Evolution of tumor cell phenotypes promotes heterogeneity and therapy resistance. Here we found that induction of CD73, the enzyme that generates immunosuppressive adenosine, is linked to melanoma phenotype switching. Activating MAPK mutations and growth factors drove CD73 expression, which marked both nascent and full activation of a mesenchymal-like melanoma cell state program. Proinflammatory cytokines like TNFα cooperated with MAPK signaling through the c-Jun/AP-1 transcription factor complex to activate CD73 transcription by binding to an intronic enhancer. In a mouse model of T-cell immunotherapy, CD73 was induced in relapse melanomas, which acquired a mesenchymal-like phenotype. We also detected CD73 upregulation in melanoma patients progressing under adoptive T-cell transfer or immune checkpoint blockade, arguing for an adaptive resistance mechanism. Our work substantiates CD73 as a target to combine with current immunotherapies, but its dynamic regulation suggests limited value of CD73 pretreatment expression as a biomarker to stratify melanoma patients. Cancer Res; 77(17); 4697–709. ©2017 AACR.
The treatment outcome of patients with metastatic melanoma has significantly improved in recent years. Small-molecule MAPK inhibitors (MAPKi) prolong survival by targeting oncogenic signaling of the mutant BRAF kinase (BRAFV600E), which is found in about half of melanomas (1, 2). Antibodies against negative immune checkpoint molecules PD-1/PD-L1 (programmed death 1/programmed death-ligand 1) and CTLA-4 (cytotoxic T-lymphocyte-associated antigen 4) enforce antitumor immune responses and achieve long-term remissions (3). Adoptive T-cell transfer (ACT) therapy is another type of immunotherapy that effectively controls tumor growth by targeting specific tumor antigens (4–6). Nevertheless, many melanoma patients do not respond or acquire resistance to these therapies. Apart from genetically hardwired mechanisms (7–9), there is growing evidence that nongenomic changes drive a coordinated coevolution of the tumor and immune cell compartment, which contributes to treatment failure (10, 11). Therefore, understanding the reciprocal crosstalk between tumor and immune cells in the microenvironment is instrumental to optimize current treatments.
Cellular plasticity plays a central role in nongenomic resistance mechanisms and phenotypic heterogeneity in melanoma (12). This has been elaborated in the context of acquired resistance to MAPKi (13–15). Dysregulated expression of MITF (microphthalmia-associated transcription factor), the master transcription factor of the melanocyte lineage, has been found to alter drug responsiveness. In particular, loss of MITF was shown to confer MAPKi resistance by activation of compensatory survival signaling through different mechanisms (13, 14). Low expression of MITF by melanoma cells has been linked to the so-called “invasive” phenotype switch, which is characterized by an epithelial-to-mesenchymal (EMT)-like transition with higher migratory capacity (16, 17). In contrast, the so-called “proliferative” phenotype is strongly driven by MITF, which controls melanocyte lineage gene expression. Recently, single-cell–based approaches provided evidence that melanoma cells exist in a phenotypic equilibrium, but the impact of driver mutations, and epigenetic and microenvironmental context remains to be determined (18, 19).
Comparatively little is known about dynamic phenotype changes of melanoma cells during immunotherapy. Using melanoma mouse models, we previously demonstrated that T-cell therapy–associated inflammation induces melanoma cell dedifferentiation, leading to therapy escape (20). Prolonged exposure of melanoma cells with proinflammatory cytokines like TNFα suppresses MITF and causes “invasive” phenotype switching. Low MITF expression causes a reciprocal activation of inflammatory pathways and instigates a feed-forward loop of chemokine expression promoting predominantly myeloid immune cell recruitment (21). Thus, melanoma phenotype switching rewires the tumor microenvironment with direct implications for cancer immunotherapy (12, 22). However, melanoma cell states other than the “invasive” and “proliferative” phenotypes remain poorly defined as well as the molecular mechanisms that govern transition.
Here we used a bioinformatic discovery approach to identify early molecular events in inflammation-induced melanoma cell plasticity. We found that expression of the immunosuppressive 5′ectonucleotidase CD73 marked both nascent and full activation of the EMT-like “invasive” melanoma cell state coordinated by the c-Jun/AP-1 transcription factor complex. We found upregulation of CD73 during immunotherapy in mouse and human melanomas. As CD73 generates immunosuppressive adenosine in the tumor microenvironment, our findings link melanoma phenotype switching to acquisition of immunosuppressive properties during immunotherapy. Our work supports the rationale of ongoing clinical trials (NCT02503774) that evaluate immune checkpoint inhibitors in combination with CD73 blockade (23).
