https://orcid.org/0000-0002-0328-8359
Abstract
The phase 3 CALGB 90203 (Alliance) trial evaluated neoadjuvant chemohormonal therapy for high-risk localized prostate cancer before radical prostatectomy. We dissected the molecular features of post-treated tumors with long-term clinical outcomes to explore mechanisms of response and resistance to chemohormonal therapy.
We evaluated 471 radical prostatectomy tumors, including 294 samples from 166 patients treated with 6 cycles of docetaxel plus androgen deprivation therapy before radical prostatectomy and 177 samples from 97 patients in the control arm (radical prostatectomy alone). Targeted DNA sequencing and RNA expression of tumor foci and adjacent noncancer regions were analyzed in conjunction with pathologic changes and clinical outcomes.
Tumor fraction estimated from DNA sequencing was significantly lower in post-treated tumor tissues after chemohormonal therapy compared with controls. Higher tumor fraction after chemohormonal therapy was associated with aggressive pathologic features and poor outcomes, including prostate-specific antigen–progression-free survival. SPOP alterations were infrequently detected after chemohormonal therapy, while TP53 alterations were enriched and associated with shorter overall survival. Residual tumor fraction after chemohormonal therapy was linked to higher expression of androgen receptor–regulated genes, cell cycle genes, and neuroendocrine genes, suggesting persistent populations of active prostate cancer cells. Supervised clustering of post–treated high-tumor-fraction tissues identified a group of patients with elevated cell cycle–related gene expression and poor clinical outcomes.
Distinct recurrent prostate cancer genomic and transcriptomic features are observed after exposure to docetaxel and androgen deprivation therapy. Tumor fraction assessed by DNA sequencing quantifies pathologic response and could be a useful trial endpoint or prognostic biomarker. TP53 alterations and high cell cycle transcriptomic activity are linked to aggressive residual disease, despite potent chemohormonal therapy.
Radical prostatectomy is a standard treatment modality for patients with clinically localized prostate cancer. High-risk features, including serum prostate-specific antigen (PSA) greater than 20 ng/mL, advanced clinical tumor stage (≥cT3a), and high tumor grade (Grade Group 4 or 5), portends a higher risk of disease progression after radical prostatectomy (1-3). The multicenter phase 3 Cancer and Leukemia Group B (CALGB) 90203 (PUNCH) trial evaluated whether neoadjuvant chemohormonal therapy (ie, the combination of androgen-deprivation therapy [ADT] with 6 cycles of docetaxel) followed by radical prostatectomy improves outcomes in men with high-risk, localized prostate cancer compared with radical prostatectomy alone (4). CALGB is now part of the Alliance for Clinical Trials in Oncology. Although the trial did not meet its primary endpoint of 3-year biochemical progression-free survival (PFS), improvements in secondary endpoints, including metastasis-free survival and overall survival, suggested that some patients may benefit from neoadjuvant chemohormonal therapy. Post–treated tissues from this trial represent a unique resource to understand molecular characteristics of hormone-sensitive prostate cancer that persists following ADT and 6 cycles of docetaxel chemotherapy.
Docetaxel chemotherapy has been used for select patients with metastatic, hormone-sensitive prostate cancer in combination with ADT and more recently as triplet therapy with more potent androgen receptor pathway inhibitors (5-7) as well as in the metastatic castration-resistant prostate cancer state (5,8). The molecular features of resistance are poorly defined because of the challenge of collecting a standardized cohort of post-treated metastatic tissues. Taxane sensitivity in metastatic prostate cancer is loosely linked with microtubule alterations, androgen receptor signaling, ETS family gene rearrangements, and PI3K/AKT signaling (9-14). Studies of tumor tissues obtained from neoadjuvant trials have proved promising for identifying potential drivers of treatment resistance [including upregulation of targetable stress-related pathways (15,16)], but few have examined chemotherapy-treated specimens. In prior phase 2 neoadjuvant studies of androgen receptor pathway inhibitors, pathologic response after treatment was associated with improved biochemical PFS (17,18). Whole-exome sequencing, RNA sequencing, or immunohistochemistry applied to diagnostic or radical prostatectomy prostate cancer tissue has also revealed an association among TP53 alteration, PTEN loss, SPOP alteration, and pathologic treatment effect with androgen receptor pathway inhibitors, but relatively small numbers of patients were analyzed (19-22).
In a preliminary evaluation of tissues from 52 patients enrolled in CALGB 90203 at 1 Canadian academic center, we demonstrated the potential for targeted DNA and RNA profiling to inform tumor biology, despite scant tissue and treatment effects (23). Here, we investigated how molecular features after neoadjuvant chemohormonal therapy are associated with pathologic changes and longer-term outcomes in the larger multicenter cohort enrolled in CALGB 90203. We hypothesized that 1) neoadjuvant docetaxel and ADT would alter the molecular landscape of localized, high-risk prostate cancer and 2) residual tumor populations would reveal correlates of intrinsic or early acquired resistance to chemohormonal therapy.
Methods
Patient sample cohort
Patients with newly diagnosed high-risk, localized prostate cancer were enrolled in the phase 3 CALGB (Alliance) 90203 trial from 2006 to 2015 across the United States and Canada (ClinicalTrials.gov. ID NCT00430183). Entry criteria were described in the primary analysis (4). Patients assigned to the neoadjuvant therapy and radical prostatectomy arm (arm A) received docetaxel at a dose of 75 mg/m2 every 3 weeks for 6 cycles, with ADT (luteinizing hormone-releasing hormone agonist or antagonist without an antiandrogen) for 18 to 24 weeks followed by radical prostatectomy. Patients assigned to the radical prostatectomy–only arm (arm B) underwent radical prostatectomy without chemohormonal therapy, although these patients were permitted to receive up to 3 months of neoadjuvant ADT before study enrollment, despite the primary endpoint of the original study being biochemical PFS. Patients in the trial were observed for a median of 6.1 years (range = 0-12.1 years). For this exploratory correlative study, we retrieved all available tissues from arm A patients from the central Alliance biorepository and randomly selected a subset of arm B cases for comparison. A total of 294 tumor tissues from 166 patients in arm A and 177 tumor tissues from 97 patients in arm B were available (1-5 per patient) (Figure 1, A; Supplementary Figure 1, available online). The objectives of this correlative study were included in the trial protocol, but no prespecified hypotheses were set. All patients provided written informed consent. The study received ethical approval at all participating hospitals and was designed and conducted in accordance with the Declaration of Helsinki.
Relationship between sequencing-based tumor fraction and pathologic features or clinical outcomes. A) Study design. B) Box plot showing the sequencing-based tumor fraction (also known as tumor purity) in samples from arm A vs arm B. At the bottom of the plot, horizontal blue or orange filled rectangles illustrate the proportion of samples in each arm that had evidence of residual tumor DNA by sequencing. C) Association between sequencing-based tumor fraction and pathologic tumor cellularity (from assessment of an hematoxylin-eosin slide) in each arm. No samples from arm B had <20% tumor cellularity on hematoxylin-eosin slides by pathology assessment. D) Association between sequencing-based tumor fraction and pathologic features in samples from arm A (for results from arm B samples, see Supplementary Figure 7, available online). E, F) Kaplan-Meier survival analysis for PSA progression-free survival and event-free survival in patients with and without sequencing-based tumor fractions detected in arm A. For patients with multiple samples, the sample with the highest sequencing-based tumor fraction was used. G) Multivariate analysis for patients in arm A. Twenty-five patients were excluded from this model for lack of annotation for intraductal carcinoma or unevaluable pathologic tumor cellularity (from sole use of core sections). P values were estimated using a Fisher exact test or Mann-Whitney U test (B, D) or univariate Cox proportional hazards regression analysis (E, F). CHT = chemohormone therapy; CI = confidence interval; HR = hazard ratio; ISUP = International Society of Urological Pathology; mRNA = messenger RNA; NR = not reported; PSA = prostate-specific antigen; pts = patients; RP = radical prostatectomy; REF = reference group.
