Extracapsular extension on MRI indicates a more aggressive cell cycle progression genotype of prostate cancer

Abstract

Purpose

To explore associations between magnetic resonance imaging (MRI) features of prostate cancer and expression levels of cell cycle genes, as assessed by the Prolaris® test.

Materials and Methods

Retrospective analysis of 118 PCa patients with genetic testing of biopsy specimen and prostate MRI from 08/2013 to 11/2015. Associations between the cell cycle risk (CCR) score and MRI features (i.e. PI-RADSv2 score, extracapsular extension (ECE), quantitative metrics) were analysed with Fisher’s exact test, nonparametric tests, and Spearman correlation coefficient. In 41 patients (34.7%), test results were compared to unfavourable features on prostatectomy specimen (i.e. Gleason group≥3, ECE, lymph node metastases).

Results

Fifty-four (45.8%), 60 (50.8%) and 4 (3.4%) patients had low, intermediate and high-risk cancers according to American Urological Association scoring system. Patients with ECE on MRI had significantly higher mean CCR scores (reader 1: 3.9 vs. 3.2, p=0.015; reader 2: 3.6 vs. 3.2, p=0.045). PI-RADSv2 scores and quantitative MRI features were not associated with CCR scores. In the prostatectomy subset, ECE on MRI (p=<0.001–0.001) and CCR scores (p=0.049) were significantly associated with unfavourable histopathologic features.

Conclusion

The phenotypic trait of ECE on MRI indicates a more aggressive genotype of prostate cancer.

INTRODUCTION

The management of localized prostate cancer (PCa) hinges on the risk of disease progression that needs to be counterbalanced against the patient’s overall fitness, life expectancy, and possible implications of cancer treatment. The American Urologic Association (AUA) guidelines base risk assessment on digital rectal examination, serum prostate specific antigen (PSA) level, and PCa differentiation and abundance on prostate biopsy specimen. The specific strength of this algorithm lies in the identification of patients at low risk for disease progression who could be safely managed conservatively through active surveillance. Outcomes in patients classified as ‘intermediate’ risk, however, are heterogeneous with reported rates of 5-year disease progression ranging between 21 and 91% in a recent systematic review [1]. Several approaches have been pursued to further increase the accuracy and customization of risk stratification algorithms, including numerical risk scores [2], nomograms [3], as well as blood and genomic biomarkers [4]. The Prolaris® genetic test (Myriad Genetics Inc, Salt Lake City, USA) measures the expression of 46 genes associated with cell cycle progression and cell division in PCa tissue. Its results are mathematically combined with a patient’s Cancer of the Prostate Risk Assessment (CAPRA) score and summarized as a numeric ‘cell cycle risk (CCR)’ score, which allows for more accurate prognostication and thus more individualized management decisions [5].

Multiparametric magnetic resonance imaging (MRI) of the prostate is routinely used for diagnosis and staging [6; 7], biopsy planning [8], and management decision support [9]. Numerous studies have reported associations between a tumor’s MRI appearance and histopathologic characteristics [10–13]; possible associations of a cancer’s MR imaging phenotype and genetic profile, however, are less thoroughly studied [14; 15]. The aim of this study was to explore potential associations between comprehensive quantitative and qualitative MRI features and the Prolaris® CCR score.

MATERIALS AND METHODS

Study Population

This institutional review board-approved, retrospective study included consecutive men with biopsy-proven prostate cancer, who underwent Prolaris® testing from August 2013 to November 2015. Men were included in the study if they met the following criteria: multiparametric 3 Tesla prostate MRI within 6 months of the genetic test, MRI performed at our institution using the protocol detailed below, and adequate biopsy tissue to perform the genetic test. 118 men met the criteria and were included in the study. No patients were excluded.

