Immunotherapy Resistance by Inflammation-Induced Dedifferentiation

Despite the high initial response rates of patients to T-cell receptor (TCR) engineered adoptive T-cell transfer therapy (ACT), most patients relapse within a few months (1–3). The prevailing hypotheses for how resistance to ACT develops in patients with melanoma are either that their T cells become ineffective, due to exhaustion or immune tolerance, or that a subset of tumor cells acquire a survival advantage, possibly by genetic alterations that increase proliferation or that result in the loss of presentation of the antigen targeted by ACT (4–9). In recent years, an alternative mechanism, inflammation-induced dedifferentiation of tumor cells to precursor cells of neural crest origin (10–12), has come to light from mouse studies and using human cell lines. Work by Landsberg and colleagues (11) in a mouse model demonstrated that ACT using T cells specific for the shared melanosomal antigen gp100 resulted in initial tumor responses followed by regrowth of cancer cells that had lost gp100 expression and dedifferentiated into a neural crest lineage. This tumor cell dedifferentiation was mediated by tumor necrosis factor alpha (TNFα) produced by the gp100 TCR transgenic T cells in response to gp100 antigen recognition.

The inflammatory cytokine-induced plasticity of tumor cells, with reversible loss of the tumor-targeting antigen (11, 13), may have widespread clinical implications on how best to target antigens in ACT. Here, we show clinical and pathologic findings from a patient who received MART1-specific ACT and developed resistance to therapy in association with a dedifferentiated tumor phenotype that lacked conventional melanocytic antigens.

A 60-year-old man with multiple pigmented nevi over his body presented with metastatic melanoma and was enrolled in a clinical trial of MART1 TCR-engineered ACT immunotherapy in combination with MART1 peptide-pulsed dendritic cells (DC) and high-dose IL2, administered after a lymphodepleting conditioning regimen (Fig. 1A; NCT00910650). Two years prior, he had noted a mole on his back that had grown and bled occasionally; however, it spontaneously regressed over time. Three months prior to presentation, he noted a right posterior neck mass, and 1 month prior to presentation, he was admitted for abdominal pain and jaundice. Endoscopic retrograde cholangiopancreatography (ERCP) revealed peripancreatic masses, and a fine-needle biopsy raised the possibility of melanoma. A whole-body PET/CT scan revealed multiple metastases in the lungs, head and tail of the pancreas, stomach wall, right lobe of the liver, and peritoneal and mesenteric regions (Fig. 1B). An incisional biopsy of the right posterior neck mass confirmed a MART1-positive metastatic melanoma. HLA typing of the patient revealed HLA-A*0201 positivity, making him a candidate for this clinical trial. This patient was treated before the approvals of immunotherapies that have shown improvement in overall survival in patients with advanced melanoma, such as ipilimumab, nivolumab, and pembrolizumab (14–16).

Per study protocol (Fig. 1A), the patient underwent a nonmyeloablative conditioning regimen of cyclophosphamide (60 mg/kg/day, days −7 and −6) given with mesna (60 mg/kg/day) followed by fludaribine (25 mg/m², days −5 to −1). On day 0, the patient received adoptive transfer of 1 × 10⁹ lymphocytes that had been retrovirally transduced with the F5 MART1-specific TCR. On days +1, +14, and +30, the patient received MART1_26-35 peptide-pulsed DC vaccines (1.8 × 10⁷, 1.8 × 10⁷, and 1.7 × 10⁷ cells, respectively). Between days +1 and +7, he received high-dose IL2 administered every 8 hours to tolerance. Over the course of the study, peripheral blood was collected at regular intervals for basic laboratory tests and immune monitoring, and he underwent baseline and on-therapy biopsies and PET/CT scans.