Pathologic evaluation
Formalin-fixed, paraffin-embedded tissue blocks from radical prostatectomy specimens were centrally reviewed by the study pathologist (J.S.) for tumor cellularity, Gleason score or International Society of Urological Pathology (ISUP) grade, treatment effect, and other morphologic features. Three different sources of formalin-fixed, paraffin-embedded tissue were available on a per-patient basis for nucleic acid extraction: standard sections, whole-mount sections, and core sections. From each available section, 10 to 20 unstained slides were taken, and tumor or adjacent benign regions were marked on a corresponding hematoxylin and eosin–stained slide. Tumor and benign regions were scraped from unstained slides and divided equally for DNA and RNA extraction. Samples for DNA extraction were prioritized when there was insufficient tissue for both DNA and RNA profiling.
Targeted DNA sequencing
Targeted DNA sequencing was performed using an established panel of prostate cancer genes at the Vancouver Prostate Centre (Supplementary Table 1, available online) (24,25); see Supplementary Methods (available online) for full experimental and bioinformatic details, including criteria for calling alterations and copy number changes. We applied 3 orthogonal methods to estimate tumor fraction in sequencing data (termed sequencing-based tumor fraction). Sequencing-based tumor fraction used the variant allele frequency of somatic mutations, deviation in the b-allele fraction of heterozygous germline single-nucleotide variations (formerly SNPs), and absolute coverage log ratio changes for prostate cancer–associated copy number alterations, including deletions of NKX3-1, TP53, PTEN, and RB1 and MYC gain (Supplementary Figure 2 and 3; Supplementary Methods, available online).
Targeted RNA profiling
Tumor messenger RNA was profiled at Weill Cornell Medicine and the Dana-Farber Cancer Institute using the nCounter Analysis System (NanoString Technologies, Inc, Seattle, WA), focusing on 155 prostate cancer–relevant genes, including AR and androgen receptor signaling genes, the AR V7 splice variant, epithelial-mesenchymal transition, plasticity, and neuroendocrine prostate cancer–associated genes, cell cycle, WNT, PI3K/AKT pathway genes, TMPRSS2-ERG fusion transcript, and control and housekeeper genes (Supplementary Table 2, available online). NanoString raw counts were normalized by a RUVSeq-based process (26) (Supplementary Figure 4, available online), and the DESeq2 package was applied to determine differentially expressed genes when comparing arm A with arm B cases and the sequencing-based tumor fraction positive to negative tumors in arm A (27). For RNA analyses, we adjusted for multiple testing using the Benjamini-Hochberg correction. Differential gene expression was considered statistically significant if the adjusted P value was less than .05 and the fold-change was more than 2 or approximately 1.4 for comparisons in arms or sequencing-based tumor fraction tumors, respectively.
Clinical outcomes and statistical analysis
PSA progression was defined as a serum PSA level above 0.2 ng/mL that increased on 2 consecutive occasions that were at least 3 months apart. PSA-progression free survival was measured from the date of randomization to the date of the first PSA level above 0.2 ng/mL or death from any cause. Event-free survival was defined as the time from date of randomization to date of PSA progression, subsequent therapy after radical prostatectomy, local or distant progression, or death. Overall survival was defined as the time from the date of randomization to the date of death from any cause. All analyses were based on the study database lock on November 4, 2019 (4).
Statistical tests and data analyses were conducted in Python, version 3.7.11, using Pandas, NumPy, SciPy, and Lifelines survival analysis (28). Categorical and continuous variables were compared using the Fisher exact test and the Mann-Whitney U test (29), respectively. Pearson r values and associated P values were calculated using linear regression. PSA progression survival functions were estimated using the Kaplan-Meier method, and hazard ratios (HRs) were calculated using univariate Cox proportional hazard regression (30,31). Multivariable Cox proportional hazards models were used to test the independent prognostic influence of sequencing-based tumor fraction and included pathologic tumor cellularity (dichotomized at median ≥50% vs <50%), ISUP grade (≥4 vs ≤3 or no grade), pathologic treatment effect (severe vs nonsevere), and intraductal carcinoma (present vs absent) as binary covariates (31). No sensitivity analyses were performed for these binary thresholds. All hypothesis tests were 2-tailed, required a P < .05 significance threshold, and were unadjusted for multiple testing, unless otherwise stated. There was no predefined statistical analysis plan.
Results
Clinical and pathologic features of the correlative cohort
Baseline preoperative characteristics for the 263 patients (arm A = 166 patients, arm B = 97 patients) with radical prostatectomy tissue samples are summarized in Table 1 (Supplementary Table 3, available online). Within each arm, there were no differences in characteristics compared with the entire trial population (Supplementary Table 4, available online). For the subgroup of patients in this exploratory correlative study, however, those in arm A had higher biopsy Gleason scores and lower diagnostic PSA than those in arm B (Table 1). Regardless, more than 90% of the patients in each arm were classified as National Comprehensive Cancer Network high risk, with no differences between the arms. Of note, 23 of 97 (23.7%) patients in arm B experienced a more than 50% reduction in PSA levels before prostatectomy, likely because patients in arm B were permitted to receive up to 3 months of ADT. Median follow-up was 7.2 years (range = 0.1-12.1 years).
