Extended Data Fig. 3: Chemotherapy triggers effector CD4⁺ and CD8⁺ T cell responses.

a, In situ mIF analysis of exhausted CD8⁺ T cells in matched NPC tumors before and after chemotherapy. Representative images (left) and quantification of changes in CD8⁺ T cell density, exhausted CD8⁺ T density and frequency after chemotherapy in responders and nonresponders (right) were shown. b, Expression of TCF7, LAG3, and HAVCR2 in exhausted CD8⁺ T (Cluster_5) before and after chemotherapy from samples in the single-cell dataset (n = 15 pairs). The bottom and top of the boxes were 25th and 75th percentiles, respectively (interquartile range). Whiskers encompassed 1.5 times the interquartile range. c, Degree of changes in b between responders and nonresponders (n = 15). The boxplot indicated median (center), 25th and 75th percentiles (bounds of box), and minimum and maximum (whiskers). d, Absolute numbers of preexisting and newly-emerged CD8⁺ and CD4⁺ T cell clones after chemotherapy in samples with scTCR-seq data (n = 3 pairs). e,f, FACS analysis of IFN-γ⁺ T_H1 (e) and T_FH (f) in NPC tumors before and after chemotherapy. Representative plot (left) and quantification of changes in T_H1 and T_FH after chemotherapy in responders and nonresponders (right) were shown. g,h, In situ mIF analysis of T_H1-like cells (g) and T_FH (h) in matched NPC tumors before and after chemotherapy. Representative images (upper) and quantification of changes in CD4⁺ T cell density, T_H1-like density and frequency, T_FH density and frequency after chemotherapy in responders and nonresponders (bottom) were shown. i, Correlations between T_H1-like (x axis, left), T_FH signature expression (x axis, right) and CTL signature expression (y axis) in Cohort_2. The correlation coefficient and two-sided P value were calculated by Spearman correlation analysis. j, Kaplan–Meier curves for DFS in NPC patients stratified by high versus low expression of T_H1-like (left) and T_FH (right) signatures in Cohort_2. Patients were dichotomously divided by median expression value. HRs and 95% CIs were calculated using a Cox regression model (two-sided P value by log-rank test). Data are the mean ± s.d. A two-sided paired-sample t-test was computed for (a,e-h), a two-sided Wilcoxon test for (b), a two-sided student’s t-test for (c).