Background: Measures of the adaptive immune response have prognostic and predictive associations in melanoma and other cancer types. Specifically, intratumoral T cell density and function have considerable prognostic and predictive value in skin cutaneous melanoma (SKCM). Less is known about the significance of tumor-infiltrating B cells in SKCM. Our goal was to understand the prognostic and predictive value of B cell phenotypic subsets in SKCM using RNA sequencing.
Methods: We used our previously published algorithm, V'DJer, to assemble B cell receptor (BCR) repertoires and estimate diversity from short-read RNA sequencing (RNA-seq). We applied machine learning-based cellular phenotype classifiers to measure relative similarity of bulk tumor sample gene expression profiles and different B cell phenotypes. We assessed these aspects of B cell biology in 473 SKCM from the Cancer Genome Atlas Project (TCGA) as well as in RNA-seq data corresponding to tumor samples procured from patients who received CTLA-4 and PD-1 inhibitors for metastatic SKCM.
Results: We found that the BCR repertoire was associated with different clinical factors, such as tumor tissue site and sex. However, increased clonality of the BCR repertoire was favorably prognostic in SKCM and was prognostic even after first conditioning on various clinical factors. Mutation burden was not correlated with any BCR measurement, and no specific mutation had an altered BCR repertoire. Lack of an assembled BCR in pre-treatment tumor tissues was associated with a lack of anti-tumor response to a CTLA-4 inhibitor in metastatic SKCM.
Conclusions: These findings suggest an important prognostic and predictive role for B cell characteristics in SKCM. This has implications for melanoma immunobiology and potential development of immunogenomics features to predict survival and response to immunotherapy.
Keywords: B cells; Cancer; Immunology; Machine learning; Melanoma.