Materials and Methods
Melanoma cells were cultured in a humidified incubator (5% CO2, 37°C) in RPMI1640 medium supplemented with 100 U/mL penicillin, 100 μg/mL streptomycin, 2 mmol/L l-glutamine, and 10 % FBS. All MaMel cell lines were established by D. Schadendorf (Essen, Germany) and provided to us between 2012 and 2013 (21). HCmel3 and HCmel3R cells were established between 2009 and 2012 by us from primary melanomas in HgfxCdk4R24C mice (20). MZ7 and SK-MEL28 cells were provided by T. Wölfel (Mainz, Germany) in 2010 (20). Oncogenic driver and other identifying mutations of the melanoma cell lines were revalidated by Sanger sequencing or NGS for authentication in our laboratory between 2009 and 2013. Virus was produced in HEK 293T cells (obtained from ATCC in 2009) cultured in complete DMEM medium. HEK 293T cells were authenticated by morphology and capability of virus production. All cell cultures were renewed at least every three months by rethawing of initial cryo stocks. All used cell lines were negative for mycoplasma and tested on a monthly basis by PCR. Reagents used were: 1,000 U/mL TNFα, 50 ng/mL HGF (Peprotech); 50 nmol/L trametinib (MEKi), 100 nmol/L SCH-772984 (ERKi), 1 μmol/L BEZ235 (Pi3Ki/mTORi), or 1 μmol/L MK-2206 (AKTi) (Selleckchem/Absource). Cells were treated daily with 5 μmol/L 5-azacytidine (Sigma) for 6 days prior to stimulations.
Flow cytometry (FACS) analysis
A total of 2 × 105 cells were stained on ice in 50 μL FACS buffer. Antibodies (all Biologend) used were: BV 421 anti-CD73mAb 1:100 (clone AD2), biotin anti-mouse CD73 mAb 1:200 (clone TY/11.8), biotin rat IgG1κ isotype control 1:200, BV 421 streptavidin 300 ng/mL. Twenty thousand events in the life cell gate were recorded on a FACSCanto II flow cytometer (BD Biosciences) and analyzed with FlowJo software.
C57BL/6 mice (H-2b) were purchased from Charles River. TCR-transgenic Pmel-1 mice expressing an αβ TCR specific for amino acids 25–33 of human and mouse gp100 presented by H2-Db were bred as described previously (20). All animal experiments were approved by the local government authorities (LANUV, NRW, Germany) and performed according to the institutional and national guidelines for the care and use of laboratory animals.
Groups of C57BL/6 mice were injected intracutaneously with 4 × 105 HCmel3 cells into the flanks. Nontreated (NT) mice were killed when tumors reached >10 mm in diameter. Mice were killed between 12 and 20 days after ACT start (early-during-treatment, EDT). Experiments were performed in groups of five and repeated twice. Late relapse (R) melanoma samples and cell lines have been described previously (20). ACT with Pmel-1 T cells and microarray gene expression analysis were carried out according to the MIAME guidelines as described previously (20, 22). Raw data are accessible through GEO (GSE40213, GSE71879, GSE99925). Details are described in Supplementary Methods.
Samples and clinical data were obtained with the approval of Institutional Ethics Committee boards (institutional review boards, IRB) and patients' consents at the respective clinical centers [UCLA/USA (UCLA IRBs 11-001918 and 11-003066), Sydney/Australia (Protocol No X10-0305 & HREC/10/RPAH/539)]. Patient studies were conducted in accordance with the ethical guidelines of the Belmont Report. For correlation of CD73 expression with clinicopathologic parameters, we retrospectively studied cases from melanoma patients and material was collected between January 1, 2000 and December 31, 2010. All patients had given their written informed consent in agreement and after approval by the local ethics committee of the University of Bonn (Bonn, Germany). Treatment and follow-up examinations were performed according to the recommendations of the German Society of Dermatology. Dacarbazine was the only approved systemic treatment for metastatic melanoma till 2011 at our institution (Department of Dermatology, University Hospital Bonn, Bonn, Germany). Patients with primary melanomas underwent surgical resection, which was largely curative. Nineteen of 126 of these patients had recurrent disease and received best supportive care or dacarbazine before death. One additional patient with recurrent disease was treated with a BRAF inhibitor. Four of the patients with cutaneous melanoma metastases were treated with an immune checkpoint inhibitor (ipilimumab, nivolumab, pembrolizumab).
Mouse melanomas were immersed in zinc-based fixative (BD Pharmingen) and human melanomas in buffered paraformaldehyde (DAKO). Paraffin embedding, hematoxylin and eosin stains and IHC were performed according to standard protocols (Supplementary Methods). Antibodies used were polyclonal rabbit-anti-human CD73 antibody (Sigma HPA017357, 1:600, antigen-retrieval pH6 10 minutes) and rabbit anti-human CD14 mAB (Clone EPR3652, LifeSpan Bio Sciences, dilution: 1:25). Rabbit-anti-CD73 mAb (clone D7F9A, #13160 Cell Signaling Technology, dilution 1:300, antigen-retrieval pH6 10 minutes) was used to detect mouse CD73 or human CD73 (Sydney cohort). Rat anti-mouse CD45 mAb (30-F11, BD Biosciences) was used to identify immune cells in mouse melanomas.
