Transcriptomic Profiling of Primary Prostate Cancers and Non-Localized Disease on Prostate-Specific Membrane Antigen PET/CT: A Multicenter Retrospective Study

Abstract

Purpose:

To characterize the relationship between Decipher genomic classifier scores and prostate-specific membrane antigen (PSMA) PET/CT-based metastatic spread.

Methods:

We identified patients from four institutions who underwent PSMA PET/CT scans pre-treatment for primary staging or post-radical prostatectomy (RP) for suspected recurrence and had Decipher transcriptomic data available from biopsy or RP specimens. PSMA PET/CT-based patterns of spread were classified as localized (miT+N0M0) or non-localized (miN1M0 or miM1a-c). We calculated the association between Decipher scores and the risk of non-localized disease on PSMA PET/CT using multivariable logistic regression for pre-treatment patients and multivariable Cox regression for post-RP patients. We also compared select transcriptomic signatures between patients with localized and non-localized disease.

Results:

586 patients were included (pre-treatment: N=329; post-RP: N=257). Higher Decipher scores were associated with non-localized disease on PSMA PET/CT both pre-treatment (odds ratio: 1.18 [95% confidence interval [CI]: 1.03–1.36] per 0.1 increase in Decipher score, P=0.02) and post-RP (hazard ratio: 1.15 [95% CI: 1.05–1.27] per 0.1 increase in Decipher score, P=0.003). In the pre-treatment setting, non-localized disease was associated with higher rates of TP53 mutations and lower rates of PAM50 luminal A subtype compared to localized disease. In the post-RP setting, overexpression of signatures related to metabolism, DNA repair, and androgen receptor signaling were associated with higher rates of non-localized disease.

Conclusion:

Higher Decipher scores were associated with non-localized disease identified on PSMA PET/CT both pre-treatment and post-RP. There were several transcriptomic differences between localized and non-localized disease in both settings.

1. Introduction

The Decipher genomic classifier is a prognostic biomarker in men with prostate cancer that estimates metastatic risk following both definitive radiotherapy and radical prostatectomy (RP).^[1–6] There is great interest in using Decipher testing to guide systemic treatment intensification and de-intensification strategies. In both the discovery study and validation studies for Decipher, metastatic disease was defined using conventional imaging (i.e., computed tomography (CT) of the abdomen and pelvis, magnetic resonance imaging (MRI) of the pelvis, and bone scans).^[7] The synchronous development of prostate-specific membrane antigen (PSMA) positron emission tomography (PET)/CT imaging, which has a greater specificity and sensitivity for detecting non-localized disease in both primary staging^[8,9] and the evaluation of biochemical recurrence (BCR),^[10,11] offers the opportunity to evaluate the prognostic capacity of Decipher in the era of advanced molecular imaging. A recent translational study comparing men with oligometastatic disease detected with advanced molecular imaging versus with conventional imaging found that men with oligometastatic disease detected with advanced imaging had fewer high-risk mutations and better survival.^[12] This underscores the notion that the population of men who have non-localized disease identified on molecular imaging may be biologically distinct from the population of men who have non-localized disease identified by conventional imaging.

In order to better understand the relationship between Decipher score and the detection of non-localized disease by PSMA PET/CT in both the primary treatment and post-RP BCR setting, we performed a multi-institutional retrospective analysis of men who had undergone PSMA PET/CT and had Decipher scores and associated transcriptomic data available.

2. Methods and Materials

Patient Cohorts

We conducted a retrospective examination of patients from four institutions: Washington University School of Medicine (St. Louis, MO), New York-Presbyterian/Weill Cornell Hospital (New York, NY), University of California, Los Angeles (UCLA, Los Angeles, CA), and Northwestern University, Feinberg School of Medicine (Chicago, IL). All patients included in this study prospectively underwent Decipher genomic classifier testing from biopsy or RP specimens (Veracyte, Inc., San Diego, CA) to guide treatment decision-making and PSMA PET/CT scans either pre-treatment (for primary radiographic staging) or post-RP at the time of BCR (to assess the radiographic extent of recurrent disease). Each site retrieved clinical data from their patients under an IRB-approved study and linked these data to each patient’s Decipher assay.

