Fig. 4: GPS correlates with CD8⁺ T cell infiltration and retains its predictive capacity in a pan-cancer ICB cohort.

a The t-SNE dimensionality reduction was applied to show the distribution and dissimilarity of all the 32 solid malignancies. b Among the pan-cancer samples, the 3 GPS-genes display a similar distribution of nonsynonymous mutation with the “category I” cancer types which are more responsive to ICB therapy. c An integrated mutational spectrum was plotted to intuitively display the mutation frequencies and distributions of the 3 GPS-genes in pan-cancer with “cancer categories (I or II)”. STK11, the only “ICB-resistant” gene in GPS, is specifically and most frequently mutated in LUAD (arrow point) compared to other cancer types. d A stacked barplot and a Sankey diagram were combined to depict their immune cell distribution and classification towards different GPS groups. e Four algorithms demonstrated CD8⁺ T cell infiltration was progressively and significantly elevated from GPS-neg to GPS-zero to GPS-pos group among all “category I” cancer samples. f A substantial number of immunotherapy-predicted pathways and immunity-related processes showed significant differences among different GPS groups. g In the Samstein cohort of 1,661 patients with various cancer types who received ICB therapy, significant differences of overall survival were observed among different GPS groups, and GPS-pos patients showed the best prognosis while GPS-neg patients the worst. h A forest plot of adjusted HRs illustrated that GPS remains as an independent risk factor for prognosis with the adjustment of various potential confounders [GPS-zero vs GPS-pos: HR = 1.365 (1.089–1.711), p = 0.007; GPS-neg vs GPS-pos: HR = 1.607 (1.113–2.320), p = 0.011]. *p < 0.05; **p < 0.01; ***p < 0.001.