Cell lineage identification of 96,505 single cells from 26 patients with lung cancer treated with ICI.

a, Workflow of sample collection and single-cell analysis of lung cancer patients treated with ICI. b, UMAP plot of 96,505 single cells from 33 samples acquired from 26 advanced lung cancer patients, colored by clusters. AMB cells, Ambiguous cells. c, Dot plot of mean expression of canonical marker genes for cell lineages. d, Proportions of the cell lineages in NSCLC tissue from core patients shown by individual samples aligned with clinical data. Labels for origins indicate LN, Metastatic lymph node; Lung, Tumor lung; PE, Malignant pleural effusion; Liver, Metastatic liver. e, Box plot of the percentage of cell lineages in responder and non-responder groups. Label represents p-value calculated via two-tailed Student’s t-test. Each box represents the median and the interquartile range (IQR, the range between the 25th and 75th percentile), whiskers indicate the 1.5 times of IQR.

Clinical overview of NSCLC patients treated with ICI.

Classification and characterization of CD4+ and CD8+ T cell subtypes.

a, Prediction strategy to classify CD4+, CD8+ T, and NK cells by applying ADT data from lung cancer. b, UMAP plot of CD4+ and CD8+ T cells, colored by clusters. c, Dot plot of mean expression of selected CD4+ (left) and CD8+ (right) T cell marker genes in each cell cluster. d, Box plot of the percentage of CD4+ and CD8+ T cell types within total CD4+ plus CD8+ cells for sample groups representing responses to ICI. Label represents p-value calculated via two-tailed Student’s t-test. Each box represents the median and the IQR, whiskers indicate the 1.5 times of IQR. e, Association of T cell functional features with clonal expansion. Dot size depicts the relative clone size of each cell, which is divided by the total number of CD4+ and CD8+ T cells, respectively. Color indicates the cell lineage. f, Comparisons of proportional changes in cell subtypes along ICI responses within each immune cell lineage. The quantitative values shown on the axis represent the mean difference in % cell proportions between sample groups. Dot size and color represent -log (p-value) for responder vs. non-responder and non-PD vs. PD, respectively. The lower left quadrant shows cell types overrepresented in the poor responder groups, while the upper right quadrant indicates cell types overrepresented in the better responder groups. p-value, two-tailed Student’s t-test.

Systemic evaluation of immune cell dynamics associated with response to ICI.

a, Heat map with unsupervised hierarchical clustering (left) and depicting significance (right) of proportional changes in cell subtypes within total immune cells. Proportional changes were compared for multiple ICI response groups. Color represents the -log (p-value) determined using two-tailed Student’s t-test. b, Distribution map for each cell type across individual samples aligned with clinical data. Color represents Ro/e score calculated using the chi-square test.

Single-cell tumor signatures associated with response to ICI.

a, Volcano plot of expression difference for responder vs. non-responder, PR vs. PD, and PR vs. SD in 12,975 malignant cells from 11 core patients. The log fold change indicates the difference in the mean expression level for each gene. The significance level was determined using two-tailed Wilcoxon Rank Sum test. b, Relative sum of loadings for all NMF factors contributed to malignant cells from 11 core patients across RECIST, tissue origins, and cancer subtypes, respectively. c, Selection of RECIST-enriched NMF programs. d, Enrichment of NMF programs for RECIST groups. Color represents the z-transformed odds ratio. e, Expression map of RECIST-specific scINSIGHT modules across individual samples aligned with clinical data. Color represents the z-transformed mean expression of genes contributing to each module. f, Enrichment of RECIST-specific gene modules for RECIST groups. Color represents the z-transformed odds ratio. g, Hierarchical clustering of pairwise similarities between tumor signatures. INT and UNION, intersection and union of DEGs for responder vs. non-responder, PR vs. PD, and PR vs. SD in Fig. 4a. h, Functional categories and i, transcription factors of the selected tumor signatures, analyzed by Metascape.

Multivariate overall survival analysis of tumor signature genes.

Combination of tumor signatures and immune index classifying the response to ICI.

a, Heat map of relative contribution of tumor signatures and immune index across individual samples aligned with clinical data. Mean expression of each tumor signature and the percentage of each immune index are divided by the maximum value across samples. INT and UNION, intersection and union of DEGs for responder vs. non-responder, PR vs. PD, and PR vs. SD in Fig. 4a. b, Univariate regression analysis of immune and tumor signatures for ICI response, together with clinical variables. c, ROC analysis of combinatorial index to classify responder and non-responder. p-value, two-tailed Wilcoxon Rank Sum test. P-values adjusted with the Benjamini-Hochberg correction; UNION.down + B/Plasma, 0.22; INT.down + B/Plasma, 0.22; UNION.down + CD4+ Treg, 0.26; INT.down + UNION.down, 0.26; INT.down, 0.26; UNION.down, 0.26.