Clinical data included age at the time of PSMA PET/CT, initial clinical or pathological International Society of Urologic Pathologists (ISUP) grade group, initial clinical or pathological TNM stage, serum prostate-specific antigen (PSA) at the time of PSMA PET/CT, time from RP to PSMA PET/CT (if PSMA PET/CT was performed post-RP), and staging information from the PSMA PET/CT report. PSMA PET/CT reports used Prostate Cancer Molecular Imaging Standardized Evaluation (PROMISE) criteria for whole-body staging using miTNM, version 1.0. A subset of 265 transcriptomic signatures representing key biological pathways and subtypes in prostate cancer was retrieved for each patient from the prospectively maintained genome-wide expression registry (NCT02609269) Genomics Resource for Intelligent Discovery (GRID) database (Veracyte, Inc., San Diego, CA). These patients’ data were de-identified in accordance with the Safe Harbor method described in the HIPAA Privacy Rule 45 CFR 164.514(b) and (c) (Veracyte, Inc., San Diego, CA) prior to analysis.

Statistical Analysis

Summary statistics were reported as medians with interquartile ranges (IQRs) for continuous variables and counts with percentages for categorical variables. Comparisons were performed between the four institutions for all baseline clinical characteristics, with statistical significance determined by Kruskal-Wallis tests for continuous variables and Pearson’s Chi-squared tests for categorical variables. For patients with pre-treatment imaging, the statistical significance of associations between genomic signatures and disease states was determined by multivariable logistic regression, adjusted for pre-treatment PSA and Gleason score. For patients with post-RP imaging, the statistical significance of associations between genomic signatures and disease states was determined by multivariable Cox proportional hazards models, incorporating the time between surgery and PSMA PET/CT and adjusted for pathologic T stage and ISUP grade group. Adjusted odds ratio (OR) and hazards ratio (HR) estimates with their associated 95% confidence intervals (CI) were reported for pre- and post-treatment comparisons, respectively. Univariable analyses were performed when comparing non-localized disease states to one another (e.g., miN1 vs. miM1a, miM1a vs. miM1b/c, or miN1 + miM1a vs. miM1b/c) due to sample size concerns, with unadjusted odds and hazards ratios reported. P-values ≤0.05 were considered statistically significant for all tests. Analyses were performed using the open-source statistical programming language R.

3. Results

Patient Characteristics

Five hundred and eighty-six patients were included in this analysis (pre-treatment: N=329; post-RP: N=257; Table 1 and Supplemental Table 1). Median age was 67 (IQR: 62–73). Median PSA was 8.0 ng/mL (IQR: 5.4–13.0 ng/mL) at the time of pre-treatment PSMA PET/CT and 0.4 ng/mL (IQR: 0.2–1.1 ng/mL) at the time of post-RP PSMA PET/CT. Median Decipher scores were 0.64 (IQR: 0.43–0.86) in the pre-treatment cohort and 0.63 (IQR: 0.39–0.87) in the post-RP cohort. On pre-treatment PSMA PET/CT, 281 patients (85%) had localized disease (miT+N0M0), 24 (7%) had nodal metastasis (miN1), and 24 (7%) had distant metastasis (miM1a-c). On post-RP PSMA PET/CT, 138 patients (54%) had a negative scan, 24 (9%) had a local recurrence (miTrN0M0), 44 (17%) had nodal metastasis (miN1), and 51 (20%) had distant metastasis (miM1a-c).

Table 1.

Patient characteristics by prostate-specific membrane antigen (PSMA) PET/CT timing.