At the time of treatment, the patient presented with diffusely metastatic disease confirmed by incisional biopsy of his left neck lesion to be stage IVc melanoma with homogenously positive MART1 (Melan-A) expression by IHC (Fig. 1B). The patient tolerated infusion of F5 TCR T cells, followed by IL2 treatments, and MART1_26-35 peptide-pulsed DC vaccines with the main adverse effect being a whole-body skin rash. The patient was able to return to work 1 month after infusion. An early follow-up PET/CT scan at 35 days after infusion demonstrated a dramatic decrease in 2[¹⁸F]fluoro-2-deoxy-d-glucose (FDG) uptake in the vast majority of the patient's baseline lesions, with increased uptake in seven lesions and the presence of two new lesions with mild FDG uptake (Fig. 1C; new lesions marked with black arrows and biopsied lesion marked with a red arrow). A subsequent PET/CT scan at 84 days after infusion demonstrated objective responses on most baseline lesions by RECIST. However, there were also several new and increasingly FDG-avid lesions (Fig. 1D; biopsied lesion marked with a red arrow) concerning for progression of disease including on his right anterior chest, liver, and paraspinal, paraaortic, and mesenteric regions, among others. The right anterior chest wall lesion was biopsied at that time for further investigation of his progressing disease. The patient's progression-free survival was 3 months, and his overall survival was 5 months.

The initial biopsy of the patient's left neck lesion (Fig. 1B, red arrow) showed an atypical malignant epithelioid neoplasm that was consistent with melanoma. IHC staining was performed and showed strong, diffuse S100 and melanocytic antigen expression, including MART1, gp100, and tyrosinase (Fig. 1E). One month after infusion of the F5 TCR T cells, the same left neck lesion was biopsied but showed a heterogeneous pattern of MART1 staining with multifocal areas of low or absent MART1 expression (Fig. 1F). S100 staining showed no difference in intensity and distribution of expression when compared with the initial biopsy; however, gp100 and tyrosinase staining patterns mimicked the multifocal MART1 pattern (Fig. 1F). At the time of progression, an anterior chest lesion biopsy revealed the complete absence of MART1 expression, along with absence of gp100 and tyrosinase, but again a normal S100 staining pattern (Fig. 1G). The global loss of melanocytic antigens, as opposed to just MART1, suggested a phenotype switch (13) within the progressing tumor.

Prior to ACT, the majority of CD8 CTLs were observed at the periphery of the tumor with few CD8 CTLs present in the central regions of the tumor parenchyma (Figs. 1E and 2A). At 35 days after infusion of F5 TCR T cells, a prominent but multifocal lymphocytic infiltrate was observed within the tumor, largely at the periphery with moderate infiltration of the center of the tumor (Figs. 1F and 2B). At the time of relapse, the anterior chest biopsy revealed a predominance of CD8 CTLs diffusely in all areas of the tumor (Figs. 1F and 2C), in direct contrast to the pattern observed on initial biopsy of pretreated tumors. Evidence supports that the presence of CD8 CTLs is indicative of an increased inflammatory milieu within the tumor microenvironment (11).

The nerve growth factor receptor (NGFR or CD271) is a member of the TNF superfamily group of receptors and has been shown to be a marker of cancer stem cells in melanoma (10). The expression of melanocytic antigens, such as MART1 and gp100, is negatively correlated with the expression of NGFR (10). Importantly, inflammatory signals, largely due to TNFα, result in increased expression of NGFR in mouse models of melanoma and human cell lines in vitro (11). In this patient, the expression of NGFR was low in pretreated tumor biopsies (Figs. 1E and 2A). However, increased multifocal expression of NGFR was observed in the tumor 35 days after infusion of F5 TCR T cells (Figs. 1F and 2B). Importantly, the expression of NGFR was negatively correlated with expression of MART1 and gp100 within the tumor and was observed predominantly in regions of increased CD8 CTL infiltration, thus suggesting that its expression was linked to regions of the tumor with increased inflammatory mediators (Figs. 1F and 2B). In the progressing tumor this phenomenon was even more apparent, as the vast majority of the tumor cells did not express MART1 and had strong NGFR expression with heavy infiltration of CD8 CTLs (Figs. 1G and 2C).

