Loss-of-function mutations in JAK1/2 can lead to acquired resistance to anti-programmed death protein 1 (PD-1) therapy. We reasoned that they may also be involved in primary resistance to anti–PD-1 therapy. JAK1/2-inactivating mutations were noted in tumor biopsies of 1 of 23 patients with melanoma and in 1 of 16 patients with mismatch repair–deficient colon cancer treated with PD-1 blockade. Both cases had a high mutational load but did not respond to anti–PD-1 therapy. Two out of 48 human melanoma cell lines had JAK1/2 mutations, which led to a lack of PD-L1 expression upon interferon gamma exposure mediated by an inability to signal through the interferon gamma receptor pathway. JAK1/2 loss-of-function alterations in The Cancer Genome Atlas confer adverse outcomes in patients. We propose that JAK1/2 loss-of-function mutations are a genetic mechanism of lack of reactive PD-L1 expression and response to interferon gamma, leading to primary resistance to PD-1 blockade therapy.
Significance: A key functional result from somatic JAK1/2 mutations in a cancer cell is the inability to respond to interferon gamma by expressing PD-L1 and many other interferon-stimulated genes. These mutations result in a genetic mechanism for the absence of reactive PD-L1 expression, and patients harboring such tumors would be unlikely to respond to PD-1 blockade therapy. Cancer Discov; 7(2); 188–201. ©2016 AACR.
See related commentary by Marabelle et al., p. 128.
This article is highlighted in the In This Issue feature, p. 115
Blocking the programmed death 1 (PD-1) negative immune receptor results in unprecedented rates of long-lasting antitumor activity in patients with metastatic cancers of different histologies, including melanoma, Hodgkin disease, Merkel cell carcinoma, and head and neck, lung, esophageal, gastric, liver, kidney, ovarian, bladder, and high mutational load cancers with defective mismatch repair, among others, in a rapidly growing list (1–8). This remarkable antitumor activity is explained by the reactivation of tumor antigen–specific T cells that were previously inactive due to the interaction between PD-1 and its ligand PD-L1 expressed by cancer cells (1, 9–12). Upon tumor antigen recognition, T cells produce interferon gamma, which through the interferon gamma receptor, the Janus kinases JAK1 and JAK2, and the signal transducers and activators of transcription (STAT) results in the expression of a large number of interferon-stimulated genes. Most of these genes lead to beneficial antitumor effects, such as increased antigen presentation through inducible proteasome subunits, transporters associated with antigen processing (TAP), and the major histocompatibility complex (MHC), as well as increased production of chemokines that attract T cells and direct tumor growth arrest and apoptosis (13). However, interferon gamma also provides the signal that allows cancer cells to inactivate antitumor T cells by the adaptive expression of PD-L1 (9), thereby specifically escaping their cytotoxic effects (12).
Acquired resistance to PD-1 blockade in patients with advanced melanoma can be associated with loss-of-function mutations with loss of heterozygosity in JAK1/2 or in beta 2-microglobulin (B2M; ref. 14). The complex genetic changes leading to acquired resistance to PD-1 blockade, wherein one JAK1/2 allele was mutated and amplified and the other was lost, suggest a strong selective pressure induced by the therapeutic immune response. Similar events leading to lack of sensitivity to interferon gamma have been reported in the cancer immune-editing process and acquired resistance to immunotherapy in mouse models (15–17) and in patients treated with the anti–CTLA-4 antibody ipilimumab who did not respond to therapy (18). Therefore, lack of interferon gamma responsiveness allows cancer cells to escape from antitumor T cells, and in the context of anti–PD-1/PD-L1 therapy, results in the loss of PD-L1 expression, the target of PD-1 blockade therapy, which would abrogate the antitumor efficacy of this approach.
In order to explore the role of JAK1 and JAK2 disruption in primary resistance to PD-1 blockade therapy, we performed a genetic analysis of tumors from patients with melanoma and colon cancer who did not respond to PD-1 blockade therapy despite having a high mutational load. We identified tumors with homozygous loss-of-function mutations in JAK1 and JAK2 and studied the functional effects of deficient interferon gamma receptor signaling that lead to a genetically mediated absence of PD-L1 expression upon interferon gamma exposure.
JAK Loss-of-Function Mutations in Primary Resistance to PD-1 Blockade in Patients with Metastatic Melanoma
Recent data indicate that tumors with a high mutational burden are more likely to have clinical responses to PD-1 blockade therapy (6, 19–21). However, in all of these series some patients failed to respond despite having a high mutational load. We performed whole-exome sequencing (WES) of 23 pretreatment biopsies from patients with advanced melanoma treated with anti–PD-1 therapy, which included 14 patients with a tumor response by immune-related RECIST (irRECIST) criteria and 9 without a response (Supplementary Table S1). Even though the mean mutational load was higher in responders than nonresponders, as reported for lung, colon, and bladder cancers (6, 19, 21), some patients with a tumor response had a low mutational load and some patients without a tumor response had a high mutational load (Fig. 1A).
