Melanoma is an immunogenic cancer and often contains many different types of immune cells. We have known for many years that the presence of these immune cells is associated with better patient outcomes and rare cases of spontaneous tumour regression. More recently these immune cells are linked with improved response to immune checkpoint inhibitors (a new suite of immunotherapies). However, resistance to these immunotherapies remains a major challenge -  up to 60% of patients do not respond to these treatments, and another 25% of patients will fail treatment after an initial response. This report examined over 90 melanoma biopsies that were collected from 68 patients, prior to and during immunotherapy.

Firstly, we set out to define expression signatures that could help clinicians predict which patients are likely to respond to an immunotherapy that inhibits an immune regulator known as PD1.  We were unable to identify a predictive signature using our patient data and when we applied previously published predictive signatures, none of these predictors accurately determined patient responses in our patient cohort. We were curious as to why these predictive markers could not be validated and we explored the response patterns in our patients in more detail. We noticed that a substantial proportion of patients underwent heterogeneous responses to therapy – in other words, within individual patients some tumours continued to grow while other tumours responded to treatment. Thus, it will be difficult to develop a single marker that accurately predicts a patient’s response to immunotherapy, based solely on the analysis of a single tumour. We need a comprehensive approach to predicting patient responses to therapy – an approach that includes clinical, patient and tumour features.

We also explored why some tumours fail to respond to immunotherapy – we looked at patients treated with PD1 inhibitors.  We knew that tumours vary in the type and number of nearby immune cells and we accounted for this variation by comparing tumours with similar immune content, based on a simple, but accurate cytolytic expression score. This analysis revealed that the downregulation of MHC class I was common in treatment resistant melanoma and that this downregulation was associated with a specific melanoma subtype. In this subtype, melanoma cells tend to show features of de-differentiation, i.e. they have reverted to a less mature cell type. Melanoma de-differentiation and treatment resistance was linked to TGFß pathway activity and we confirmed that TGFß signalling diminished the expression of MHC class I and reduced the recognition of melanoma cells by their matched immune cells.

Collectively, our study provides important insights into the limitations of developing predictive signatures, highlights the importance of considering the heterogenous nature of melanoma and adds valuable new information on a common mechanism of immunotherapy resistance in melanoma. Reversing the de-differentiation of melanoma and restoring MHC class I expression is a critical step in providing new treatment opportunities for melanoma patients.