Predicting Prostate Cancer Molecular Subtype With Deep Learning on Histopathologic Images.

Microscopic examination of prostate cancer has failed to reveal a reproducible association between molecular and morphologic features. However, deep-learning algorithms trained on hematoxylin and eosin (H&E)-stained whole slide images (WSI) may outperform the human eye and help to screen for clinically-relevant genomic alterations. We created deep-learning algorithms to identify prostate tumors with underlying ETS-related gene (ERG) fusions or PTEN deletions using the following 4 stages: (1) automated tumor identification, (2) feature representation learning, (3) classification, and (4) explainability map generation. A novel transformer-based hierarchical architecture was trained on a single representative WSI of the dominant tumor nodule from a radical prostatectomy (RP) cohort with known ERG/PTEN status (n = 224 and n = 205, respectively). Two distinct vision transformer-based networks were used for feature extraction, and a distinct transformer-based model was used for classification. The ERG algorithm performance was validated across 3 RP cohorts, including 64 WSI from the pretraining cohort (AUC, 0.91) and 248 and 375 WSI from 2 independent RP cohorts (AUC, 0.86 and 0.89, respectively). In addition, we tested the ERG algorithm performance in 2 needle biopsy cohorts comprised of 179 and 148 WSI (AUC, 0.78 and 0.80, respectively). Focusing on cases with homogeneous (clonal) PTEN status, PTEN algorithm performance was assessed using 50 WSI reserved from the pretraining cohort (AUC, 0.81), 201 and 337 WSI from 2 independent RP cohorts (AUC, 0.72 and 0.80, respectively), and 151 WSI from a needle biopsy cohort (AUC, 0.75). For explainability, the PTEN algorithm was also applied to 19 WSI with heterogeneous (subclonal) PTEN loss, where the percentage tumor area with predicted PTEN loss correlated with that based on immunohistochemistry (r = 0.58, P = .0097). These deep-learning algorithms to predict ERG/PTEN status prove that H&E images can be used to screen for underlying genomic alterations in prostate cancer.

PTEN Loss Is Associated with Adverse Outcomes in the Setting of Salvage Radiation Therapy.

BACKGROUND: Salvage radiation therapy (SRT) is a mainstay of treatment for biochemical relapse following radical prostatectomy; however, few studies have examined genomic biomarkers in this context. OBJECTIVE: We characterized the prognostic impact of previously identified deleterious molecular phenotypes-loss of PTEN, ERG expression, and TP53 mutation-for patients undergoing SRT. DESIGN, SETTING, AND PARTICIPANTS: We leveraged an institutional database of 320 SRT patients with available tissue and follow-up. Tissue microarrays were used for genetically validated immunohistochemistry assays. INTERVENTION: All men underwent SRT with or without androgen deprivation therapy OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS: Univariable and multivariable Cox-proportional hazard models assessed the association of molecular phenotypes with biochemical recurrence-free (bRFS) and metastasis-free (MFS) survival after SRT. RESULTS AND LIMITATIONS: Loss of PTEN (n = 123, 43%) and ERG expression (n = 118, 39%) were common in this cohort, while p53 overexpression (signifying TP53 missense mutation) was infrequent (n = 21, 7%). In univariable analyses, any loss of PTEN portended worse bRFS (hazard ratio [HR] 1.86; 95% confidence interval 1.36-2.57) and MFS (HR 1.89; 1.21-2.94), with homogeneous PTEN loss being associated with the highest risk of MFS (HR 2.47; 1.54-3.95). Similarly, p53 overexpression predicted worse bRFS (HR 1.95; 1.14-3.32) and MFS (HR 2.79; 1.50-5.19). ERG expression was associated with worse MFS only (HR 1.6; 1.03-2.48). On the multivariable analysis adjusting for known prognostic features, homogeneous PTEN loss remained predictive of adverse bRFS (HR 1.82; 1.12-2.96) and MFS (HR 2.08; 1.06-4.86). The study is limited by its retrospective and single-institution design. CONCLUSIONS: PTEN loss by immunohistochemistry is an independent adverse prognostic factor for bRFS and MFS in prostate cancer patients treated with SRT. Future trials will determine the optimal approach to treating SRT patients with adverse molecular prognostic features. PATIENT SUMMARY: Loss of the PTEN tumor suppressor protein is associated with worse outcomes after salvage radiotherapy, independent of other clinical or pathologic patient characteristics.