We then assessed whether loss-of-function mutations in interferon receptor signaling molecules, which would prevent adaptive expression of PD-L1, might be present in tumors with a relatively high mutational load that did not respond to therapy. A melanoma biopsy from the patient with the highest mutational load among the 9 nonresponders (patient #15) had a somatic P429S missense mutation in the src-homology (SH2) domain of JAK1 (Fig. 1B). WES of an early passage cell line derived from this tumor (M431) showed an amplification of chromosome 1p, including the JAK1 locus, and a 4:1 mutant:wild-type allele ratio was observed at both the DNA and RNA level (Supplementary Fig. S1A–S1E and Supplementary Database S1). None of the tumors from the other 22 patients had homozygous loss-of-function mutations or deletions in the interferon receptor pathway. Rather, the other JAK2 mutations found in biopsies of responders had low variant allele frequency (VAF) as shown in Fig. 1B and were likely heterozygous. These mutations would not carry the same functional significance, as signaling would still occur upon interferon exposure through the wild-type JAK protein from the nonmutated allele. Two nonresponders had IFNGR mutations, also of low allele frequency and therefore uncertain significance. We also analyzed potential mutations in genes involved in the antigen-presenting machinery and did not find any loss-of-function mutations that were homozygous (Supplementary Fig. S2).
As expected, tumors from patients who responded had a higher density of CD8 cells and PD-L1 in the center and invasive tumor margin (Fig. 1C and D). In contrast, the baseline biopsy from patient #15 with a high mutational load but with the JAK1P429S missense mutation had undetectable CD8 infiltrates, PD-1 and PD-L1 expression (Supplementary Fig. S3). The amplification of PD-L1, PD-L2, and JAK2 (PDJ amplicon), which has been associated with a high response rate in Hodgkin disease (4), was noted only in patient #16, who did not respond to PD-1 blockade therapy despite having the second highest mutational load and a high level of PD-L1 expression (Fig. 1B, D and E).
Functional Analyses of the Role of JAK Loss-of-Function Mutations in Regulating PD-L1 Expression
We next sought to characterize the interferon response of M431, the melanoma cell line established from a biopsy of patient #15 with high mutational load and no response to therapy. First, we optimized flow cytometry conditions in selected human melanoma cell lines (Supplementary Figs. S4A–S4F, S5A–S5D, S6A–S6H, and S7A–S7C). PD-L1 expression increased less than 1.5-fold interferon gamma exposure in M431 (Fig. 2A), versus 5.1-fold in M438, a cell line established from patient #8 used as a positive control in this same series. Phosphorylated STAT1 (pSTAT1) was induced at 30 minutes in M431, but the signal dissipated at 18 hours, faster than in cell lines with more durable responses to interferon gamma leading to PD-L1 upregulation (Fig. 2B, C compared with Supplementary Fig. S8A–S8C). These data are consistent with the 4:1 JAK1 mutant:wild-type allele frequency in the M431 cell line (Supplementary Fig. S1A–S1E).
We then screened a panel of 48 human melanoma cell lines for absolute absence of PD-L1 induction by either type I (alpha and beta) or type II (gamma) interferons. Among the three interferons, interferon gamma most potently induced PD-L1 expression (Fig. 2D; Supplementary Fig. S9A and S9B for type I interferons). Two cell lines had JAK1/2 homozygous loss-of-function mutations and did not respond to interferon gamma with upregulation of surface PD-L1 expression. M368 had a mutation in JAK2 (20 out of 22 reads, VAF = 0.91) that is predicted to disrupt and shift the D313 splice-site acceptor in exon 8 by one nucleotide, changing the reading frame, and had loss of the wild-type allele (Fig. 3A; Supplementary Fig. S10A and S10B). M395 had an inactivating JAK1D775N kinase domain mutation in exon 17 and loss of the other allele (140 out of 143 reads, variant allele frequency 0.98; Fig. 3B).
We then analyzed signaling in response to interferon alpha, beta, and gamma in these two cell lines. M368, which harbored the JAK2 loss-of-function mutation, maintained signaling in response to interferon alpha and beta, but did not respond to interferon gamma (Fig. 3C), which resulted in the ability of M368 to upregulate PD-L1 when exposed to interferon alpha and beta, but not to interferon gamma (Fig. 3C; Supplementary Fig. S9A and S9B). M395, which harbored the JAK1 loss-of-function mutation, did not respond to downstream signaling to interferon alpha, beta, or gamma (Fig. 3D), and equally did not upregulate PD-L1 in response to any of these cytokines (Fig. 3D; Supplementary Fig. S9A and S9B). We were able to retrieve the tumor from which the cell line M395 had been established, and this tumor exhibited an absence of CD8 infiltration similar to the finding in patient #15 with a JAK1 loss-of-function mutation who did not respond to anti–PD-1 therapy (Supplementary Fig. S11). Taken together, these data are consistent with the knowledge that JAK1 (disabled in M395) is required to propagate signaling downstream of the interferon alpha/beta and gamma receptors, whereas JAK2 (disabled in M368) is required for signaling downstream only from the interferon gamma receptor (22–24).