	PSMA PET/CT Timing
	Pre-treatment N = 329	Post-RP N = 257	Overall N = 586
Age at scan: Median (Q1, Q3) #	69 (64, 75)	66 (61, 71)	67 (62, 73)
PSA (ng/mL) at scan: Median (Q1, Q3) §	8.0 (5.4, 13.0)	0.4 (0.2, 1.1)	5.2 (0.5, 9.7)
Clinical/pathological stage: N (%) ¶
T1	254 (83%)	-	254 (46%)
T2	47 (15%)	84 (34%)	131 (24%)
T3a	3 (1%)	91 (37%)	94 (17%)
T3b-T4	2 (<1%)	70 (29%)	72 (13%)
Clinical/pathological ISUP grade group: N (%)
1	12 (4%)	7 (3%)	19 (3%)
2	91 (28%)	75 (29%)	166 (28%)
3	117 (36%)	82 (32%)	199 (34%)
4	53 (16%)	26 (10%)	79 (13%)
5	56 (17%)	67 (26%)	123 (21%)
PSMA PET/CT finding; N (%)
Prostate (miT+N0M0)	281 (86%)	-	281 (48%)
Negative	-	138 (54%)	138 (24%)
Local recurrence (miTrN0M0)	-	24 (9.3%)	24 (4.1%)
Pelvic nodal metastasis (miN1M0)	24 (7%)	44 (17%)	68 (12%)
Distant metastasis (miM1a-c)	24 (7%)	51 (20%)	75 (13%)
Decipher score: Median (Q1, Q3)	0.64 (0.43, 0.86)	0.63 (0.39, 0.87)	0.64 (0.42, 0.86)
Decipher risk group: N (%)
Low	86 (26%)	77 (30%)	163 (28%)
Intermediate	58 (18%)	43 (17%)	101 (17%)
High	185 (56%)	137 (53%)	322 (55%)

^#Missing 1 value.

^§Missing 24 values.

^¶Missing 35 values.

Abbreviations: ISUP: International Society of Urologic Pathologists; PSMA=prostate-specific membrane antigen; RP=radical prostatectomy; Q=quartile; PSA=prostate-specific antigen.

Genomic Classifier Score Analysis

In patients with a pre-treatment PSMA PET/CT, median Decipher score was 0.63 (IQR: 0.43–0.84) for miT+N0M0 disease, 0.86 (IQR: 0.46–0.93) for miN1 disease, 0.92 (IQR: 0.56–0.97) for extra-pelvic nodal metastasis (miM1a), 0.62 (IQR: 0.46–0.84) for bone metastasis (miM1b), and 0.87 (IQR: 0.85–0.88) for visceral metastasis (miM1c) (Figure 1A). In patients with a post-RP PSMA PET/CT, median Decipher score was 0.59 (IQR: 0.37–0.81) for negative scans, 0.76 (IQR: 0.54–0.89) for local recurrences (miTrN0M0), 0.70 (IQR: 0.44–0.87) for miN1 disease, 0.38 (IQR: 0.38–0.66) for miM1a disease, 0.82 (IQR: 0.55–0.92) for miM1b disease, and 0.97 (IQR: 0.90–0.99) for miM1c disease (Figure 1B).

Figure 1.

Boxplots of Decipher scores by prostate-specific membrane antigen (PSMA) PET/CT findings for the (A) pre-treatment and (B) post-radical prostatectomy (RP) cohorts. Values superimposed over each boxplot represent the sample size and median (first quartile, third quartile).

In patients with a pre-treatment PSMA PET/CT, higher Decipher scores were significantly associated with higher odds of non-localized disease (OR per 0.1 increase: 1.18, 95% CI: 1.03–1.36; P=0.02). Higher scores were associated with metastatic (miM1a-c) disease but did not meet conventional levels of statistical significance (OR per 0.1 increase: 1.16, 95% CI: 0.96–1.40; P=0.13) (Table 2 and Figure 2A). In patients with a post-RP PSMA PET/CT, higher Decipher scores were significantly associated with higher rates of both non-localized (HR per 0.1 increase: 1.15, 95% CI: 1.05–1.27; P=0.003) and metastatic disease (HR per 0.1 increase: 1.22, 95% CI: 1.07–1.40; P=0.003) (Table 2 and Figure 2B).

Table 2.

