Mohammed Alshalalfa

72 Background: Active surveillance (AS) is becoming standard of care for men with low-risk prostate cancer; however a need exists for better tools to assess which men are optimal candidates for AS. In this study we compare genomic expression profiles of AS candidates against higher-risk radical prostatectomy (RP) patients to characterize the genomics of clinically low-risk prostate cancer. Methods: Biopsies from 473 UCSF patients potentially suitable for AS (stage ≤ cT2N0M0, PSA ≤ 10 ng/ml, Gleason 3+3 or low-volume 3+4 ) were profiled using the Affymetrix HuEx microarray to generate RNA expression data. These cases were compared to 2043 RP cases previously profiled on the same microarray platform. Scores for 21 published prognostic signatures were calculated and pathway associated genes were summarized to provide levels of patient risk and pathway activity. Results: Of the 473 AS biopsies profiled, 408 (86%) passed quality control and were used for analysis. Based on the quartiles of average scores for 21 prognostic signature risk models, 49%, 36%, 11%, and 4%, respectively, were classified into the 1st, 2nd, 3rd, or 4th score quartiles. Considering only the clinically low-risk patients at diagnosis, 356 (87%) were low, 45 (11%) were intermediate and 7 (2%) were high risk. Genomic risk was positively associated with cell cycle related pathways (p < 0.001) and negatively associated with apical junction (p < 0.001), epithelial−mesenchymal transition (p < 0.001), and androgen receptor (p < 0.05) pathways. Clustering of patients based on the expression of 36 pathways revealed two biologic groups corresponding to putative basal and luminal subtypes. Compared to higher risk RP patients, the low risk prostate cancer tumors at diagnosis were enriched for basal-like tumors (20% vs 33%, p < 0.001). Conclusions: Although only 2% of low risk AS candidates have high risk genomic characteristics, very substantial genomic heterogeneity exists in this population, and pathway activation overlaps significantly with higher-risk RP patients. These results suggest that even in potential AS candidates, genomic profiling could eventually be used to better guide management.

3 Background: There is a clear need to develop a clinically relevant molecular subtyping approach for prostate cancer. We hypothesized that prostate cancer can be subtyped based on luminal versus basal lineage. Methods: We applied the PAM50 classifier, which is used clinically to identify luminal and basal cancers in breast cancer, to subtype a total of 3,782 prostate cancer samples using a high-density microarray platform run in a CLIA-certified laboratory. We examined the associations of these subtypes and clinical outcomes. Results: We demonstrate that PAM50 segregates prostate cancer into three reproducible subtypes in both retrospective cohorts and on prospective validation: luminal A (33.3%-34.3%), luminal B (28.5%-32.6%), and basal (34.1%-37.1%). Luminal B prostate cancers exhibited the worst clinical prognoses, followed by basal and luminal A subtypes (10-year biochemical recurrence-free survival: 29/39/41%; distant metastasis-free survival: 53/73/73%; prostate cancer-specific survival: 78/86/89%; overall survival: 69/80/82% respectively) on both univariable and multivariable analyses accounting for standard clinicopathologic prognostic factors. Known luminal lineage markers, such as NKX3.1 and KRT18, and the basal lineage CD49f signature, were enriched in luminal- and basal-like cancers respectively, demonstrating the connection between these subtypes and established prostate cancer biology. While both luminal-like subtypes were associated with increased AR expression and signaling, only luminal B prostate cancers were significantly associated with post-operative response to androgen deprivation therapy (ADT) in a subset analysis matching patients based on clinicopathologic variables (interaction p = 0.006, luminal B 10-year metastasis: 33% (treated) vs. 55% (untreated), non-luminal B: 37% (treated) vs. 21% (untreated)). Conclusions: These findings contribute novel insight into the biology of prostate cancer, and provide translatable clinical tools for personalizing post-operative ADT for patients with prostate cancer. Similar to breast cancer, these findings suggest that luminal/basal subtyping may be useful in treatment selection in prostate cancer.