To assess a causal relationship between loss of adaptive PD-L1 expression and loss-of-function JAK mutations, we transduced the M395 and M431 cell lines with a lentivirus vector expressing JAK1 wild-type (Supplementary Fig. S12A–S12C). Reintroduction of wild-type JAK1 rescued PD-L1 expression in M395 cells, which exhibited a 4-fold increase in PD-L1 surface expression after interferon gamma exposure (Fig. 3E). For M431, the magnitude of change in PD-L1 expression after 18-hour interferon gamma exposure for M431 was modest after reintroducing the JAK1 wild-type protein (approximately 2-fold, compared with a 1.5-fold in the untransduced cell line; Fig. 3F). However, the difference between untransduced and JAK1 wild-type transduced M431 was more distinct when observed over a longer time course (Fig. 3G and 3H).
JAK Loss-of-Function Mutations in Primary Resistance to PD-1 Blockade in Patients with Metastatic Colon Carcinoma
To determine whether JAK1/2 loss-of-function mutations are present and relate to response to PD-1 blockade therapy in another cancer histology, we analyzed WES data from 16 biopsies of patients with colon cancer, many with a high mutational load resultant from mismatch-repair deficiency (6). One of the biopsies of a rare patient with high mutational load with neither an objective response nor disease control with anti–PD-1 had a homozygous JAK1W690* nonsense loss-of-function mutation, expected to truncate the protein within the first kinase domain, and an accompanying loss of heterozygosity at the JAK1 locus (Fig. 4A–D). No mutations in antigen presentation machinery were detected in this sample (Supplementary Fig. S13). Although we observed other interferon pathway and antigen presentation mutations in the high mutational load patients with a response to therapy in this cohort, they appeared to be heterozygous by allele frequency (adjusted VAF < 0.6) after adjustment for stromal content. Most were splice-site mutations or frameshift insertions/deletions unlikely to create a dominant-negative effect. Several samples bore two mutations in JAK1/2 or B2M, but either retained at least one wild-type copy (subjects #4 and #5), were too far apart to determine cis versus trans status (subject #6), or were of uncertain significance (subject #1, both near c-terminus).
Frequency of JAK Loss-of-Function Mutations in Cell Lines of Multiple Histologies
We then analyzed data from the Cancer Cell Line Encyclopedia (CCLE) from cBioPortal to determine the frequency of homozygous putative loss-of-function mutations in JAK1/2 in 905 cancer cell lines (25). For this analysis, we considered a homozygous mutation when the VAF was 0.8 or greater, as previously described (26). Approximately 0.7% of cell lines have loss-of-function mutations that may predict lack of response to interferons (Fig. 5A and 5B). The highest frequency of mutations was in endometrial cancers, as described previously (26). None of these cell lines had POLE or POLD1 mutations, but microsatellite instability and DNA-damage gene mutations were present in the JAK1/2 mutant cell lines (Supplementary Fig. S14). The frequency of JAK1/2 mutations across all cancers suggests that there is a fitness gain with loss of interferon responsiveness.
JAK1/2 Loss-of-Function Alterations in The Cancer Genome Atlas
Analysis of WES, RNA sequencing (RNA-seq), and reverse-phase protein array (RPPA) data from tissue specimens from 472 patients in The Cancer Genome Atlas (TCGA) Skin Cutaneous Melanoma dataset revealed that 6% (28 of 472) and 11% (50 of 472) harbored alterations in JAK1 and JAK2, respectively. These include loss-of-function alterations in either JAK1 or JAK2 that would putatively diminish JAK1 or JAK2 signaling (homodeletions, truncating mutations, or gene or protein downregulation).
There was no survival difference in patients in the TCGA Skin Cutaneous Melanoma dataset harboring any JAK1 or JAK2 alteration (Fig. 6A). However, when considering only loss-of-function JAK1 or JAK2 alterations (homodeletions, truncating mutations, or gene or protein downregulation), patients with tumors that had JAK1 or JAK2 alterations had significantly decreased overall survival (P = 0.009, log-rank test). When considered separately, the 8 patients with truncating mutations in JAK1 or JAK2 and the 18 patients with JAK1 or JAK2 gene or protein downregulation also had significantly decreased overall survival (P = 0.016 and P < 0.001, respectively).
To assess the relevance of these findings in a broader set of malignancies, we examined the frequency of JAK1 and JAK2 alterations and their association with clinical outcome in TCGA datasets for four common malignancies (breast invasive carcinoma, prostate adenocarcinoma, lung adenocarcinoma, and colorectal adenocarcinoma). Similar to findings in melanoma, alterations in JAK1 were found in 6%, 8%, 10%, and 10% of patients with breast invasive carcinoma, prostate adenocarcinoma, lung adenocarcinoma, and colorectal adenocarcinoma, respectively. Likewise, alterations in JAK2 were found in 12%, 7%, 12%, and 5% of these respective malignancies.