Clinical characteristics of patients in the correlative study cohort
| Arm A (n = 166) | Arm B (n = 97) | Pa | |
|---|---|---|---|
| Age at diagnosis, y | .697 | ||
| Median | 62 | 62 | |
| Interquartile range | 58-66 | 58-66 | |
| Clinical tumor stage,b No. (%) | .361 | ||
| T1 | 47 (28.3) | 21 (21.6) | |
| T2 | 95 (57.2) | 57 (58.8) | |
| T3 | 24 (14.5) | 19 (19.6) | |
| Biopsy Gleason score,b No. (%) | .001 | ||
| 7 | 10 (6.0) | 19 (19.6) | |
| 8 | 62 (37.3) | 35 (36.1) | |
| 9 | 83 (50) | 38 (39.2) | |
| 10 | 11 (6.6) | 5 (5.2) | |
| Prostate-specific antigen level at diagnosis, ng/mL | .007 | ||
| Median | 8.7 | 13.2 | |
| Interquartile range | 5.3-16.9 | 6.2-27.5 | |
| Prostate-specific antigen level before radical prostatectomy, ng/mL | <.001 | ||
| Median | 0.4 | 6.4 | |
| Interquartile range | 0.2-1.2 | 2.4-19.7 | |
| Prostate-specific antigen level change rate, %c | <.001 | ||
| Median | ‒95.2 | ‒10.3 | |
| Interquartile range | ‒97.8 to ‒90.0 | ‒74.9 to 12.9 | |
| National Comprehensive Cancer Network risk group, No. (%) | .226 | ||
| Intermediate | 4 (2.4) | 6 (6.2) | |
| High | 162 (97.6) | 91 (93.8) | |
| Clinical outcomes, No. (%) | |||
| Prostate-specific antigen progression | 21 (13.1) | 17 (17.5) | .282 |
| Eventd | 90 (56.3) | 58 (59.7) | .440 |
| Death | 12 (7.5) | 14 (14.4) | .085 |
| Arm A (n = 166) | Arm B (n = 97) | Pa | |
|---|---|---|---|
| Age at diagnosis, y | .697 | ||
| Median | 62 | 62 | |
| Interquartile range | 58-66 | 58-66 | |
| Clinical tumor stage,b No. (%) | .361 | ||
| T1 | 47 (28.3) | 21 (21.6) | |
| T2 | 95 (57.2) | 57 (58.8) | |
| T3 | 24 (14.5) | 19 (19.6) | |
| Biopsy Gleason score,b No. (%) | .001 | ||
| 7 | 10 (6.0) | 19 (19.6) | |
| 8 | 62 (37.3) | 35 (36.1) | |
| 9 | 83 (50) | 38 (39.2) | |
| 10 | 11 (6.6) | 5 (5.2) | |
| Prostate-specific antigen level at diagnosis, ng/mL | .007 | ||
| Median | 8.7 | 13.2 | |
| Interquartile range | 5.3-16.9 | 6.2-27.5 | |
| Prostate-specific antigen level before radical prostatectomy, ng/mL | <.001 | ||
| Median | 0.4 | 6.4 | |
| Interquartile range | 0.2-1.2 | 2.4-19.7 | |
| Prostate-specific antigen level change rate, %c | <.001 | ||
| Median | ‒95.2 | ‒10.3 | |
| Interquartile range | ‒97.8 to ‒90.0 | ‒74.9 to 12.9 | |
| National Comprehensive Cancer Network risk group, No. (%) | .226 | ||
| Intermediate | 4 (2.4) | 6 (6.2) | |
| High | 162 (97.6) | 91 (93.8) | |
| Clinical outcomes, No. (%) | |||
| Prostate-specific antigen progression | 21 (13.1) | 17 (17.5) | .282 |
| Eventd | 90 (56.3) | 58 (59.7) | .440 |
| Death | 12 (7.5) | 14 (14.4) | .085 |
Unadjusted P values are provided to help contextualize the genomic results.
Fisher exact tests were performed using the following categories for clinical tumor stage and biopsy Gleason score, respectively: T1/2 vs T3 and ≤7 and ≥8.
Percentage change in prostate-specific antigen values between diagnosis and to before radical prostatectomy.
An event was defined as prostate-specific antigen progression, subsequent therapy after radical prostatectomy, local or distant progression, or death.
Clinical characteristics of patients in the correlative study cohort
| Arm A (n = 166) | Arm B (n = 97) | Pa | |
|---|---|---|---|
| Age at diagnosis, y | .697 | ||
| Median | 62 | 62 | |
| Interquartile range | 58-66 | 58-66 | |
| Clinical tumor stage,b No. (%) | .361 | ||
| T1 | 47 (28.3) | 21 (21.6) | |
| T2 | 95 (57.2) | 57 (58.8) | |
| T3 | 24 (14.5) | 19 (19.6) | |
| Biopsy Gleason score,b No. (%) | .001 | ||
| 7 | 10 (6.0) | 19 (19.6) | |
| 8 | 62 (37.3) | 35 (36.1) | |
| 9 | 83 (50) | 38 (39.2) | |
| 10 | 11 (6.6) | 5 (5.2) | |
| Prostate-specific antigen level at diagnosis, ng/mL | .007 | ||
| Median | 8.7 | 13.2 | |
| Interquartile range | 5.3-16.9 | 6.2-27.5 | |
| Prostate-specific antigen level before radical prostatectomy, ng/mL | <.001 | ||
| Median | 0.4 | 6.4 | |
| Interquartile range | 0.2-1.2 | 2.4-19.7 | |
| Prostate-specific antigen level change rate, %c | <.001 | ||
| Median | ‒95.2 | ‒10.3 | |
| Interquartile range | ‒97.8 to ‒90.0 | ‒74.9 to 12.9 | |
| National Comprehensive Cancer Network risk group, No. (%) | .226 | ||
| Intermediate | 4 (2.4) | 6 (6.2) | |
| High | 162 (97.6) | 91 (93.8) | |
| Clinical outcomes, No. (%) | |||
| Prostate-specific antigen progression | 21 (13.1) | 17 (17.5) | .282 |
| Eventd | 90 (56.3) | 58 (59.7) | .440 |
| Death | 12 (7.5) | 14 (14.4) | .085 |
| Arm A (n = 166) | Arm B (n = 97) | Pa | |
|---|---|---|---|
| Age at diagnosis, y | .697 | ||
| Median | 62 | 62 | |
| Interquartile range | 58-66 | 58-66 | |
| Clinical tumor stage,b No. (%) | .361 | ||
| T1 | 47 (28.3) | 21 (21.6) | |
| T2 | 95 (57.2) | 57 (58.8) | |
| T3 | 24 (14.5) | 19 (19.6) | |
| Biopsy Gleason score,b No. (%) | .001 | ||
| 7 | 10 (6.0) | 19 (19.6) | |
| 8 | 62 (37.3) | 35 (36.1) | |
| 9 | 83 (50) | 38 (39.2) | |
| 10 | 11 (6.6) | 5 (5.2) | |
| Prostate-specific antigen level at diagnosis, ng/mL | .007 | ||
| Median | 8.7 | 13.2 | |
| Interquartile range | 5.3-16.9 | 6.2-27.5 | |
| Prostate-specific antigen level before radical prostatectomy, ng/mL | <.001 | ||
| Median | 0.4 | 6.4 | |
| Interquartile range | 0.2-1.2 | 2.4-19.7 | |
| Prostate-specific antigen level change rate, %c | <.001 | ||
| Median | ‒95.2 | ‒10.3 | |
| Interquartile range | ‒97.8 to ‒90.0 | ‒74.9 to 12.9 | |
| National Comprehensive Cancer Network risk group, No. (%) | .226 | ||
| Intermediate | 4 (2.4) | 6 (6.2) | |
| High | 162 (97.6) | 91 (93.8) | |
| Clinical outcomes, No. (%) | |||
| Prostate-specific antigen progression | 21 (13.1) | 17 (17.5) | .282 |
| Eventd | 90 (56.3) | 58 (59.7) | .440 |
| Death | 12 (7.5) | 14 (14.4) | .085 |
Unadjusted P values are provided to help contextualize the genomic results.
Fisher exact tests were performed using the following categories for clinical tumor stage and biopsy Gleason score, respectively: T1/2 vs T3 and ≤7 and ≥8.
Percentage change in prostate-specific antigen values between diagnosis and to before radical prostatectomy.
An event was defined as prostate-specific antigen progression, subsequent therapy after radical prostatectomy, local or distant progression, or death.
The central pathology review of prostatectomy tissue sections in this study is summarized in Table 2 (Supplementary Table 5, available online). As expected, tissues from arm A had lower pathologic tumor cellularity (ie, percentage of tumor cells) than tissues from arm B (median, 40% vs 80%, P < .001, Mann-Whitney U test). Sixty percent of evaluated tumor tissues from arm A showed evidence of severe treatment effect (Supplementary Figure 5, available online). The proportion of tissues with high ISUP grade (ie, ≥4) was lower in arm A than in arm B (37.2% [105/282] vs 57.1% [100/175], P < .001), although 28% of tumor tissues in arm A could not be assigned a grade because of treatment effect.
