Integrated Genomic Analysis of Primary Prostate Tumor Foci and Corresponding Lymph Node Metastases Identifies Mutations and Pathways Associated with Metastasis

Abstract

Prostate cancer is a highly heterogeneous disease and mortality is mainly due to metastases but the initial steps of metastasis have not been well characterized. We have performed integrative whole exome sequencing and transcriptome analysis of primary prostate tumor foci and corresponding lymph node metastases (LNM) from 43 patients enrolled in clinical trial. We present evidence that, while there are some cases of clonally independent primary tumor foci, 87% of primary tumor foci and metastases are descended from a common ancestor. We demonstrate that genes related to oxidative phosphorylation are upregulated in LNM and in African-American patients relative to White patients. We further show that mutations in TP53, FLT4, EYA1, NCOR2, CSMD3, and PCDH15 are enriched in prostate cancer metastases. These findings were validated in a meta-analysis of 3929 primary tumors and 2721 metastases and reveal a pattern of molecular alterations underlying the pathology of metastatic prostate cancer. We show that LNM contain multiple subclones that are already present in primary tumor foci. We observed enrichment of mutations in several genes including understudied genes such as EYA1, CSMD3, FLT4, NCOR2, and PCDH15 and found that mutations in EYA1 and CSMD3 are associated with a poor outcome in prostate cancer.

Keywords: cancer genomics; metastasis; prostate cancer; tumor heterogeneity.

Figure 1

RNAseq analysis of 165 prostate cancer patient samples identifies oxidative phosphorylation as differentially expressed between AA and White patients and between LNM and benign lymph nodes. (A) Heat map of 704 genes that trend consistently upward or downward from normal to primary tumor foci to LNM. (B) GSEA plot of the oxidative phosphorylation gene set from a comparison of LNM to benign lymph nodes. (C) GSEA plot of the oxidative phosphorylation gene set from analysis of African-American vs. White patients with Gleason scores of 7. (D) OXPHOS score increases in from normal prostate to primary tumor to LNM. (E) FGFR3 is highly correlated with OXPHOS gene IDH1. (F) FGFR3 is expressed at higher levels in AA PCa than EA PCa in our cohort. (G) FGFR3 mRNA is significantly higher in patients with >50% African ancestry in the TCGA PRAD dataset. (H) Ecotyper analysis of 131 samples assigned to one of ten unique carcinoma ecotypes (CE).

Figure 2

WES analysis of 137 prostate cancer patient samples identifies recurrent mutations associated with metastasis. (A) Graphical depiction of the top ten most frequently mutated genes in this study. Each column represents an individual tissue sample. Percentages on the right indicate the percentage of samples with mutations in that gene. Tumor mutational burden (TMB) for each sample is plotted along the top. (B) Lollipop plot of mutations detected in the SPOP gene. (C) Forest plot of mutations preferentially detected in patients with metastases. * <0.05; ** <0.01 (D) Signaling pathways most highly impacted by the detected mutations.

Figure 3

Analysis of metastatic and non-metastatic clones identifies oxidative phosphorylation as associated with metastasis. (A) Heatmaps of all identified mutations clustered based on allele frequencies. LNM are clonally related to most primary tumor foci. Some patients (PCM034 and PCM003) contained primary tumor foci that were not clonally related to others, although most patients did not. (B) Heatmaps of genes with significantly increased allele frequencies in LNM relative to primary tumor foci within the same patients indicating potential clonal selection. (C) Venn diagram of sets of significantly differentially expressed genes comparing normal prostate to metastatic primary tumor foci, normal prostate to non-metastatic primary tumor foci, and metastatic primary tumor foci to non-metastatic primary tumor foci. (D) Volcano plot of over-representation analysis of the 810 genes in the center of the Venn Diagram in panel C. Oxidative phosphorylation was the only significantly overrepresented pathway.

Figure 4

HATCHet analysis of tumor heterogeneity for patient PCM034 identifies subclones with differential gains in chr8. Gains and losses for each chromosome are indicated as red or blue. The abundance of subclones is indicated with lighter colors showing less abundance and darker colors showing greater abundance. Four subclones were identified, with clone 1 containing a loss of chr8p and gain of chr8q being most prevalent in the LNM samples. Primary tumor foci were enriched in the normal subclone.

Figure 5

Validation of mutations using external datasets analyzed in cBioPortal demonstrates EYA1 and CSMD3 mutations are associated with patient outcome. (A) Enrichment of mutations in metastases (n = 2721) or primary tumors (n = 3929). Asterisk (*) indicates significant q-value based on Fisher exact tests. (B) Lollipop plot of EYA1 mutations in metastases (top) compared to primary tumors (bottom). (C) Lollipop plot of CSMD3 mutations in metastases (top) compared to primary tumors (bottom). (D) Kaplan–Meier plot of disease-specific survival comparing EYA1 mutant cases (n = 34) to EYA1 wild-type cases (n = 458). (E) Kaplan–Meier plot of disease-free survival comparing EYA1 mutant TCGA cases (n = 18) to EYA1 wild-type TCGA cases (n = 316). (F) Kaplan–Meier plot of progression-free survival comparing EYA1 mutant TCGA cases (n = 34) to EYA1 wild-type TCGA cases (n = 460). (G) Kaplan–Meier plot of disease-specific survival comparing CSMD3 mutant TCGA cases (n = 63) to CSMD3 wild-type TCGA cases (n = 429). (H) Kaplan–Meier plot of disease-free survival comparing CSMD3 mutant TCGA cases (n = 39) to CSMD3 wild-type TCGA cases (n = 295). (I) Kaplan–Meier plot of progression-free survival comparing CSMD3 mutant TCGA cases (n = 63) to CSMD3 wild-type TCGA cases (n = 431).