Consistent with our findings in melanoma, JAK1 or JAK2 alterations as a whole were not associated with a difference in survival in any of the four additional TCGA datasets. However, for patients with breast invasive carcinoma harboring truncating mutations, there was an association with decreased survival (P = 0.006, log-rank test; Fig. 6B). Likewise, patients with prostate adenocarcinoma harboring truncating mutations had worse overall survival (P = 0.009, log-rank test; Fig. 6C), with a similar trend noted in patients harboring any loss-of-function JAK1 or JAK2 alterations (P = 0.083, Fig. 6C). We did not observe differences in survival in patients with lung adenocarcinoma or colorectal adenocarcinoma harboring JAK1 or JAK2 loss-of-function alterations, when considered either separately or as a whole (Supplementary Fig. S15A and S15B).
For this work, we hypothesized that if cancer cells evolved to disable inducible PD-L1 expression upon interferon exposure due to selective immune pressure as demonstrated in preclinical models of cancer immune-editing (15, 16), then it would be superfluous to attempt to treat these cases with anti–PD-1/PD-L1 antibody therapy (Supplementary Fig. S16A and S16B). The premise of therapy with anti–PD-1– or anti–PD-L1–blocking antibodies is that T cells with specificity for cancer antigens recognize their target on cancer cells and produce interferon gamma. The cancer cell then finds a way to specifically protect itself from the T-cell attack by reactively expressing PD-L1 upon interferon gamma signaling. This reactive process is termed adaptive immune resistance, and it requires signaling through the interferon gamma receptor (12). By understanding this process, it is then logical to anticipate that a genetically acquired insensitivity to interferon gamma signaling could represent an immune resistance mechanism; these tumors would be expected to be incapable of upregulating either antigen-presenting machinery or PD-L1 even in the presence of a robust preexisting repertoire of tumor-specific T cells. With a genetic mechanism of lack of interferon gamma signaling, a T-cell response with interferon gamma production would not lead to reactive PD-L1 expression and therefore these would be cases that would be considered constitutively PD-L1 negative.
JAK kinases mediate signaling from many cytokine receptors, but the commonality between JAK1 and JAK2 homozygous loss-of-function mutations is that they are both required for signaling upon exposure to interferon gamma (27). Interferon gamma is a major cytokine produced by T cells upon recognizing their cognate antigen, and it has multiple effects on target cells. In the setting of acquired resistance to PD-1 blockade therapy in patients who progressed while on continuous anti–PD-1 therapy, the tumor's insensitivity to interferon gamma provides a selective advantage for the relapsed cancer to grow, as it no longer is sensitive to the antiproliferative effects of interferon gamma (14). In that setting, T cells continued to recognize cancer cells with JAK1 or JAK2 mutations despite the known role of interferon gamma signaling in upregulating a series of genes involved in the antigen-presenting machinery. However, as the baseline expression of MHC class I, proteasome subunits and TAP transporters is unchanged, tumor antigen presentation to T cells was not impaired (14).
In primary resistance to checkpoint blockade therapy with the anti–CTLA-4 antibody ipilimumab, there is a higher frequency of mutations in the several molecules involved in the interferon signaling pathway (18). It is hypothesized that cancer cells lacking interferon receptor signaling would have a selective advantage because they evade T cells activated by CTLA-4 blockade, in particular through decreased antigen presentation and resistance to the antiproliferative effects of interferons. The same processes may have an important role in the lack of response to anti–PD-1 therapy in the cancers with JAK1/2 loss-of-function mutations in our series, as antitumor T cells would be anticipated to have lower ability to recognize and kill cancer cells. Loss-of-function mutations in JAK1/2 would likewise prevent the antitumor activity of any immunotherapy that results in the activation of T cells to attack cancer cells. But in the setting of anti–PD-1/PD-L1 therapy, it has the additional important effect of preventing PD-L1 expression upon interferon gamma exposure, thereby making it futile to pharmacologically inhibit the PD-L1/PD-1 interaction.
As the interferon gamma receptor pathway downstream of JAK1/2 controls the expression of chemokines with a potent chemoattractant effect on T cells, such as CXCL9, CXCL10, and CXCL11 (28), it is possible that an important effect of JAK1/2 loss may result in a lack of T-cell infiltrates. Indeed, both the patient in the melanoma series with a JAK1 loss of function and the biopsy from which we had derived a melanoma cell line with a JAK1 mutation were completely devoid of T-cell infiltrates. As preexisting T cells in the tumor are a requisite for response to anti–PD-1 therapy (11), a JAK1/2 mutation may result in lack of response not only because PD-L1 cannot be reactively expressed but also because the cancer fails to attract T cells due to lack of chemokine production.
Beyond a genetic mutation that prevented expression of JAK1/2, it is also possible that epigenetic silencing of JAKs could result in lack of response to interferon gamma, as previously reported for the LNCaP cell line (29). In this case, loss of JAK1/2 expression could then be corrected with exposure to a demethylating agent. This evidence suggests that the frequency of loss of function in JAK1/2 may be higher than can be estimated by exome-sequencing analyses, as it could occur epigenetically, and in these cases it would provide an option for pharmacologic intervention.