To determine if F5 TCR-transduced T cells are sufficient to induce dedifferentiation of human melanoma cells in vitro, we cultured the human melanoma cell line M397 with conditioned media from the coculture of F5 TCR T cells with M397 cells. We were unable to derive cell lines from the patient's biopsy samples; thus, the M397 line was chosen due to its HLA-A2 positivity and expression of MART1. M397 cells cultured with conditioned media generated from coculture with F5 TCR transduced T cells demonstrated decreased surface expression of MART1 and upregulation of NGFR (Fig. 3A). No effect was seen on M397 cells that remained untreated or that were treated with conditioned media from untransduced T cells (Fig. 3A), thus suggesting that conditioned media from F5 TCR T cells are sufficient to induce dedifferentiation of human melanoma cells.

A report in preclinical models had suggested that inflammation-induced dedifferentiation in mouse models of melanoma and in human cell lines is mediated by TNFα (11). To test if F5 TCR T cells produce TNFα upon antigen encounter, we performed intracellular staining of F5 TCR T cells cocultured with M397 cells and found significantly increased TNFα expression in these cells compared with untransduced cells (Fig. 3B).

To verify TNFα-induced dedifferentiation in human melanoma cell lines, we treated 8 previously described patient-derived MART1-expressing melanoma cell lines (17) with either TNFα or DMSO for 3 days, and analyzed them for surface expression of MART1 and NGFR (Fig. 4A and B). Consistent with previous findings, we found that TNFα treatment led to dedifferentiation characterized by decreased surface MART1 expression and upregulation of NGFR (Fig. 4A and B). To better understand the molecular mechanism underlying this phenotypic transformation, we performed RNA sequencing of three different melanoma cell lines, M229, M263, and M297, treated with either TNFα or DMSO for 3 days. Gene set enrichment analysis (GSEA) of differentially expressed genes between TNFα- and DMSO-treated cells (Supplementary Tables S1 and S2) revealed that TNFα-treated cells were enriched for genes characteristic of the TNFα inflammatory response [normalized enrichment score (NES) = 2.21, FDR < 0.001], epithelial-to-mesenchymal transition (EMT; NES = 1.92, FDR < 0.001) and neural crest stem cells (NES = 1.49, FDR = 0.016; Fig. 4C). Moreover, genes involved in the microphthalmia-associated transcription factor (MITF) pathway (18), including melanocytic antigens such as MART1, tyrosinase, and PMEL (same as gp100), were enriched in DMSO samples and downregulated in TNFα-treated cells (NES = −2.38, FDR < 0.001; Fig. 4C; Supplementary Table S1). Interestingly, TNFα treatment also led to the enrichment of several pathways characteristic of the innate anti–PD-1 resistance (IPRES) gene signature (ref. 19; Supplementary Table S2). This included enrichment of genes involved in EMT, angiogenesis, and hypoxia, among others (Supplementary Table S2), which is reflective of a more invasive melanoma phenotype and that has been associated with increased NGFR expression (20, 21).

Among the differentially expressed genes with TNFα treatment of our melanoma cell lines were LIF and IL8, both of which are known to promote EMT and tumor resistance (22, 23), thus suggesting that autocrine signaling through these proteins may play a role in tumor dedifferentiation. We thus measured the secretion of IL8 and LIF in the supernatant of M397 cells treated with TNFα or cocultured with untransduced or F5 TCR T cells. Both TNFα and F5 TCR T-cell treatment of M397 cells resulted in increased IL8 and LIF secretion (Supplementary Fig. S1A). However, neither IL8 nor LIF treatment of M397 cells was sufficient to induce dedifferentiation (Supplementary Fig. S1B), thus suggesting that the melanoma resistance phenotype observed in this study is a direct effect of TNFα or may be mediated by factors other than IL8 and LIF.

To test the reversibility of inflammation-induced dedifferentiation in our melanoma cell lines, we treated M397 cells for 3 days with TNFα and subsequently removed the inflammatory media for 7 days (Fig. 4D). Withdrawal of inflammatory media resulted in increased surface MART1 expression and loss of NGFR expression (Fig. 4D), thus suggesting that inflammation-induced dedifferentiation is reversible and that persistent TNFα exposure is required to maintain this phenotypic state.

