Assessing the Molecular Heterogeneity of Prostate Cancer Biopsy Sampling: Insights from the MAST Trial

Abstract

Background and objective:

Prostate cancer (PC) heterogeneity can result in sampling discrepancies during biopsy, leading to inaccurate molecular classifications that affect treatment decisions. We evaluated transcriptomic profile variability between multiparametric magnetic resonance imaging (mpMRI)-targeted biopsy (TBx) and systematic biopsy (SBx) methods using the Decipher GRID platform.

Methods:

The study included 205 men from the MAST trial. We analyzed 408 biopsy samples, of which 149 were TBx and 259 were SBx samples. Three prognostic signatures—the Decipher genomic classifier (DGC), cell cycle progression (CCP), and Genomic Prostate Score—were assessed in relation to grade group (GG) and MRI phenotype. Multivariable linear regression was conducted to adjust for the confounding effects of GG and tumor purity.

Key findings and limitations:

Unpaired analysis revealed that TBx samples had higher derived GPS and CCP scores than SBx samples (p < 0.05), but the difference was no longer significant after multiple-test adjustment. There was no significant difference in scores between SBx and TBx samples in the subgroup with GG 1 disease. For TBx cores, higher genomic scores were associated with higher Prostate Imaging-Reporting and Data System (PI-RADS) scores in the overall cohort, but not in the GG 1 subgroup. Multivariable analysis revealed significant associations between DGC and CCP scores and PI-RADS scores (p < 0.01). Higher DGC score concordance between TBx and SBx lesions was observed in the low-risk subgroup. A limitation of the study is the small sample size, so further validation is required.

Conclusions and clinical implications:

TBx samples yield higher genomic scores than SBx samples, with grade influencing the association between PI-RADS score and genomic risk. For the GG 1 subgroup, there was no correlation between PI-RADS and genomic scores. These findings need further validation to assess the impact of TBx on genomic risk assessment in active surveillance.

Patient summary:

We examined the effectiveness of two different biopsy methods in assessing the risk of prostate cancer (PC) progression. We found that while biopsy samples guided by MRI (magnetic resonance imaging) scans often showed higher genetic risk scores than biopsy samples without MRI guidance, the difference was not significant for men with lower-grade PC. Our findings suggest that MRI targeting for biopsy might not always provide additional information about cancer aggressiveness for patients with low-risk PC.

Magnetic resonance imaging (MRI)-targeted prostate biopsies often have higher genomic risk scores than systematic template biopsies, but this difference is not significant in low-grade cancers. This suggests that MRI targeting may not improve risk assessment in active surveillance for patients with low-risk prostate cancer.

1. Introduction

Prostate cancer (PC) is a heterogeneous and often multifocal disease that is often undersampled on systematic biopsy (SBx) [1,2]. For many men with low-risk or favorable intermediate-risk PC, active surveillance (AS) may be an option, which balances the need for intervention against the risks of overtreatment. However, patient selection relies heavily on accurate risk assessment and timely detection of progression [3]. Central to this approach is the use of multiparametric magnetic resonance imaging (mpMRI), which is increasingly being used in men considering AS to reduce the risk of underestimating the true grade and extent of tumor within the prostate. Repeat biopsy has also become an essential component of AS [4], with nearly one-third of patients reclassified on the basis of a first confirmatory biopsy [5].

The current standard of care is combined MRI-targeted biopsy (TBx) and a systematic biopsy (SBx) template that samples the entire gland [6]. However, there is increasing use of tissue-based genomic classifiers (GCs) such as the Decipher GC, the Genomic Prostate Score (GPS) [7], and the Prolaris cell cycle progression (CCP) score [8], which provide molecular insights into tumor biology and help in guiding decisions regarding the need for and intensity of treatment [9]. Previous studies have shown that higher biopsy grades and MRI risk levels correlate with higher GC scores [10,11]. However, there is a paucity of reports on the difference in GC scores between TBx and SBx cores sampled from men on AS. Our aim was to address this question in a single-center prospective cohort of men on AS undergoing serial TBx and SBx of the prostate.