38 Background: The Androgen-receptor (AR) signaling plays a pivotal role in prostate cancer initiation and progression. AR amplification and overexpression have been associated with lethal PCa and CRPC; however, much of the impact of these events on cancer progression is poorly understood with an unclear prognostic significance of down-stream targets of AR. Methods: A total of 2,555 RP expression profiles were extracted from the Decipher GRID database; 262 of which were from retrospective natural history cohort with known metastasic outcomes and the remaining from anonymized prospective cases with basic demographic and pathological data available. We built a 9 gene AR-output signature (ARoS) score based on canonical androgen regulated genes to represent AR-output activity in RP tissues. The 9 genes were weighted based on their distribution skewness in the prospective data (n = 2,293) and then scores were calculating by summing the weighted expression of the 9 genes. Results: AR expression and ARoS score were not strongly correlated with approximately 50% of samples with low ARoS have relatively high AR expression. In a prospective cohort of 2,293 patients, low ARoS was associated with higher Decipher metastasis risk scores (OR: 3.1, p < 0.001), higher Gleason grade (OR: 2.4, p < 0.001), ERG-fusion negative (OR: 1.8, p < 0.001) and SVI (OR: 2.1, p < 0.001). Additionally, low ARoS was associated with higher expression of neuroendocrine biomarkers (NCAM1, ENO2). In a large retrospective cohort with long term clinical follow-up, patients with low ARoS score had a higher probability of 10-year metastasis rate (p = 0.002). In 55 patients from this cohort who received ADT post metastasis, the ARoS signature derived from primary tissue was lower in patients who developed CRPC (p < 0.001). Furthermore, ARoS was significantly lower in mCRPC samples (p < 0.001) compared to primary PCa tissues. Conclusions: ARoS derived from primary prostate cancer tissue is correlated with metastatic outcomes and progression to CRPC. With further validation, such a signature may better predict those patients which require early intensification of systemic therapy.

Epigenetic reprogramming including altered transcription factor binding and altered patterns of chromatin and DNA modifications are now accepted as the hallmark of aggressive cancers. We show that global changes in chromatin structure and chromatin accessibility in prostate tumour tissue can define castrate-resistant prostate cancer and be used to inform the discovery of gene-level classifiers for therapy. In addition, we show that the androgen receptor overexpression alone, which is a hallmark of progression on its own, is a primary driver for chromatin relaxation, which can lead to promiscuous binding of other key transcription factors important to disease progression. We identify mediators of chromatin relaxation that can be used as disease biomarker and show that such mediators identify DNA stretches enriched in low p-value GWAS-significant disease/tissue-specific susceptibility loci. Presenting Author: Urbanucci Alfonso Urbanucci, Centre for Molecular Medicine Norway, Nordic European Molecular Biology Laboratory Partnership, Forksningsparken, University of Oslo, Oslo, Norway and Department of Molecular Oncology, Institute for Cancer Research, Oslo University Hospital, Oslo, Norway. Email: alfonso.urbanucci@gmail.com

5018 Background: Androgen deprivation therapy (ADT) is one of the main treatment options for locally advanced, recurrent and metastatic prostate cancer. Neuroendocrine prostate cancer (NEPC) is inherently less sensitive to ADT. NEPC can be observed de novo (e.g., small cell prostate cancer) but more commonly arises after exposure to ADT. In some patients, even short exposure to ADT may ‘unmask’ a NEPC phenotype. We hypothesized that a gene expression signature of NEPC when measured in primary tumor specimens (RP) may be useful for predicting patients with innate resistance to ADT. Methods: Expression profiles of 786 PCa patients treated with RP were obtained from the Decipher GRID database. These were split into training (n=360) and validation (n=426) sets and stratified by the receipt of adjuvant ADT (n=222) or no adjuvant ADT (n=564). A literature review of ADT resistance and neuroendocrine genes identified 1,632 genes as candidates. This set was further filtered, using logistic regression and forward feature selection to select a genomic ADT resistance signature (ARS). ARS was trained using a generalized linear model with lasso regularization. Survival c-index, Kaplan Meier and Cox regression analysis was used to compare survival differences between treated and untreated patients with high and low ARS scores (defined by median split). Results: In validation cohorts, the ARS was predictive of metastasis in cohorts receiving adjuvant ADT (10-year metastasis free survival c-index of 0.64 (95% CI 0.55-0.71) as compared to 0.50 (95% CI 0.36-0.63) in patients not treated with ADT). Among ADT treated patients, those with low ARS scores had a 10 year MFS of 86%, versus 68% in those with high ARS scores (p=0.03). In multivariable analysis adjusting for confounding variableswith metastasis as an endpoint, ARS validated with a significant interaction with ADT treatment (p=0.02) which demonstrates ARS’s adjuvant ADT predictive capability. Conclusions: An ADT resistance signature validated as a predictive biomarker with a significant interaction term for predicting metastasis after hormonal treatment. ARS may allow for identification of patients that may be optimal candidates for trials of novel systemic agents

