Transcriptional Profiling Reveals Mesenchymal Subtypes of Small Cell Lung Cancer with Activation of the Epithelial-to-Mesenchymal Transition and Worse Clinical Outcomes

Abstract

While molecular subtypes of small cell lung cancers (SCLC) based on neuroendocrine (NE) and non-NE transcriptional regulators have been established, the association between these molecular subtypes and recently recognized SCLC-inflamed (SCLC-I) tumors is less understood. In this study, we used gene expression profiles of SCLC primary tumors and cell lines to discover and characterize SCLC-M (mesenchymal) tumors distinct from SCLC-I tumors for molecular features, clinical outcomes, and cross-species developmental trajectories. SCLC-M tumors show elevated epithelial-to-mesenchymal transformation (EMT) and YAP1 activity but a low level of anticancer immune activity and worse clinical outcomes than SCLC-I tumors. The prevalence of SCLC-M tumors was 3.2-7.4% in primary SCLC cohorts, which was further confirmed by immunohistochemistry in an independent cohort. Deconvoluted gene expression of tumor epithelial cells showed that EMT and increased immune function are tumor-intrinsic characteristics of SCLC-M and SCLC-I subtypes, respectively. Cross-species analysis revealed that human primary SCLC tumors recapitulate the NE-to-non-NE progression murine model providing insight into the developmental relationships among SCLC subtypes, e.g., early NE (SCLC-A and -N)- vs. late non-NE tumors (SCLC-M and -P). Newly identified SCLC-M tumors are biologically and clinically distinct from SCLC-I tumors which should be taken into account for the diagnosis and treatment of the disease.

Keywords: developmental trajectories; epithelial-mesenchymal transformation; mesenchymal tumors; molecular taxonomy; small cell lung cancers.

Figure 1

Transcriptional analysis to define SCLC subtypes. (A) Boxplots showing ranges of cophenetic correlation coefficients (y-axis) measured across K values (2 to 10; x-axis). Gray area indicates 90 percentiles (5% and 95%) of cophenetic scores per K value as obtained from permutation tests. (B) For six NMF clusters (NMF1 to NMF6), the top three enriched molecular functions (Hallmark gene sets in MSigDB) are shown with enrichment scores. (C) Cluster-specific transcriptional regulators are shown per NMF cluster. The level of significance was estimated by the VIPER algorithm and is shown on the x-axis (−log₁₀p). (D) Six NMF clusters (NMF1 to NMF6) are annotated SCLC-M, -I, -A, -P, -N, and -H, respectively, according to the level of expression of four transcriptional regulators (ASCL1, NEUROD1, YAP1, and POU2F3) and the signature scores of three molecular functions (EMT, immune activity, and hypoxia). (E) Kaplan–Meier survival curves of overall survival are shown for the six SCLC subtypes. Significance was estimated by log-rank test.

Figure 2

Molecular features of SCLC subtypes. Various molecular and cellular features of the six SCLC subtypes are shown as heatmaps. Along with the primary dataset used for subtype discovery (George et al., left), results for two additional cohorts (Rudin et al. (middle) and CCLE (right)) are presented. Levels of the four transcriptional regulators and three molecular functions indicated in Figure 1D are shown. Stromal-immune scores (‘ESTIMATE’) and the abundance of 10 immune and stromal cell types (‘MCPcounter’) are also shown. Relative immune activities are shown for 14 immune genes (‘Immune markers’) and six immune pathways (‘Immune signatures’). In addition, expression levels of eight EMT-related (‘EMT markers’), seven genes associated with SCLC tumorigenesis (‘NE/non-NE markers’), and three molecular functions (‘Signaling’) are shown.

Figure 3

Tumor-intrinsic features of SCLC-M and SCLC-I tumors. (A) Ten molecular features are shown as those either up-regulated or down-regulated in deconvoluted SCLC tumor epithelial cells of SCLC-M compared with those of SCLC-I. (B) Enrichment plots of the selected functions of SCLC-M and SCLC-I are shown. (C) For the seven SCLC-M cell lines, tumor-intrinsic M and I scores are shown (red and blue lines, respectively). Bar plots indicate differential M and I scores. (D) The heatmap shows selected genes highly correlated with the differential M and I scores for cell lines and primary SCLC tumors (left and right, respectively).

Figure 4

The prevalence of SCLC-M subtype in primary SCLC tumor cohort. (A) Mean CD8+ TIL scores of vimentin (+) tumors and vimentin (−) tumors. CD8+ TIL score of vimentin (+) tumors divided into SCLC-I and SCLC-M subtypes. (B) Representative immunohistochemical stains of CD8+ TIL high/Vimentin (+) (SCLC-I) and CD8+ TIL low/Vimentin (+) (SCLC-M) tumors. (scale bar; 100 μm).

Figure 5

Cross-species analysis of NE to non-NE progression of human and mouse SCLC tumors. (A) Pseudotime trajectory of murine SCLC single cell RNA-seq data shows the ordering of cells from early to late time points (Day 4 to Day 21). Black arrow shows an increase in pseudotime. (B) Cell abundance is plotted according to pseudotime and segregated cells harvested early (Days 4–7) and late (Days 11–21). (C,D) Similarly, the trajectory of human SCLC tumor epithelial cells and cellular abundance along the pseudotime are plotted. Black arrow shows an increase in pseudotime. (E) A heatmap shows the correlation in gene expression between human SCLC epithelial cells (x-axis; 81 samples) and mouse single cells (y-axis, 19,366 cells), both sorted in order of pseudotime. NE scores representing the transcriptional shift from NE to non-NE decreased consistently with pseudotime. (F) Levels of expression (y-axis, log2-scaled) are shown for NE and EMT markers as well as those belonging to Hippo/YAP1 and Notch/REST signaling pathways. Dashed lines in the Notch signaling pathway indicate Notch-inhibitory genes.