Clinical testing of transcriptome-wide expression profiles in high-risk localized and metastatic prostate cancer starting androgen deprivation therapy: an ancillary study of the STAMPEDE abiraterone Phase 3 trial

Abstract

Metastatic and high-risk localized prostate cancer respond to hormone therapy but outcomes vary. Following a pre-specified statistical plan, we used Cox models adjusted for clinical variables to test associations with survival of multi-gene expression-based classifiers from 781 patients randomized to androgen deprivation with or without abiraterone in the STAMPEDE trial. Decipher score was strongly prognostic (p<2×10^-5) and identified clinically-relevant differences in absolute benefit, especially for localized cancers. In metastatic disease, classifiers of proliferation, PTEN or TP53 loss and treatment-persistent cells were prognostic. In localized disease, androgen receptor activity was protective whilst interferon signaling (that strongly associated with tumor lymphocyte infiltration) was detrimental. Post-Operative Radiation-Therapy Outcomes Score was prognostic in localized but not metastatic disease (interaction p=0.0001) suggesting the impact of tumor biology on clinical outcome is context-dependent on metastatic state. Transcriptome-wide testing has clinical utility for advanced prostate cancer and identified worse outcomes for localized cancers with tumor-promoting inflammation.

Keywords: Decipher score; Prostate cancer; STAMPEDE trial; high-risk localized; interferon signalling; mRNA; prognosis; synchronous metastatic; transcriptome-wide profiling.

Figure 1:. Cohort and signature overview.

A. Experimental and analytical workflow showing total eligible patients in full trial cohort (FTC) and progress to our Abi781 cohort (additional details in Supplemental Figure 1) with breakdown of Abi781 by treatment arm and disease burden. Patients with high-risk localized (M0) disease were stratified by nodal status at randomization and patients with metastatic (M1) disease were assigned burden status following post-hoc central review, where scans were available (M1x indicates cases where burden status could not be assigned). B. Stacked barplots and box-violin plots comparing proportion or range of baseline clinical features that were adjusted for in prognostic analyses: treatment assignment, disease burden, Gleason score (GS), World Health Organisation (WHO) performance status (PS), age and serum prostate specific antigen (PSA) obtained prior to androgen deprivation therapy (ADT) of the FTC to Abi781. C. Spearman correlation plot of 57 (continuous) signatures (detailed in Supplemental Table 2) included in the outcome analyses, ordered by first principal component in M0 cohort (matched in M1 cohort) and annotated for signature biology group (a) and cancer type the signature was developed in (b). M0 cohort in the bottom-left triangle and M1 in the top-right triangle. Size of circle represents level of correlation. M0N0: high-risk localized node-negative, M0N1: high-risk localized node-positive, M1LV: metastatic low-volume, M1HV: metastatic high-volume, RT: radiation therapy, AAP: abiraterone acetate and prednisolone.

Figure 2:. Signature score distribution (continuous or categorical) by disease burden.

A. Diverging barplots of 15 signatures which are statistically-significantly differently distributed (p<0.001) in M1 high-volume tumors collected after versus prior to LHRHa, ordered by difference in median signature score and coloured by signature biology (detailed in Supplemental Figure 7). B. Arc diagram of the 18 signatures whose scores are statistically-significantly different (p<0.001) by disease burden (M0N0, M0N1, M1 low-volume and M1 high-volume) in the Abi781* and Abi723* cohorts and denoting between which states they are different. Colored by signature biology and ordered by which cohort they are significant in (both, Abi723 then Abi781). Red lines indicate the score decreases with increasing disease burden (direction of arrow) whilst green lines indicate the score increases. Dotted lines represent significant associations in Abi723, dashed lines in Abi781 and solid lines in both Abi723 and Abi781. C. Box-violin plots showing the distribution by disease burden of Decipher scores in Abi723* and the Wilcoxon rank sum test p-values. Threshold for low (0.45) and high Decipher score (>0.60) shown. D. Stacked barplots showing the frequency distribution by disease burden in Abi723* of the three categories for PAM50 (basal [B], luminal A [LA] and luminal B [LB]). E. Stacked barplots showing the frequency distribution by disease burden in Abi723* of the four categories for PSC (basal immune [BI], basal neuroendocrine [BN], luminal differentiated [LD], luminal proliferating [LP]). LHRHa: luteinizing hormone-releasing hormone agonist or antagonist. M0N0: high-risk localized node-negative, M0N1: high-risk localized node-positive, M1LV: metastatic low-volume, M1HV: metastatic high-volume, PSA: prostate serum antigen. *excluding 19 M1 cases where disease burden is unknown.

Figure 3:. Primary and secondary outcome analyses.

A. Kaplan-Meier plot of overall survival in patients with metastatic disease in Abi781 receiving either ADT or ADT + AAP split by Decipher score above or below median (0.77 on scale of 0–1). B. Same as for A but showing localized disease patients and metastasis-free survival. C. Line plot showing the absolute difference in overall survival (as a percentage) of addition of AAP to ADT in patients with metastatic disease in Abi781* split by Decipher score above or below median with 95% confidence intervals adjusted for clinical and pathological variables. D. Same as for C but showing localized disease patients and metastasis-free survival. E. Prognostic forest plots of Decipher, AR-A, PAM50 and PSC with primary and secondary outcomes by metastatic state and adjusted for clinical and pathological variables, scaled to fit the range of 95% CI for each signature for Abi781*. Full line for M0 and dashed line for M1 with size of hazard ratio square relative to level of statistical significance of association. ADT: androgen deprivation therapy, AAP: abiraterone acetate and prednisolone, M0: high-risk localized, M1: metastatic, HR: hazard ratio, CI: confidence interval. *excluding 18 M1 cases where disease burden is unknown, three where Gleason score is unknown and one where both are unknown.

Figure 4:. Exploratory outcome analyses.

A. Scatter plot of hazard ratios (adjusted by one standard deviation of signature score) and p values resulting from prognostic testing adjusted for clinical and pathological variables of overall survival in 57 signatures (with continuous scores) in patients with metastatic disease*. Dots representing statistically significantly prognostic (p<0.00045) signatures in Abi781 and Abi723 are colored in dark pink and light pink respectively. Signatures which are prognostic in M0 are colored in the outer circle in Abi781 and Abi723 in blue and light blue respectively. Signatures which are not prognostic in Abi781 or Abi723 are colored in grey. Dotted lines connect values for the same signature. Dashed line marks p=0.00045. B. As for A but in M0 patients. C. Kaplan-Meier plot of overall survival in patients** receiving ADT split by metastatic state (M0 and M1) and PORTOS score (above or below median in Abi781: −0.31) with M1 patients further split by disease volume (low and high). D. As for E but for patients** receiving ADT+AAP. E. Forest plot of PORTOS, indicating the p value for interaction with metastatic state, and hazard ratio, 95% CI and p value in metastatic states for Abi781*. Full line for M0, dashed line for M1 with size of hazard ratio square relative to level of statistical significance of association. F. Box and violin plot of IFN_HM score split by tumor infiltrating lymphocytes (four groups: 0–5%, 5.1–10%, 10.1–20%, >20.1%) with added p values for significance. N=731, does not include 43 cases which were TURPs and seven biopsies that could not be assessed for technical reasons. HR: hazard ratio, M0: high-risk localized, M1: metastatic. ADT: androgen deprivation therapy, AAP: abiraterone acetate and prednisolone. *excluding 18 M1 cases where disease burden is unknown, three where Gleason score is unknown and one where both are. **excluding 19 M1 cases where disease burden is unknown.