Heterogeneity in the inter-tumor transcriptome of high risk prostate cancer

Abstract

Background: Genomic analyses of hundreds of prostate tumors have defined a diverse landscape of mutations and genome rearrangements, but the transcriptomic effect of this complexity is less well understood, particularly at the individual tumor level. We selected a cohort of 25 high-risk prostate tumors, representing the lethal phenotype, and applied deep RNA-sequencing and matched whole genome sequencing, followed by detailed molecular characterization.

Results: Ten tumors were exposed to neo-adjuvant hormone therapy and expressed marked evidence of therapy response in all except one extreme case, which demonstrated early resistance via apparent neuroendocrine transdifferentiation. We observe high inter-tumor heterogeneity, including unique sets of outlier transcripts in each tumor. Interestingly, outlier expression converged on druggable cellular pathways associated with cell cycle progression, translational control or immune regulation, suggesting distinct contemporary pathway affinity and a mechanism of tumor stratification. We characterize hundreds of novel fusion transcripts, including a high frequency of ETS fusions associated with complex genome rearrangements and the disruption of tumor suppressors. Remarkably, several tumors express unique but potentially-oncogenic non-ETS fusions, which may contribute to the phenotype of individual tumors, and have significance for disease progression. Finally, one ETS-negative tumor has a striking tandem duplication genotype which appears to be highly aggressive and present at low recurrence in ETS-negative prostate cancer, suggestive of a novel molecular subtype.

Conclusions: The multitude of rare genomic and transcriptomic events detected in a high-risk tumor cohort offer novel opportunities for personalized oncology and their convergence on key pathways and functions has broad implications for precision medicine.

Figures

Figure 1

Therapy response and resistance in neo-adjuvant hormone treated tumors. (A) Breakdown of the patient cohort. (B) Principal component analysis using RNA-seq derived transcript expression (after removal of three outliers; see Additional file 2: Figure S1) demonstrating the global split between hormone-naive tumors and tumors treated with neo-adjuvant hormone therapy (NHT). (C) Representative genes that were significantly differentially expressed between hormone-naive and NHT tumors (DESeq comparison; Benjamini-Hochberg corrected P values). (D) Expression levels of the neuroendocrine prostate cancer marker CHGA in the cohort, showing elevated levels in NHT tumors. † indicates the unique hybrid adenocarcinoma-neuroendocrine tumor described in-depth in a separate study (ref 45). (E-G) CHGA protein staining showing diffuse positivity (<10% of cells overall) in selected NHT tumors with elevated CHGA mRNA expression levels. (H) Dual AR (brown) and CHGA (pink) staining in tumor T20 (with the highest mRNA expression of CHGA in (D)) showing evidence of neuroendocrine transdifferentiation, potentially in response to 8 months of NHT. The top panel shows the diagnostic biopsy, prior to treatment, showing few (pink) CHGA positive cells but predominant (brown) AR staining, while the bottom panel shows tumor T20 (at radical prostatectomy (RP)), demonstrating small cell morphology, widespread (pink) CHGA positivity and scant remaining AR positive foci.

Figure 2

Significant pathway enrichment of transcripts with outlier gene expression. The heatmap in the top panel provides the pathway enrichment scores for representative canonical pathways in each tumor sample and benign tissue showing that outlier transcript sets from different tumors converge on distinct cellular functions (pathway score = -log10(Benjamini–Hochberg adjusted P value); only pathway scores >1 (that is, B-H <0.1) are shown). Note pathway names truncated from 1Role of CHK Proteins in Cell Cycle Checkpoint Control, 2Cell Cycle: G2/M DNA Damage Checkpoint Regulation, 3Role of Pattern Recognition Receptors in Recognition of Bacteria and Viruses, 4Altered T Cell and B Cell Signaling in Rheumatoid Arthritis. The bottom panel provides explorations of several key pathways, highlighting the expression distribution of representative genes within those pathways across the entire tumor cohort (full list of outlier gene sets within enriched pathways in Additional file 1: Table S8). Tumors with outlier gene enrichment within a given pathway are represented by colored circles. Immunohistochemistry images show high Ki67 indices in three tumors reflecting the probable high proliferation rate of tumors with outlier gene enrichment within cell cycle-related pathways (for additional images with larger fields see Additional file 2: Figure S3).

Figure 3

The landscape of non-ETS fusion genes and complex genome rearrangements in high-risk prostate cancer. (A) Depicts the number of expressed fusion genes (‘Fusion load’) and complex genome rearrangements (‘Complex’) detected in each tumor sample, and indicates selected genes involved in fusion events. Only genes with putative links to cancer are shown (full list of involved genes in Additional file 1: Table S9). Genes marked with * indicates a putative gain-of-function. † highly complex chromothripsis-driven rearrangements; ‡ involvement in complex rearrangement but not expressed; § DNA sequence coverage too low to elucidate complex rearrangements. (B) Schematic of the complex genome rearrangement (also known as chromoplexy [6]) in tumor T20 which lead to the disruption of TP53. Breakpoints in the six genes involved are indicated on the left, together with the final configuration on the right. This particular rearrangement led to the expression of just one fusion transcript (i.e. MATR3-HMG3P22). (C-F) Further schematics of complex genome rearrangements (C) shows the same rearrangement as (B)]. Green nodes indicate a gene is disrupted by rearrangement while red and gray indicate potential activating or neutral effects, respectively. Full edges represent a DNA rearrangement, and dotted lines indicate a rearrangement that was also detected in the RNA sequence data (that is, a fusion transcript was expressed). (G) Schematics of selected fusion genes with putative gain-of-function. Tumor ID is provided in each box, and major protein domains are annotated.

Figure 4

A tandem duplication genotype in high-risk prostate cancer. (A) Copy number profile of hormone-naïve tumor T4, showing focal gains across the entire genome, distinctive of tandem duplications. Key events discussed in the text are annotated. (B) Schematic illustrating how individual tandem duplications present at the genomic level. (C) Serial tandem duplications across the MDM2 loci in T4. The copy number plot shows a focal high gain, with the colored lines representing segments that have been duplicated. Tandem duplication ID is indicated next to each colored line and in brackets is the estimated number of copies of each tandem duplication. For example, the genetic breakpoint of the most focal tandem duplication (green line) was detectable at a high frequency indicating multiple copies, suggesting that the breakpoint itself had been subsequently duplicated, potentially by the broader tandem duplications (pink and blue lines). Note that one breakpoint (purple line) was predicted at <1 suggestive of sub-clonality and highlighting the potential for continual evolution. (D) Transcript expression of MDM2 across the high-risk cohort showing elevated expression in tumor T4 relative to the other tumors and benign samples. (E, F) Chromogenic in situ hybridization of MDM2 in benign tissue (E) and tumor (F) from T4 confirming MDM2 amplification. Note the ‘clumps’ of staining in the tumor cells suggesting MDM2 amplifications are proximally located (consistent with tandem duplication). (G) Copy number profiles from two tumors in a public dataset, which appear to harbor the distinctive pattern of focal gains across the entire genome (see Additional file 2: Figure S9 for further examples). Oncogenes within focal amplifications are annotated.