Gene regulation assays
qPCR, chromatin immunoprecipitation (ChIP)-qPCR, and CRISPR-Cas9 approach for gene regulation by conditionally expressed GFP-tagged transcription factors were performed as described previously (21). Used DNA oligos and protocol adjustment are described in Supplementary Table S1 and Supplementary Methods. For bisulfite conversion of 500 ng of genomic DNA the EZ DNA-Methylation Direct Kit (Zymo Research) was used according to the manufacturer's instructions. Used DNA oligos and NGS-based analysis are described in Supplementary Methods.
Cells were lysed in Laemmli buffer and immunoblots performed according to standard procedures as described previously (21). Used antibodies are listed in Supplementary Methods.
Additional methods in detailed descriptions are found in Supplementary Methods.
CD73 marks nascent “invasive” melanoma phenotype switching
To identify markers and mechanisms of nascent “invasive” phenotype switching in melanoma, we used gene expression data from a well-characterized cell line panel (n = 88) (24). We ranked the cell lines by expression of the EMT-like “invasive” gene signature described by Verfaillie and colleagues (17) and analyzed the correlation with the expression level of MITF mRNA. This approach clearly separated MITFlow “invasive” melanoma cell lines from the others (Fig. 1A). Importantly, the “invasive” signature strongly correlated with our recently described melanoma TNF response signature (21), which supports the notion that inflammation-induced dedifferentiation and EMT-like “invasive” phenotype switching converge (Fig. 1B; ref. 25). Next, we searched for individual TNF response genes showing a high variance and thus a broad range of expression level in MITFhigh cells. We hypothesized that high expression of such a gene in a MITFhigh cell line could indicate intrinsic inflammatory pathway activation and nascent “invasive” phenotype switching (Fig. 1C). This analysis identified a small list of genes and among those, NT5E was of particular interest, because it encodes the 5′ectonucleotidase and mesenchymal stem cell marker CD73 (26–28). CD73 generates immunosuppressive adenosine in the microenvironment, which defines the CD73-adenosinergic pathway as an emerging target for combination immunotherapy (29, 30). In comparison to the typical “invasive” phenotype markers WNT5A or AXL (13, 31), NT5E (CD73) showed robust expression in MITFhigh cells with a nascent “invasive” phenotype as well as MITFlow cells with a fully established “invasive” phenotype (Fig. 1D and E). On the basis of this broad range expression of NT5E (CD73) in melanoma cell lines, we expected variable level of CD73 in human melanoma tissues. We established a protocol for sensitive and specific IHC detection of CD73 protein (Supplementary Fig. S1A and S1B) and we developed a CD73 IHC scoring system (0–3) to categorize CD73 expression level (Fig. 1F). In a cohort of primary cutaneous human melanomas with known sentinel lymph node status (n = 126), CD73 expression was significantly associated with tumor thickness, ulceration, and a positive sentinel lymph node (Fig. 1G; Supplementary Table S2). Most melanomas expressed a low or intermediate level of CD73 (Supplementary Table S2). Absence of CD73 expression (IHC score 0) was only found in thin melanomas, which had a better progression-free survival. Otherwise, we found no significant association between CD73 and progression-free or overall survival in patients with primary melanomas (Fig. 1H) or with cutaneous melanoma metastases (n = 70), respectively (Supplementary Fig. S1C). The distribution of CD73 IHC scores was comparable between primary melanomas and cutaneous melanoma metastases (Supplementary Fig. S1D; Supplementary Table S2). Taken together, CD73 showed variable expression in human melanomas, and gene signature analyses defined that CD73 expression associated with both a nascent and fully established EMT-like “invasive” phenotype.
