Identification of a candidate prognostic gene signature by transcriptome analysis of matched pre- and post-treatment prostatic biopsies from patients with advanced prostate cancer

Patient samples for gene expression analysis (RNA-Seq) were collected as part of the GenTax (Tumour profiling in an open-labelled, two-arm study investigating the tolerability and efficacy of Taxotere in patients with hormone-naïve high-risk prostate cancer) study by Newcastle upon Tyne Hospitals National Health Service (NHS) Foundation Trust [15]. All patients with a clinical suspicion of advanced PCa were subjected to TRUSS-guided prostatic biopsy (BK Medical, 8818) for histopathological assessment by Gleason Sum score [16] of Haematoxylin and Eosin (H&E)-stained tissue. Radiological staging investigations were performed according to national guidelines [17]. Patient eligibility criteria were cT3/T4 [18] PCa, Prostate Specific Antigen (PSA) ≥50 ng/ml or Gleason Sum score ≥8, or metastatic disease to be commenced on ADT. Further eligibility for study inclusion were Karnofsky Performance status (KPS) Score [19] ≥ 70%; a life expectancy of ≥ 3 months; and adequate haematological, hepatic, and renal function. All patients received ADT, which consisted of the goserelin 3.6 mg on a q28-day schedule with anti-androgen “flare” protection and 6 cycles of docetaxel (Taxotere^®) 75 mg/m² on a q21-day schedule [15]. Further material for RNA-Seq was taken by TRUSS-guided biopsy prior to commencement of chemotherapy and again at ~22 weeks following initiation of treatment. Biopsies were specifically taken from tumour-rich areas of the prostate, where typically over 60% of the initial diagnostic cores taken were occupied by tumour. All patient material was anonymized and stored at -80°C. Serum PSA was measured ~3-weekly until ~22 weeks and then 3-monthly, and repeat radiological staging undertaken at ~6 months after diagnosis for patients with N+ and/or M+ disease to assess the radiological response. PSA progression was defined as two consecutive rises in PSA above nadir at least 2 weeks apart, although whether patients subsequently fulfilled the European Association of Urology (EAU) criteria for castration resistant PCa disease [1] is not known. Written informed consent to participate was obtained from all subjects. Ethical approval was granted from the local research and ethics committee (Northumberland, Tyne and Wear NHS Strategic Health Authority Local Research Ethics Committee Ref: 2003/11).

Patient samples for RNA-Seq were analysed as previously described [20]. Total RNA was extracted from pre- and post-treatment samples using the RNeasy Mini Kit (QIAgen, 74104) according to manufacturer’s instructions. The NanoDrop 2000 spectrophotometer (Thermo Scientific) and 2100 Bioanalyzer (Agilent) were used to assess RNA quantity and quality, with calculation of RNA integrity number (RIN) [21]. Samples were included for RNA-Seq if RIN > 6 and total RNA > 500 ng. Illumina RNA-Seq was performed according to manufacturer’s instructions, with cDNA sample library normalization using the Illumina DSN (Duplex-specific Nuclease) protocol prior to cluster generation and library sequencing on the HiSeq™ 2000 (Illumina) with a paired-end sequencing strategy. The read length was set at 90 nt with an expected library size of 200 bp.

The FastQC package (http://​www.​bioinformatics.​babraham.​ac.​uk/​projects/​fastqc) was used to assess the quality of raw reads, which were then mapped to human genome assembly hg19 using TopHat version 1.4.1 [22] with a junctions library derived from Ensembl version 68. Quality control was performed on all samples by examining the following parameters: (a) the percent of reads uniquely mapping to the genome; (b) the percent of reads mapping to known protein coding sequence; (c) the number of exon junctions identified; (d) the percent of spliced reads; and (e) the number of genes with 90% base coverage (Additional file 1: Table S1). TopHat-Fusion version 0.1.0 [23] was used to identify gene fusions. HTSeq version 0.5.3 (http://​www.​huber.​embl.​de/​users/​anders/​HTSeq) was used to identify differentially-expressed genes by counting the number of reads mapping to each gene from Ensembl version 68. The TMM method was used to normalise read counts and differential expression tested for using a paired generalized linear model design with the Bioconductor version 2.11 edgeR package [24]. The Circos plot was generated using RCircos version 1.1.2 [25]. Correlations were identified using Pearson’s product moment correlation coefficient (p < 0.05). Enriched KEGG (Kyoto Encyclopedia of Genes and Genomes) pathways [26] were identified by downloading gene pathways associations and testing each pathway for enrichment in significantly up- and down-regulated genes (FDR < 0.05) with a transcript length-corrected Wallenius approximation as implemented by the GOSeq package for Bioconductor 3.0 [27]. Pathways were deemed to be enriched if the enrichment over background was at least 2-fold and the FDR < 0.05. Gene lists were uploaded to cBioPortal (http://​www.​cbioportal.​org) [28, 29] to study gene expression changes in all prostate tumours with mRNA expression data (n = 150) from the Memorial Sloan Kettering Cancer Center (MSKCC) Prostate Oncogenome Project dataset [30] using a mRNA Z-score threshold of ± 1.6 as compared with normal prostate samples. Genes altered in a significant number of tumours (>25%) were considered for associations with disease-free survival though the cBioPortal software using the Kaplan–Meier method with log rank testing with p < 0.05 taken to indicated statistical significance. Raw sequencing data have been deposited at Gene Expression Omnibus (http://​www.​ncbi.​nlm.​nih.​gov/​geo/​) under accession number GSE51005 and all details are MIAME compliant.