Regression results for predicting non-localized or metastatic disease on pre-treatment and post-radical prostatectomy prostate-specific membrane antigen (PSMA) PET/CT. Multivariable logistic regression was used for pre-treatment patients and multivariable Cox regression was used for post-radical prostatectomy patients.

Pre-Treatment Cohort
Endpoint	Non-localized disease		Metastatic disease
Variable	OR (95% CI)	P-value	OR (95% CI)	P-value
Decipher score (per 0.1)	1.18 (1.03 – 1.36)	0.02*	1.16 (0.96 – 1.40)	0.13
Log2PSA	1.23 (0.92 – 1.63)	0.16	1.25 (0.85 – 1.82)	0.26
GG 3 vs. 1–2	1.00 (0.43 – 2.36)	1	1.12 (0.37 – 3.38)	0.8
GG 4–5 vs. 1–2	1.60 (0.70 – 3.64)	0.26	1.12 (0.36 – 3.44)	0.8
Post-Radical Prostatectomy Cohort
Endpoint	Non-localized disease		Metastatic disease
Variable	HR (95% CI)	P-value	HR (95% CI)	P-value
Decipher score (per 0.1)	1.15 (1.05 – 1.27)	0.003*	1.22 (1.07 – 1.40)	0.003*
GG 3 vs. 1–2	1.74 (0.95 – 3.19)	0.07	2.72 (1.11 – 6.70)	0.03*
GG 4–5 vs. 1–2	1.14 (0.61 – 2.14)	0.7	1.39 (0.54 – 3.54)	0.5
T3b/T4 vs. T2/T3a	1.71 (1.05 – 2.77)	0.03*	2.17 (1.10 – 4.28)	0.03*

^*P≤0.05.

Abbreviations: CI=confidence interval; GG=grade group; HR=hazard ratio; Log2PSA=PSA in ng/ml on the log-base-2 scale; OR=odds ratio; PSA=prostate-specific antigen.

Figure 2.

(A) Stacked probability of nodal and distant metastatic findings in pre-treatment prostate-specific membrane antigen (PSMA) PET/CT by Decipher score. (B) Stacked cause-specific cumulative incidence of non-localized disease in post-RP PSMA PET/CT by Decipher risk group.

Exploration of Transcriptomic Signatures

Among patients with a pre-treatment PSMA PET/CT scan, higher scores on prognostic risk models (e.g., Decipher, Penney 2011, and cell cycle progression) were associated with a higher risk of non-localized (miN1M0 or miM1a-c) disease (Figure 3A and Supplemental Table 2). Patients with non-localized disease had higher rates of predicted TP53 mutations (47% vs. 16% in non-localized vs. localized disease, respectively), higher staging signatures for seminal vesicle invasion (SVI) and lymph node invasion (LNI) (Figure 3B and Supplemental Table 2), and lower rates of PAM50 luminal A subtype (21% vs 36%) compared to patients with localized (miT+N0M0) disease (Figure 3B). Higher DNA repair (e.g., TP53 mutations and homologous recombination deficiency) and neuroendocrine signatures were associated with a higher risk of osseous or visceral metastases (miM1b-c disease) (Figure 3C and Supplementary Table 3). When only analyzing pre-treatment patients with non-localized disease (miN1M0 or miM1a-c disease: N=48 patients), higher scores for TP53 mutations, hallmarks of cancer (e.g., glycolysis and fatty acid metabolism), and neuroendocrine signatures were associated with a higher risk of miM1b-c disease versus miN1 or miM1a disease (Supplementary Table 4).

Figure 3.

(A) Odds-ratio (OR) of having non-localized (miN1 or miM1a-c) disease on the pre-treatment prostate-specific membrane antigen (PSMA) PET/CT scan based on 265 genomic signatures available from the GRID database. (B) Stacked bar charts of the genomic signatures of luminal differentiated subtype, p53 mutations, seminal vesicle invasion (SVI), and lymph node invasion (LNI) staging by pre-treatment PSMA PET/CT findings. (C) OR of having miM1b-c disease versus miM0-M1a disease on the pre-treatment PSMA PET/CT based on 265 genomic signatures available from the GRID database. Abbreviation: SD=standard deviation.