5055 Background: Small cell (SC) neuroendocrine carcinoma of the prostate has very poor prognosis and does not respond well to androgen receptor (AR)-targeted therapies. Neuroendocrine prostate carcinomas are increasingly recognized to show a spectrum of morphologic changes, and may not always be distinguishable from adenocarcinoma based on morphology alone. Here we hypothesize that SC carcinoma harbor a unique gene expression signature that, when measured in primary prostatic adenocarcinoma, may help guide subsequent management. Methods: Whole genome expression profiles of12 morphologically-diagnosed SC carcinoma samples and 12 high risk adenocarcinoma retrieved from the Decipher GRID to define SC Genomic Fingerprint (SCGFt) through differential expression analysis. The SCGFt was first validated in mCRPC and NE cohorts, then translated to a genomic score (SCGS) using UVA to weight SCGFt components. An additional 2,400 RP adenocarcinomas from the Decipher GRID were used to evaluate the utility of the SCGS for predicting outcome. Results: Based on genomic prognostic tests calculated using the Decipher assay including Decipher, Penney and microarray-derived Cell Cycle Progression, SC carcinomas have very poor prognosis compared to high grade adenocarcinomas. A total of 25 genes defined the SCGFt. PEG10, cell proliferation and RB1-loss program are the most overexpressed genes in SC and AR-related genes the most down-regulated. SCGS was generated for Prostate cancer cell lines and showed to be predictive of Abiraterone response. Evaluating SCGS in RP adenocarcinoma cohorts showed that patients with high SCGS are at higher risk of developing metastasis after RP in a natural history cohort or after adjuvant hormonal therapy based on Kaplan Meier analysis (both p < 0.001). SCGS was then generated for > 2,000 RP samples from prospective Decipher test showing 2% are likely to harbour small cells or NE-like expression profiles. Conclusions: SC carcinoma has a distinct genomic profile that may be useful for treatment management of patients with adenocarcinoma.

303 Background: Small cell (SC) neuroendocrine carcinoma of the prostate has very poor prognosis and does not respond well to androgen receptor (AR)-targeted therapies. Neuroendocrine prostate carcinomas are increasingly recognized to show a spectrum of morphologic changes, and may not always be distinguishable from adenocarcinoma (adeno) based on morphology alone. Here we hypothesize that SC carcinoma harbor a unique gene expression signature that, when measured in primary prostatic adenocarcinoma, may help guide subsequent management. Methods: In total, 617 whole genome expression profiles were retrieved from the Decipher GRID providing gene expression data for 1.4 million markers. 17 morphologically-diagnosed SC carcinoma samples were compared to 32 Gleason 8-10 and 77 Gleason 6 RP adeno samples with no prior treatment and no evidence of metastasis, and this comparison was used to define SC Genomic Fingerprint (SCGFt) through adjusted median fold difference and Wilcoxon test. An additional 493 RP adeno from Mayo Clinic and John Hopkins Hospital were used to evaluate the utility of the SCGFt for predicting outcome. Results: Based on genomic prognostic tests calculated using the Decipher assay including Decipher, Penney and microarray-derived Cell Cycle Progression, SC carcinomas have very poor prognosis compared to high grade adeno. A total of 356 genes defined the SCGFt with 258 gene upregulated in SC and 98 down-regulated. PEG10, HELLS and RB1-loss program are the most overexpressed genes in SC and AR-related genes the most down-regulated. As expected, SC lacked ERG and alternative ETS expression and showed relatively low SPINK1 expression. A median summarization of the 356 genes in SCGFt was used to build a small cell genomic score (SCGS). Evaluating SCGS in RP adeno cohorts showed that patients with high SCGS are at higher risk of developing metastasis after RP in a natural history cohort or after adjuvant hormonal therapy based on Kaplan Meier analysis (both p< 0.001). When restricted to patients with high Decipher scores, SCGS provided independent prognostic information. Conclusions: SC carcinoma has a distinct genomic profile that may be useful for treatment management of patients with adenocarcinoma.