Pathologic characteristics of samples in the correlative study cohorta
| Arm A (n = 294)b | Arm B (n = 177)b | P | |
|---|---|---|---|
| Tumor cells, % | <.001 | ||
| Median | 40 | 80 | |
| Interquartile range | 30-70 | 60-80 | |
| Treatment effect,b No. (%) | <.001 | ||
| None | 17 (6.0) | 111 (63.4) | |
| Minimal | 12 (4.3) | 21 (12) | |
| Mild | 38 (13.5) | 22 (12.6) | |
| Moderate | 48 (17.0) | 7 (4) | |
| Severe | 167 (59.2) | 11 (6.3) | |
| Unknown | — | 3 (1.7) | |
| Radical prostatectomy ISUP grade,b No. (%) | <.001 | ||
| 1 | 5 (1.8) | 5 (2.9) | |
| 2 | 34 (12.1) | 27 (15.4) | |
| 3 | 58 (20.6) | 41 (23.4) | |
| 4 | 20 (7.1) | 18 (10.3) | |
| 5 | 85 (30.1) | 82 (46.9) | |
| No grade | 80 (28.4) | — | |
| Unknown | — | 2 (1.1) | |
| Intraductal carcinoma,c No. (%) | .071 | ||
| Present | 49 (17.4) | 43 (24.6) | |
| Absent | 175 (62.1) | 96 (54.9) | |
| Unknown | 58 (20.6) | 36 (20.6) |
| Arm A (n = 294)b | Arm B (n = 177)b | P | |
|---|---|---|---|
| Tumor cells, % | <.001 | ||
| Median | 40 | 80 | |
| Interquartile range | 30-70 | 60-80 | |
| Treatment effect,b No. (%) | <.001 | ||
| None | 17 (6.0) | 111 (63.4) | |
| Minimal | 12 (4.3) | 21 (12) | |
| Mild | 38 (13.5) | 22 (12.6) | |
| Moderate | 48 (17.0) | 7 (4) | |
| Severe | 167 (59.2) | 11 (6.3) | |
| Unknown | — | 3 (1.7) | |
| Radical prostatectomy ISUP grade,b No. (%) | <.001 | ||
| 1 | 5 (1.8) | 5 (2.9) | |
| 2 | 34 (12.1) | 27 (15.4) | |
| 3 | 58 (20.6) | 41 (23.4) | |
| 4 | 20 (7.1) | 18 (10.3) | |
| 5 | 85 (30.1) | 82 (46.9) | |
| No grade | 80 (28.4) | — | |
| Unknown | — | 2 (1.1) | |
| Intraductal carcinoma,c No. (%) | .071 | ||
| Present | 49 (17.4) | 43 (24.6) | |
| Absent | 175 (62.1) | 96 (54.9) | |
| Unknown | 58 (20.6) | 36 (20.6) |
Note that 12 tumor samples in arm A and 2 tumor samples in arm B had no pathologic information. ISUP = International Society of Urological Pathology.
Fisher exact tests were performed with the following classifications: “severe” and other effects for treatment effect and “≤2” and “≥3” for radical prostatectomy ISUP grade. Tumor samples with no radical prostatectomy ISUP grade (n = 80 in arm A) were also excluded from the analysis.
Tumor samples with no evaluation of intraductal carcinoma were excluded from Fisher exact tests.
Pathologic characteristics of samples in the correlative study cohorta
| Arm A (n = 294)b | Arm B (n = 177)b | P | |
|---|---|---|---|
| Tumor cells, % | <.001 | ||
| Median | 40 | 80 | |
| Interquartile range | 30-70 | 60-80 | |
| Treatment effect,b No. (%) | <.001 | ||
| None | 17 (6.0) | 111 (63.4) | |
| Minimal | 12 (4.3) | 21 (12) | |
| Mild | 38 (13.5) | 22 (12.6) | |
| Moderate | 48 (17.0) | 7 (4) | |
| Severe | 167 (59.2) | 11 (6.3) | |
| Unknown | — | 3 (1.7) | |
| Radical prostatectomy ISUP grade,b No. (%) | <.001 | ||
| 1 | 5 (1.8) | 5 (2.9) | |
| 2 | 34 (12.1) | 27 (15.4) | |
| 3 | 58 (20.6) | 41 (23.4) | |
| 4 | 20 (7.1) | 18 (10.3) | |
| 5 | 85 (30.1) | 82 (46.9) | |
| No grade | 80 (28.4) | — | |
| Unknown | — | 2 (1.1) | |
| Intraductal carcinoma,c No. (%) | .071 | ||
| Present | 49 (17.4) | 43 (24.6) | |
| Absent | 175 (62.1) | 96 (54.9) | |
| Unknown | 58 (20.6) | 36 (20.6) |
| Arm A (n = 294)b | Arm B (n = 177)b | P | |
|---|---|---|---|
| Tumor cells, % | <.001 | ||
| Median | 40 | 80 | |
| Interquartile range | 30-70 | 60-80 | |
| Treatment effect,b No. (%) | <.001 | ||
| None | 17 (6.0) | 111 (63.4) | |
| Minimal | 12 (4.3) | 21 (12) | |
| Mild | 38 (13.5) | 22 (12.6) | |
| Moderate | 48 (17.0) | 7 (4) | |
| Severe | 167 (59.2) | 11 (6.3) | |
| Unknown | — | 3 (1.7) | |
| Radical prostatectomy ISUP grade,b No. (%) | <.001 | ||
| 1 | 5 (1.8) | 5 (2.9) | |
| 2 | 34 (12.1) | 27 (15.4) | |
| 3 | 58 (20.6) | 41 (23.4) | |
| 4 | 20 (7.1) | 18 (10.3) | |
| 5 | 85 (30.1) | 82 (46.9) | |
| No grade | 80 (28.4) | — | |
| Unknown | — | 2 (1.1) | |
| Intraductal carcinoma,c No. (%) | .071 | ||
| Present | 49 (17.4) | 43 (24.6) | |
| Absent | 175 (62.1) | 96 (54.9) | |
| Unknown | 58 (20.6) | 36 (20.6) |
Note that 12 tumor samples in arm A and 2 tumor samples in arm B had no pathologic information. ISUP = International Society of Urological Pathology.
Fisher exact tests were performed with the following classifications: “severe” and other effects for treatment effect and “≤2” and “≥3” for radical prostatectomy ISUP grade. Tumor samples with no radical prostatectomy ISUP grade (n = 80 in arm A) were also excluded from the analysis.
Tumor samples with no evaluation of intraductal carcinoma were excluded from Fisher exact tests.
The DNA and RNA yield from all tissue sources was lower in arm A than in arm B, consistent with the pathologic reduction in cellularity after chemohormonal therapy (Supplementary Figure 6, A; Supplementary Table 6, available online). Targeted DNA sequencing was completed in 92.5% of tumor samples (431/466; median read depth of 362×), and completion rate did not differ by arm (Supplementary Figure 6, B and C; Supplementary Table 7, available online). After RNA extraction, however, 46.0% (133/289) of tumor samples in arm A and 33.0% (38/115) in arm B had insufficient RNA yield or quality for downstream profiling, as expected because RNA is more susceptible to degradation than DNA.
Residual tumor fraction correlation with pathologic features and clinical outcomes
The median tumor fraction, as assessed by targeted DNA sequencing across all radical prostatectomy tumor tissue samples, was 18.1% (Supplementary Figure 3, C; Supplementary Table 8, available online). The sequencing-based tumor fraction was lower in samples from arm A than in arm B (median = 7.8% vs 32.6%, P < .001, Mann-Whitney U test) (Figure 1, B; Supplementary Table 4, available online). The proportion of samples with a positive sequencing-based tumor fraction estimate (ie, any evidence of residual tumor DNA) was also lower in arm A (154/272) than in arm B (137/159) (56.6% vs 86.2%, P < .001, Fisher exact test) (Figure 1, B). For the 22 evaluable arm B patients who experienced a more than 50% PSA reduction before radical prostatectomy, the sequencing-based tumor fraction was lower than the remainder of patients in arm B (median for all evaluable samples = 20.8% vs 38.5%, P < .001, Fisher exact test), but the proportion of samples with evidence of residual tumor DNA was similar (83.7% vs 87.1%) (Supplementary Figure 7, A, available online).
