Associations of Luminal and Basal Subtyping of Prostate Cancer With Prognosis and Response to Androgen Deprivation Therapy

Abstract

Importance: There is a clear need for a molecular subtyping approach in prostate cancer to identify clinically distinct subgroups that benefit from specific therapies.

Objectives: To identify prostate cancer subtypes based on luminal and basal lineage and to determine associations with clinical outcomes and response to treatment.

Design, setting, and participants: The PAM50 classifier was used to subtype 1567 retrospectively collected (median follow-up, 10 years) and 2215 prospectively collected prostate cancer samples into luminal- and basal-like subtypes.

Main outcomes and measures: Metastasis, biochemical recurrence, overall survival, prostate cancer–specific survival, associations with biological pathways, and clinicopathologic variables were the main outcomes.

Results: Among the 3782 samples, the PAM50 classifier consistently segregated prostate cancer into 3 subtypes in both the retrospective and prospective cohorts: luminal A (retrospective, 538 [34.3%]; prospective, 737 [33.3%]), luminal B (retrospective, 447 [28.5%]; prospective, 723 [32.6%]), and basal (retrospective, 582 [37.1%]; prospective, 755 [34.1%]). Known luminal lineage markers, such as NKX3.1 and KRT18, were enriched in luminal-like cancers, and the basal lineage CD49f signature was enriched in basal-like cancers, demonstrating the connection between these subtypes and established prostate cancer biology. In the retrospective cohort, luminal B prostate cancers exhibited the poorest clinical prognoses on both univariable and multivariable analyses accounting for standard clinicopathologic prognostic factors (10-year biochemical recurrence-free survival [bRFS], 29%; distant metastasis-free survival [DMFS], 53%; prostate cancer-specific survival [PCSS], 78%; overall survival [OS], 69%), followed by basal prostate cancers (10-year bRFS, 39%; DMFS, 73%; PCSS, 86%; OS, 80%) and luminal A prostate cancers (10-year bRFS, 41%; DMFS, 73%; PCSS, 89%; OS, 82%). Although both luminal-like subtypes were associated with increased androgen receptor expression and signaling, only luminal B prostate cancers were significantly associated with postoperative response to androgen deprivation therapy (ADT) in a subset analysis in our retrospective cohorts (n = 315) matching patients based on clinicopathologic variables (luminal B 10-year metastasis: treated, 33% vs untreated, 55%; nonluminal B 10-year metastasis: treated, 37% vs untreated, 21%; P = .006 for interaction).

Conclusions and relevance: Luminal- and basal-like prostate cancers demonstrate divergent clinical behavior, and patients with luminal B tumors respond better to postoperative ADT than do patients with non–luminal B tumors. These findings contribute novel insight into prostate cancer biology, providing a potential clinical tool to personalize ADT treatment for prostate cancer by predicting which men may benefit from ADT after surgery.

Conflict of interest statement

Conflict of Interest Disclosures: Dr Zhao reported receiving travel expenses from GenomeDx and holding patent 14/491,679: Compositions and methods for the analysis of radiosensitivity, in the process of being licensed to PFS Genomics. Dr Chang reported being an employee of PFS Genomics. Mr Erho reported holding patent 13/968,838: Cancer diagnostics using biomarkers. Dr Speers reported holding patent 14/491,679: Compositions and methods for the analysis of radiosensitivity, in the process of being licensed to PFS Genomics, and reported having an ownership interest in PFS Genomics. Dr Cooperberg reported receiving honoraria from Takeda; serving as a consultant for Myriad, Astellas, Dendreon, and Janssen; and receiving research funding from Genomic Health, Myriad, and GenomeDx. Dr. Kim reported serving as a consultant for Dendreon, Genentech, Pfizer, and Bayer and receiving research funding from Novartis. Dr Freedland reported receiving research funding from GenomeDx and serving as a consultant for GenomeDx, Astellas, Medivation, Bayer, Sanofi, Janssen, Dendreon, Armune, Parallel 6, Singulex, Boston Scientific, and Churchill. Dr Sandler reported serving as a consultant for Janssen, Ferring, Genentech, Dendreon, and NantHealth. Dr Klein reported serving as a consultant for Berg and GenomeDx, serving on the speakers bureau for Genomic Health, and receiving research funding from GenomeDx and Genomic Health. Dr Black reported serving on advisory boards for Novartis, AbbVie, Astellas, Janssen, Biocancell, Merck, Cubist, Sitka, Bayer, Sanofi, Spectrum, Lilly, and Biosyen; serving as a speaker for Ferring and RedLeaf Medical; receiving grant funding from New B Innovation and iProgen; having clinical trials with Janssen, Ferring, Astellas, and Genentech; holding patent 62/448,921 with GenomeDx; and receiving honorarium from Pendopharm. Dr Tomlins reported receiving honoraria from Ventana Medical Systems; serving as a consultant for Ventana Medical Systems, Astellas/Medivation, Janssen, and Almac Diagnostics; receiving research funding and travel expenses from Thermo Fisher Scientific and GenomeDx; and holding patent 13/483,176 on ETS gene fusions in prostate cancer, the diagnostic field of use has been licensed to Hologic/Gen-Probe Inc, who has sublicensed some rights to Ventana Medical Systems/Roche. Dr Chinnaiyan reported serving on the advisory board for Wafergen and ThermoFisher and holding patent 13/797,106 on SChLAP1, licensed to GenomeDx, and long-noncoding RNAs. Dr Jenkins reported holding patent 13/968,838: Cancer diagnostics using biomarkers. Dr Davicioni reported being an employee of and having leadership of and ownership in GenomeDx and holding patent 13/968,838: Cancer diagnostics using biomarkers. Dr Ross reported serving as a consultant for GenomeDx, having an ownership interest in GenomeDx, and receiving research funding from Merck, Novartis, and Metamark. Dr Schaeffer reported serving as a consultant for GenomeDx and Myriad Genetics. Dr Nguyen reported serving as a consultant for Medivation, GenomeDx, Ferring, Nanobiotix, and Dendreon and receiving research funding from Astellas. Dr Carroll reported receiving honoraria from Takeda Pharmaceuticals and Genomic Health and serving as a consultant for Janssen Pharmaceuticals and Medivation. Dr Karnes reported receiving research funding and travel expenses from GenomeDx. Dr Feng reported having leadership and ownership in PFS Genomics; consulting for Medivation/Astellas, GenomeDx, Celgene, Sanofi, Dendreon, and Merck; receiving research funding from Varian Medical Systems; and holding patent 14/491,679: Compositions and methods for the analysis of radiosensitivity, in the process of being licensed to PFS Genomics, and long-noncoding RNAs. Drs Zhao, Chang, and Feng reported having patent 62/237,178 pending with GenomeDx Biosciences on a signature to predict radiation response in prostate cancer (different work than what is found in this article) and provisional patent 62/469,174 pending with GenomeDx Biosciences on the work in this article. No other disclosures were reported.

