Baseline metabolic signatures predict clinical outcomes in immunotherapy-treated melanoma patients: a pilot study

Abstract

Background: Immune checkpoint inhibitors (ICIs) have improved the metastatic melanoma (MM) treatment. However, a significant proportion of patients show resistance to immunotherapy, and predictive biomarkers for non-responders or high-risk recurring patients are currently lacking. Recent studies have shown that tumor-related metabolic fingerprints can be useful in predicting prognosis and response to therapy in various cancer types. Our study aimed to identify serum-derived metabolomic signatures that could predict clinical responses in MM patients treated with ICIs.

Patients and methods: ¹H-NMR (proton nuclear magnetic resonance) was used to analyze the serum metabolomic profiles from 71 MM patients undergoing anti-PD-1 therapy (43 patients as first-line, 27 as second-line, 1 as third-line). Feature selection was applied to identify key metabolites within these profiles, to develop risk score models predicting overall survival (OS) and progression-free survival (PFS).

Results: A multivariable model was used to identify distinct prognostic factors for OS. Negative factors included glucose, high-density lipoprotein (HDL) cholesterol, and apolipoprotein B-very low-density lipoprotein (ApoB-VLDL), whereas glutamine and free HDL cholesterol emerged as positive factors. They were then used to construct a risk score model able to stratify patients in prognostic groups. Similarly, a separate predictive risk score model for PFS was developed, focusing solely on glucose and apolipoprotein A1 (ApoA1) HDL. Threefold cross validation resulted in mean concordance indices of 0.72 and 0.74 for PFS and OS, respectively. Importantly, this analysis was replicated in patients who received first-line ICIs. Interestingly, the prognostic score for OS included glutamine, glucose, and LDL (low-density lipoprotein) triglycerides, whereas only glucose negatively influenced PFS. In this subset, the concordance indices increased to 0.81 and 0.9 for PFS and OS, respectively.

Conclusions: Our data identified glycolipid signatures as robust predictors of distinct therapeutic outcomes in MM patients treated with ICIs. These results could pave the way for novel therapeutic approaches.

Keywords: NMR; immune checkpoint inhibitors; immunotherapy-treated melanoma patients; separate predictive risk score model; serum metabolomic profiles; tumor-related metabolic fingerprints.

Figure 1

Impact of the univariate significant features on OS. (A) Multivariate Cox-hazard regression model; (B) threefold cross validation of OS RiskScore depicting density of the concordance index; (C) Kaplan–Meier survival curves stratifying patients according to the risk score calculated using the multivariate model.

Figure 2

Impact of the univariate significant features on PFS. (A) Multivariate Cox-hazard regression model; (B) threefold cross validation of PFS RiskScore depicting density of the concordance index; (C) Kaplan–Meier survival curves stratifying patients according to the risk score calculated through the multivariate model.

Figure 3

Prognostic role of the different metabolites in patients treated with first-line ICI. (A) Multivariable Cox-hazard regression model; (B) threefold cross validation of OS RiskScore depicting density of the concordance index; (C) Kaplan–Meier survival curves comparing OS of patients stratified according to the risk score.

Figure 4

Predictive role of the different metabolites in patients treated with first-line ICI. (A) Multivariable Cox-hazard regression model; (B) threefold cross validation of PFS RiskScore depicting density of Concordance index; (C) Kaplan–Meier survival curves comparing PFS of patients stratified according to the risk score.