Next generation RNA sequencing (RNA-Seq) was performed on 12 paired pre- and post-docetaxel plus ADT samples from 6 patients with locally-advanced/metastatic PCa (Table 1). The post-treatment samples from Patients 2 and 3 performed markedly worse on multiple quality control measures, and so all samples from both patients were excluded from further quantitative expression analysis (Additional file 1: Table S1). The remaining 8 samples matched our previously-published dataset on the ADT-only control arm of the GenTax study [20] on two key quality control measures: At least 50 million 90 bp paired-end reads were obtained per sample with at least 40% coverage of transcripts sequenced (Additional file 1: Table S1).

Genomic rearrangements involving ETS-family transcription factors are implicated in PCa with the most common gene fusion product TMPRSS2/ERG reported in >50% cases [31]. We searched for expression of transcripts derived from such gene fusions in our datasets. We observed the intra-chromosomal TMPRSS2/ERG gene fusion product in only the pre-treatment sample from 1 patient (Patient 3), which was actually excluded from the quantitative expression analysis (Additional file 2: Table S2). However, we observed 3 further novel intra-chromosomal gene fusions: two products were derived from a fusion between DOPEY2 and ERG genes within chromosome 21 (Fusion event 7), and 2 different gene fusions were observed within chromosome 22 (Fusion events 3 and 4) (Figure 1A). A further five novel fusion transcripts were identified (Figure 1A and Additional file 2: Table S2). In three patients, identical inter- and intra-chromosomal gene fusions (CCNY/LRCC49, PVT1/CPNE4, and DOPEY2/ERG) were identified in both pre- and post-treatment samples.

Across the genome, we observed a total of 298 genes up-regulated and 277 genes similarly down-regulated at least 2-fold (False Discovery Rate [FDR] <0.05) in response to docetaxel plus ADT (Figure 1A, Table 2 and Additional file 3: Table S3). The levels of expression of KLK3, which encodes PSA (Prostate Specific Antigen), detected by RNA-Seq of the docetaxel plus ADT arm correlated as expected with serum PSA levels (r² = 0.927; p = 0.037) (Figure 1B). A number of other known androgen-regulated genes (including those encoding kallikreins) were also consistently down-regulated in the docetaxel plus ADT arm (Figure 1C) suggesting that ADT in combination with docetaxel had the expected action on androgen-regulated gene expression.

Based on the full gene list (Log2 fold change ≥2/≤ - 2; FDR < 0.05) (Table 2 and Additional file 3: Table S3), we ranked genes according to the magnitude of their fold changes, regardless of whether they were up- or down-regulated. The 10 top-ranking genes differentially-regulated by docetaxel plus ADT were arbitrarily selected (range of fold changes -9.96 to 9.86) for further downstream knowledge-based validation. From these 10 genes, we selected genes that exhibited expression changes consistent in direction in at least 3 out of 4 patients. We identified 6 differentially-expressed genes (Figure 2A) including ORM1, which had the highest average level of differential expression of all transcripts in our dataset (Log₂ fold change = -9.96; FDR < 0.05). This gene encodes an acute phase plasma protein that has been identified as a putative biomarker of chemo-resistance to docetaxel and doxorubicin in breast cancer [13].

