‘High proliferative cribriform prostate cancer’ defines a patient subgroup with an inferior prognosis

A cribriform pattern, reactive stroma (RS), PTEN, Ki67 and ERG are promising prognostic biomarkers in primary prostate cancer (PCa). We aim to determine the relative contribution of these factors and the Cancer of the Prostate Risk Assessment Postsurgical (CAPRA‐S) score in predicting PCa prognosis.


Introduction
Prostate cancer (PCa) patients treated with radical prostatectomy (RP) have diverse outcomes, especially among intermediate and high-risk PCa.Hence, there is an unmet need for additional prognostic markers to improve stratification of PCa patients regarding choice of treatment and follow-up.
Histopathological evaluation plays a fundamental role in determining PCa prognosis, and the importance of specific morphological features has gained increasing attention due to their associations with poorer prognosis.In particular, a cribriform pattern [intraductal carcinoma (IDC)] and/or invasive cribriform carcinoma (ICC)]) 1,2 and reactive stroma (RS) 2 have been highlighted as potential prognostic biomarkers, and are included or being considered for inclusion in contemporary PCa histopathological grading.
Several molecular biomarkers have emerged as potential prognostic biomarkers in PCa, including Ki67, PTEN and ERG.5][6][7][8] Ki67 is a cell proliferation marker, 9 whereas PTEN is the protein encoded by the tumour suppressor gene PTEN, and loss of PTEN expression is among the most frequently identified molecular aberrations in PCa. 10,11Fusion genes involving ERG or other ETS genes are the most common molecular abberrations in localised PCa, with TMPRSS2::ERG being the most prevalent (present in approximately 50% of PCas). 10,12onsequently, much research has been directed at whether ERG is a useful PCa biomarker; however, its prognostic value is still debated 13 and, as we recently demonstrated, this may be due to the substantial spatial heterogeneity of ERG expression. 14here is a need for studies directly comparing multiple morphological and/or molecular biomarkers as prognosticators. 3,15Identification of which biomarkers provide the strongest prognostic information is needed to optimise risk stratification and determine which features pathologists should evaluate and report.However, any assessed biomarker must be compared to standard clinicopathological parameters that are used for risk stratification today, and one approach is to use the Cancer of the Prostate Risk Assessment Postsurgical (CAPRA-S) score. 16This tool combines information from pre-and postoperative clinicopathological factors to risk stratify patients after RP and aid clinicians in making well-informed decisions on whether to give additional therapy.
Although a cribriform pattern is emphasised as being associated with a poorer prognosis, not all patients with this feature will experience relapse.Consequently, molecular biomarkers represent one approach towards identifying subgroups of patients who will experience relapse.In one study, PTEN loss in association with IDC and/or ICC did not have an impact on prognosis, and it was suggested that other molecular aberrations contribute to the poorer prognosis for these patients. 17To the best of our knowledge, the prognostic impact of Ki67 or ERG expression in patients with a cribriform pattern has not been explored.
The aim of this study was to determine which of the biomarkers: cribriform pattern, RS, PTEN, Ki67 or ERG, were prognostic predictors in primary PCa and should be implemented in pathology reports.Further, we aim to evaluate if the assessment of PTEN, Ki67 and ERG could aid in stratification of patients with a cribriform pattern with regard to clinical outcome.

S T U D Y C O H O R T
9][20] The current study originally included 515 patients with a median follow-up of 8.7 years [interquartile range (IQR) = 7.7-9.8years].A flowchart for patient inclusion is shown in Supporting information, Figure S1.
Informed written consent was obtained by all included patients.The project is approved by the Regional Committees for Medical Research Ethics South East Norway (number 2013/595/REK southeast A).
The study is reported according to the REporting recommendations for tumour MARKer prognostic studies 21 (REMARK; Supporting information, Table S1).