In conclusion, we propose that JAK1/2 mutations that lead to loss of interferon gamma signaling and prevent adaptive PD-L1 expression upon interferon gamma exposure represent an immunoediting process that defines patients with cancer who would not be good candidates for PD-1 blockade therapy. This mechanism would add to other multiple explanations that may lead to primary resistance to PD-1 blockade therapy, including a tumor that lacks antigens that can be a target for a T-cell response, the presence of immune suppressive factors in the tumor microenvironment that exclude T cells in tumors or that lead to alteration of T-cell function, presence of immune suppressive cells such as T regulatory or myeloid-derived suppressor cells, or cancers that have specific genetic signaling or transcriptomes that are not permissive to T-cell infiltrates (20, 30, 31). The recognition that JAK1/2 loss-of-function mutations would lead to lack of response to PD-1 blockade therapy could be incorporated in oncogenic sequencing panels used to select patients for precision cancer treatments.
Tumor biopsies were obtained from a subset of patients enrolled in a phase I expansion clinical trial with pembrolizumab after signing a written informed consent (32). Patients were selected for this analysis by having adequate tumor biopsy samples and clinical follow-up. Baseline biopsies of metastatic tumors were obtained within 30 days of starting on treatment, except for one in a patient with an eventual complete response (Fig. 3B, subject #4) collected after 84 days on treatment. Samples were immediately fixed in formalin followed by paraffin embedding, and when there was an additional sterile piece of the tumor, processed for snap-freezing in liquid nitrogen and to establish a cell line as previously described (33–35). Tumor biopsy and peripheral blood cell collection and analyses were approved by UCLA Institutional Review Boards 11-001918 and 11-003066.
Treatment and Response Assessment
Patients received single-agent pembrolizumab intravenously in one of three dosing regimens: 2 mg/kg every 3 weeks (2Q3W), 10 mg/kg every 3 weeks (10Q3W), or 10 mg/kg every 2 weeks (10Q2W; ref. 32). Tumor responses to pembrolizumab were evaluated at 12 weeks after the first infusion (confirmed at 16 weeks), and every 12 weeks thereafter. The RECIST version 1.1 was used to define objective clinical responses. The protocol was allowed to proceed beyond initial progression at the restaging scans at 12 weeks and have repeated imaging scans 4 weeks later following the immune-related response criteria (irRC; ref. 36).
For CD8 T-cell density, 5 of the 11 cases were reanalyzed blindly from IHC samples already used in our prior work (11), and the other 6 cases were newly stained cases also analyzed blindly. Slides were stained with hematoxylin and eosin, S100, CD8, CD68, PD-1, and PD-L1 at the UCLA Anatomic Pathology IHC Laboratory. Immunostaining was performed on Leica Bond III autostainers using Leica Bond ancillary reagents and the REFINE polymer DAB detection system as previously described (11). Cell density (cells/mm2) in the invasive margin or intratumoral area was calculated using the Indica Labs Halo platform as previously described (11).
Cell Lines, Cell Culture, and Conditions
Patient-derived melanoma cell lines were generated as reported previously and characterized for their oncogenic mutational status (33–35). Each melanoma cell line was thawed and maintained in RPMI-1640 medium supplemented with 10% FBS, 100 units/mL penicillin, and 100 μg/mL streptomycin at 37°C in a humidified atmosphere of 5% CO2. Cells were subject to experimental conditions after reaching two passages from thawing. Cell lines were periodically authenticated using GenePrint 10 System (Promega) and were matched with the earliest passage cell lines. Selected melanoma cell lines were subjected to Mycoplasma tests periodically (every 2–3 months) with the MycoAlert Mycoplasma Detection Kit (Lonza).
Surface Flow Cytometry Analysis for PD-L1 and MHC Class I
Melanoma cells were seeded into 6-well plates on day 1, ranging from 420,000 to 485,000 depending on their doubling time, targeting 70% to 80% of confluence at the time of trypsinization after 18 hours of exposure to interferons. For 48-hour exposure, 225,000 to 280,000 cells were seeded, and 185,000 to 200,000 cells were seeded for 72-hour exposure. After trypsinization, cells were incubated at 37°C for 2 hours with media containing different concentrations of interferons. Concentrations of each interferon were determined after optimization process (dose–response curves were generated with representative cell lines as shown in Supplementary Fig. S5B–S5D). After 2 hours of incubation, the media were removed by centrifugation and cells were resuspended with 100% FBS and stained with APC anti–PD-L1 antibody on ice for 20 minutes. The staining was halted by washing with 3 mL of PBS, which was removed by centrifugation at 500 × g for 4 minutes. The cells were resuspended with 300 μL of PBS, and 7-AAD for dead cell discrimination was added to samples prior to data acquisition by LSRII. The data were analyzed by FlowJo software (Version 10.0.8r1, Tree Star Inc.). Experiments were performed at least twice for each cell line; some cell lines with high assay variability were analyzed three times.