MAPK and proinflammatory signaling cooperatively induce CD73
Next, we searched for signaling pathways that enhance CD73 expression and a nascent “invasive” phenotype in MITFhigh melanoma cells (Fig. 2A). We used gene-set enrichment analysis (GSEA) and found signatures related to c-MET (rank#2), HGF (rank#11), and NF-κB (rank#13) signaling among the top hits (Fig. 2B and C), as well as other gene sets related to mitogenic and proinflammatory signaling (Supplementary Table S3). The identification of NF-κB signaling was confirmatory, because CD73 is part of the identified melanoma TNF response signature (21) correlating with NF-κB activity. HGF (hepatocyte growth factor) is the ligand for the receptor tyrosine kinase c-MET and a potent activator of MAPK and AKT signaling (32). Thus, this analysis suggested that mitogenic (e.g., HGF) and inflammatory signals (e.g., TNFα) cooperatively induce CD73 expression and “invasive” phenotype switching. To test this hypothesis by functional studies, we used a panel of human melanoma cell lines with known driver mutation status (33). CD73 protein expression was high in MITFlow but also several MITFhigh cell lines (Fig. 2D), as predicated by our transcriptional analyses above and confirmed by FACS analyses (Fig. 2E, top). Negativity for CD73 in two of the tested cell lines could be explained by CpG island methylation in the promoter region of the CD73 gene consistent with a previous study describing this mode of regulation (Fig. 2E, bottom; Supplementary Fig. S2; ref. 34). Of note, the two cell lines MaMel.71 and MaMel.15 without activating BRAF or NRAS mutations exhibited low levels of CD73 expression despite the lack of CD73 promoter methylation. We hypothesized that this could be due to insufficient MAPK signaling. Indeed, HGF induced CD73 expression in both cell lines in a MEK-dependent manner and TNFα had a cooperative effect (Fig. 2F). Costimulation with HGF and TNFα enforced MITF downregulation (Supplementary Fig. S3A and S3B). In contrast, cell lines with CD73 promoter methylation (e.g., MaMel.67a) failed to upregulate CD73 by HGF/TNFα costimulation, but became responsive upon prior exposure to the DNA-demethylating agent 5-azacytidine (Fig. 2G). Finally, we dissected the contribution of different mitogenic signaling pathways to high CD73 baseline expression in the remaining cell lines. We found that the MEK inhibitor (MEKi) trametinib or the ERK inhibitor (ERKi) SCH-772984 suppressed CD73 protein and mRNA levels in most cell lines tested, in comparison with the Pi3K/mTOR inhibitor BEZ235 or the AKT inhibitor MK-2206 (Fig. 2H and I). Taken together, we found that MAPK signaling cooperates with the proinflammatory cytokine TNFα to induce CD73 in melanoma cells.
c-Jun/AP-1 induces CD73 expression via binding to an intronic enhancer
Next, we aimed to investigate the molecular mechanisms that control transcription of CD73 in melanoma. We integrated ChIP-Seq data from the ENCODE database to identify transcription factors with reported binding in the CD73 genomic region (Supplementary Table S4) that showed significant co-expression with CD73 mRNA in melanoma cells. On the basis of this approach, the top ranking transcription factors were IRF1, JUN (c-Jun), and NR3C1 (Fig. 3A). JUN was of particular interest, because others and we have recently shown that c-Jun/AP-1 is involved in melanoma phenotype switching (15, 17, 21, 35). c-Jun is a member of the heterodimeric AP-1 transcription factor complex that regulates cellular responses to mitogens, inflammation, and diverse stress conditions downstream of MEK-ERK and JNK signaling (36, 37). Classical heterodimerization partners of c-Jun are c-Fos, FOSL1 (FRA-1), and FOSL2 (FRA-2), which are also regulated by MEK–ERK signaling through transcriptional and posttranslational mechanisms. Thus, the AP-1 complex was a promising candidate to integrate mitogenic and inflammatory signaling in the regulation of CD73 expression. Indeed, stimulation of MaMel.71 cells (BRAFwt/NRASwt) with TNFα, HGF, or a combination of both led to cooperative induction of c-Jun (Fig. 3B). Accumulation of c-Fos and FOSL1 was largely dependent on HGF. To dissect the genomic context of CD73 transcriptional regulation by the AP-1 complex, we aimed to use conditional gene expression approaches for c-Jun, because c-Jun can induce heterodimerization partners like FOSL1. Immunoblots confirmed the reciprocal expression of c-Jun and MITF in our melanoma cells as previously described (Supplementary Fig. S3C; refs. 13, 21). FOSL1 expression was notably lower in the two BRAFwt/NRASwt cell lines MaMel.71 and MaMel.15, suggesting that lower MEK–ERK pathway activity impeded accumulation of FOSL1, as FOSL1 expression was dependent on MEK signaling in other cell lines (Supplementary Fig. S3D). Therefore, we reasoned that a CD73low cell line with an activating BRAF or NRAS mutation (e.g., MaMel.79b) would be suitable for doxycycline-dependent conditional c-Jun expression. Interestingly, we found that a transient expression (achieved by low-dose doxycycline 25 ng/mL) of c-Jun tagged with the GFP variant citrine was sufficient to induce CD73, which persisted for several days (Fig. 3C). As MITF was only transiently suppressed, pulsed c-Jun expression established a MITFhigh/CD73high phenotype, which marks nascent activation of the “invasive” program. Having characterized this experimental system, we next looked for potential c-Jun/AP-1–binding sites in the CD73 gene using ENCODE ChIP-Seq data (Fig. 3D). A ChIP-qPCR approach revealed prominent binding of c-Jun at the AP-1 sites #1–2 and #5–6 (Fig. 3E) coinciding with enhancer regions (H3K27Ac) identified by the ENCODE ChIP-seq project (Fig. 3D).
