A prognostic signature for lung adenocarcinoma in people who have never smoked

Abstract

Knowledge of tumor cell dynamics can inform prognosis and treatment yet is largely lacking for lung adenocarcinoma in people who have never smoked (NS-LUAD). With RNA-seq data from 684 NS-LUAD and validation in an independent dataset, we identified three subtypes with distinct phenotypic traits and cell compositions. Additional genomic and histological data further characterized the subtypes. 'Steady', marked by low proliferation, high alveolar cell fraction, moderate-to-well differentiation, and fewer driver genes' alterations, is linked to prolonged survival and low immune evasion. 'Proliferative' shows high proliferation markers, TP53 mutations, and gene fusions. 'Chaotic', with high epithelial-to-mesenchymal transition markers, has the worst prognosis even within stage I tumors. Lacking known molecular or histological characteristics, this aggressive subtype is solely identified by transcriptomic data. A 60-gene signature recapitulates the overall classification and strongly predicts survival even within subgroups based on tumor stage or known genomic features, emphasizing its potential for improving NS-LUAD prognostication in clinical settings.

Fig. 1. Transcriptomic clusters in the Sherlock cohort.

(A) Consensus matrices for NMF rank 2 to 10. Cophenetic coefficient from consensus matrix (left) and silhouettes from basis, coefficient and consensus matrices (right) were computed from 80 runs for each rank value. (B) Clustering of subtype-specific genes in the three expression subtypes are defined by NMF. The top panel shows cluster assignment, driver mutations and fusions, data set, genetic ancestry, sex and S-LUAD subtypes. The right panel indicates the subtype specific genes.

Fig. 2. Comparison of clinical and histopathological features across NS-LUAD expression subtypes.

A-C, Comparison of (A) tumor stages, (B) histologic grades, and (C) signet ring cell features across the NS-LUAD expression subtypes. P-values for comparison of morphological features from chi-squared test are shown. D-E, Violin plots of (D) LUAD histology scores and (E) mix-histological lineage scores across the NS-LUAD expression subtype. Mean values are indicated by the yellow lines. P-values from two-sided Mann–Whitney U-test are shown.

Fig. 3. Cell type deconvolution in the Sherlock cohort.

A, Summary of the average proportions of cell types in NS-LUAD expression subtypes. B-C, Violin plots of the proportions of (B) epithelial cells and (C) fibroblasts across NS-LUAD expression subtypes. D, Summary of the average proportions of epithelial cell types relative to all epithelial cells. E-F, Violin plots of the proportions of (E) AT2 and (F) basal cells relative to all epithelial cells across NS-LUAD expression subtypes. G-I, Comparison of the relative proportions of (G) COL10A1+, (H) COL4A1+, and (I) non-malignant lung fibroblasts across NS-LUAD expression subtypes. The fibroblast estimates are relative differences in abundances and cannot be compared across cell types. Mean values are indicated by the yellow lines in the violin plots. P-values from two-sided Mann-Whitney U-test are shown.

Fig. 4. Gene fusion events in the Sherlock cohort.

(A) Comparison of the total numbers of gene fusions per sample across NS-LUAD expression subtypes. Mean values are indicated by the yellow lines. P-value from two-sided Mann-Whitney U-test is shown. (B) Summary of types of all detected gene fusions. (C) Summary of frequent gene fusions between pairs of protein-coding genes. The bar plot indicates the number of tumors containing gene fusion events. Colors indicate fusion types. Recurrent in-frame fusions of protein-coding genes are indicated by the star signs(*). (D) Summary of genes most frequently involved in fusions between protein-coding genes. The bar plot indicates the number of tumors in which gene fusions involving the listed genes are detected. Colors indicate fusion types. (E) Summary of genes most frequently involved in fusions between protein-coding genes within NS-LUAD expression subtypes. The bar plots indicate the percentage of tumors in which gene fusions involving the listed genes are detected. Colors indicate fusion types.

Fig. 5. Association between NS-LUAD expression subtypes and overall survival in the Sherlock cohort.

A-D, Kaplan-Meier survival curves for overall survival stratified by NS-LUAD expression subtypes in (A) all patients, (B) stage I patients, (C) patients classified as TRU subtype²⁶, and (D) TP53-wt patients. P-values and hazard ratios (HR) were calculated using Cox proportional hazards models adjusting for age, sex, and tumor stage (the analysis restricted to tumor stage I did not include tumor stage in the model).

Fig. 6. Association between NS-LUAD expression subtypes and overall survival in the Genome Institute of Singapore (GIS) cohort²².

A-B, Kaplan-Meier survival curves for overall survival stratified by NS-LUAD expression subtypes in (A) all patients and (B) TP53-wt patients. P-values and hazard ratios (HR) are calculated using Cox proportional hazards models adjusting for age, sex, and tumor stage. (C) C-indexes of Cox models predicting the overall survival using different sets of predictors in the training (Sherlock) and testing (GIS) data sets. P-values testing for the effect of adding the expression subtypes to models of single modality were calculated using likelihood ratio test.