Multi-parametric MRI

All images were acquired on a 3 Tesla scanner (Discovery MR750w or Signa HDXT, GE Healthcare), and consisted of axial T1-weighted (repetition time, 450–750 msec; echo time, 7–12 msec; section thickness, 5 mm; intersection gap, 1 mm; field of view, 28–36 cm; matrix, 256×256–512*512), high-resolution axial/coronal/sagittal T2-weighted sequences (repetition time, 2500–6500 msec; echo time, 100–120 msec; section thickness, 3–5 mm; intersection gap, 0–1 mm; field of view, 14–24 cm; matrix, 256×192– 512*512), and diffusion weighted imaging with multiple b-values (repetition time, 3500–8500 msec; echo time, 60–100 msec; field of view, 16–20 cm; section thickness, 3–5 mm; intersection gap 0–1mm; field of view 14–24 cm; matrix 256 × 256; b values between 0 and 1000 sec/mm2). Apparent diffusion coefficient (ADC) maps were generated from DWI using a monoexponential model. Two board-certified radiologists (N.L.R., A.G.W.) with 4 and 6 years of experience in interpreting prostate MRI, who were blinded to all clinical data and the patients’ CCR scores, read all MRIs. Recorded MRI features, including PI-RADS version 2 score [16], extracapsular tumor extension (ECE), and quantitative metrics, are listed in Table 1.

Table 1:

Recorded MRI features

Feature	Description	Range
PI-RADS score	According to the Prostate Imaging-Reporting and Data System version 2	1–5
Extracapsular Tumor Extension (ECE)	Tumor spread beyond the prostate gland, assessed on a 5-point Likert scale; 1–3: absent, 4–5: present Criteria on MRI: Capsular contact/abutment; loss of T2 hypointense capsule; capsular bulging, irregularity, and/or retraction; neurovascular bundle retraction and/or thickening; presence of measurable extraprostatic tumor tissue; obliteration of recto-prostatic angle;	absent/present
Maximal Tumor Diameter	Maximum diameter of a dominant lesion on T2-weighted images (any plane) and ADC maps, respectively. ‘0’, if no dominant lesion was present.	Continuous
Mean ADC*	Mean ADC value of a dominant lesion; or, if no dominant lesion was present, mean ADC of normal appearing peripheral zone tissue	Continuous
Overall Tumor Volume#	Volume of the largest dominant lesion on ADC maps. ‘0’, if no dominant lesion was present.	Continuous
Markedly Diffusion Restricted Tumor Volume#	Tumor volume with ADC <1,000 mm2/s. ‘0’, if no dominant lesion was present.	Continuous

^*region of interest (ROI) placed in the centre of a dominant lesion being as large as possible, but remaining confined to the tumor margins to limit partial volume effects.

^#measured with TeraRecon^® software. The reader contoured the outline of each tumor on every slice (free-hand), using the ADC map derived from the b-0-1000 DWI sequences. The histogram analysis tool was then applied, which calculated the volume of tumor within two pre-defined ADC bins (i.e., <1,000 and ≥1,000 mm²/s).

Prolaris® gene expression testing

All patients underwent systematic transrectal ultrasound-guided prostate biopsy, the number of systematic cores per patient was ≤10 (n=4), 12 (n=57), 14 (n=46), or >14 (n=11). Thirty-seven patients (31.4%) also underwent targeted biopsy, the number of targeted cores ranged between one to four. The Prolaris® test (Myriad Genetics Inc, Salt Lake City, UT, USA) was applied to biopsy specimens performed at our institution in all study subjects. The test measures the expression level of 31 cell cycle progression (CCP) genes and 15 housekeeper genes to generate a CCP score. [17]. The CCP score is mathematically combined using a validated algorithm with the patient’s CAPRA score (13), resulting in the CCR score. Each 1-unit increase in CCR score represents a doubling of risk for PCa-specific mortality, with a higher score representing more aggressive cancers. Furthermore, the test result reports an individual patient’s relative risk compared to other patients within the same AUA risk category (i.e. “less aggressive”, “consistent”, or “more aggressive” than the average patient within the same AUA risk category). A subgroup analysis of patients undergoing radical prostatectomy was performed to investigate the association of genetic test results and MRI features with unfavourable histopathologic features, i.e. Gleason group≥3, ECE, lymph node metastases.

Statistical Analysis

We examined associations of MRI features with the genetic test results and AUA risk group in all patients, and of MRI features with unfavourable findings on prostatectomy specimen using Fisher’s exact test, exact Wilcoxon rank-sum test and exact KruskalWallis test. Spearman correlation coefficient was estimated between CCR and continuous tumor length or volume on MRI. The bias-corrected 95% confidence intervals (CI) for correlation coefficients were estimated using nonparametric bootstrapping method with 2000 replicates. Agreement was assessed using kappa (or weighted kappa with squared weights) on categorical features, intraclass correlation coefficients (ICC) and limits of agreement on continuous variables. The bias-corrected 95% confidence intervals (CI) were estimated for kappa and ICC.