Figures

Figure 1.. PAM50 Clustering and Clinical Outcomes in Prostate Cancer

A, The PAM50 genes cluster prostate cancer samples into 3 subtypes, luminal A (LumA), luminal B (LumB), and basal, in the pooled prostate cancer cohorts (Mayo Clinic I and II, Cleveland Clinic, Thomas Jefferson University, Johns Hopkins University, and Durham Veterans Affairs) using hierarchical clustering of the genes. Each column represents a patient sample, and each row represents a gene. B, Kaplan-Meier curves showing that the PAM50 clusters risk stratify biochemical recurrence-free survival, distant metastasis–free survival, prostate cancer–specific survival, and overall survival.

Figure 2.. Association of Basal and Luminal Subtypes With Basal and Luminal Lineage Markers

A, Examination of the prostate basal lineage 91-gene CD49f signature shows higher scores in the basal subtype (P < .001). B and C, Conversely, prostate luminal lineage has been characterized by high expression of NKX3.1 (P < .001) and KRT18 (P < .001), and the expression of these genes is higher in the luminal subtypes. D, Androgen receptor expression (also a luminal lineage marker) is increased in the luminal subtypes (P < .001). E, On Gene Set Enrichment Analysis, androgen response targets are positively enriched in the luminal samples (P < .001). Bar graphs show the mean (SE) of median-centered gene expression, and P values are determined via analyses of variance. LumA indicates luminal A subtype; LumB, luminal B subtype.

Figure 3.. Prospective Validation in GRID

A, PAM50 clusters demonstrate the same pattern of expression in a prospective validation cohort of 2215 prostate cancer samples run on a commercial clinical platform, with 3 subtypes: luminal A (LumA), luminal B (LumB), and basal. B, As in the retrospective cohorts, AR is increased in luminal samples (P < .001). C, The basal lineage CD49f signature is increased in the basal subtype (P < .001). D, NKX3-1 (P < .001) and KRT18 (P < .001) are increased in the luminal subtypes. Bar graphs show the mean (SE) of median-centered gene expression, and P values are determined via analyses of variance.

Figure 4.. Predicting Response to Androgen Deprivation Therapy (ADT)

A, A matched cohort was obtained from the Mayo Clinic I (MCI) and II (MCII) cohorts, which matched ADT-untreated and ADT-treated patients in a 2:1 ratio based on Gleason score, prostate-specific antigen (PSA), lymph node invasion (LNI), extracapsular extension (ECE), seminal vesicle invasion (SVI), positive surgical margin status (SM), and radiotherapy (RT), resulting in 315 total patients. B, Kaplan-Meier curves for the patients with luminal B (LumB) and non–luminal B subtypes, which group the patients with luminal A (LumA) and basal subtypes. Patients with luminal B subtype who received ADT have lower rates of metastasis than patients who did not receive ADT. However, in the patients with non–luminal B subtype, this trend is reversed. C, Comparison of the 10-year metastasis rates for treated and untreated patients in the patients with luminal B and non–luminal B subtypes (with the interaction term Wald P = .006).