Proteogenomic analysis of lung adenocarcinoma reveals tumor heterogeneity, survival determinants, and therapeutically relevant pathways

Abstract

We present a deep proteogenomic profiling study of 87 lung adenocarcinoma (LUAD) tumors from the United States, integrating whole-genome sequencing, transcriptome sequencing, proteomics and phosphoproteomics by mass spectrometry, and reverse-phase protein arrays. We identify three subtypes from somatic genome signature analysis, including a transition-high subtype enriched with never smokers, a transversion-high subtype enriched with current smokers, and a structurally altered subtype enriched with former smokers, TP53 alterations, and genome-wide structural alterations. We show that within-tumor correlations of RNA and protein expression associate with tumor purity and immune cell profiles. We detect and independently validate expression signatures of RNA and protein that predict patient survival. Additionally, among co-measured genes, we found that protein expression is more often associated with patient survival than RNA. Finally, integrative analysis characterizes three expression subtypes with divergent mutations, proteomic regulatory networks, and therapeutic vulnerabilities. This proteogenomic characterization provides a foundation for molecularly informed medicine in LUAD.

Keywords: TP53; cancer; immune; lung adenocarcinoma; proteogenomics; proteomics; subtype; survival; transcriptomics; whole genome.

Graphical abstract

Figure 1

Subtyping LUAD by whole-genome somatic signatures (A) Clustering of LUAD tumors by somatic single-nucleotide variant (SNV), insertion or deletion (indel), and structural variant (SV) signatures. Columns are tumors, and rows are somatic signature values or patient/tumor features (n = 87). Patients and tumor features are tested for association with somatic signature subtypes: ANOVA (∗p < 0.05); χ² test (ˆp < 0.05). Di-atc and ragged-anast refer to dispersed intra-alveolar tumor cells pattern and ragged-anastomosing glands pattern, respectively. (B–G) Transition/transversion ratios, SV deletions, SV inversions, TP53 and phospho-TP53 expression from RPPA, and mutant TP53 RNA expression scores compared across somatic signature subtypes. Boxplot lines indicates 25%, 50%, and 75% percentiles, and points are tumors with horizontal jitter added for visualization. p' refers to Wilcoxon rank-sum test on structurally altered versus transition subtype. p'' refers to Wilcoxon rank sum on structurally altered versus transversion subtype. p refers to Wilcoxon rank-sum test on structurally altered versus other subtypes. See also Figures S1–S4 and Table S1.

Figure 2

Correlation of gene-wise and tumor-wise RNA and protein expression (A) Gene-wise RNA and protein expression correlations in the APOLLO cohort: 87 tumors and 7,472 co-detected genes. (B) Gene-wise RNA and protein correlation comparison between APOLLO and CPTAC cohorts: 106 tumors, over 6,729 common, expressed genes between the cohorts. p refers to Spearman correlation test. (C) Pathway enrichments according to gene-wise RNA and protein expression correlation in the APOLLO cohort. (D) Tumor-wise RNA:protein expression correlation in APOLLO (n = 87) and in CPTAC cohorts (n = 105). Columns indicate individual tumors. Rows are molecular features except for manual slide review features of tumor cellularity percentage, stroma percentage, and grade. Tumor features tested for association with tumor-wise RNA:protein correlation by Spearman correlation tests for continuous variables and by Kruskal-Wallis tests for categorical variables. See also Figure S5 and Table S1.

Figure 3

RNA and protein expression determinants of patient survival (A) Comparison of log hazard ratios between RNA expression and protein expression on matched genes in APOLLO cohort (n = 87). Points outside the axis scale (less than 2 or greater than 2) are plotted as −2 and 2, respectively. ρ and p refer to the Spearman rank correlation coefficient and p value, respectively. “Other” refers to genes not associated with survival. (B) Overall and metastasis-free survival in APOLLO cohort (n = 83 with follow up), with high and low referring to a 50th percentile split on the respective signature score. p refers to Cox proportional hazards Wald test of the continuous signature score. Top panels are signatures based on survival proteins and survival RNAs, and bottom panels are signatures based on survival RNA proteins. (C and E) Gene-wise RNA:protein correlation across survival gene sets compared by Kruskal-Wallis tests, p∗. (D) CPTAC cohort survival following same layout as (B). 50th percentile split for visualization was based on entire cohort, n = 106, and plotted for those with follow up, n = 101. See also Figure S6 and Table S2.

Figure 4

Molecular subtype characteristics and survival outcomes (A) RNA expression subtypes. Tumors (n = 87) appear in columns and clinical and genomic features in rows. Protein refers to MS proteomics expression. RNA refers to RNA-seq expression. mut refers to non-silent gene mutations. Immune and stromal scores refer to RNA-based ESTIMATE scores. Continuous features analyzed by Kruskal-Wallis tests. Categorical features analyzed by Fisher’s exact tests. RNA and protein expression compared by Spearman correlation tests. (B) Proteogenomic expression analysis of RNA expression subtypes. Columns indicate molecular enrichment (RNA, protein, phosphoprotein by subtype), and rows indicate gene sets. Phosphoprotein expression is from combined RPPA and MS platforms. (C) Survival outcomes of RNA expression subtypes and histological subtypes, analyzed by log rank tests (p). See also Figure S6.

Figure 5

Proteogenomic network characterization of subtypes (A) Kinase enrichments based on known kinase-substrate links to measured MS-based proteomics and RPPA phosphoresidues from APOLLO cohort (n = 87). Triangles indicate significant kinase enrichments in either PI, TRU, or PP subtypes (combined FDR < 0.01). (B–D) Regulatory networks of subtypes. Box to right indicates network layout (top) and node/edge shape, size, and color schemes (bottom): node shapes indicate molecule types or pathways; red outlines identify nodes significantly associated with the subtype (gray otherwise); blue-to-red shading indicates node association/enrichment with subtype (gray denotes no measured data); enlarged diamonds and “vee” shapes indicate enriched kinases and mutated genes, respectively; red outlined triangles with italic text labels indicate TFs identified from TF enrichment analysis; and edge color represents types of protein-protein or protein-pathway links. Values and data sources for each network node are listed in Table S3. See also Figure S7.

Figure 6

Proteogenomic features associated with subtype networks Individual features associated with LUAD subtypes and networks (related to Figure 5). Sample-wise somatic alterations in KEAP1, STK11, SMARCA4, TP53, KRAS, and EGFR with black triangles to the right indicating significant enrichment of molecular alterations in the given subtype (Fisher’s exact test p < 0.05) and black diamonds indicating significantly recurrent somatic mutations in the subtype (MutEnricher FDR < 0.1). Additional panels display select individual molecular features associated with the subtypes (see STAR Methods for molecular type statistics). Asterisks indicate feature measurement from the RPPA platform.