RESULTS

The study population consisted of 118 men with a mean age of 61.4±7.4 years. According to the AUA criteria, 54 (45.8%) men were classified as low-risk, 60 (50.8%) as intermediate, and 4 (3.4%) as high-risk patients. The median CCR score was 3.3 (range, 1.6–5.5). In 57 (48.3%) patients, the genetic relative risk was “consistent” with the average, while 33 (28.0%) had “less aggressive” PCa, and 28 (23.7%) had “more aggressive” disease than the average within the same AUA category. On MRI, no patient was classified as PI-RADS 1, while a PI-RADS score of 2, 3, 4, and 5 was assigned in 21 and 20 (17.8% and 16.9%), 32 and 38 (27.1% and 32.2%), 53 and 52 (44.9% and 44.1%), and 12 and 8 (10.2% and 6.8%) cases by reader 1 and 2, respectively; inter-reader agreement for PIRADS scores was moderate (kappa 0.789). ECE was deemed present in 20 (16.9%) patients by reader 1, and in 18 (15.3%) cases by reader 2; inter-reader agreement for ECE on MRI was strong (kappa 0.81). Detailed descriptive statistics for all metric MRI-derived variables (i.e. tumor length on T2-weighted images and ADC maps, mean ADC, tumor volumes) including inter-reader agreement statistics, are provided in Supplement Table 1.

Association between AUA risk categorization and MRI features

Higher AUA risk categorization was significantly associated with higher PI-RADS scores (p<0.001 for both readers), a higher likelihood of having ECE on MRI (p<0.001 for both readers) and larger tumor diameters on T2-weighted images (p=0.002–0.004) and ADC maps (p=0.002–0.004). Also, mean tumor volume on ADC maps (p=0.002 for both readers) and the volume of markedly diffusion restricted tumor volume (p=0.0010.002) were higher in higher AUA risk categories. Higher AUA risk categorization was also associated with lower mean ADC values (p=0.007–0.008). Detailed numbers of these analyses are given in Supplement Table 2.

Association between MRI features and Genetic Test Results

As detailed in Table 2 and Figure 1, there was a significant association of the CCR score and the presence of ECE on MRI (reader 1: p=0.015, reader 2: 0.045), while there were no significant associations observed between CCR score and PI-RADS score or any of the quantitative MRI features. As listed in Table 3, there was no significant association between the evaluated MRI features and the likelihood of the genetic test indicating “less aggressive” or “more aggressive” disease.

Table 2:

Associations between MRI features and CCR score

Qualitative MRI Features	Reader 1		Reader 2
	Median CCR score (Range)	p-Value	Median CCR score (Range)	p-Value
PI-RADS v2 scores		0.289		0.214
2	2.9 (1.8, 5.5)		2.85 (1.8, 5.5)
3	3.25 (2.1, 4.7)		3.3 (2, 4.7)
4	3.3 (2.2, 4.7)		3.3 (1.6, 4.7)
5	3.75 (1.6, 4.6)		3.75 (2.6, 4.6)
Extracapsular Extension		0.015		0.045
Absent	3.2 (1.6, 5.5)		3.2 (1.6, 5.5)
Present	3.9 (2.3, 4.6)		3.6 (2.3, 4.6)
Quantitative MRI Features	Correlation Coefficent with (95%CI)	p-Value	Correlation Coefficent with (95%CI)	p-Value
Tumor Length on T2w images	0.127 (−0.049, 0.314)	0.169	0.105 (−0.074, 0.29)	0.257
Tumor Length on ADC maps	0.117 (−0.068, 0.298)	0.207	0.084 (−0.101, 0.268)	0.368
Tumor Volume on ADC maps	0.132 (−0.049, 0.318)	0.154	0.136 (−0.045, 0.321)	0.143
Tumor ADC value	−0.154 (−0.33, 0.046)	0.096	−0.106 (−0.279, 0.083)	0.254
Tumor Volume with Markedly Restricted Diffusion	0.144 (−0.036, 0.324)	0.120	0.173 (−0.014, 0.356)	0.061

Abbreviations: ADC Apparent Diffusion Coefficient, AUA American Urologic Association, CCR Cell Cycle Risk, PI-PI-RADSv2 Prostate Imaging Reporting and Data System version 2

Figure 1:

Distribution of the Cell Cycle Risk score amongst PI-RADSv2 scores (A), and amongst patients with and without extracapsular extension on MRI (B), respectively; data is shown separately for reader 1 (blue) and reader 2 (red).