Using cBioPortal [28, 29], we interrogated the MSKCC Prostate Oncogenome Project dataset (n = 150) [30] for changes in expression of the above 6 genes in treatment-naive prostate tumours as compared with normal controls. We observed alterations in expression of all 6 genes, with FAM72B and ADAM7 exhibiting significant alterations (Figure 2B). Survival analysis identified a statistically-significant reduction in disease-free survival of patients with tumours exhibiting alterations in expression of this geneset (p = 0.023) (Additional file 4: Figure S1A) which was lost when FAM72B and ADAM7 were removed from the geneset (p > 0.05) (data not shown). Using only FAM72B and ADAM7, survival analysis demonstrated a statistically-significant disease-free survival advantage in patients with no alterations in gene expression (p = 0.001) (Figure 2C). Taken together, these data suggest that alterations in expression of FAM72B and ADAM7 are associated with early treatment relapse and hence may be biomarkers with prognostic value in treatment-naïve PCa.

To identify biological pathways perturbed by combined docetaxel chemotherapy with ADT, we performed an enrichment analysis on our lists of up- and down-regulated genes (FDR < 0.05) using 3 different pathways analysis tools: the KEGG (Kyoto Encyclopedia of Genes and Genomes) database [26]; IPA “Core Analysis” function; and Metacore (Figure 3 and Additional file 5: Figure S2, Additional file 6: Table S4 and Additional file 7: Table S5). The KEGG terms “Cell Cycle” (n = 11/124; enrichment = 5.89-fold; FDR = 0.0014) and “Steroid Biosynthesis” (n = 5/19; enrichment = 17.63-fold; FDR = 0.0014) were enriched greater than 2-fold in the down-regulated gene list (Additional file 6: Table S4), while no pathways were significantly enriched in the up-regulated gene list. Genes within the KEGG term “Cell Cycle” included the key positive cell cycle regulators CCNB1, CCNB2, CDK1 and CDC25A (Figure 3A and Additional file 8: Table S6), the expression of which was down-regulated following docetaxel plus ADT. The Ingenuity Pathway Analysis “Core Analysis” function also identified the “Cell Cycle” as the highest-ranking network containing clusters of docetaxel and ADT-regulated genes (Additional file 5: Figure S2A and Additional file 7: Table S5). Metacore analysis of docetaxel and ADT-regulated genes identified Cell cycle “The metaphase checkpoint” as the 2^nd top enriched pathway after Cytoskeleton remodelling “Keratin filaments”, which is consistent with the known actions of docetaxel (Additional file 5: Figure S2B).

The observed enrichment for cell cycle-related genes, including down-regulation of expression of positive regulators of cell cycle progression, is in keeping with the known actions of docetaxel in vitro on the induction of G2/M arrest [5]. In the light of evidence suggesting that androgen withdrawal may diminish docetaxel-induced apoptosis in vitro[32], we wished to ensure that our in vivo observations were consistent with the mechanism of action of docetaxel in vitro in the absence of androgens. We used the LNCaP PCa cell line grown in steroid-depleted medium as a model for non-CRPCa treated with ADT. Reassuringly, we observed statistically-significant induction of G2/M arrest (p < 0.05) following treatment with docetaxel (at 10 nM, 100 nM or 1 μM doses) (Additional file 5: Figure S2C).

Finally, we used cBioPortal [28, 29] to interrogate the MSKCC Prostate Oncogenome Project dataset [30] for changes in expression of the genes enriched within the KEGG term “Cell Cycle” in clinical PCa and observed alterations in expression of all genes in a large (78%) proportion of cases (Additional file 5: Figure S2D), suggesting that expression of these transcripts is associated with prostate tumourigenesis. Survival analyses did not identify any statistically-significant associations between disease-free survival time in patients with tumours exhibiting alterations in expression of these genes as compared with patients with tumours exhibiting no alterations in expression (p > 0.05) (data not shown). However, when genes exhibiting alterations in high (>25%) proportion of tumours only were included in this geneset (Figure 3B), we observed statistically-significant reduction in disease-free survival of patients with tumours exhibiting alterations in expression of this geneset (p = 0.024) (Figure 3C). Using a combined geneset of these 5 remaining cell cycle-related genes (BUB1B, CCNB1, CCNB2, TTK and CDK1) as well as ADAM7 and FAM72B, we also observed a statistically-significant reduction in disease-free survival of patients with tumours exhibiting alterations in gene expression (p = 0.015) (Figure 3D). Our observations suggest that these 7 genes in combination could form a panel of biomarkers associated with early relapse from treatment in clinical PCa.