We report a case of melanoma dedifferentiation as a mechanism of immune escape in a patient treated with cancer immunotherapy, a phenomenon previously described only in mouse models and human cell lines (11, 13, 20). The notion of phenotypic switching of melanoma is consistent with our current model of melanoma resistance; that is, we see the loss of the target antigen, in this case MART1, in progressing tumors, which is a phenomenon frequently encountered by pathologists. The global loss of melanocytic markers, including gp100 and tyrosinase, in conjunction with the expression of the inflammation-induced neural crest marker NGFR, suggests that the tumor cells acquired a dedifferentiated state that is reflective of earlier stages of embryologic development of melanocytes, cells that originally arise from within the neural crest. Although it has been suggested that the plasticity of melanoma cells is reversible in mouse models (11), this cannot readily be tested in human patients. Our in vitro data using human melanoma cell lines suggest, however, that the same phenomenon may apply to humans, with cells dedifferentiating in the setting of inflammation with the possibility of redifferentiating after inflammation resolves.

Our RNA-sequencing analysis of melanoma cell lines treated with TNFα suggests that the pathways of inflammation-induced dedifferentiation may overlap with those of innate anti–PD-1 resistance (Supplementary Table S2). In particular, the enrichment of genes involved in EMT suggests that dedifferentiation may reflect a more invasive phenotype. This has parallels to resistance patterns observed with BRAF inhibition and may suggest that the concept of phenotypic switching applies more broadly as a resistance mechanism to other modes of melanoma therapy (13, 18, 20, 21, 24, 25). In particular, several hypotheses suggest that the expression of MITF serves as a rheostat for melanoma cell phenotypes, with high levels of MITF promoting differentiation, moderate levels promoting proliferation, and low levels promoting invasiveness (26). MITF is a critical factor for expression of melanocyte pigment genes, and loss of MITF and MART1 (18) with elevated expression of AXL receptor tyrosine kinase and NGFR (20, 21, 27) has been implicated in switching of melanoma cells from a proliferative to an invasive phenotype. Given that invasive cells are more resistant to BRAF inhibition (28), this phenotypic switching may serve as a mechanism to develop resistance to such inhibitors. Single-cell studies of human melanomas at different stages of treatment have revealed that most tumors contain heterogeneous subsets of low and high MITF-expressing cells, with the relative composition of such cells shifting toward low MITF-expressing cells in BRAF or MEK inhibitor–treated patients (27).

This case report suggests that inflammation-induced dedifferentiation of melanoma cells may play a role in evading ACT in a subgroup of patients. In the clinical trial, 13 patients were evaluable for tumor response and 9 patients (69%) showed evidence of tumor regression; however, they all had disease progression within 6 months (29), similar to other TCR-engineered ACT trials (30). There were 4 patients with biopsies at progression, and only this case had evidence of dedifferentiation. This highlights the critical need for more studies investigating the underlying mechanisms of ACT resistance in human patients. These data are important in suggesting that future ACT protocols may benefit from the simultaneous targeting of multiple tumor antigens using mixed T-cell populations and from broadening of the response by concurrent administration of checkpoint inhibitors, so that dedifferentiation and loss of critical melanocytic antigens such as MART1 and gp100 do not contribute to treatment resistance. In addition, concurrent reduction of the T cell–induced inflammatory milieu that is associated with elevated NGFR expression may also enhance therapeutic efficacy.

This was a pilot clinical trial for patients with metastatic melanoma (Clinicaltrials.gov number NCT00910650; refs. 31–33). The study protocol and its amendments were approved by the University of California, Los Angeles, Institutional Review Board. The patient provided written informed consent.

The patient underwent leukapharesis to collect peripheral blood mononuclear cells (PBMC). T cells and DCs were manufactured as previously described (31–33). Briefly, PBMCs were activated using anti-CD3 antibodies (OKT3) and IL2 to enrich for T cells and retrovirally infected with the MSGV1-F5AfT2AB vector. Cells were infused fresh. DCs were produced by ex vivo differentiation of PBMCs using granulocyte/monocyte colony stimulating factor and IL4 and pulsed with the MART1_26-35 anchor–modified immunodominant peptide in the HLA-A2.1 allele (29). Full details are provided in the Supplementary Appendix.