Next, we devised a CRISPR/Cas9 approach to clarify the functional relevance of the individual c-Jun/AP-1–binding sites. MaMel.79b cells with inducible c-Jun were transiently transfected with Cas9-sgRNA constructs targeting these binding sites, as well as nearby control sites with no c-Jun/AP-1 motif. This was followed by CD73 induction through c-Jun, FACS analysis, and sorting of CD73low versus CD73high subfractions (lowest 10% versus highest 10% as unbiased cutoff) coupled to next-generation sequencing (NGS)-based analysis of mutagenesis frequencies at the respective c-Jun/AP-1 or control sites (Fig. 3F). We hypothesized that cells with successfully Cas9-mutagenized c-Jun/AP-1 sites should become enriched in the CD73low subfraction and depleted from CD73high subfraction if a particular AP-1–binding site was critical for CD73 expression. Strong enrichment of AP-1 site mutagenesis frequencies in the CD73 low subfractions (52% vs. 4% in CD73 lowest 10% vs. highest 10%) was observed only in the cell population transfected with the sgRNA against the c-Jun/AP-1 site #5 (Fig. 3G and H) being consistent with reduced CD73 surface expression (Supplementary Fig. S4A).
Next, we used the same approach to investigate whether the c-Jun/AP-1 site #5 was also driving high CD73 baseline expression in the melanoma cell lines SK-MEL28, MaMel.54a, and MaMel.85. Again, only targeting of the c-Jun/AP-1 site #5 resulted in a strong enrichment of the mutagenesis frequency in the CD73 low subfraction (73%) with a reciprocal reduction in the CD73 high subfraction (3%) considering a mutagenesis frequency of only 19% in the nonsorted total SK-MEL28 cell population (Fig. 3I). Similar results were obtained for MaMel.54a and MaMel.85 (Supplementary Fig. S4B). We then monitored CD73 surface expression in the total SK-MEL28 cell populations as well as FACS-sorted subfractions for several days (days 0, 3, and 7 after FACS sorting). This revealed a persistent reduction of CD73 surface expression only in the CD73 low FACS-sorted subfraction of cells transfected with the sgRNA against the c-Jun/AP-1 site #5 (Fig. 3J), being in line with the AP-1 site mutagenesis frequency of 73% (Fig. 3I). We corroborated this finding by showing a corresponding reduction of CD73 mRNA and protein level (Fig. 3K). Taken together, we identified a critical c-Jun/AP-1–binding site in an intronic enhancer of the CD73 gene that controls its c-Jun/AP-1–dependent transcriptional activation downstream of mitogenic MAPK and inflammatory cytokine signaling.
CD73 is expressed in “invasive” and inflammatory mouse melanomas
We revealed multiple regulatory pathways for CD73 expression in human melanoma as a marker of tumor cell plasticity. Using mouse models, we previously dissected the interaction between inflammation and melanoma cell plasticity (20–22). Therefore, we aimed to investigate the regulation of CD73 in mouse melanomas with inflammation-driven tumor microenvironments. Recently, we described that BrafV600ExCdk4R24C mice simultaneously develop amelanotic and pigmented melanomas (Fig. 4A; ref. 22). Amelanotic melanomas were infiltrated by myeloid immune cells (22) and this inflammatory microenvironment was also evident by broad chemokine gene expression (Fig. 4B). CD73 mRNA level were significantly higher in inflamed amelanotic melanomas and coexpressed with typical “invasive” marker genes like Axl. Amelanotic CD73high melanomas also exhibited higher levels of the AP-1 components c-Jun and Fosl2 together with a reciprocal reduction of Mitf (Fig. 4C). GSEA using the hallmark gene set collection (38) showed elevated inflammatory (TNFα signaling) and mitogenic (KRAS signaling) pathway activity in amelanotic melanomas (Fig. 4D; Supplementary Table S5). Immunohistochemical stains confirmed increased CD73 protein level broadly expressed throughout amelanotic BrafV600ExCdk4R24C melanomas, whereas CD73 expression in pigmented BrafV600ExCdk4R24C melanomas was localized and found at sites of ulceration or necrosis (Fig. 4E). This observation was recapitulated in pigmented primary melanomas from our related HgfxCdk4R24C mouse model (Fig. 4F). Human melanomas also showed intense CD73 expression around necrotic areas. In 32 of 70 (45.7%) human cutaneous melanoma metastases, necrotic areas could be detected and from that 23 of 32 (71.9 %) showed intense CD73 expression around necrotic areas (Fig. 4G, top), which suggested hypoxia as another critical inducer of CD73 in melanoma. Indeed, hypoxia is known to promote “invasive” melanoma phenotype switching (39, 40) and drives CD73 expression in other cell types (41, 42). We also stained for CD14 to exclude that a potential recruitment of myeloid immune cells accounted for CD73 expression at the necrotic rim (Fig. 4G, bottom). In summary, the divergent phenotypic evolution of melanomas in the BrafV600ExCdk4R24C mouse model demonstrated that CD73 expression is linked to an inflamed and hypoxic microenvironment.