We observed a broad relationship between sequencing-based tumor fraction and pathologic characteristics in arm A. For example, 60.9% of samples in arm A with 20% or greater pathologic tumor cellularity had a positive sequencing-based tumor fraction compared with only 15.4% of samples with less than 20% cellularity (P < .001, Fisher exact test) (Figure 1, C). Twenty percent prostate cancer cellularity by pathology is a common minimum threshold for clinical genomics tests (32). Samples with a positive sequencing-based tumor fraction in arm A were more likely to exhibit disease with ISUP grade 4 or higher, no evidence of severe pathologic treatment effect, and to harbor intraductal carcinoma (Figure 1, D). Sequencing-based tumor fraction as a continuous variable was also positively related to these pathologic features (Figure 1, D). In arm B, no samples had a pathologic tumor cellularity below 20%, and there was no association between sequencing-based tumor fraction and pathologic features, except for ISUP grade (Supplementary Figure 7, B, available online).
PSA progression after radical prostatectomy was observed in 21 patients in arm A, and patients with sequencing-based tumor fraction–positive samples had shorter PSA-PFS than those with only sequencing-based tumor fraction–negative samples (HR = 6.3, 95% confidence interval [CI] = 1.9 to 21.5, P < .005) (Figure 1, E). Because 43.1% of arm A patients received additional treatment before biochemical recurrence, we examined event-free survival, which was also shorter in patients with sequencing-based tumor fraction–positive samples (HR = 2.1, 95% CI = 1.4 to 3.3, P < .005) (Figure 1, F). There was no association between PSA-PFS or event-free survival and pathologic tumor cellularity (Supplementary Figure 8, A and B, available online), although ISUP grade 4 or higher and intraductal carcinoma were associated with modestly poorer outcomes (Supplementary Figure 8, C-H, available online). The relationship between arm A sequencing-based tumor fraction and clinical outcomes remained statistically significant in a multivariable analysis, including pathologic characteristics from the matched tissue section (Figure 1, G). In arm B, sequencing-based tumor fraction was not associated with PSA-PFS or event-free survival (Supplementary Figure 7, C and D, available online). There were too few events (12 in arm A and 14 in arm B) to interpret overall survival differences (Supplementary Figure 9, available online).
Enrichment of TP53 alterations after chemohormone therapy
The most frequently mutated genes in evaluable samples (with positive sequencing-based tumor fraction) were TP53 and SPOP (Figure 2, A). These alterations were broadly nonoverlapping, although 1 patient in arm B exhibited an SPOP alteration in 2-sampled foci and a TP53 alteration in a third, suggesting independent tumors (Supplementary Figure 10, A, available online). Twenty-two of 91 (24.2%) evaluable patients in arm A exhibited somatic TP53 mutations in their radical prostatectomy samples compared with 13 of 82 (15.9%) in arm B (P = .190, Fisher exact test). Only 8 of 91 (8.8%) evaluable patients in arm A harbored SPOP alterations compared with 12 of 82 (14.6%) patients in arm B (P = .245, Fisher exact test) (Supplementary Figure 10, B, available online). Interestingly, SPOP-mutant tumors in arm A had a lower sequencing-based tumor fraction than those in arm B (median sequencing-based tumor fraction = 18.2% vs 50.0%, P = .001), while there was no strong difference in TP53-mutant tumors (34.3% in arm A vs 44.2% in arm B, P = .063 (Figure 2, A)). We did not observe an association between TP53 or SPOP alterations and pathologic treatment effect in either arm (Supplementary Figure 10, D and E, available online). There was evidence of an association between TP53 alterations and poor clinical outcomes in both arms, even after separately stratifying patients where sequencing-based tumor fraction was not detected (Figure 2, B-D; Supplementary Figure 11 and 12, available online). No potential relationships with outcomes were observed for SPOP alterations (Supplementary Figure 13, available online).
![Common genomic alterations present in residual cancers after chemohormone therapy. A) Driver genomic landscape of residual high-risk, localized prostate cancer after treatment with neoadjuvant chemohormonal therapy before radical prostatectomy (arm A) vs prostate cancers not exposed to therapy (arm B). Only patients who had molecular residual disease are shown (ie, detected sequencing-derived tumor fraction: 91 in arm A and 82 in arm B). The highest sequencing-derived tumor fraction sample from each patient is represented if multiple same-patient tumor samples were available. Patients were sorted by the presence or absence of alterations in TP53 or SPOP and sequencing-derived tumor fraction (high to low). Fourteen genes associated with prostate cancer were provided (see Supplementary Figure 13 for additional genes, available online). Per-arm sequencing-derived tumor fraction estimates in SPOP-mutant and TP53-mutant tumors were compared using the Mann-Whitney U test. B, C, D) Kaplan-Meier survival analysis of PSA progression-free survival, event-free survival, and overall survival in patients according to TP53 alteration status (patients without residual tumor [ie, undetected sequencing-derived tumor fraction] who are unevaluable for TP53 alteration status are represented by the gray line). Statistical significance was measured using multivariate Cox proportional hazards regression analysis. Seq-TF = sequencing-based tumor fraction; CI = confidence interval; HR = hazard ratio; PSA = prostate-specific antigen; NR = not reached; REF = reference group.](https://oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/jnci/116/1/10.1093_jnci_djad184/1/m_djad184f2.jpeg?Expires=1748078279&Signature=o72OXn1zpiu2iyVWCV0m4KqLyjLnNwewsCdPun99Ew9r5jsUBtXQNGj72HjsSWqc9TicxDcRq3QCIWmbv6cc7nQjgKgaOUZk~MrzPXuhONuFwj1PUP8Cpu2gt7Gc-BbRqs7Zz6h758rnErkZfYFBLJzkMsZ4nK8l5n9JHZb1Ul4wb5KlAhGMjrJKx4tB92pxPyysypkGEB5~D6HcRNNhtiKmqD8bEeH0y3Cf0c3eJgr~SX9Z6G0XQF9ORj3ZdmSDDD7z2AgnsLgVdk9712Fvd2VZPJsukSkxynDbiWdGwSloYlMQi2wLNDHm5FgpWLklQUh2-HEVQVtNTXljjGk76w__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
Common genomic alterations present in residual cancers after chemohormone therapy. A) Driver genomic landscape of residual high-risk, localized prostate cancer after treatment with neoadjuvant chemohormonal therapy before radical prostatectomy (arm A) vs prostate cancers not exposed to therapy (arm B). Only patients who had molecular residual disease are shown (ie, detected sequencing-derived tumor fraction: 91 in arm A and 82 in arm B). The highest sequencing-derived tumor fraction sample from each patient is represented if multiple same-patient tumor samples were available. Patients were sorted by the presence or absence of alterations in TP53 or SPOP and sequencing-derived tumor fraction (high to low). Fourteen genes associated with prostate cancer were provided (see Supplementary Figure 13 for additional genes, available online). Per-arm sequencing-derived tumor fraction estimates in SPOP-mutant and TP53-mutant tumors were compared using the Mann-Whitney U test. B, C, D) Kaplan-Meier survival analysis of PSA progression-free survival, event-free survival, and overall survival in patients according to TP53 alteration status (patients without residual tumor [ie, undetected sequencing-derived tumor fraction] who are unevaluable for TP53 alteration status are represented by the gray line). Statistical significance was measured using multivariate Cox proportional hazards regression analysis. Seq-TF = sequencing-based tumor fraction; CI = confidence interval; HR = hazard ratio; PSA = prostate-specific antigen; NR = not reached; REF = reference group.