2. Patients and methods

2.1. Trial design and procedures

The MAST trial (Miami MRI selection for Active Surveillance versus Treatment; NCT02242773) was a prospective, single-arm interventional study to assess the molecular heterogeneity of PC biopsy sampling among patients with low- or favorable intermediate-risk PC on AS. In the present study, we specifically compared the efficacy of TBx versus SBx in evaluating expression data and genomic risk. We included a total of 205 patients with an initial diagnostic biopsy indicating low- or intermediate-risk PC with a prostate-specific antigen level of <20 ng/ml and no history of prior PC treatment or pelvic radiation. Patients on AS had diagnostic, confirmatory, and surveillance biopsies during follow-up. Biopsy sampling was performed using MRI guidance for any target lesion with a Prostate Imaging-Reporting and Data System (PI-RADS) score of ≥3 (TBx) and was combined with 12-core SBx. mpMRI scans were obtained before confirmatory and surveillance biopsies and were interpreted by genitourinary radiologists using the most up-to-date version of PI-RADS at the time. All prostate biopsies were read by a fellowship-trained genitourinary pathologist at our institution.

2.2. Specimen processing and genomic analysis

Two formalin-fixed, paraffin-embedded cores from each biopsy session were analyzed using the Decipher assay (San Diego, CA, USA). Specimen collection and tumor processing were based on the Veracyte standard operating procedures, with selection of biopsy cores having at least 1 mm of tumor and the highest grade group (GG) from both the TBx and SBx samples sent for transcriptomic analysis. In cases with only GG 1 disease on TBx and SBx, cores with the longest linear tumor length were selected. Quality control checks and normalization methods were applied to raw data from the microarray assay to yield transcriptome-wide expression data. Gene-wise expression data were then summarized into a plethora of signatures delivered as the Decipher Genomic Resource Information Database (GRID) platform, broadly separated into prognostic risk scores and biological pathways including distinct GCs. Among these GCs, Decipher (v1.2) score is obtained from a 22 gene signature analysis, and it ranges from 0 to 1. This score is computed using the locked Clinical Laboratory Improvement Amendments–accredited commercial assay. This signature is the commercially available assay, for which scores are categorized as low (0–0.45), intermediate (0.45–0.6), or high (>0.6) genomic risk. GRID also contains other signature scores, including the cell cycle progression (CCP) score [8] and Genomic Prostate Score (GPS) [7], derived from expression levels of the particular genes in the signatures. A score (prescaled to 0–1) for tumor cell purity was calculated according to the ESTIMATE algorithm [12] and stored in the Decipher GRID database. Differential gene expression analysis transcriptomic data for paired samples was performed to assess alterations in hallmark pathways.

2.3. Outcomes and data analysis

Each biopsy core with a genomic profile was categorized as either TBx or SBx. For TBx cores, the PI-RADS score for the target was also recorded. The primary analysis compared GC scores between TBx (visible lesion) and SBx (non-visible). For each GC signature, we compared the scores between TBx and SBx cores in either patient paired design (comparing targeted and systematic biopsies within each patient; Wilcoxon signed-rank test was employed) or un-paired design (comparing targeted and systematic biopsies regardless of the source patients; Wilcoxon rank-sum test was used). In addition, for TBx samples we evaluated the correlation between GC and PI-RADS scores using the Kruskal-Wallis test.

The same analyses were performed in the subgroup of patients with GG 1 disease. We also conducted linear regression in two multivariable analyses (MVAs) to rigorously adjust for potential cofounding effects of GG and tumor purity. The first MVA analyzed the effect of binary biopsy type (TBx vs SBx) on GC scores, while the second assessed the effect of PI-RADS score (3, 4, or 5) versus SBx on the scores. A separate analysis for each of the three GC signatures was conducted using these two MVA models.

Finally, we evaluated gene expression related to specific molecular pathways across TBx versus SBx samples and by PI-RADS score.