Phosphoflow Signaling Analyses
Cells were seeded into two 6-well plates for each cell line for single phospho-proteomics study. After 30-minute or 18-hour exposure to interferon alpha, beta, or gamma, cells were trypsinized and resuspended with 1 mL of PBS per 1 to 3 million cells and stained with live/dead agent at room temperature in the dark for 30 minutes. Cells were then fixed with paraformaldehyde at room temperature for 10 minutes in the dark, permeabilized by methanol, and stained with pSTAT1. Cells were incubated at room temperature in the dark for 30 minutes, washed with phospho-flow cytometry buffer, and resuspended with 300 to 500 μL of the same buffer and analyzed with an LSRII. The flow cytometry standard (FCS) files obtained by LSRII were analyzed using the online flow cytometry program (Cytobank; ref. 37). The raw FCS files were deconvoluted into four different conditions, three of which were exposed to interferon alpha, beta, and gamma and compared with an untreated condition at each time point. Data represented as Arcsinh ratio, which is one of transformed ratio of cytometry data (inverse hyperbolic sine) analyses; each data point was compared with its control [Value = arcsinh((x − control)/scale_argument)].
Western Blot Analyses
Selected melanoma cells were maintained in 10-cm cell culture dishes and exposed to interferon alpha, beta, or gamma (same concentrations as above) for 30 minutes or 18 hours. Western blotting was performed as described previously (38). Primary antibodies included pJAK1 (Tyr1022/1023), pJAK2 (Tyr221), pSTAT1 (Tyr701), pSTAT3 (Tyr705), pSTAT5 (Tyr695), and their total proteins; PIAS1, IRF1, SOCS1, and GAPDH (all from Cell Signaling Technology). Antibodies were diluted to 1:1,000 ratio for each blot. Immunoreactivity was revealed with an ECL-Plus Kit (Amersham Biosciences Co.), using the ChemiDoc MP system (Bio-rad Laboratories).
Lentiviral Vector Production and Gene Transfer
Lentivirus production was performed by transient cotransfection of 293T cells (ATCC). The lentiviral vectors pLenti-C-mGFP and pLenti-C-JAK1-mGFP were purchased from Origen (cat# RC213878L2). In brief, T175 tissue culture flasks coated with poly-l-lysine (Sigma Aldrich) containing 6 × 106 293T cells were used for each transfection. The constructs required for the packaging of third-generation self-inactivating lentiviral vectors pLenti-C-mGFP and pLenti-C-JAK1-mGFP (60 μg), pMDLGg/p (39 μg), pRSV–REV (15 μg), and pMD.G (21 μg) were dissolved in water in a total volume of 2.7 mL. A total of 300 μL of 2.5 mol/L CaCl2 (Sigma Aldrich) was added to the DNA mixture. A total of 2.8 mL of the DNA/CaCl2 mix was added dropwise to 2.8 mL of 2× HBS buffer, pH 7.12 (280 nmol/L NaCl, 1.5 mmol/L Na2HPO4, 100 mmol/L HEPES). The DNA/CaPO4 suspension was added to each flask and incubated in a 5% CO2 incubator at 37°C overnight. The next morning, the medium was discarded, the cells were washed, and 15 mL DMEM with 10% FBS containing 20 mmol/L HEPES (Invitrogen) and 10 mmol/L sodium butyrate (Sigma Aldrich) was added, and the flask was incubated at 37°C for 8 to 12 hours. After that, the cells were washed once, and 10 mL fresh DMEM medium with 20 mmol/L HEPES was added onto the 293T cells, which were further incubated in a 5% CO2 incubator at 37°C for 12 hours. The medium supernatants were then collected, filtered through 0.2 μmol/L filters, and cryopreserved at minus 80°C. Virus supernatant was added at different concentrations into 6-well plates containing 5 × 105 cells per well. Protamine sulphate (Sigma Aldrich) was added at a final concentration of 5 μg/mL, and the transduction plates were incubated at 37°C in 5% CO2 overnight.
Exon capture and library preparation were performed at the UCLA Clinical Microarray Core using the Roche Nimblegen SeqCap EZ Human Exome Library v3.0 targeting 65 Mb of genome. Paired-end sequencing (2 × 100 bp) was carried out on the HiSeq 2000 platform (Illumina) and sequences were aligned to the UCSC hg19 reference using BWA-mem (v0.7.9). Sequencing for tumors was performed to a target depth of 150× (actual min. 91×, max. 162×, mean 130×). Preprocessing followed the Genome Analysis Toolkit (GATK) Best Practices Workflow v3, including duplicate removal (PicardTools), indel realignment, and base quality score recalibration.