Formalin-fixed paraffin-embedded baseline and on-therapy biopsies were cut and stained with hematoxylin and eosin, anti-S100, anti-Melan-A, anti-CD8, and anti-NGFR antibodies at the UCLA Anatomic Pathology Immunochemistry and Histology Laboratory as previously described (34).

Human melanoma cell lines were cultured as previously described (17). For in vitro dedifferentiation experiments, MART1-specific T cells and untransduced T cells were cocultured with M397 cells, and supernatant was obtained after 24 hours. The conditioned media were diluted 1:1 in fresh culture media for use in subsequent cultures. One million M397 cells were then seeded and treated with TNFα (1,000 IU/mL), conditioned media from MART1 F5 TCR-transduced T cells, conditioned media from mock untransduced T cells, or no treatment for 3 days. To test the reversibility of dedifferentiation, the treatment media were removed from some replicates on day 3 and replaced with fresh culture media for 7 days. All cells were stained with anti-NGFR and anti-MART1 fluorescently conjugated antibodies for analysis by flow cytometry. Intracellular staining was performed using anti-TNFα antibodies after fixation using the Fixation/Permeabilization Solution Kit (BD Biosciences). Samples were analyzed using an LSR-II flow cytometer (BD Biosciences), and the data were analyzed using FlowJo software (TreeStar, Inc.). Full details of all in vitro assays are provided in the Supplementary Appendix.

RNA was extracted for three melanoma cell lines treated with either DMSO or TNFα and processed for library preparation. Samples were sequenced by 50-bp single-end sequencing on an Illumina HiSeq 2500 sequencer. Reads were mapped using TopHat2 v2.0.9 (35), and gene expression was quantified using Cufflinks v2.2.1 and Cuffnorm (36). GSEA was used to assess enrichment across several gene sets on the Molecular Signatures Database (MSigDB) and the pathways for IPRES (19). Full details are provided in the Supplementary Appendix.

No potential conflicts of interest were disclosed.

Conception and design: A. Mehta, Y.J. Kim, P.C. Tumeh, C. Puig-Saus, A. Ribas

Development of methodology: A. Mehta, Y.J. Kim, L. Robert, B. Comin-Anduix, B. Berent-Maoz, A.J. Cochran, P.C. Tumeh, A. Ribas

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Mehta, L. Robert, B. Berent-Maoz, A.J. Cochran, J.S. Economou, P.C. Tumeh, A. Ribas

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Mehta, Y.J. Kim, L. Robert, J. Tsoi, B. Comin-Anduix, A.J. Cochran, P.C. Tumeh, C. Puig-Saus, A. Ribas

Writing, review, and/or revision of the manuscript: A. Mehta, Y.J. Kim, L. Robert, B. Comin-Anduix, A.J. Cochran, P.C. Tumeh, C. Puig-Saus, A. Ribas

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y.J. Kim, A.J. Cochran, A. Ribas

Study supervision: C. Puig-Saus, A. Ribas

This study was funded in part by the UCLA/Caltech Medical Scientist Training Program (to A. Mehta and Y.J. Kim), NIH National Research Service Award CA183220 (to A. Mehta), NIH grants R35 CA197633 and P01 CA168585, the Parker Institute for Cancer Immunotherapy, the Ressler Family Fund, the Samuels Family Fund, the Grimaldi Family Fund, and the Garcia-Corsini Family Fund (to A. Ribas). We thank the JCCC Flow Cytometry Core Facility supported by National Institutes of Health awards P30 CA016042 and 5P30 AI028697. We would like to thank Nathan Phillips and the UCLA Technology Center for Genomics and Bioinformatics Core Facility for their assistance in processing samples for genomic analysis. We would also like to thank Daniel E. Speiser for his helpful discussions about this work and critical review of our manuscript.

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.