Integrative clinical genomics of metastatic cancer

Abstract

Metastasis is the primary cause of cancer-related deaths. Although The Cancer Genome Atlas has sequenced primary tumour types obtained from surgical resections, much less comprehensive molecular analysis is available from clinically acquired metastatic cancers. Here we perform whole-exome and -transcriptome sequencing of 500 adult patients with metastatic solid tumours of diverse lineage and biopsy site. The most prevalent genes somatically altered in metastatic cancer included TP53, CDKN2A, PTEN, PIK3CA, and RB1. Putative pathogenic germline variants were present in 12.2% of cases of which 75% were related to defects in DNA repair. RNA sequencing complemented DNA sequencing to identify gene fusions, pathway activation, and immune profiling. Our results show that integrative sequence analysis provides a clinically relevant, multi-dimensional view of the complex molecular landscape and microenvironment of metastatic cancers.

Extended Data Figure 1. Demographics of the MET500 cohort and summary of common genetic aberrations

a. Gender distribution of the MET500 cohort. b. Age distribution of the MET500 cohort. c. Bubble plot of clinically actionable genetic aberrations. Genes have been divided by putative gain-of-function (oncogene, red) or loss-of-function (tumor suppressor, blue) status. Common aberrations are defined as those observed in 5 or more MET500 analysis cohorts (Fig 1c), restricted aberrations are found in less than five analysis cohorts. Bubble area is proportional to the observed frequency of the aberration across the MET500 cohort. d. Comparison of genetic aberration frequencies (SNVs, indels, amplifications, predicted homozygous deletions) between primary (TCGA) and metastatic (MET500) tumors for select tumor-suppressors (left panels) and oncogenes (right panels). TCGA data for the primary cancer cohorts have been obtained from the cBio portal. Nominal statistical significance is based on the Fisher’s exact test. Statistically significant differences in frequencies following correction for multiple dependent tests using the Benjamini-Yekutieli procedure are indicated as circles, insignificant differences are shown as triangles.

Extended Data Figure 2. Analysis of pan-cancer metastatic transcriptomes

a. Structural rearrangements in metastatic genomes. Distribution of the number of fusions per case is plotted across the MET500 by analysis cohort (see Fig. 1c for cancer abbreviations). Y-axis is truncated at 100 fusions. b. Summary circos diagrams of predicted inactivating fusions for select tumor suppressor genes across the cohort. Arc end positions indicate the chimeric junctions; colors indicate type of rearrangement. Black: tandem duplication, blue: translocation, red: inversion, gray: signifies that multiple close junctions were detected. c. t-SNE plot for the TCGA pan-cancer meta-cohort (a random selection of cases from each primary tumor type) based on the expression of tumor-type specific marker genes (same genes as in Fig. 4a). d. t-SNE plot for the MET500 samples colored by biopsy site (same samples as in Fig. 4a, there colored by cancer type). e. Average percentile expression of tissue-specific genes in normal tissues, primary cancers, and metastases. Error-bars indicate standard deviation. Significance test have been carried out for all normal-primary and primary-mets pairs of samples, all comparisons were significant (p<0.01) according to a two-tailed t-test, with the exception of those indicated with (NS). Abbreviations: ACC, Adrenocortical Carcinoma; BLCA, Bladder Urothelial Carcinoma; BRCA, Breast Invasive Carcinoma; CESC, Cervical Squamous Cell Carcinoma and Endocervical Adenocarcinoma; CHOL, Cholangiocarcinoma; COAD, Colon Adenocarcinoma; DLBC, Lymphoid Neoplasm Diffuse Large B-cell Lymphoma; ESCA, Esophageal Carcinoma; GBM, Glioblastoma Multiforme; HNSC, Head and Neck Squamous Cell Carcinoma; KICH, Kidney Chromophobe; KIRC, Kidney Renal Clear Cell Carcinoma; KIRP, Kidney Renal Papillary Cell Carcinoma; LAML, Acute Myeloid Leukemia; LGG, Brain Lower Grade Glioma; LIHC, Liver Hepatocellular Carcinoma; LUAD, Lung Adenocarcinoma; LUSC, Lung Squamous Cell Carcinoma; MESO, Mesothelioma; OV, Ovarian Serous Cystadenocarcinoma; PAAD, Pancreatic Adenocarcinoma; PCPG, Pheochromocytoma and Paraganglioma; PRAD, Prostate Adenocarcinoma; READ, Rectal Adenocarcinoma; SARC, Sarcoma; SKCM, Skin Cutaneous Melanoma; STAD, Stomach Adenocarcinoma; TGCT, Testicular Germ Cell Tumors; THYM, Thymoma; THCA, Thyroid Carcinoma; UCS, Uterine Carcinosarcoma; UCEC, Uterine Corpus Endometrial Carcinoma; UVM, Uveal Melanoma.