Among patients with a post-RP PSMA PET/CT scan (N=257 patients), higher scores on prognostic risk models (e.g., Decipher, Penney 2011, and cell cycle progression), metabolism (e.g., glycolysis, Krebs cycle, and ribose sugar metabolism), DNA repair (e.g., homologous recombination deficiency, base excision repair, and chromosomal instability), hallmarks of cancer (e.g., MYC targets, E2F targets, and mTOR signaling), and androgen receptor signaling (e.g., ARv7 splice variant and abiraterone drug response score) were associated with a higher risk of non-localized (miN1M0 or miM1a-c) disease (Figure 4A and Supplementary Table 5). Patients with luminal differentiated disease by PSC classification had less frequent and longer intervals to detection of non-localized disease, with a median imaging event-free survival time of 6 years compared to 4 years in patients classified as luminal proliferating, basal immune, or basal neuroendocrine subtypes (Figure 4B). Higher scores on signatures related to immune suppression (e.g., T cell regulatory, T cell exhaustion, and IL-8 inflammatory pathway), ADT response, SPOP mutation, and luminal differentiated subtype were associated with a lower risk of non-localized (miN1M0 or miM1a-c) disease (Supplemental Table 5). Higher scores on prognostic risk models (e.g., Decipher and cell cycle progression), metabolism (e.g., Krebs cycle and ribose sugar metabolism), DNA repair (e.g., mismatch repair, homologous recombination, non-homologous end joining, and chromosomal instability), androgen receptor biology (e.g., ARv7 splice variant), mTOR signaling, and pTEN loss were associated with higher risk of osseous or visceral metastasis (miM1b-c disease) (Figure 4C and Supplementary Table 6). Higher scores on signatures related to immune suppression (e.g., T cell regulatory, T cell exhaustion, T cell dysfunction) were associated with a lower risk of osseous or visceral metastases (miM1b-c disease) (Supplementary Table 6). When only analyzing post-RP patients with non-localized disease (miN1M0 or miM1a-c disease: N=95 patients), higher scores for prognostic risk models (e.g., Decipher), TP53 mutations, neuroendocrine signatures, and metabolism (e.g., Krebs cycle and ribose sugar metabolism) were associated with a higher risk of miM1b-c disease versus miN1M0 or miM1a disease (Supplementary Table 7).

Figure 4.

(A) Hazard-ratio (HR) of having non-localized (miN1 or miM1a-c) disease on the post-radical prostatectomy (RP) prostate-specific membrane antigen (PSMA) PET/CT based on 265 genomic signatures available from the GRID database. (B) Stacked cause-specific cumulative incidence of non-localized disease in the post-RP PSMA PET/CT scans by luminal differentiated subtype. (C) HR of having miM1b-c disease versus miM0-M1a disease on the post-RP PSMA PET/CT. P<0.05 and P<0.001 thresholds are shown using gray dotted lines based on 265 genomic signatures available from the GRID database. Abbreviation: SD=standard deviation.

4. Discussion

To our knowledge, this represents one of the most comprehensive studies evaluating the association between PSMA PET/CT-based identification of non-localized disease in the pre-treatment and post-RP setting and transcriptomic profiles, including the independently prognostic Decipher score. We found that higher Decipher scores were associated with non-localized disease on both pre-treatment and post-RP PSMA PET/CT. In the pre-treatment setting, non-localized disease was associated with higher rates of predicted TP53 mutations, higher staging signatures (SVI and LNI), and lower rates of PAM50 luminal A subtype compared with localized disease. In the post-RP setting, signatures related to metabolism, DNA repair, and androgen receptor signaling were associated with higher rates of non-localized disease.