CD73 is induced in response to adoptive T-cell therapy
As cancer immunotherapy causes inflammation and damage in tumor tissues, we hypothesized that CD73 might be induced as part of an adaptive regenerative response (12). We have previously established an adoptive T-cell transfer (ACT) therapy that achieves remissions and long-term tumor control of primary HgfxCdk4R24C melanomas or transplantable HCmel3 melanomas (20). Adoptively transferred Pmel-1 CD8 T cells target the endogenous antigen gp100 (Pmel) expressed by melanoma cells. To study CD73 regulation and phenotypic coevolution under Pmel-1 T-cell therapy, we treated HCmel3 melanoma-bearing mice with our ACT protocol. Melanoma tissues were harvested for microarray gene expression analysis from mice early during ACT treatment (EDT, day 12–20 after ACT) and compared with nontreated (NT) controls and late relapse melanomas (R, day 140–210 after ACT; Fig. 5A; ref. 20). The induced antitumor immune response following ACT was evidenced by the increase of Cd8a T cell and Ifng transcript level in EDT melanomas, which declined in relapse melanomas (Fig. 5B). In contrast, the myeloid marker transcript Itgam (Cd11b) persisted in relapse melanomas, indicating a sustained myeloid immune cell infiltration.
Nt5e (CD73) mRNA expression was highest in relapse melanomas, but was already increased in EDT melanomas (Fig. 5B, right). We also confirmed high CD73 protein expression in relapse melanomas by IHC (Fig. 5C) and by FACS on isolated cell cultures from relapse melanomas (Fig. 5D). Careful histomorphologic analysis and stains for CD45+ immune cells also supported predominant CD73 expression by tumor cells in relapse melanomas in situ (Supplementary Fig. S5A and S5B). HCmel3-R cell lines also had moderately increased level of active phosphorylated ERK and AKT, but lacked expression of the melanocyte marker protein tyrosinase (Fig. 5E). Microarray gene expression analysis showed that inflammation-induced dedifferentiation and downregulation of pigmentation-related genes was progressive from EDT to relapse (Fig. 5F). This was accompanied by a reciprocal induction of “invasive” genes including Axl (Fig. 5F) and corresponding gene sets (Supplementary Fig. S6A).
Consistent with the decrease in CD8 T cells in relapse melanomas, we noticed a concomitant reduction of IFN targets (e.g., Irf1, Irg1, Cd274, Ifi205) including genes encoding for MHC class II molecules and immunoproteasome components (Psmb8, Psmb9; Fig. 5F). GSEA also confirmed that ACT-induced IFNγ pathway activation in EDT melanomas (Fig. 5G top left; Supplementary Table S6) decreased in relapse melanomas (Fig. 5G, top right; Supplementary Table S7) in line with a diminishing antitumor T-cell response. In contrast, ACT-induced TNFα/NF-κB pathway activation in EDT melanomas (Fig. 5G, bottom left; Supplementary Table S6) was not downregulated in relapse melanomas (Supplementary Table S7). We further found that lowered IFNγ pathway activity in relapse melanomas was associated with a reciprocal upregulation of cell-cycle–related gene sets including E2F (Fig. 5G, bottom right) and MYC (Supplementary Fig. S6B). Of note, the Verfaillie “proliferative” melanoma gene signature (17) was actually downregulated in relapse melanomas, because it largely contains MITF target genes including those involved in pigmentation or melanocyte differentiation (Supplementary Fig. S6B and S6C) and had almost no overlap with the E2F or MYC hallmark gene sets (Supplementary Fig. S6D). Importantly, inactivating mutations in the IFNγ pathway emerge as genetic determinants of primary and acquired resistance to checkpoint immunotherapy in melanoma patients (8, 9). Lowered IFNγ pathway activity in ACT relapse versus EDT mouse melanomas supported a critical role for IFNγ pathway inactivation in resistance. Furthermore, our findings defined that CD73 was preferentially induced in relapse melanomas. CD73 may facilitate therapy escape, as CD73-adenosinergic signaling limits immunotherapy in many solid cancer mouse models (30, 43–45).
Therefore, we asked whether we could find evidence for adaptive upregulation of CD73 in melanoma patients undergoing a similar type of ACT immunotherapy. In an independent study (Ribas and colleagues, manuscript in preparation), serial biopsies from six melanoma patients, which were treated and initially responded to adoptive transfer of MART-1 T-cell receptor transgenic lymphocytes and dendritic cell vaccination (NCT00910650; ref. 6), were screened for acquired loss of MART-1 (MLANA) and other melanocyte marker proteins. Indeed, one case (patient F5-1 in ref. 6) showed dedifferentiation with loss of the target antigen MART-1 upon progression. In this particular case, CD73 was reciprocally expressed and low before treatment, but induced upon MART-1 ACT therapy and persisted at progression (Fig. 5H), and thus similar to the relapse phenotype seen in our ACT-treated mouse melanoma model.