Somatic mutations were called by comparison to sequencing of matched normals for the PD1-treated whole-tumor patient samples. Methods were modified from ref. 39; specifically, the substitution the GATK-HaplotypeCaller (HC, v3.3) for the UnifiedGenotyper. gVCF outputs from GATK-HC for all 23 tumor/normal exomes, and cell lines M395 and M431, were jointly genotyped and submitted for variant quality score recalibration. Somatic variants were determined using one-sided Fisher exact test (P value cutoff ≤ 0.01) between tumor/normal pairs with depth >10 reads. Only high-confidence mutations were retained for final consideration, defined as those identified by at least two out of three programs [MuTect (v1.1.7; ref. 40), Varscan2 Somatic (v2.3.6; ref. 41), and the GATK-HC] for single nucleotide variants, and those called by both Varscan2 and the GATK-HC for insertions/deletions. Variants were annotated by Oncotator (42), with nonsynonymous mutations for mutational load being those classified as nonsense, missense, splice_site, or nonstop mutations, as well as frame_shift, in_frame_, or start_codon altering insertions/deletions. Adjusted variant allele frequency was calculated according to the following equation:
This is an algebraic rearrangement of the equation used in the clonal architecture analysis from McGranaham and colleagues (43) to calculate the fraction of mutated chromosomal copies while adjusting for the diluting contribution of stromal chromosomal copies. Local tumor copy number (CNt,) tumor fraction (purity, or p) and stromal fraction (1 – p) were produced by Sequenza (44), which uses both depth ratio and SNP minor B-allele frequencies to estimate tumor ploidy and percent tumor content, and perform allele-specific copy-number variation analysis.
PDJ amplification was considered tumor/normal depth ratio ≥ 2 standard deviations above length-weighted genome average. BAM files for the 16 colorectal cases were previously mapped to hg18, and sequencing and analysis were performed at Personal Genome Diagnostics. After preprocessing and somatic variant calling, positions were remapped to hg19 using the Ensembl Assembly Converter before annotation.
M431 and M395 were compared with matched normal samples, the other 47 cell lines lacked a paired normal sample. For detection of potential JAK1 or JAK2 mutations, variants were detected using the Haplotype Caller, noted for membership in dbSNP 146 and allele frequency from the 1000 Genomes project, and confirmed by visual inspection with the Integrated Genomics Viewer.
Forward 5′-AACCTTCTCACCAGGATGCG-3′ and reverse 5′-CTCAGCACGTACATCCCCTC-3′ primers were designed to perform RT-PCR (700 base pair of target PCR product to cover the P429 region of the JAK1 protein) on the M431 cell line. Total RNA was extracted by the mirVana miRNA Isolation Kit, with phenols as per the manufacturer's protocol (Thermo Fischer Scientific). RT-PCR was performed by utilizing ThermoScript RT-PCR Systems (Thermo Fisher Scientific, cat# 11146-057). PCR product was subject to Sanger sequencing at the UCLA core facility.
To determine the relevance of JAK1 and JAK2 alterations in a broader set of patients, we queried the TCGA skin cutaneous melanoma provisional dataset for the frequency of genetic and expression alterations in JAK1 and JAK2. We then extended our query to the breast invasive carcinoma, prostate adenocarcinoma, lung adenocarcinoma, and colorectal adenocarcinoma provisional TCGA datasets. We then examined the association of various JAK1 and JAK2 alterations with overall survival for each dataset. The results are based upon data generated by the TCGA Research Network and made available through the NCI Genomic Data Commons and cBioPortal (45, 46).
The mutation annotation format (MAF) files containing JAK1 and JAK2 mutations in the TCGA datasets were obtained from the Genomic Data Commons. In addition, mutations, putative copy-number alterations, mRNA expression, protein expression, and survival data were obtained using the cBioPortal resource. The putative copy-number alterations (homodeletion events, in particular) available in cBioPortal were obtained from the TCGA datasets using Genomic Identification of Significant Targets in Cancer (GISTIC; ref. 47). The mRNA expression data available in cBioPortal were obtained from the TCGA datasets using RNA-seq (RNA Seq V2 RSEM). Upregulation and downregulation of JAK1 and JAK2 mRNA expression were determined using an mRNA z-score cutoff of 2.0. Protein expression data available in cBioPortal were obtained from the TCGA dataset using RPPA, with a z-score threshold of 2.0.
Mutation data between the MAF files and data from cBioPortal were combined. Genetic and expression alterations were characterized in one of six categories: amplifications, homodeletions, single-nucleotide polymorphisms, truncating mutations (stop codons and frameshift insertions and deletions), mRNA or protein downregulation, and mRNA or protein upregulation. The frequency of JAK1 and JAK2 alterations was determined using combined data from the *.MAF file and cBioPortal. Kaplan–Meier survival curves were generated in R, using the “survminer” package and the “ggsurvplot” function. Overall survival was determined using log-rank analysis.
Statistical comparisons were performed by the unpaired two-tailed Student t test (GraphPad Prism, version 6.0 for Windows). Mutational load was compared by unpaired two-sided Mann–Whitney test. R programming was utilized to generate arrow graphs of PD-L1/MHC class I expression upon interferon exposures and the CCLE JAK1/2 mutation frequency graph.