FASN Gene Methylation is Associated with Fatty Acid Synthase Expression and Clinical-genomic Features of Prostate Cancer.

Fatty acid synthase (FASN) catalyzes the synthesis of long-chain saturated fatty acids and is overexpressed during prostatic tumorigenesis, where it is the therapeutic target in several ongoing trials. However, the mechanism of FASN upregulation in prostate cancer remains unclear. Here, we examine FASN gene CpG methylation pattern by InfiniumEPIC profiling and whole-genome bisulfite sequencing across multiple racially diverse primary and metastatic prostate cancer cohorts, comparing with FASN protein expression as measured by digitally quantified IHC assay and reverse phase protein array analysis or FASN gene expression. We demonstrate that the FASN gene body is hypomethylated and overexpressed in primary prostate tumors compared with benign tissue, and FASN gene methylation is significantly inversely correlated with FASN protein or gene expression in both primary and metastatic prostate cancer. Primary prostate tumors with ERG gene rearrangement have increased FASN expression and we find evidence of FASN hypomethylation in this context. FASN expression is also significantly increased in prostate tumors from carriers of the germline HOXB13 G84E mutation compared with matched controls, consistent with a report that HOXB13 may contribute to epigenetic regulation of FASN in vitro. However, in contrast to previous studies, we find no significant association of FASN expression or methylation with self-identified race in models that include ERG status across two independent primary tumor cohorts. Taken together, these data support a potential epigenetic mechanism for FASN regulation in the prostate which may be relevant for selecting patients responsive to FASN inhibitors. SIGNIFICANCE: Here, we leverage multiple independent primary and metastatic prostate cancer cohorts to demonstrate that FASN gene body methylation is highly inversely correlated with FASN gene and protein expression. This finding may shed light on epigenetic mechanisms of FASN regulation in prostate cancer and provides a potentially useful biomarker for selecting patients in future trials of FASN inhibitors.

Comparison of Pathologist and Artificial Intelligence-based Grading for Prediction of Metastatic Outcomes After Radical Prostatectomy.

Gleason grade group (GG) is the most powerful prognostic variable in localized prostate cancer; however, interobserver variability remains a challenge. Artificial intelligence algorithms applied to histopathologic images standardize grading, but most have been tested only for agreement with pathologist GG, without assessment of performance with respect to oncologic outcomes. We compared deep learning-based and pathologist-based GGs for an association with metastatic outcome in three surgical cohorts comprising 777 unique patients. A digitized whole slide image of the representative hematoxylin and eosin-stained slide of the dominant tumor nodule was assigned a GG by an artificial intelligence-based grading algorithm and was compared with the GG assigned by a contemporary pathologist or the original pathologist-assigned GG for the entire prostatectomy. Harrell's C-indices based on Cox models for time to metastasis were compared. In a combined analysis of all cohorts, the C-index for the artificial intelligence-assigned GG was 0.77 (95% confidence interval [CI]: 0.73-0.81), compared with 0.77 (95% CI: 0.73-0.81) for the pathologist-assigned GG. By comparison, the original pathologist-assigned GG for the entire case had a C-index of 0.78 (95% CI: 0.73-0.82). PATIENT SUMMARY: Artificial intelligence-enabled prostate cancer grading on a single slide was comparable with pathologist grading for predicting metastatic outcome in men treated by radical prostatectomy, enabling equal access to expert grading in lower resource settings.