H I S T O P A T H O L O G I C A L A S S E S S M E N T
RP specimens were reviewed by two investigators (U.A. and M.B.) according to the 2014 International Society of Urological Pathology (ISUP) consensus conference on Gleason grading of prostatic carcinoma. 22he grade group for the entire prostate was determined in addition to the presence of a cribriform pattern, RS and multifocality.
A cribriform pattern was defined as the presence of IDC and/or ICC.IDC was defined as proliferation of malignant epithelial cells within predefined acini/ ducts and with at least a partially preserved basal cell layer. 23Only IDC with cribriform morphology was included in the current study.ICC was defined as sheets of cells with multiple glandular lumina without intervening stroma. 24A patient was classified as having a cribriform pattern if a cribriform pattern was detected in any tissue section from that patient.If no cribriform pattern was identified, the patient was classified as having non-cribriform PCa.
RS was evaluated according to morphology as described by De Vivar et al. 25  in Supporting information, Figure S2.A patient was classified as having a RS if RS was detected in any tissue section from that patient.
The index tumour 26 was defined as the malignant focus with the highest pathological tumour stage.If multiple foci had the same pathological tumour stage, the focus with the highest Gleason score was defined as the index tumour.In cases where multiple foci had both the same pathological tumour stage and the same Gleason score, the largest focus in diameter was defined as the index tumour.Intrafocal heterogeneity was defined as heterogeneous expression of a molecular biomarker within the same malignant focus.Interfocal heterogeneity was defined as heterogeneous expression of a molecular biomarker between different malignant foci from the same patient.
For each patient, the CAPRA-S score was calculated 16 (Supporting information, Table S2).The CAPRA-S score is a tool used to predict BCR and cancer-specific mortality after RP. 16 In brief, the CAPRA-S score ranges from 0 to 12 points, and includes information on the pre-operative prostatespecific antigen (PSA; 0-3 points) and the postoperative pathological variables surgical margin status (0-2 points), seminal vesicle invasion (0-2 points), extraprostatic extension (0-1 point), lymph node invasion (0-1 point) and Gleason score (0-3 points) from the RP specimen.For two patients, the CAPRA-S score could not be determined due to missing information about the pre-operative PSA level or pathological tumour stage.Patients where a lymph node dissection was not performed (pNX) were classified as without lymph node invasion.For statistical analyses, patients with CAPRA-S scores of 0-2 were classified as low-risk, 3-5 as intermediate-risk and ≥ 6 as high-risk.
The tissue microarray (TMA) construction was performed as previously described. 14Regions of interest for TMA sampling included the highest Gleason score, areas with cribriform pattern, RS and areas from different malignant foci.In total, 477 patients had at least one malignant sample that could be evaluated on the TMAs for PTEN, Ki67 and ERG expression (Supporting information, Figure S1 and Table S3).In total, 1960 samples (1391 malignant, 569 benign) could be evaluated, with a median of three malignant samples per patient.In total, 304 patients had multiple samples from the same malignant focus represented on the TMAs and were included in the assessment of intrafocal heterogeneity of PTEN and Ki67 expression.A total of 149 patients had samples from multiple malignant foci represented on the TMAs, and were included in the assessment of interfocal heterogeneity of PTEN and Ki67 expression.Heterogeneity in ERG expression has been described previously. 14TEN staining was performed using Dako Autostainer Link (Agilent, Santa Clara, CA, USA), with antihuman PTEN (clone 6H2.1; dilution 1:100; Agilent) and Dako EnV FLEX+, High pH (Link) detection kit (product number: K8002; Agilent).Controls included PTEN-negative endometrioid adenocarcinoma, PTENpositive clear cell adenocarcinoma and tonsillar tissue.The stroma and/or surrounding benign tissue were used as internal positive controls.27 Ki67 staining was performed using the Ventana Benchmark Ultra system (Roche Tissue Diagnostics, Tucson, AZ, USA), with anti-human Ki67 antigen (clone MIB-1, dilution 1:100; Agilent) and Ventana UltraView Universal DAB detection kit (product number: 760-500; Roche).Controls included tissue from tonsil, liver, pancreas and appendix.ERG staining was performed as previously described.14 PTEN expression was visually scored and dichotomised as 'intact' or 'loss' in malignant cells.27 A TMA sample was classified as 'loss' if any percentage of the malignant cells demonstrated a significant reduction in staining intensity or was entirely negative when compared to benign glands and stroma.27 Ki67 expression was evaluated semi-quantitatively based on the percentage of malignant cells with positive (brown) nuclear staining and classified as either 'low' (< 5%) or 'high' (≥ 5%).4,5,28 ERG expression was evaluated as 'positive' or 'negative', as previously described.14 A patient was classified as having PTEN loss, high Ki67 or being ERG-positive if any malignant sample from that patient displayed PTEN loss, high Ki67 or was ERG-positive, respectively.

