BACKGROUND Localized high-risk prostate cancer (PCa) often recurs despite neoadjuvant androgen deprivation therapy (ADT). We sought to identify baseline molecular programs that predict pathologic response and reveal targetable vulnerabilities.METHODS We profiled 147 biopsy foci from 48 MRI-visible lesions in 37 patients before 6 months of ADT plus enzalutamide and radical prostatectomy. Residual cancer burden (RCB) at prostatectomy was the primary outcome. Analyses incorporated PTEN loss, TMPRSS2:ERG status, and HER2/androgen receptor (AR) immunohistochemistry on baseline and posttreatment tissues. Findings were evaluated in an external transcriptional cohort (n = 121) and by multiplex immunostaining in an independent cohort (n = 61). Functional assays tested enzalutamide-responsive enhancers near ERBB2 and sensitivity to HER2 inhibition.RESULTS A baseline, HER2-associated transcriptional program correlated with higher RCB and inversely with AR activity, independent of PTEN and ERG. Exceptional responders had lower HER2 protein levels in pretreatment biopsy specimens. The inverse AR-HER2 relationship recurred across data sets and multiplex immunostaining, which revealed coexisting AR-high/HER2-low and HER2-high/AR-low subpopulations. Enzalutamide inhibited AR-mediated repression of ERBB2. HER2-high/AR-low cells present before therapy resisted ADT yet were sensitive to HER2 inhibitors; combining HER2 inhibitors with enzalutamide increased tumor cell killing. These findings were reproduced in the external cohort and orthogonal assays.CONCLUSION Baseline HER2 activity marks intrinsic resistance to neoadjuvant ADT in localized high-risk PCa and identifies a preexisting, targetable AR-low subpopulation. HER2-directed therapy, alone or with AR blockade, warrants clinical evaluation.TRIAL REGISTRATION ClinicalTrials.gov registration: NCT02430480.FUNDING Prostate Cancer Foundation; Department of Defense Prostate Cancer Research Program; National Institutes of Health.
Scott Wilkinson, Anson T. Ku, Rosina T. Lis, Isaiah M. King, Daniel Low, Shana Y. Trostel, John R. Bright, Nicholas T. Terrigino, Anna Baj, Emily R. Summerbell, Kayla E. Heyward, Sumeyra Kartal, John M. Fenimore, Chennan Li, Cassandra Singler, BaoHan Vo, Caroline S. Jansen, Huihui Ye, Nichelle C. Whitlock, Stephanie A. Harmon, Nicole V. Carrabba, Rayann Atway, Ross Lake, David Y. Takeda, Haydn T. Kissick, Peter A. Pinto, Peter L. Choyke, Baris Turkbey, William L. Dahut, Fatima Karzai, Adam G. Sowalsky
Usage data is cumulative from September 2025 through August 2026.
| Usage | JCI | PMC |
|---|---|---|
| Text version | 3,615 | 337 |
| 876 | 141 | |
| Figure | 1,202 | 0 |
| Supplemental data | 724 | 55 |
| Citation downloads | 290 | 0 |
| Totals | 6,707 | 533 |
| Total Views | 7,240 | |
Usage information is collected from two different sources: this site (JCI) and Pubmed Central (PMC). JCI information (compiled daily) shows human readership based on methods we employ to screen out robotic usage. PMC information (aggregated monthly) is also similarly screened of robotic usage.
Various methods are used to distinguish robotic usage. For example, Google automatically scans articles to add to its search index and identifies itself as robotic; other services might not clearly identify themselves as robotic, or they are new or unknown as robotic. Because this activity can be misinterpreted as human readership, data may be re-processed periodically to reflect an improved understanding of robotic activity. Because of these factors, readers should consider usage information illustrative but subject to change.