Extended Data Fig. 5: Human brain metastatic SCLC contains a subpopulation enriched for neural gene signatures.

a, Integrated UMAP embeddings of all cells from a cohort of n = 12 patients with SCLC brain metastases, colored by their assigned cell-types (left) or patient (right). b, as in a, but of cells from n = 16 patients with lung primary, recurrent, and non-brain-metastatic SCLC lesions (data made publicly available by Chan et al., 2021). c, Spearman correlation of cell programs classifying malignant cells in a, grouped into 6 distinct metaprograms (indicated by boxes within the co-correlation plot). d, as in c, but for malignant cells in b grouped into 9 distinct metaprograms. e, Normalized gene contribution of top defining genes (rows) for each respective metaprogram in the brain-metastatic cohort (columns). Annotated biological functions corresponding to these genes, individual metaprograms, or groups of metaprograms are shown on the right. Metaprogram 3 is defined by genes driving neural-like differentiation. f, Scaled correlation of brain-metastatic cohort metaprograms (columns) with annotated transcriptional hallmarks of tumor heterogeneity derived by Gavish et al.⁴⁶ (rows). Metaprogram 3 demonstrates a strong correlation with neural-like hallmarks. g, as in e for the primary, recurrent, and non-brain metastatic cohort. h, as in f for the primary, recurrent, and non-brain metastatic cohort.