The association between upstaging on PSMA PET/CT and Decipher genomic classifier scores has been previously studied, albeit indirectly.^[13] In a cohort of exclusively high-risk and very high-risk prostate cancer at the time of initial diagnosis, Smith et al. reported that Decipher scores were significantly correlated with the risk of upstaging on PSMA PET/CT (Spearman correlation, 0.42 [95% CI: 0.39–0.44]; P<0.001), though actual PSMA results were not available and the data focused on patients with high-risk disease in the upfront setting. In our study, which consisted of patients from all risk groups in both the primary setting and the post-RP BCR setting, we confirmed a significant association between higher Decipher and higher rates of non-localized disease on both pre-treatment and post-RP PSMA PET/CT. Similar to the conclusions of the Smith et al. study, the present data support the idea that patients with high Decipher scores have a worse prognosis with respect to the development of metastatic disease by conventional imaging because of occult non-localized disease identified with advanced imaging.

The results of our exploratory transcriptomic analysis are broadly consistent with the hypothesis that non-localized disease captured on PSMA PET/CT generally reflects a more biologically aggressive natural history. This is important, as upstaging based solely on PSMA PET/CT has not yet been shown to allow for improved clinical outcomes by influencing treatment. There is concern that utilizing this modality could lead to inappropriate changes in management.^[13,14] In the present study, DNA repair and neuroendocrine signatures were increased in patients with de novo miM1b-c disease. An even broader subset of signatures was associated with recurrent miM1b-c disease, including DNA repair, metabolism, androgen receptor signaling, pTEN loss, and mTOR signaling. Notably, a recent study of 295 patients with de novo oligometastatic disease (123 with advanced molecular imaging and 172 with conventional imaging) found significantly fewer pathogenic mutations in TP53 among those with lesions identified on advanced molecular imaging.^[12] The present results are not inconsistent; Sutera et al. concluded that recurrent oligometastatic disease detected on advanced molecular imaging likely reflected a more indolent disease than disease detected on conventional imaging but did not directly compare either to localized disease. Further, that study used next generation sequencing data and evaluated pre-specified mutational signatures of interest for metastatic hormone-sensitive prostate cancer, whereas the present study evaluated the Decipher classifier, a validated prognostic biomarker in the localized and post-RP BCR setting.

There were several limitations to this study. First, multiple factors affected whether patients ultimately underwent both Decipher testing and PSMA PET/CT imaging, such as patient characteristics, provider preference, or patient preference. This likely introduced further confounding factors and potentially ascertainment bias, as Decipher testing results could influence the likelihood of undergoing PET/CT. Second, our analysis of transcriptomic differences was exploratory and we cannot definitively draw conclusions about specific mutational differences.

In conclusion, higher Decipher scores were associated with non-localized disease on PSMA PET/CT in both the pre-treatment and the post-RP setting. In the pre-treatment setting, non-localized disease was associated with higher rates of p53 mutations and lower rates of PAM50 luminal A subtype compared to localized disease. In the post-RP setting, signatures related to metabolism, DNA repair, and androgen receptor signaling were associated with higher rates of non-localized disease. Taken together, these data suggest that high Decipher scores are associated with the presence of non-localized disease, and that PSMA PET/CT-identified non-localized disease harbors transcriptomic characteristics associated with a more aggressive natural history. Further prospective studies are needed to determine whether treatment intensification (or potentially de-intensification) based on Decipher score and/or PSMA PET/CT improves outcomes.

Context Summary:

Key Points:

Key objective:

What is the relationship between Decipher genomic classifier scores and prostate-specific membrane antigen (PSMA) PET/CT-based metastatic spread?

Knowledge generated:

In the pre-treatment setting, for every 0.1 increase in Decipher score, the hazard ratio for non-localized disease on PSMA PET/CT was 1.18 (95% confidence interval: 1.03–1.36; P=0.02). In the post-radical prostatectomy setting, for every 0.1 increase in Decipher score, the hazard ratio for non-localized disease on PSMA PET/CT was 1.15 (95% confidence interval: 1.05–1.27; P=0.003).

Relevance:

In the era of readily accessible PSMA PET/CT scans, higher Decipher scores are associated with the detection of non-localized disease on PSMA PET/CT. This applies both in the pre-treatment and post-radical prostatectomy setting and can be used to risk-stratify patients.

Data Sharing Statement:

Research data are stored in an institutional repository and will be shared upon request to the corresponding author.