Abstract LB-274: Primary tumor gene expression signature predicts long-term outcomes in primary melanoma: A prospective multicenter study

• Published in: Cancer research (Chicago, Ill.) Vol. 80; no. 16_Supplement; p. LB-274
• Main Authors: Garg, Manik, Courturier, Dominique-Laurent, Fonseca, Nuno A., Wongchenko, Matthew, Yan, Yibing, Nsengimana, Jeremie, Bishop, Tim, Newton-Bishop, Julia, Middleton, Mark, Corrie, Pippa, Adams, David J., Brazma, Alvis, Rabbie, Roy
• Format: Journal Article
• Language: English
• Published: 15-08-2020
• ISSN: 0008-5472; 1538-7445
• DOI: 10.1158/1538-7445.AM2020-LB-274

Abstract
Purpose: Adjuvant therapies prolong survival in patients with stage III melanoma. However, biomarkers are needed to stratify patients with primary melanoma at highest risk for metastases which could help minimize exposure to potentially irreversible toxicities and allow for rational clinical trial designs in the adjuvant setting.
Methods: We analyzed data from 194 RNA-sequenced primary cutaneous melanomas from patients with stage IIB-IIIC disease recruited to the multicenter AVAST-M phase III randomized trial. By undertaking covariate-corrected differential expression between patients experiencing distant metastasis (n=89) versus no-metastases (n=105), we identified metastasis-associated genes of which 121 were externally validated and made up our predictive signature, “Cam_121”. Several machine learning classification models were trained using nested leave-one-out cross validation (LOOCV) to test the signature's capacity to predict metastases. Univariate and multivariate Cox proportional hazard regression survival analyses were performed. The signatures' predictive accuracy was further externally validated in an independent population-controlled cohort study measuring melanoma-specific survival (Leeds Melanoma Cohort, n=687).
Results: The signature distinguished patients with distant recurrence from those without across multiple machine learning models (sensitivity=0.64, specificity=0.79, accuracy=0.72, kappa=0.43) and performed significantly better than any of the models trained with the clinical covariates alone (pAccuracy =4.92x10-3), as well as those trained with predictive signatures selected from two published datasets (Decision-Dx MelanomaTM and Leeds Melanoma Cohort 150 genes). The signature also correlated with progression-free survival (PFS), overall survival (OS) and melanoma-specific survival (MSS) while retaining its predictive accuracy following multivariate correction (PFS: HR=0.49 (0.35-0.69), p=2.8x10-5, OS: HR=0.6 (0.42-0.86), p=0.005 and MSS: HR=0.57, p=8x10-5). Importantly, we found that the median signature expression score positively correlated with measures of immune cell infiltration, with a lower score representing a poorer tumor lymphocytic infiltration and worse long-term prognosis.
Conclusions: We have identified Cam_121 a primary melanoma expression signature that outperforms currently available predictive signatures. The signature confirms (using unbiased approaches) the central prognostic importance of immune cell infiltration in long-term patient outcomes and could help identify primary melanoma patients at highest risk of metastases and poor survival who might benefit most from adjuvant therapies.
Citation Format: Manik Garg, Dominique-Laurent Courturier, Nuno A. Fonseca, Matthew Wongchenko, Yibing Yan, Jeremie Nsengimana, Tim Bishop, Julia Newton-Bishop, Mark Middleton, Pippa Corrie, David J. Adams, Alvis Brazma, Roy Rabbie. Primary tumor gene expression signature predicts long-term outcomes in primary melanoma: A prospective multicenter study [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr LB-274.