Relative expression orderings based prediction of treatment response to Anti-PD-1 immunotherapy in advanced melanoma

Abstract

Programmed cell death protein 1 (PD-1) plays a critical role in immune tolerance and evasion within the tumor microenvironment, and anti-PD-1 immunotherapy has shown efficacy in treating advanced melanoma. However, response rates vary significantly among patients, necessitating the identification of reliable biomarkers to predict treatment efficacy. Based on within-sample relative expression orderings, we analyzed RNA sequencing data from melanoma patients to construct a predictive model comprising gene pairs associated with treatment response. The model's performance was validated across multiple independent datasets and assessed for correlations with immune infiltration and survival outcomes. The constructed 15-pair model achieved a prediction accuracy of 100% in training datasets and 89.47% in validation sets. Validation in melanoma patients lacking treatment response data revealed significant differences between predicted responders and non-responders across datasets, with the model being an independent prognostic factor. Increased immune cell infiltration was observed in responders, correlating with higher expression levels of key immune checkpoint genes. The relative expression orderings-based model shows promise as a tool for predicting responses to anti-PD-1 therapy in melanoma patients, supporting personalized treatment strategies.

Keywords: Biomarker; Melanoma; REOs; anti-PD-1 immunotherapy.

Fig. 1

Kaplan-Meier (KM) survival analysis and Cox regression analysis for responder and non-responder groups. (a,b) KM survival analysis comparing survival differences between responders and non-responders predicted in the TCGA-SKCM and GSE22153 datasets. (c,d) Forest plot based on multivariate Cox regression analysis using the TCGA-SKCM and GSE22153 datasets. (e,f) Survival differences between responders and non-responders predicted by the random forest model and TIDE model in the TCGA dataset. The light shaded area displays the confidence interval of the survival curve. NR: non-responders; R: responders; CI: confidence interval.

Fig. 2

Radar plot showing differences in immune subpopulation infiltration between responder and non-responder groups. (a) TCGA-SKCM. (b) GSE22153. p-values were estimated using the Wilcoxon test: *indicates p < 0.05; **indicates p < 0.01; ***indicates p < 0.001; ****indicates p < 0.0001. NR: non-responders; R: responders.

Fig. 3

Boxplots showing expression differences of immune checkpoints between responder and non-responder groups. (a) TCGA-SKCM, (b) GSE22153. ns indicates p > 0.05; * indicates p < 0.05; ** indicates p < 0.01; *** indicates p < 0.001; **** indicates p < 0.0001. NR: non-responders; R: responders.