Extended Data Figure 3. Global activity of oncogenic signatures

a. Activity of signatures is calculated relative to a normal tissue baseline, i.e. activity scores are compared to a compendium of 36 normal tissues. Therefore, this plot represents a comparison of pathway activities between metastatic tissues and normal tissues. Increased activity (positive difference, red) or decreased activity (negative difference, blue) indicates that the signature genes are on average more (or less) expressed in a metastatic tumor sample relative to the baseline (in average percentile point difference labeled “% diff activity”). Samples (columns) are ordered from left to right by decreasing average activity difference (column averages, i.e. the aggregate score in panel b) b. Boxplots summarizing the aggregate scores (column averages of “% diff activity”) in a. Analysis cohorts are ordered left-to-right by median aggregate scores.

Extended Data Figure 4. Relative activity of oncogenic signatures

Hierarchically clustered heatmap of activity scores for the most variable oncogenic signatures. In contrast to (Supp. Fig. 7), here activity scores are computed intrinsically, i.e. relative to other samples in the MET500 (like ssGSEA or GSVA), which represents a relative comparison between different patients / samples. Red indicates that a signature is more active (in percentile points) for a given sample relative to the median activity, blue indicates that a signature is less active for a given sample.

Extended Data Figure 5. Activity of cancer hallmarks in metastatic cancers

Clustered heatmaps of activity scores for the 50 MSigDB cancer hallmarks are shown. a. Gene expression patterns of cancer hallmark pathways. Average increase (red) or decrease (blue) in the relative expression levels (percentiles) of transcriptional signatures associated with the hallmarks of cancers are illustrated. b. Activity scores are calculated relative to a compendium of 36 normal tissues, which represent a comparison of hallmark activities between metastatic tissues and normal tissues (analogous to Supp. Fig. 7 but for a different gene sets). Red indicates that a signature is more active (in percentile points) for a given sample relative to the median activity, blue indicates that a signature is less active for a given sample.

Extended Data Figure 6. Discovery of oncogenic meta-signatures

Relative activity scores were computed for all experimental signatures in the MSigDB database across the MET500 cohort. The signatures were clustered into 25 meta-signatures based on their activity profiles across the MET500. For each of the 25 meta-signature clusters, the 5 most variable signatures are selected. Red indicates that a signature is more active (in percentile points) for a given sample relative to the median activity across the MET500. Blue indicates that a signature is less active for a given sample.

Extended Data Figure 7. Activity of the oncogenic meta-signatures

a. Relative activity of EMT and proliferation signatures across the TCGA analysis meta-cohort. b. Relative activity of the 25 meta-signatures across MET500 samples from different biopsy sites. Red indicates that a signature is more active for a given biopsy site relative to the median activity, blue indicates that a signature is less active for a given biopsy site. c. Relative activity of the 25 meta-signatures across samples from different normal tissues. Red indicates that a signature is more active (in percentile points) for a given tissue relative to the median activity, blue indicates that a signature is less active for a given tissue. d. Correlations between the 25 meta-signatures. Correlation heatmap and hierarchical clustering showed similarities (red) and dissimilarities (blue) in the transcriptional activity of computationally derived aggregate sets of MSigDB signatures, i.e. “meta-signatures” across samples from the MET500 stratified by the most common primary tumor type (left panels) and biopsy site (right panels).

Extended Data Figure 8. Prediction of immune infiltration in cancer tissues

a. Correlation between the MImmScore, a measure of absolute immune infiltration in a tumor samples, with tumor content estimated from exome DNA sequencing using CNVs and SNVs. b. Correlation between MImmScores and an analogous score for tumor-stromal infiltration. c. Correlation between a T-cell expression score summarizing the expression levels (RNA-seq based) of marker genes: CD3D, CD3E, CD3G, CD6, SH2D1A, TRAT1 and the estimated number of T-cells based on T-cell repertoire profiling (DNA-based). d. Number of T-cells based on T-cell repertoire profiling for index cases stratified into MImmScore low (<0) or MImmScore high (>0). Significance levels of Spearman rank correlation coefficient test: * (p=0.05–0.001), ** (p=0.001-1e-6), *** (p<1e-6) test and Wilcoxon test.