All statistical analyses were performed using R v4.2.2. (R Foundation for Statistical Computing, Vienna, Austria). All tests were two-sided. Because the statistical tests were repeated for three GC signatures in parallel, we applied the Bonferroni principle for multiple test adjustment and considered p < 0.016 as the threshold for statistical significance.

3. Results

3.1. Descriptive analysis

A total of 205 patients were included in the MAST trial, of whom 159 had transcriptomic profiles available from at least one of the positive biopsy cores at baseline or during AS. The total number of cores analyzed was 408, of which 259 were obtained via systematic sampling and 149 via targeted sampling. Of the TBx samples, 25% corresponded to PI-RADS 3, 58% to PI-RADS 4, and 17% to PI-RADS 5 lesions.

Among the total biopsy cores, 323 (79%) had GG 1, 53 (13%) had GG 2, 15 (4%) had GG 3, and 17 (4%) had GG ≥4 PC. Differences in GC score by GG were statistically significant (Supplementary Fig. 1). Most of the positive cores were GG 1, with detection rates of 83% for SBx samples, 75% for PI-RADS 3 lesions, 77% for PI-RADS 4 lesions, and 42% for PI-RADS 5 lesions (Table 1).

Table 1 –

Characteristics of the patients included in the study

Parameter	Overall cohort	TBx group	SBx group
Number of patients	159	91	143
Number of biopsies	279	130	223
Total number of cores analyzed	408	149	259
Median number of cores analyzed per patient (IQR)	2 (1–3)	1 (1–2)	1 (1–2)
Median number of cores analyzed per biopsy (IQR)	1 (1–2)	1 (1–1)	1 (1–1)
Tumor purity score (standard deviation)	0.78 (0.12)	0.79 (0.11)	0.78 (0.13)
Grade group, n (%)
GG 1	323 (79)	106 (71)	217 (83)
GG 2	53 (13)	29 (19.4)	24 (9.2)
GG 3	15 (3.6)	6 (4)	9 (3.4)
GG 4	10 (2.4)	4 (2.6)	6 (2.3)
GG 5	7 (1.7)	4 (2.6)	3 (1.1)
PI-RADS score, n (%)
3	-	37 (25)	-
4	-	86 (58)	-
5	-	26 (17)	-
Decipher risk category, n (%)
Low	314 (77)	114 (76.5)	200 (77.2)
Intermediate	38 (9)	11 (7.3)	27 (10.5)
High	56 (14)	24 (16.2)	32 (12.3)

IQR = interquartile range; PI-RADS = Prostate Imaging-Reporting and Data System; SBx = systematic biopsy; TBx = targeted biopsy.

3.2. Correlation between GC scores and MRI phenotype

Comparison of TBx and SBx via unpaired analysis revealed significantly higher derived GPS (p = 0.047) and CCP scores (p = 0.019) for TBx samples (Wilcoxon test), but the difference was no longer significant after adjustment for multiple testing (Fig. 1A). Paired analysis revealed no significant difference in any of the genomic scores between TBx and SBx samples (Fig. 1B). Similar findings were obtained when limiting the paired analysis to GG 1 lesions (Fig. 1D).

Fig. 1 –

Comparison of genomic classifier scores for systematic versus targeted biopsies in the overall cohort according to (A) unpaired and (B) paired analysis, and the subgroup with grade group 1 disease on (C) unpaired and (D) paired analysis. The p values are from two-group Wilcoxon tests. CCP = cell cycle progression; GPS = Genomic Prostate Score.

Among PI-RADS 3, 4, and 5 lesions, there were significant differences in Decipher (p = 0.007) and derived CCP scores (p = 0.002) between TBx and SBx samples (Kruskal-Wallis test) in the overall cohort (Fig. 2A) but not in the subgroup of patients with GG 1 cancer (Fig. 2B).

Fig. 2 –

Comparison of the main genomic classifiers (Decipher, derived GPS, derived CCP) for magnetic resonance imaging phenotypes (PI-RADS 3, 4, and 5) in (A) the overall cohort and (B) the subgroup with grade group 1 disease. The p values are from Kruskal-Wallis tests. CCP = cell cycle progression; GPS = Genomic Prostate Score; PI-RADS = Prostate Imaging-Reporting and Data System.