Other genomic alterations were typical of high-risk, localized prostate cancer, including FOXA1 and CDK12 alterations; loss of PTEN, TP53, RB1, CHD1, and NKX3-1; and gain of MYC (Supplementary Figure 14; Supplementary Tables 9 and 10, available online). Copy number alterations (especially deep deletions) in these and other genes, however, were likely underdetected because of low sequencing-based tumor fraction in a large proportion of the cohort. Nevertheless, the frequency of alterations in major prostate cancer–related genes was comparable between arm B (ie, tumor samples not exposed to neoadjuvant chemohormone therapy) and public whole-exome sequencing data from localized prostate cancer, suggesting that genomic features within the CALGB 90203 correlative cohort broadly reflect established patterns for localized prostate cancer (Supplementary Figure 10, B, available online). CDK12-mutant tumors had comparable sequencing-based tumor fraction between the 2 arms, indicating that CDK12 alterations may be associated with resistance to chemohormonal therapy (27.6% in arm A vs 37.5% in arm B, P = .383) (Supplementary Figure 10, C, available online). Two samples harbored germline mismatch repair defects (PMS2 or MSH2) and associated hypermutation, while germline BRCA2 (n = 2), ATM (n = 2), or BRCA1 (n = 1) alterations were detected in only 5 patients in arm A (3.1%) (Supplementary Figure 15, available online). Less than 5% of samples showed alterations at the AR locus, with no gene amplifications (only single-copy gains were observed) and no difference between the arms, suggesting that up to 6 months of ADT plus docetaxel is not sufficient to result in clonally expanded AR genomic resistance mechanisms in patients with clinically high-risk, localized prostate cancer. For 144 patients, multiple tumor samples were analyzed, and paired samples were largely concordant for alterations and copy number changes (Supplementary Figure 16, available online).
Transcriptomic subtyping further stratifies residual disease
Unsupervised clustering of 155 core transcripts demonstrated clear segregation of arm A and arm B samples (Supplementary Figure 17, available online). Analysis of differentially expressed genes in arm A vs arm B (log2 fold change >1, P = .05 adjusted) suggested upregulation of steroid receptors (ESR1, ESR2, PGR), MET, and a subset of neuroendocrine and plasticity genes (NSE, CD56, NEUROD1) after exposure to chemohormonal therapy (Figure 3, A and B). As expected, AR and AR splice variants such as ARv567 were upregulated, and downstream androgen receptor target genes (eg, KLK2, KLK3, TMPRSS2) were downregulated after chemohormone therapy (Figure 3, B). Further clustering of arm A samples using the 155 core transcripts identified 2 distinct subgroups (Figure 3, C). Tumor samples in the smaller subgroup were associated with higher sequencing-based tumor fraction, ISUP grade 4 or higher, and no evidence of severe pathologic treatment effect, but there was no strong difference in clinical outcomes between patients in either subgroup (Supplementary Figure 18, available online).
![Transcriptomic changes in prostate cancer samples after chemohormonal therapy. A) Differentially expressed genes between arm A and arm B. The volcano plot shows the fold-change (x-axis) vs the significance (y-axis) of 155 core transcripts. Differentially expressed genes are established at change >2 and FDR < 0.05. Red dots indicate the genes that were significantly upregulated in arm A. Blue dots indicate the genes that were significantly downregulated in arm A. Black dots represent the remaining genes in the integrated panel that were not significantly changed after treatment therapy. B) Differences in the expression of AR, AR-variant 7 (AR-V7), FOLH1 (PSMA), androgen receptor–targeted genes (KLK2, KLK3, TMPRSS2), and a subset of neuroendocrine and plasticity genes (CD56, CHGB [chromogranin B], ENO2 [NSE], NEUROD1, SPDEF, SST [somatostatin]) in arm A vs arm B. C) Heatmap of gene expression levels (z-score) showing hierarchically clustered genes (rows; 155 core transcripts) and samples in arm A (columns), with dendrograms based on all core transcripts. TP53 and SPOP alteration status is provided for each sample, along with tumor fraction from targeted DNA sequencing. The dendrogram for samples is shown on top of the heat map. Values are measured by euclidean distance, with a complete linkage clustering algorithm. FC = fold change; FDR = false discovery rate; mRNA = messenger RNA; Seq-TF = sequencing-based tumor fraction.](https://oup.silverchair-cdn.com/oup/backfile/Content_public/Journal/jnci/116/1/10.1093_jnci_djad184/1/m_djad184f3.jpeg?Expires=1748078279&Signature=rnnTODL~Ebto6qfvBmKXubAxFuR7NvXukgYIBjWvPAydEi9jqfWRrIUhUtrJrx0fivvEgvk0dMUPMXNM5ack~MmQh-exBvdzkrj0VKdTj9MKItsJjT~Yvf1ua8iB0A1lAu2wrexesSTPo-DFRuM7DkKsXf~qzwfq6DVRVsHTLMEB5-LZPC-ICqv17-D5mgD2o5PtmuK5hqrH148tXYqqtl9uWy6uYCvJz~FrIIVeRXZZe0zxB2oaOXZwVSUCu1DIVmVI8wZu9KGVczBoCxLIgNY8htAdiPdY0uQ25ZopqAx9OBvbe0OS1LkDnEIN0KK4oWb4keH9djKpj5om7VcALg__&Key-Pair-Id=APKAIE5G5CRDK6RD3PGA)
Transcriptomic changes in prostate cancer samples after chemohormonal therapy. A) Differentially expressed genes between arm A and arm B. The volcano plot shows the fold-change (x-axis) vs the significance (y-axis) of 155 core transcripts. Differentially expressed genes are established at change >2 and FDR < 0.05. Red dots indicate the genes that were significantly upregulated in arm A. Blue dots indicate the genes that were significantly downregulated in arm A. Black dots represent the remaining genes in the integrated panel that were not significantly changed after treatment therapy. B) Differences in the expression of AR, AR-variant 7 (AR-V7), FOLH1 (PSMA), androgen receptor–targeted genes (KLK2, KLK3, TMPRSS2), and a subset of neuroendocrine and plasticity genes (CD56, CHGB [chromogranin B], ENO2 [NSE], NEUROD1, SPDEF, SST [somatostatin]) in arm A vs arm B. C) Heatmap of gene expression levels (z-score) showing hierarchically clustered genes (rows; 155 core transcripts) and samples in arm A (columns), with dendrograms based on all core transcripts. TP53 and SPOP alteration status is provided for each sample, along with tumor fraction from targeted DNA sequencing. The dendrogram for samples is shown on top of the heat map. Values are measured by euclidean distance, with a complete linkage clustering algorithm. FC = fold change; FDR = false discovery rate; mRNA = messenger RNA; Seq-TF = sequencing-based tumor fraction.
