoblotting using a Flag-specific antibody (Sigma). The point mutation was introduced using Quikchange mutagenesis (Agilent) according to the manufacturer's protocol and the following primers: JAK1G600Wfo: 5 0 -ACACACATCTATTCTTGGACCCTGA TGGATTA-3 0 and JAK1G600Wre: 5 0 -TAATCCATCAGGGTCCAAGAATAGAT GTGTGT-3 0 ). The intended point mutations were verified by DNA sequencing and protein expression was examined by immune blotting. Lipofectamine (Life Technologies) was used for plasmid transfection of melanoma cells. After 48 h, cells were harvested and subjected to further analyses or treated with G418 for enrichment of transfectants.
Real-time proliferation assay (xCELLigence). For background measurement 50 ml medium was added to an E-Plate 96 (Roche). Subsequently, melanoma cells were seeded in an additional volume of 100 ml medium. Cell attachment was monitored using the RTCA SP (Roche) instrument and the RTCA software Version 1.1 (Roche). After 20-24 h cells were treated with IFNg (500 U ml À 1 ) or left untreated, followed by incubation for 7 d at 37°C. All experiments were performed in duplicates. Changes in electrical impedance were expressed as a dimensionless cell index value, which derives from relative impedance changes corresponding to cellular coverage of the electrode sensors, normalized to baseline impedance values with medium only.
Expansion of autologous tumour-reactive T cells. Tumour-reactive T cells were expanded following a previously described protocol 47 . Briefly, CD8 þ T lymphocytes were isolated from cryopreserved peripheral blood mononuclear cells using anti-CD8 MicroBeads (Miltenyi Biotech). Isolated T cells (1 Â 10 6 ) were co-cultured in 24-well culture plates with 1 Â 10 5 irradiated (100 Gy or 120 Gy) autologous tumour cells per well in 2 ml of AIM-V (GIBCO/BRL) supplemented with 10% (vol/vol) human AB serum. Medium was supplemented with IL-2 (250 U ml À 1 ) on day 3. CD8 þ T cells were restimulated at weekly intervals with irradiated melanoma cells. After two rounds of restimulation, T cells were subjected to ELISpot assays.