Dynamic regulation of CD73 in human melanomas during anti-PD-1 therapy
Next, we addressed whether dynamic regulation of CD73 also occurred in melanoma patients treated with checkpoint immunotherapy. Examining on-treatment biopsies from melanoma patients (n = 25, 16/25 cutaneous metastases, UCLA cohort) under pembrolizumab (anti-PD-1) therapy (46) revealed stable expression of CD73 in 13 cases (52%; Fig. 6A; Supplementary Table S8). In six cases (24%) all with a low CD73 baseline level (IHC score 1), we detected upregulation of CD73. In another six cases, of which three had highest CD73 baseline level (IHC score 3), we found downregulation of CD73. Notably, patients with progressive disease did not downregulate CD73, in fact the highest proportion of patients increasing CD73 displayed progressive disease (Fig. 6B). One case with a complete remission as best response showed upregulated CD73 in a recurrent lesion 306 days after start of anti-PD-1 treatment (Fig. 6C). We obtained similar findings in an independent cohort of melanoma patients (n = 8, Sydney cohort) treated with anti-PD-1 (pembrolizumab or nivolumab; Supplementary Table S9). Four cases were negative for CD73 expression across all longitudinal samples. Notably, three of these cases had received prior BRAFi+MEKi therapy. The remaining four cases had not received prior MAPKi therapy and were all CD73 positive at progression. Two of these cases demonstrated robust upregulation of CD73 at progression (case#7, acquired resistance; case#8 primary resistance; Fig. 6D and E). Together, dynamic upregulation of CD73 argues for an adaptive resistance mechanism and therefore CD73 blockade should not be restricted to melanoma patients with high CD73 expression in pretreatment biopsies.
Here we identified the 5′-ectonucleotidase CD73 as a marker of melanoma cell plasticity. CD73 expression is associated with the EMT-like “invasive” phenotype, which is characterized by low expression of the melanocyte lineage transcription factor MITF (16, 17). Coexpression of CD73 with EMT markers in melanoma cells has been noted previously (47), but the mechanistic basis was not addressed. Importantly, we found that CD73 is strongly expressed by MITFlow but also by a subset of MITFhigh melanoma cells, where it marks nascent activation of the EMT-like invasive program, which makes CD73 a distinct “invasive” marker (Supplementary Fig. S7). We demonstrated that MAPK signaling and the proinflammatory cytokine TNFα cooperatively induce CD73 expression through the c-Jun/AP-1 transcription factor complex, a central node in cellular stress signaling (36). This implements CD73 upregulation in a melanoma cell response to acute or chronic stress, which includes therapy-induced inflammation, hypoxia and amino acid starvation (12, 20, 48). Thus, CD73-adenosinergic signaling emerges as a stress-dependent regulator of melanoma phenoytpes and immune cell interactions.
Our and the accompanying work by Young and colleagues show that MAPK pathway activity promotes CD73 expression, as reported previously in A375 melanoma cells (49). In our concerted effort, we demonstrate that CD73 is downregulated in melanoma patients treated with MAPK inhibitors and sustained CD73 suppression correlates with a favorable response. MAPK-dependent regulation of CD73 has important implications for the development of treatment strategies, because combined inhibition of CD73 and A2A adenosine receptor (A2AR) has additional immunotherapeutic efficacy (30). Indeed, the accompanying work by Young and colleagues shows that MAPK inhibitors potentiate A2AR antagonism through CD73 downregulation.
Using our HCmel3 syngeneic inoculation model, we studied the phenotypic evolution of melanomas at different stages during adoptive T-cell therapy. CD73 was progressively induced and reached the highest level in dedifferentiated relapse melanomas arguing that immunosuppressive CD73 facilitates therapy escape. As noted in our previous studies, mouse melanomas recapitulate phenotype transitions found in human melanomas (20–22, 50). Of note, EMT, hypoxia and wound-healing related gene signatures were shown to be upregulated in melanoma patients with poor response to anti-PD-1 therapy (11). Similar to our mouse models, we also found that CD73 was induced during MART-1 ACT or anti-PD-1 checkpoint immunotherapy in a subset of melanoma patients. Larger patient cohorts are needed to clearly define the relationship between CD73 modulation, treatment outcome and the type of resistance (primary versus adaptive). Cases with CD73 downregulation are of particular interest for future studies, because none of the patients in our limited cohort had a primary resistance.