Table 3:

Associations between MRI features and Prolaris® relative risk.

	Reader 1				Reader 2
MRI Features	Prolaris® Relative Risk			p-Value	Prolaris® Relative Risk			p-Value
	Less Aggressive (n=33)	Consistent (n=57)	More Aggressive (n=28)		Less Aggressive (n=33)	Consistent (n=57)	More Aggressive (n=28)
PI-RADS v2 scores				0.475				0.617
2	7 (21.2)	10 (17.5)	4 (14.3)		8 (24.2)	8 (14.0)	4 (14.3)
3	11 (33.3)	12 (21.1)	9 (32.1)		12 (36.4)	16 (28.1)	10 (35.7)
4	13 (39.4)	30 (52.6)	10 (35.7)		12 (36.4)	29 (50.9)	11 (39.3)
5	2 (6.1)	5 (8.8)	5 (17.9)		1 (3.0)	4 (7.0)	3 (10.7)
Extracapsular Extension (%)	3 (9.1)	8 (14.0)	9 (32.1)	0.054	4 (12.1)	8 (14.0)	9 (32.1)	0.300
Median Tumor Length on T2w images (mm)[range]	7.0 [0.0, 25.4]	8.5 [0.0, 23.2]	8.7 [0.0, 35.4]	0.468	6.1 [0.0, 26.8]	7.8 [0.0, 25.3]	8.3 [0.0, 30.9]	0.729
Median Tumor Length on ADC maps (mm)[ [range]	6.0 [0.0, 27.8]	8.7 [0.0, 26.0]	7.6 [0.0, 36.4]	0.462	6.2 [0.0, 27.0]	7.8 [0.0, 24.6]	6.8 [0.0, 37.1]	0.546
Median Tumor Volume on ADC maps (cm3)[ [range]	0.1 [0.0, 3.9]	0.3 [0.0, 4.9]	0.2 [0.0, 8.9]	0.368	0.1 [0.0, 0.6]	0.3 [0.0, 0.8]	0.2 [0.0, 0.9]	0.245
Median Tumor ADC values (mm2/s) [range]	1.2 [0.4, 1.9]	1.0 [0.5, 2.5]	1.0 [0.4, 1.9]	0.223	1.2 [0.9, 1.5]	0.9 [0.8, 1.5]	1.0 [0.7, 1.5]	0.491
Median Tumor Volume with Markedly Restricted Diffusion (cm3) [range]	0.0 [0.0, 3.5]	0.1 [0.0, 3.3]	0.0 [0.0, 7.5]	0.287	0.0 [0.0, 0.2]	0.10 [0.0, 0.5]	0.10 [0.0, 0.5]	0.144

Abbreviations: ADC Apparent Diffusion Coefficient, PI-RADSv2 Prostate Imaging Reporting and Data System version 2

Associations with Adverse Histopathologic Features

Of the 118 study subjects, 41 (34.7%) underwent radical prostatectomy, of which 18/41 (43.9%) were found to have at least one unfavourable finding on surgical histopathology (i.e. Gleason score 4+3 or higher in 9, ECE in 15, lymph node metastases in 5 patients). We observed significant associations of unfavourable surgical histopathology findings with the CCR score (p=0.049), and the presence of ECE on MRI (reader1: p<0.001, reader 2: p=0.001). We found a significant association of PIRADS scores and unfavourable pathology features for reader 1 (p=0.036), but not for reader 2 (p=0.779). No significant associations were observed between unfavourable pathology features and patients’ AUA risk category or any of the quantitative MRI features. Detailed numbers of these analyses are provided in Table 4, representative case studies are presented in Figure 2.

Table 4:

Associations of MRI features and unfavorable findings on prostatectomy histopathology.