Extended Data Figure 9. Differential immune infiltration in various cancer types

a. Distribution of MImmScores, a measure of the magnitude of immune infiltration in a tumor sample, for MET500 samples/patients grouped by tumor biopsy site. b. Distribution of MImmScores across the TCGA meta-cohort, grouped by primary cancer designation. Hematological malignancies (DLBC, LAML) are included as positive control. c. Percentage of patients in each of the MET500 analysis cohorts that has a high MImmScore defined here as >80th percentile across the whole MET500. The total number of cases with high MImmScore is indicated above each bar. d. Same as c but for the TCGA meta-cohort. e. Correlation between the total number of T-cells (templates) based on T-cell repertoire (DNA) sequencing of the T-cell receptor CDR3 sequence, and the number of expanded clones, an expanded T-cell clone is defined as having more than 30 cells with the same CDR3 sequence. f. Ratio of expression levels for markers of CD8+ T-cells (CD8A, CD8B) and T regs (FOXP3) as a function of the total number of T-cells. Significance levels of Spearman rank correlation coefficient: * (p=0.05–0.001), ** (p=0.001-1e-6), *** (p<1e-6).

Extended Data Figure 10. Genomic correlates of immune infiltration

a. Association between the MImmScore and mutation status (hypermutated samples have been defined here as having >250 non-synonymous mutations). Statistical significance of this association was done using logistic regression. b–c. The chi-square test for independence is used to determine whether the clusterings of samples based on T-cell and APC markers are independent. Enrichment or depletion is calculated as the Pearson residual. Red indicates (positive enrichment) that the clusters overlap significantly. Blue indicates (depletion) that clusters tend to be mutually exclusive. Clustered heatmap of enrichment levels (chi-square table cell residuals) is shown in b. Enrichment levels for clusters for the active Tcell-1 and Tcell-4 clusters and all APC clusters (APC-1,4 active) are shown in c.

Figure 1. Landscape of molecular alterations in metastatic cancer

a. Cancer types in the MET500 cohort. Number of cases indicated for each cancer type. b. Site of biopsies. c Mutational burden across tumor types from the MET500 and corresponding primary TCGA cohorts. Transparent boxplots signify insignificant differences (Wilcoxon rank-sum test FDR >= 0.1) d and e. Landscape of molecular alterations in the MET500 cohort. Each cell represents the mutation status of an individual gene for a select patient. Putative oncogenes are represented on panel d and putative tumor suppressor genes in panel e. The percentage of mutations across the MET500 cohort is represented by vertical histograms.

Figure 2. Putative pathogenic germline variants in metastatic cancers

a. Pathogenic germline alterations identified in the MET500 cohort. DNA repair pathway related variants are indicated in shades of blue while “other” alterations are indicated in shades of red. b. Gene level schematic of pathogenic germline variants identified in the MET500 cohort.

Figure 3. Diverse classes of gene fusions identified in metastatic cancers

a. Fusions classified by underlying structural aberrations. b. Functional gene fusions identified in the MET500 cohort. c. Molecular structure of novel, potentially activating, gene fusions in the MET500 cohort..

Figure 4. Diverse transcriptional profiles of metastatic cancers

a. Global gene expression patterns of the MET500 identify poorly differentiated cancers as illustrated by a t-SNE projection of the MET500 samples. Position of samples within the plot reflects the relative similarity in the expression of cancer-specific markers. Samples are color-coded based on their assigned analysis cohort. b. Correlation heatmap and hierarchical clustering showing similarities (red) and dissimilarities (blue) in the transcriptional activity of computationally derived aggregate sets of signatures across the MET500 “meta-signatures”. c. Negative correlation between signatures of EMT and proliferation (S-phase of the cell cycle, FANCA pathway). All MET500 samples are shown.

Figure 5. The immune microenvironment of metastatic cancers

a. Magnitude of immune (leukocyte) infiltration (MImmScore) across the MET500 analysis cohort. b. Hierarchical clustering of samples by their predicted immune infiltrates. c–d. T-cell receptor profiling by TCRβ DNA deep-sequencing. Correlation of estimated T-cell numbers (templates): c. with clonal expansion (high clonality indicates that many T-cells have the same TCRβ sequence) (rho - Spearman’s rank-correlation coefficient). d. with number of mutations. e. Clusters of patients based on the normalized expression levels of APC (left) or T-cell (right) surface molecules. f–h. genomic correlates for patients grouped by their membership in immunologically active clusters: TIL-5, APC-1, Tcell-1 (silent=none, partial=some, complete=all). f. expression of PD-L1 (t-test). g. number of non-synonymous mutations (Wilcoxon test). h. response-score based on a predictive gene expression signature to immunotherapy (t-test). Significance levels: * (p=0.05–0.001), ** (p=0.001–1e-6), *** (p=<1e-6).