Fig. 3: Molecular features of metastasis and drug resistance are co-expressed in pre-existing NSCLC cell populations.

A Scheme of in vivo brain metastatic selection yielding PC9-BrM4, C2, and R2 bulk cell populations from the parental PC9 line. B scRNA-seq Uniform Manifold Approximation and Projection (UMAP) analysis of ~6000 cells from each of the bulk cell populations in A. C Same cells in B were re-colored based on distinct cell sub-populations as defined by UMAP analysis. D Stacked graph indicating the % of the sub-populations identified in C within the parental PC9, PC9-BrM4, C2, and R2 bulk populations. E Bar graph (log- scaled) plotting the % of C2 or R2 enriched sub-populations across the PC9, PC9-BrM4, C2, and R2 bulk populations. F Pathways most significantly upregulated in R2 enriched cell sub-populations (1, 5, and 8) compared to the C2 cells based on scRNA-seq analysis. Enrichment score is calculated by Metacore and plotted as –log₁₀(P-value). G Differential mean expression of the dual Metastasis-Resistance (MetRes) signature is shown as a violin plot for single tumor cells collected from EGFR-mutant NSCLC patients (N = 20 patients)²⁴. TN = treatment-naïve patients (N = 457 cells). PR = Patients with partial response after systemic therapy (N = 557 cells). PD = patients with progressive disease after systemic therapy (N = 1088 cells). P-value calculated by Welch t-test (two-sided). H Human LUADs from TCGA (N = 449) were classified as “high” or “low” based on whether expression of the MetRes signature was above or below the median, respectively. Kaplan–Meier curves were generated for the incidence of death of “high” vs. “low” groups. P-value calculated by log-rank test. The MetRes signature was generated for each sample (single cells in G; bulk tumors in H) by calculating the average expression of all MetRes genes identified in Supplementary Data 2.