Variable		No Adverse Feature (n=23)		Any Adverse Feature (n=18)		p-Value
AUA Risk (%)	High	0 (0.0)		2 (11.1)		0.161
	Intermediate	13 (56.5)		12 (66.7)
	Low	10 (43.5)		4 (22.2)
Prolaris® Results	Median CCR score [range]	3.1 [2.2, 5.5]		3.8 [1.8, 4.7]		0.049
	Relative Risk (%)					0.311
	Less Aggressive	6 (26.1)		4 (22.2)
	Consistent	12 (52.2)		6 (33.3)
	More Aggressive	5 (21.7)		8 (44.4)
Magnetic Resonance Imaging Features		Reader 1			Reader 2
		No	Any	p-Value	No	Any	p-Value
	PI-RADSv2 scores (%)			0.036			0.779
	2	2 (8.7)	1 (5.6)		3 (13.0)	1 (5.6)
	3	6 (26.1)	5 (27.8)		7 (30.4)	5 (27.8)
	4	15 (65.2)	7 (38.9)		12 (52.2)	10 (55.6)
	5	0 (0)	5 (27.8)		1 (4.3)	2 (11.1)
	Extracapsular Extension (%)	1 (4.3)	9 (50.0)	0.001	0 (0.0)	9 (50.0)	<0.001
	Median Tumor Length on T2w images (mm)[range]	7.0 [0.0, 17.9]	11.1 [0.0, 29.3]	0.070	7.4 [0.0, 16.6]	11.2 [0.0, 26.8]	0.166
	Median Tumor Length on ADC maps (mm)[ [range]	7.5 [0.0, 18.0]	11.6 [0.0, 28.1]	0.052	7.3 [0.0, 17.2]	10.6 [0.0, 27.0]	0.201
	Median Tumor Volume on ADC maps (cm3)[ [range]	0.28 [0.0. 3.51]	0.54 [0.0, 3.28]	0.293	0.26 [0.0, 3.38]	0.60 [0.0, 3.67]	0.253
	Median Tumor ADC values (mm2/s) [range]	1.04 [0.6, 1.73]	0.89 [0.42, 1.87]	0.331	0.94 [0.58, 1.9]	0.95 [0.42, 1.88]	0.854
	Median Tumor Volume with Markedly Restricted Diffusion (cm3) [range]	0.07 [0.0, 2.26]	0.21 [0.0, 2.88]	0.125	0.10 [0.0, 2.23]	0.15 [0.0, 3.25]	0.255

Abbreviations: ADC Apparent Diffusion Coefficient, AUA American Urologic Association, CCR Cell Cycle Risk, PI-RADSv2 Prostate Imaging Reporting and Data System version 2

Figure 2:

Representative T2-weighted images (T2w, column 1), high b-value diffusion weighted images (DWI, column 2) and apparent diffusion coefficient maps (ADC, column 3) of 5 patients (A-E), which highlight the strengths and imperfections of the investigated risk assessment tools, i.e. AUA criteria, Prolaris® test, and MRI.

• Patient A: Concordant high-risk features by AUA criteria, genetic test, and MRI (arrow indicates a dominant lesion in the left posterior peripheral zone with ECE); high-grade cancer with unfavourable features was verified on surgical histopathology.

• Patient B: PSA 13.1 ng/ml, otherwise low-risk features on clinical examination, biopsy, genetic test, and MRI (no measurable lesion); surgical histopathology showed ECE and positive surgical margins.

• Patient C: Low-risk features by AUA criteria and the genetic test, MRI showed an anterior lesion with ECE (white arrows), which was verified on the surgical specimen.

• Patient D: Intermediate risk by AUA criteria with more aggressive gene expression profile on the genetic test, MRI was equivocal for a measurable lesion and did not show ECE; surgical pathology revealed ECE and positive surgical margins.

• Patient E: Intermediate risk by AUA criteria with more aggressive gene expression profile on the genetic test, MRI was negative for a measurable lesion or ECE; low-grade tumor without unfavourable features on the prostatectomy specimen.