In the linear regression MVAs, GG ≥2 in both models was significantly associated with GC scores for all three classifiers. Tumor purity was significantly correlated with the derived GPS and the Decipher scores, but not with derived CCP scores. By contrast, in both the binary analysis (TBx vs SBx, model A) and the categorical analysis (PI-RADS 3, 4, and 5 vs SBx, model B; Table 2), the only statistically significant associations were for the derived CCP score (TBx vs SBx: p = 0.015; PI-RADS 4 vs SBx: p = 0.001; PI-RADS 5 vs SBx: p = 0.008).

Table 2 –

Multivariable regression results from models for (A) binary TBx versus SBx and (B) discrete PI-RADS scores versus SBx

Model	Factor	Decipher GC		Derived GPS		Derived CCP
		Coefficient	p value	Coefficient	p value	Coefficient	p value
Model A	Intercept	0.82	<0.001	0.27	<0.001	0.46	<0.001
	TBx vs SBx	0.02	0.239	0.00	0.832	0.03	0.015 *
	GG ≥2 vs GG 1	0.12	<0.001	0.07	<0.001	0.06	<0.001
	Tumor purity	−0.68	<0.001	−0.18	<0.001	−0.09	0.061
Model B	Intercept	0.82	<0.001	0.27	<0.001	0.46	<0.001
	PI-RADS 3 vs SBx	−0.05	0.108	−0.02	0.280	−0.04	0.062
	PI-RADS 4 vs SBx	0.05	0.039	0.01	0.277	0.05	0.001 *
	PI-RADS 5 vs SBx	0.05	0.203	−0.01	0.775	0.07	0.008 *
	GG ≥2 vs GG 1	0.11	<0.001	0.07	<0.001	0.06	<0.001
	Tumor purity	−0.68	<0.001	−0.18	<0.001	−0.10	0.052

CCP = cell cycle progression; GC = genomic classifier; GG = grade group; GPS = Genomic Prostate Score; PI-RADS = Prostate Imaging-Reporting and Data System; SBx = systematic biopsy; TBx = targeted biopsy.

^*Statistically significant.

Variation in DGC scores for TBx samples was also observed, but the scores were concordant with those for SBx samples in paired analyses, with no major low-risk versus high-risk discordance between TBx and SBx samples in the overall cohort or in the GG 1 subgroup (Fig. 3). However, subjects with intermediate-risk DGC scores for SBx samples tended to have low-risk scores for TBx samples. Supplementary Figure 2 compares the concordance of SBx and TBx scores between GG 1 TBx–GG ≥2 SBx and GG 1 TBx–GG 1 SBx pairs, with higher scores for the GG 1 TBx–GG ≥2 SBx pairs across all three GCs.

Fig. 3 –

Sankey diagram of the flow of risk group according to the Decipher genomic classifier score between systematic biopsy and targeted biopsy in paired analysis for (A) the overall cohort and (B) the subgroup with grade group 1 disease.

3.3. Genomic expression analysis

The principal pathways with differential gene expression between TBx and SBx samples and by PI-RADS score are shown in Supplementary Figure 3. TBx samples were associated with upregulation of the PAM 50 luminal B signature, cell metabolism pathways, and the E2F pathway, and with downregulation of the cell adhesion pathway (apical junction). Among the TBx samples, highly suspicious lesions were significantly associated with upregulation of cell-cycle and plasticity pathways (E2F targets, G2M checkpoint), androgen receptor and DNA repair pathways, and the PAM 50 luminal B signature.