Disclosure of Potential Conflicts of Interest
B. Chmielowski reports receiving speakers bureau honoraria from Genentech and Janssen and is a consultant/advisory board member for Merck, Genentech, Eisai, Immunocore, BMS, and Amgen. D.T. Le reports receiving commercial research grants from Merck and BMS, and is a consultant/advisory board member for Merck. D.M. Pardoll reports receiving a commercial research grant from BMS. L.A. Diaz has ownership interest (including patents) in Personal Genome Diagnostics and PapGene, and is a consultant/advisory board member for Merck and Cell Design labs. No potential conflicts of interest were disclosed by the other authors.
Conception and design: D.S. Shin, A. Garcia-Diaz, R.S. Lo, B. Comin-Anduix, A. Ribas
Development of methodology: D.S. Shin, H. Escuin-Ordinas, A. Garcia-Diaz, S. Sandoval, D.Y. Torrejon, G. Abril-Rodriguez, L.A. Diaz, Jr, P.C. Tumeh, R.S. Lo, A. Ribas
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D.S. Shin, J.M. Zaretsky, S. Hu-Lieskovan, D.Y. Torrejon, B. Chmielowski, G. Cherry, E. Seja, I.P. Shintaku, D.T. Le, D.M. Pardoll, L.A. Diaz, Jr, P.C. Tumeh, A. Ribas
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D.S. Shin, J.M. Zaretsky, S. Hu-Lieskovan, A. Kalbasi, C.S. Grasso, W. Hugo, N. Palaskas, B. Chmielowski, D.M. Pardoll, L.A. Diaz, Jr, P.C. Tumeh, T.G. Graeber, R.S. Lo, B. Comin-Anduix, A. Ribas
Writing, review, and/or revision of the manuscript: D.S. Shin, J.M. Zaretsky, S. Hu-Lieskovan, A. Kalbasi, C.S. Grasso, W. Hugo, D.Y. Torrejon, N. Palaskas, G. Parisi, B. Chmielowski, D.T. Le, D.M. Pardoll, P.C. Tumeh, T.G. Graeber, R.S. Lo, B. Comin-Anduix, A. Ribas
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Escuin-Ordinas, S. Sandoval, B. Berent-Maoz, P.C. Tumeh, A. Ribas
Study supervision: B. Comin-Anduix, A. Ribas
Other (assisted with Western blot work): A. Azhdam
This study was funded in part by the NIH grants R35 CA197633 and P01 CA168585, the Parker Institute for Cancer Immunotherapy (PICI), the Dr. Robert Vigen Memorial Fund, the Ruby Family Fund, and the Garcia-Corsini Family Fund (to A. Ribas); and P01 CA168585, the Ressler Family Fund, the Samuels Family Fund, and the Grimaldi Family Fund (to A. Ribas and R.S. Lo). D.S. Shin was supported by the Oncology (5T32CA009297-30), Dermatology (5T32AR058921-05), and Tumor Immunology (5T32CA009120-39 and 4T32CA009120-40) training grants, a 2016 Conquer Cancer Foundation ASCO Young Investigator Award, and a Tower Cancer Research Foundation Grant. A. Ribas and D.M. Pardoll were supported by a Stand Up To Cancer – Cancer Research Institute Cancer Immunology Dream Team Translational Research Grant (SU2C-AACR-DT1012). Stand Up To Cancer is a program of the Entertainment Industry Foundation administered by the American Association for Cancer Research. J.M. Zaretsky is part of the UCLA Medical Scientist Training Program supported by NIH training grant GM08042. S. Hu-Lieskovan was supported by a Conquer Cancer Foundation ASCO Young Investigator Award, a Conquer Cancer Foundation ASCO Career Development Award, and a Tower Cancer Foundation Research Grant. G. Parisi was supported by the V Foundation-Gil Nickel Family Endowed Fellowship in Melanoma Research and also in part by the Division of Medical Oncology and Immunotherapy (University Hospital of Siena). T.G. Graeber was supported by an American Cancer Society Research Scholar Award (RSG-12-257-01-TBE), a Melanoma Research Alliance Established Investigator Award (20120279), and the National Center for Advancing Translational Sciences UCLA CTSI Grant UL1TR000124. R.S. Lo was supported by the Steven C. Gordon Family Foundation and the Wade F.B. Thompson/Cancer Research Institute CLIP Grant. L.A. Diaz and D.T. Le were funded from the Swim Across America Laboratory and the Commonwealth Fund at Johns Hopkins. W. Hugo was supported by the 2016 Milstein Research Scholar Award from the American Skin Association and by the 2016 AACR-Amgen, Inc., Fellowship in Clinical/Translational Cancer Research (16-40-11-HUGO).
We acknowledge the UCLA Translational Pathology Core Laboratory (TPCL); Rongqing Guo, Jia Pang, and Wang Li from UCLA for blood and biopsy processing; and Matt Klinman, Thinle Chodon, and Charles Ng for establishing some of the cell lines.
Note: Supplementary data for this article are available at Cancer Discovery Online (http://cancerdiscovery.aacrjournals.org/).
- Received November 1, 2016.
- Revision received November 28, 2016.
- Accepted November 28, 2016.
- ©2016 American Association for Cancer Research.