Within arm A, higher sequencing-based tumor fraction was associated with higher expression of androgen receptor–regulated (ADT target) genes (eg, FOLH1 and TMPRSS2), cell cycle–related genes, and neuroendocrine prostate cancer–related genes (eg, CHGA) (Supplementary Figure 19, available online) pointing to potential early resistance pathways. Supervised clustering of arm A samples using genes that were differentially expressed according to sequencing-based tumor fraction status identified 2 clusters, with cluster 1 demonstrating higher expression of cell cycle genes (MK167 [Ki67], AURKB, PLK1, CCNB1) and the EZH2 gene (Figure 4, A). The median sequencing-based tumor fraction was higher in samples from cluster 1 than from cluster 2 (22.9% vs 3.5%, P = .007, Mann-Whitney U test) (Figure 4, B). TP53 alterations were also enriched in cluster 1 samples (46.2% [6/13] in cluster 1 vs 11.5% [15/131] in cluster 2, P = .004, Fisher exact test), while no cluster 1 samples harbored SPOP alterations (Figure 4, A). Samples in cluster 1 were more likely to exhibit disease with ISUP grade 4 or higher (Figure 4, C). Differences in clinical outcomes of arm A cluster 1 were observed, including shorter PSA-PFS (HR = 3.32, 95% CI = 0.94 to 11.77; P = .06 (Figure 4, D) and event-free survival (HR = 2.59, 95% CI = 1.31 to 5.10; P = .01) (Figure 4, E) than cluster 2.
Cluster 1 and cluster 2 gene expression subgroups based on sequencing-based tumor fraction. A) Hierarchical clustering heatmap analysis of differentially expressed genes between sequencing-based tumor fraction–detected (n = 77) vs not detected (n = 67) within arm A revealed 2 clusters (1 and 2) (P = .03857). Red in the heatmap denotes upregulation, while blue denotes downregulation. The horizontal axis refers to the samples in arm A, and the vertical axis denotes the differentially expressed genes based on sequencing-based tumor fraction status. On top of the heatmap are the same annotations as Figure 3, C. The dendrogram values are measured by euclidean distance, with a complete linkage clustering algorithm. B) Box plot shows the sequencing-based tumor fraction in samples from arm A, cluster1; arm A, cluster2; and arm B. At the bottom of the plot, horizontal blue, orange, or green filled rectangles illustrate the proportion of samples in each group that had evidence of residual tumor DNA by sequencing. C) Association between clusters and clinical outcomes in patients with high-risk, localized prostate cancer following exposure to chemohormonal therapy. D, E) Kaplan-Meier survival analysis for prostate-specific antigen progression-free survival and event-free survival in patients between 2 clusters, identified based on sequencing-based tumor fraction status in arm A. P values were estimated using a Pearson χ2 test with Yates continuity correction, Fisher exact test, Mann-Whitney U test (A, B, C), or univariate Cox proportional hazards regression analysis (D, E). CI = confidence interval; HR = hazard ratio; Seq-TF = sequencing-based tumor fraction.
Discussion
This study took advantage of a large phase 3 trial with long-term follow-up data to provide new insights into the genomic and transcriptomic characteristics of residual high-risk, localized prostate cancer after neoadjuvant chemohormonal therapy. Importantly, several molecular features of radical prostatectomy specimens were linked with mature clinical outcomes, suggesting that post-treatment tumor testing has potential prognostic utility.
We identified a strong correlation between radical prostatectomy tumor fraction assessed by DNA sequencing (sequencing-based tumor fraction) and both PSA-PFS and event-free survival in patients treated with neoadjuvant chemohormonal therapy. Few established biomarkers can prognosticate or help determine a treatment strategy after radical prostatectomy following neoadjuvant therapy for localized, high-risk prostate cancer (22). Currently, the gold standard for tumor response assessment in neoadjuvant studies is through pathologic evaluation, and pathologic complete response (CR) has been used as a surrogate marker to support approval of neoadjuvant treatment in breast cancer (33). In prostate cancer, the pathologic CR rate is generally low, with 9.4% of patients treated with neoadjuvant ADT plus androgen receptor pathway inhibitors and no patients in the CALGB 90203 trial achieving pathologic CR (4,18,34,35). Hence, recent studies have included minimum residual disease, defined as residual tumor largest cross-section dimension 5 mm or larger, as part of the response assessment. Minimum residual disease is observed in 20% of patients treated with neoadjuvant ADT plus androgen receptor pathway inhibitors and associated with a decreased risk of biochemical recurrence (17,18), but associations between minimum residual disease and metastasis-free survival or overall survival have not yet been reported. Pathologic tumor response is often challenging in neoadjuvant studies because of the limitation in the number of slides per patient and the variability in tumor regression and pathologic response at the 3-dimensional level, especially in the context of severe treatment effect. Our approach mirrored pathologic assessment in other, prior neoadjuvant studies, focusing on the most representative tumor-rich regions based on central hematoxylin-eosin review. About 40% of patients treated with neoadjuvant chemohormonal therapy (but only 15% in the surgery-only arm) exhibited low sequencing-based tumor fraction, despite visible tumor cells on a 2-dimensional slide. It is possible that some areas annotated as tumor in these samples contained minimal or no residual disease. Several reports have demonstrated that sequencing-derived tumor fraction estimates are often lower than pathology-based tumor fraction estimates, even for tumors not receiving systemic therapies, which may contribute to the low sequencing-based tumor fraction detection rates observed in the chemohormonal therapy arm (36,37). Regardless of these limitations, we demonstrated that sequencing-based tumor fraction was a more favorable marker beyond pathologic variables alone. The potential added utility of sequencing-based tumor fraction as an independent prognostic biomarker warrants investigation in other neoadjuvant trials, such as the PROTEUS trial (ClinicalTrials.gov ID NCT03767244) and integration with ongoing efforts to quantify and standardize pathology assessments using artificial intelligence–based models (38,39).
Recurrent missense alterations in the MATH domain of the SPOP gene occur in approximately 10% of clinically localized prostate cancer and contribute to tumorigenesis by activating the androgen receptor signaling pathway (40-43). SPOP alterations have been associated with favorable response to ADT and androgen receptor pathway inhibitors in metastatic prostate cancer (44-47). In a smaller study of patients with high-risk, localized prostate cancer treated with neoadjuvant ADT plus androgen receptor pathway inhibitors, SPOP alterations were exclusively observed in exceptional responders (21). In our study, SPOP alterations were less frequent in the chemohormonal therapy–exposed tissues, and post-treatment tumors harboring SPOP alterations exhibited low tumor fraction, again suggesting stronger responses to treatment. These findings provide further evidence of SPOP alterations as a potential biomarker for favorable response to standard prostate cancer therapies.
TP53 alterations appeared enriched in tumors treated with neoadjuvant chemohormonal therapy compared with untreated tumors and approached the frequency observed in diagnostic biopsy studies in metastatic hormone-sensitive prostate cancer (48,49). This observation suggests that clones harboring TP53 alterations may be selected during chemohormonal therapy, although without testing the pretreatment diagnostic biopsy specimens it is also possible that the apparent enrichment merely represents an unbalanced cohort. Regardless, the poor prognostic weight conferred by TP53 alterations (independent of tumor fraction) further underscores a link with aggressive, treatment-resistant disease and matches findings from metastatic castration-sensitive prostate cancer and castration-resistant prostate cancer (44,49). Other genomic aberrations were as expected for this high-risk, localized prostate cancer population and at similar frequencies in treated vs untreated tumors; findings were consistent across lesions in our multifocal analyses, providing insights into possible future neoadjuvant or adjuvant approaches (eg, poly(ADP-Ribose) polymerase inhibitor therapy for BRCA1/2-altered tumors, immunotherapy for mismatch repair deficiency), such as being tested in the ongoing neoadjuvant, phase 2 Genomic Biomarker-Selected Umbrella Neoadjuvant Study (GUNS) trial (ClinicalTrials.gov ID NCT04812366).