98 Background: Prostate cancers patient management has been enhanced with commercially available genomic prognostic tests such as the Decipher prostate cancer classifier that are useful for making treatment decision-making. In addition to being the most validated predictor of metastasis in prostate cancer, Decipher is also a genome-wide assay that measures the expression of many druggable targets. Methods: Decipher GRID (Genomic Resource Information Database), was queried to assess the expression patterns of 14 genes from 5 biological pathways (Table) in 1,850 patients from previously published Decipher validation studies. The frequency of high (or low) expression of each gene was ascertained using a standard and more conservative thresholds based on the median absolute deviation (MAD) metric. For the standard threshold, genes whose high expression is of clinical relevance, patients with gene expression above the median + 1.48*MAD were annotated as high expression and for genes whose low expression is of clinical relevance, patients with gene expression below the median - 1.48*MAD were annotated as low expression. For the conservative threshold, median +/- 2*1.48*MAD was used. Results: See Table. Conclusions: Since every patient receiving the Decipher test also has a genome-wide expression profile, the Decipher GRID will allow researchers to evaluate on a systematic population-level the expression of genes that may be targeted with existing therapies. Such information may be useful for selection of optimal systemic therapy and inclusion into clinical trials of novel targeted agents. This rich genomic resource is being made available on a research use only basis to prostate cancer researchers and to clinicians seeking to uncover individualized genomic insights for patients to advance precision medicine. [Table: see text]

256 Background: B7-H3 (CD276), part of the B7 superfamily, has been shown to play an immunomodulatory role, however its regulation, receptor and mechanism of action remain unclear. Protein levels of B7-H3 have been previously shown to relate to prostate cancer outcomes and currently, humanized monoclonal antibodies are being developed for clinical use (MGA271, Macrogenics). Here we use genomic expression data to examine the relationship of B7-H3 to prostate cancer outcomes and molecular subtypes. Methods: Prostatectomy tissue from 905 patients were profiled using the Affymetrix HuEx 1.0 ST microarray. Kruskal-Wallis tests were used to identify significant associations of B7-H3 expression with clinico-pathologic variables, and survival analysis were used to evaluate the prognostic value of B7-H3. Pearson’s correlation analyses were also performed to assess the relationship of B7-H3 expression with molecular subtypes and individual transcripts. Androgen receptor (AR) occupancy of promoter regions was derived in silico from chromosomal immune-precipitation (ChIP) data. Results: B7-H3 expression was positively associated with Gleason score (p < 0.01) and tumor stage (p < 0.01). High B7-H3 expression also correlated with the development of metastasis and prostate cancer specific mortality (HR of 3.4 and 2.4 respectively, p < 0.05 for both), but this was not significant on multi-variable analysis. B7-H3 was positively associated with ERG+ disease (n = 670, r = 0.85, p < 0.05) and AR expression (n = 670, r = 0.46, p < 0.001). B7-H3 was found to be one of the most correlated genes with AR (95th percentile) and ChIP analysis revealed AR binding upstream of B7-H3, suggesting potential androgen dependent regulation. Conclusions: B7-H3 expression correlates with high Gleason grade and advanced prostate cancer stage with higher quartiles of expression portending poor oncologic outcomes in two independent prostatectomy cohorts. B7-H3 expression appears to relate to the androgen receptor.