4. Discussion

Use of mpMRI has changed the diagnostic and therapeutic management of PC, especially in the AS setting, as an additional tool to improve risk assessment. Given the heterogeneity of PC, MRI can be helpful in identifying the index lesion, which is likely to harbor the highest grade and therefore have the most impact on the risk of progression. The use of genomics in PC has allowed a better understanding of disease biology and could aid in enhancing personalized treatment strategies, especially in AS, where it could help in decisions not only on whether to treat or not but also on the intensity of surveillance. Several studies have looked at the impact of combining MRI identification of the index lesion with GC signatures and have seen a benefit in using MRI targeting to sample the most aggressive part of the prostate and the most appropriate tissue for genomic risk assessment [11]. However, few have directly addressed this issue in a prospective AS cohort. In this study, we assessed the correlation between tissue-based GC scores and MRI phenotypes in an AS cohort, with adjustment for grade. As hypothesized, patients with higher cancer grade on biopsy had higher genomic risk, regardless of their MRI phenotype.

PI-RADS scores were correlated with GC scores in our overall cohort, but not in the low-risk group. This finding suggests that the molecular profile of high-risk tumors is correlated more to grade, and that the features of the prostate observable on imaging are not associated with higher genomic risk in the case of low-grade disease.

It has been demonstrated that MRI visibility is correlated with the genomic expression that is associated with tumor metastasis and disease prognosis [13,14]. However, higher-grade tumors are associated with higher PI-RADS scores. TBx of MRI-visible lesions is more indicative of tumor aggressiveness than SBx in GG >1 disease. This correlation is not event in GG 1 disease. The variation in genomic risk could be because of the TBx correlation with Gleason score rather than the MRI phenotype per se. However, whether the genomic risk for the index lesion correlates with the overall genomic risk remains to be answered. Radkte et al found 83% concordance for Decipher risk scores between biopsy and radical prostatectomy specimens [15].

Several studies have analyzed the correlation between MRI lesions and genomic assessment of PC and demonstrated an association between radiomics features and GC scores [10,16,17]. However, many of these studies involved surgical cohorts and the generalization to AS populations may yield inconsistent findings [18]. Our results are consistent with other studies in which MRI findings were not correlated with genomic risk in low- to intermediate-risk PC [19], which calls into question the incremental benefit of MRI in improving genomic risk stratification beyond detecting higher-grade disease. Nevertheless, MRI guidance remains an important tool for adequate tumor sampling, while GCs provide information about disease biology.

Significant advances in genomic assays have increased their utility for decision-making in PC management, but uptake is much slower than for MRI and these tools are mostly used in the radiation setting (primary and salvage) rather than in AS. The principal use of these tests is for prostate biopsy and prostatectomy specimens. However, a limitation of genomic assays is intratumoral and intertumoral heterogeneity [2]. Our paired analysis of biopsy cores revealed alterations in different pathways, explaining the molecular basis underlying genomic heterogeneity.

Our study has several limitations. First, while the paired sample design is more accurate as it adjusts for confounding patient-associated clinical factors, we had a limited cohort of 62 patients contributing 74 samples with both TBx and SBx cores. The unpaired analysis involved a much larger sample size for the comparator group, but between-patient variability was not taken into account. Second, correlation with a final prostatectomy specimen was not possible for the study population. Whole-mount analysis of a prostatectomy specimen would allow better evaluation of index and nonindex lesions and of the presence of adverse features and overall oncological risk. Nevertheless, the study cohort was an AS population and any focus on patients who underwent surgery would have been impacted by selection bias.

Strengths of the study include the prospective analysis in an AS cohort and use of the Decipher score on the same basis as in routine clinical practice. While GPS and CCP scores were derived from quantitative reverse transcription-polymerase chain reaction data for gene expression instead of the proprietary algorithms, the Decipher results reflect routine use of this commercial assay. Genomic assessment of the core with the highest grade and volume from both TBx and SBx samples, which is not the standard of care when using these tests in clinical practice, allowed us to address this question.

5. Conclusions

Although TBx results for lesions observed on MRI correlate with clinically significant PC, GC scores were not associated with PI-RADS scores in the subgroup with low-risk disease. PC heterogeneity was lower in low-risk disease, but we did observe variation in molecular pathway expression between TBx and SBx samples. Incorporation of GC scores in AS protocols could add a potential tool for risk classification and assessment of disease aggressiveness.