DISCUSSION

In this study, we explored associations between qualitative and quantitative prostate MRI features and the results of the Prolaris® genetic test in patients with low-, intermediate-, and high-risk PCa. The presence of ECE on MRI was associated with a higher CCR score. We did not find significant associations of the PI-RADS score or quantitative MRI features with the results of the genetic test. Renard-Penna et al. have previously analysed possible associations between (semi-)quantitative MRI features and cell cycle gene expression in prostate cancer [14; 15]. In concordance with our results, they did not find an association of tumor size and gene expression profile. For tumor ADC values, they reported a significant association with the CCP score in a cohort of 106 patients [15], but could not find this association in another cohort of 67 men [14], the latter being in concordance with our results. When assessed on a 5-point Likert scale [14], the cell cycle progression score was not associated with the MRI appearance of prostate cancer. However, when applying a 15-point scale derived from the PI-RADS version 1 classification algorithm, the authors found a significant association with the CCP score [15]. In neither of these studies, the authors evaluated the presence of ECE on MRI, which was associated with cell cycle gene expression in our cohort. In our subgroup analysis of patients undergoing prostatectomy, the CCR score and the presence of ECE on MRI were associated with unfavourable histopathologic features on prostatectomy specimen.

The Prolaris® genetic test measures the expression of 31 genes associated with cell cycle progression and cell division, as well as 15 housekeeping genes in prostate cancer tissue specimen. Its results are represented by a proliferative index, expressed as ‘cell cycle progression’ score, which is then mathematically combined with a patient’s CAPRA score and summarized as a numeric CCR score. The prognostic potential of this test has been evaluated in tissue specimen from transurethral resections [18], prostate biopsies [19], and prostatectomy specimen [20], and in regards to different clinical endpoints, including biochemical recurrence after prostatectomy [21] and radiation therapy [22], development of metastases [20], and mortality [23]. The potential of prostate MRI features for predicting biochemical tumor recurrence have been suggested in the postprostatectomy [24; 25] and post-radiotherapy [26] settings; in the latter two studies, the presence of ECE was associated with a higher likelihood of biochemical recurrence, independently of the cancer’s Gleason grade and pre-treatment PSA levels, respectively [25; 26]. The present study provides one possible molecular foundation for the poor prognostic implications of ECE on MRI. Both the genetic test results and the presence of ECE on MRI are associated with a higher risk of tumor recurrence after treatment, thus it will be interesting to assess the combined ability of these tools for prognostication in patients with prostate cancer. Theoretically, MRI might provide macroscopic phenotypic information and be less affected by sampling bias and genetic tumor heterogeneity [27], while the genomic test could compensate for some limitations of MRI, e.g. limited sensitivity for low-volume and/or low-grade malignancy [28], artefact and distortion [29], and inter-observer variability [30]. Due to the short follow-up period, the number of events (biochemical recurrence) limits statistical testing of this hypothesis in the current study.

Our study has several limitations, most importantly its retrospective design. By blinding the reader to all clinical data and the results of the genetic test, we aimed to minimize the probability of a confirmation bias. By including all consecutive patients undergoing genetic testing and prostate MRI, we tried to minimize selection bias; the relatively low proportion of patients with AUA high risk disease indicates that the latter might still be present. Also, we were not able to objectively assess the indication for genetic testing to incorporate into our analysis, which is a possible additional source of selection bias. At our institution, men with newly diagnosed prostate cancer routinely undergo MRI for staging, risk stratification, and treatment planning, decreasing the likelihood that the indication for MRI poses a substantial source of bias in this study. The lack of histopathological prove for tumor length, tumor volume, and presence of ECE in parts of our study population is another limitation of our study. Due to the small number of patients undergoing prostatectomy, we were not able to conduct multivariate statistical modelling to directly compare MRI and genetic features in their ability to predict unfavourable pathology features.

CONCLUSION

The phenotypic trait of extraprostatic tumor extension on MRI indicates a more aggressive genotype of prostate cancer, as quantified by the Prolaris® cell cycle risk score. Extracapsular extension on MRI and the genetic test result are associated with adverse histopathologic features on prostatectomy specimen.

Specific Remarks.

• Steve Stone is an employee of Myriad Genetics, Salt Lake City, UT, USA

• Michael K Brawer is a former employee of Myriad Genetics, Salt Lake City, UT, USA

• Andreas G Wibmer was supported by the Peter Michael Foundation