Gene expression changes were variable after neoadjuvant therapy, with an overall decrease in androgen receptor signaling in chemohormonal therapy–exposed tumors, as expected, and a subset demonstrating increase in steroid receptor and plasticity gene expression. Although some of these genes and pathways have been linked to mechanisms of castration-resistant prostate cancer resistance (50-52), their timing has not been well established. These data point to adaptive responses, even after short-term therapy, that may drive future resistance pathways, providing additional insights into the short-term molecular response that occurs after docetaxel and ADT. Unfortunately, there was insufficient remaining tumor tissue in this trial to confirm transcriptomic features at the protein level by immunohistochemistry or to evaluate intratumoral heterogeneity with digital spatial profiling, and the molecular patterns associated with relapse are not known. These observations warrant further exploration in future cohorts of treatment-exposed specimens.
We found gene expression of cell cycle genes and higher Ki67 in treated tumors with higher tumor fraction and in those harboring TP53 alterations. Two distinct clusters based on gene expression were identified in treated cases, with cluster 1 associated with higher tumor fraction and inferior PSA-PFS and event-free survival than cluster 2. These data provide new insights into short-term gene expression changes that occur in residual tumors treated with chemohormonal therapy and are distinguished from acquired resistance.
This study has several limitations. Our analysis was limited to prostatectomy specimens; without pretreatment biopsies available, we were not able to assess tumor evolution in individuals before and after therapy or identify predictive biomarkers. Furthermore, although pathologic assessment mirrored prior neoadjuvant studies by focusing on tumor-rich regions based on central hematoxylin-eosin review, our analysis was still restricted to select regions. Given the few unstained slides available per patient, we could not use immunohistochemistry to better identify residual tumor regions for molecular profiling, and this limitation may contribute to the discordance between pathologic assessments of tumor cellularity and sequencing-based tumor fraction. Despite mature long-term follow-up data, there was an overall low event rate and high variability across patients, making correlations with long-term outcomes challenging. Because of the design of CALGB 90203, metastasis-free survival has been established as a surrogate endpoint for overall survival and would be an important correlate for future studies (53). Based on the limited amount of formalin-fixed, paraffin-embedded tumor tissue collected in this study, our analyses were restricted to targeted DNA and RNA panels, preventing broad characterization or new discovery, and we were not able to systematically assess spatial tumor heterogeneity in post-treated tumors. Finally, an important caveat of our study is that we cannot determine whether 1 individual treatment (ADT or docetaxel) is more strongly associated with the observed molecular changes. This limitation is relevant because previous neoadjuvant studies that did not use chemotherapy have identified similar post-treatment molecular features (19-22). Despite these limitations, this study provides a large dataset to support future studies characterizing the adaptive responses after chemohormonal therapy and sets the stage for future correlative analyses in neoadjuvant trials.
In summary, we demonstrated the spectrum of genomic and transcriptome alterations that occur in radical prostatectomy specimens with or without treatment with neoadjuvant chemohormonal therapy in the CALGB 90203 trial in conjunction with mature long-term clinical outcomes data. Evidence of both clonal selection (eg, TP53) and adaptive responses were observed in residual tumor tissues. Although pathologic complete responses after chemohormonal therapy were not identified in this study, pathologic tumor cellularity was variable and correlated with quantified sequencing-based tumor fraction. Importantly, sequencing-based tumor fraction was associated with longer-term clinical outcomes, including both PSA-PFS and event-free survival, suggesting that this prognostic variable may help guide subsequent therapy.
Data availability
Participants did not agree for data to be shared publicly to a biorepository as part of the informed consent process. Structured data underlying the results presented in this paper are provided in the Supplementary Tables (available online). Deidentified genomic data presented in the study can be available on request to the senior authors.
Author contributions
Takayuki Sumiyoshi, MD (Data curation; Formal analysis; Investigation; Writing—original draft; Writing—review & editing), Martin E Gleave, MD (Conceptualization; Funding acquisition; Investigation; Supervision; Writing—review & editing), Michael J. Morris, MD (Investigation; Project administration; Writing—review & editing), Susan Halabi, PhD (Funding acquisition; Investigation; Writing—review & editing), Jeffrey Simko, MD (Data curation; Formal analysis; Investigation; Writing—review & editing), Mary-Ellen Taplin, MD (Investigation; Project administration; Writing—review & editing), James Eastham, MD (Investigation; Writing—review & editing), Ladan Fazli, MD (Investigation; Writing—review & editing), Sheng-Yu Ku, PhD (Investigation; Writing—review & editing), Kei Mizuno, MD PhD (Investigation; Writing—review & editing), Kevin Beja, PhD (Investigation; Writing—review & editing), Michael Sigouros, BS (Investigation; Writing—review & editing), Matti Annala, PhD (Investigation; Writing—review & editing), Andrea Sboner, PhD (Data curation; Formal analysis; Investigation; Writing—review & editing), Evan W. Warner, PhD (Investigation; Writing—review & editing), Xiaofei Wang, PhD (Data curation; Formal analysis; Writing—review & editing), Alexander W. Wyatt, PhD (Conceptualization; Data curation; Formal analysis; Funding acquisition; Investigation; Methodology; Project administration; Resources; Supervision; Writing—original draft; Writing—review & editing), Himisha Beltran, MD (Conceptualization; Data curation; Formal analysis; Funding acquisition; Investigation; Methodology; Project administration; Supervision; Writing—original draft; Writing—review & editing).
Funding
Research reported in this publication was supported by the National Cancer Institute of the National Institutes of Health under Award No. U10CA180821 and U24CA196171 (to the Alliance for Clinical Trials in Oncology), UG1CA233180, UG1CA233253, UG1CA233290, and U10CA180863; and CCS No. 707213 (Canadian Cancer Trials Group). See also https://acknowledgments.alliancefound.org. Additional funding support was provided by the Prostate Cancer Foundation, Canadian Institutes of Health Research, Department of Defense (W81XWH-17-1-0653), NCI R37 CA24148601A1, NIH 5 U01 CA157703, R01 CA256157, R01 CA249279, P50CA211024, P30 CA008748, a Terry Fox New Frontiers Program Project Grant, and by Sanofi. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Conflicts of interest
A.W. reports advisory roles and/or speaking engagements for Astellas, AstraZeneca, Bayer, EMD Serono, Janssen, Merck, and Pfizer and a contract research agreement with ESSA Pharma. H.B. has served as consultant/advisory board member for Janssen, Astellas, Merck, Pfizer, Foundation Medicine, Blue Earth Diagnostics, Amgen, Bayer, Oncorus, LOXO, Daiichi Sankyo, Sanofi, Curie Therapeutics, Astra Zeneca, Novartis, and Fusion Pharma and has received research funding (to institution) from Janssen, AbbVie/Stemcentrx, Eli Lilly, Astellas, Millennium, Bristol Myers Squibb, Circle Pharma, and Daiichi Sankyo.
Acknowledgements
We would like to thank Yujia Wen and Linda McCart from the Alliance Biospecimen Repository for their coordination of samples for this study as well as the Alliance Statistics and Data Center for sharing of deidentified clinical data. The funder had no role in the design of the study; the collection, analysis, or interpretation of the data; or the writing of the manuscript and decision to submit it for publication.
References
Author notes
Alexander W. Wyatt and Himisha Beltran contributed equally to this work.