Our experiments revealed downregulation of IFNγ signaling in mouse melanomas that escaped from Pmel-1 ACT therapy in line with a reduction of IFNγ-producing T cells at relapse. The reciprocal upregulation of proliferation-associated E2F/MYC gene signatures suggests that antiproliferative IFNγ signaling significantly contributes to immunotherapeutic efficacy. Previously, it was shown that IFNγ and TNFα released by infiltrating T cells promote senescence-like responses of tumor cells (51), and our analysis suggests that IFNγ signaling is critical for this senescence program. In support of this notion, inactivating mutations in the JAK1/JAK2 kinases were recently identified in human melanomas with primary or acquired resistance to anti-PD-1 therapy (8, 9). JAK1/JAK2 deficiency blocked IFNγ responses and conferred insensitivity to its antiproliferative effects. However, another report found that persistent IFN signaling activates an adaptive resistance program to checkpoint immunotherapy through the induction of several T-cell–inhibitory ligands including PD-L1 (52). These studies underscore that IFNγ signaling can exert opposing functions in the context of cancer immunotherapy and it remains to be investigated how this context dependency impacts on the CD73-adenosinergic pathway.
Combination immunotherapy is a strategy to limit primary or acquired resistance (53), but a better understanding of the tumor and immune cell crosstalk is needed (11). CD73-directed treatments also target inhibitory immune cells populations like regulatory T cells (27, 54, 55), which coevolve during immunotherapy. In melanomas, we showed that induction of immunosuppressive CD73 by mitogenic and inflammatory signals related to an EMT-like “invasive” phenotype as adaptive response to immunotherapy. As blocking CD73-adenosinergic signaling is effective in preclinical cancer models (30, 43, 44), our results support CD73 as a target to combine with current melanoma immunotherapies (23). This approach could restrain the production of immunosuppressive adenosine that may facilitate therapy escape. Finally, the dynamic induction of CD73 seen in subsets of melanoma patients during immunotherapy suggests that adjuvant CD73 blockade should not be restricted to patients with high CD73 expression in pretreatment biopsies. Measuring soluble CD73 (sCD73) in the blood of patients could help to identify cases with adaptive CD73 upregulation during immunotherapy. Our work has important implications for the development of biomarkers for anti-CD73 combination immunotherapies.
Disclosure of Potential Conflicts of Interest
G.V. Long is a consultant/advisory board member for Bristol-Myers Squibb, Amgen, Novartis, Merck, Roche, Pierre-Fabre, and Array. M.J. Smyth reports receiving a commercial research grant from Bristol Myers Squibb, Corvus Pharmaceuticals, and Aduro Biotech, is a consultant/advisory board member for Arcus Biosciences, has provided expert testimony for Corvus Pharmaceuticals. No potential conflicts of interest were disclosed by the other authors.
Conception and design: J. Reinhardt, P.C. Tumeh, T. Tüting, M. Hölzel
Development of methodology: J. Reinhardt, J. Madore
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Reinhardt, J. Landsberg, B.B. Ramis, T. Bald, N. Glodde, D. Lopez-Ramos, A. Young, S.F. Ngiow, D. Nettersheim, H. Schorle, W. Kolanus, G.V. Long, J. Madore, R.A. Scolyer, P.C. Tumeh, T. Tüting
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Reinhardt, J. Landsberg, J.L. Schmid-Burgk, B.B. Ramis, T. Bald, N. Glodde, A. Young, S.F. Ngiow, D. Nettersheim, T. Quast, W. Kolanus, D. Schadendorf, G.V. Long, J. Madore, R.A. Scolyer, A. Ribas, M.J. Smyth, P.C. Tumeh, T. Tüting, M. Hölzel
Writing, review, and/or revision of the manuscript: J. Reinhardt, J. Landsberg, T. Bald, N. Glodde, A. Young, S.F. Ngiow, H. Schorle, W. Kolanus, D. Schadendorf, G.V. Long, R.A. Scolyer, A. Ribas, M.J. Smyth, P.C. Tumeh, T. Tüting, M. Hölzel
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Bald, P.C. Tumeh, T. Tüting, M. Hölzel
Study supervision: M.J. Smyth, T. Tüting, M. Hölzel
We thank P. Aymans for managing the mouse colony, Sandra Ferring-Schmitt for help with dermatohistology, N. Fricker, K. Keppler, S. Sivalingam, A.Heimbach for help with microarray and NGS (UKB Genomics & NGS Core Facility, Department of Human Genetics), E. Endel, P. Wurst and An. Dolf for help with FACS sorting (UKB FACS core facility).
M.J. Smyth was supported by National Health and Medical Research Council of Australia (NHMRC) Senior Research Fellowship (1078671) and Project Grant (1120887). J.L. Schmid-Burgk was supported by the German Academic Scholarship Foundation. T. Bald was funded in part by an EMBO Long-Term Fellowship (ALTF 945-2015) and the European Commission (Marie Curie Action LTFCOFUND2013, GA-2013-609409). This research was funded in part by grants from the DFG (TU 90/8-1, A22 in the SFB704 and A27 in the SFB854; to T. Tüting) and in part by grants from the Deutsche Krebshilfe (111084 to M. Hölzel) and DFG (HO4281/2-1 to M. Hölzel).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.