S T A T I S T I C A L A N A L Y S E S
Statistical analyses were performed using R (version 4.1.1)and RStudio (version 1.4.1717).A P-value of < 0.05 was used as the threshold for statistical significance.
Associations between the biomarkers and also with CAPRA-S were assessed with the v 2 test of independence, Fisher's exact test (categorical) or Wilcoxon's rank-sum test (continuous).In total, 475 patients Morphological and molecular biomarkers in PCa 855 with complete clinicopathological and molecular data were included in these analyses.In addition, on the TMA sample level the association between molecular biomarker expression and the morphological growth patterns cribriform pattern and non-cribriform Gleason pattern 4 was assessed using the v 2 test.
For time-to-event analyses, 418 patients without a post-RP persistent PSA and who did not receive adjuvant treatment were included.A persistent PSA was defined as a PSA of ≥ 0.10 ng/ml at 4-8 weeks after RP. Biochemical recurrence (BCR) and clinical recurrence (CR) were used as endpoints.Metastatic castration-resistant PCa (mCRPC) and PCa-specific mortality were not included as endpoints due to few events.BCR was defined as PSA recurrence ≥ 0.20 ng/ml in two consecutive blood samples (≥ 1 week apart), and the first date was used as the time-point for BCR.CR was defined as local recurrence in the prostatic bed or metastasis to lymph nodes or distant sites; and the date of treatment commencement was used as the time-point for CR.Patients with positive pelvic lymph nodes at the time of RP were not defined as having CR.Cases with incomplete follow-up data were censored at the last known follow-up date.Follow-up data were collected continuously through review of hospital records or correspondence with general practitioners.In addition, PSA values were collected from F€ urst Medical Laboratories (Oslo, Norway).Date and cause of death were recorded based on data collected from the Norwegian Cause of Death Registry (Oslo, Norway).
The 'survival' and 'survminer' R packages were used for time-to-event analyses.Kaplan-Meier plots and log-rank test were used to visualise and compare survival curves, respectively.Survival curves were truncated when the number at risk within a particular group was less than five.Univariable and multivariable Cox regression analyses were performed to estimate hazard ratios (HRs) with 95% confidence intervals (CIs) for BCR and CR.The proportional hazards assumption was assessed with the Schoenfeld test using the cox.zph function implemented in the 'survival' package.
In multivariable analyses the prognostic value of the individual biomarkers: cribriform pattern, RS, PTEN, Ki67 and ERG, in addition to the CAPRA-S risk group, were assessed.In addition, a separate analysis assessing the prognostic value of a cribriform pattern when adjusting for grade group was performed.To evaluate and compare which biomarker contributed the most prognostic information, the relative proportion of explained variation for BCR and CR was calculated for each variable included in the multivariable models using the 'survMisc' package. 29,30n multivariable analyses assessing the prognostic value of the constructed variable of cribriform pattern in combination with PTEN, Ki67 or ERG status, the CAPRA-S risk group was included as a covariable.Interactions between a cribriform pattern and PTEN, Ki67 or ERG were also assessed.The likelihood-ratio test was used to determine whether interaction terms should be included in the multivariable model.

C L I N I C O P A T H O L O G
The cohort was comprised of 475 patients with complete clinicopathological data and with at least one malignant sample on the TMAs that could be evaluated for PTEN, ERG and Ki67 status.According to the CAPRA-S risk groups, 36% of patients were classified as low-risk, 41% as intermediate-risk and 23% as high-risk (Table 1 and Supporting information, Table S4).Of the assessed biomarkers, a cribriform pattern was identified in 57% of patients, PTEN loss in 55%, ERG expression in 51%, RS in 39% and high Ki67 in 9% (Table 1 and Figure 1A,B).

A C R I B R I F O R M P A T T E R N A N D H I G H K I 6 7 A R E A S S O C I A T E D W I T H A H I G H E R C A P R A -S R I S K G R O U P , P R E S E N C E O F A P O S T -R P P E R S I S T E N T P S
Patients with a cribriform pattern and/or RS were more frequently classified as being in a higher CAPRA-S risk group and had PTEN loss, high Ki67 and ERG expression compared to patients without these morphological features.Furthermore, patients with a cribriform pattern more often had RS, a post-RP persistent PSA and developed mCRPC compared to patients with non-cribriform PCa (Figure 1C and Supporting information, Table S5).
Patients with loss of PTEN or high Ki67 expression more frequently had PCa classified in higher CAPRA-S risk groups, whereas a significant association was not identified for ERG expression.Patients with a high Ki67 expression more frequently also had apost-RP persistent PSA and developed mCRPC compared to patients with low Ki67 expression (Figure 1C and Supporting information, Table S6).
The heterogeneity of PTEN and Ki67 expression was also assessed to determine whether heterogeneity may impact the accurate assessment of these biomarkers.The heterogeneity of ERG expression has previously been assessed and is not further described here. 14For patients with PTEN loss or a high Ki67, and with at least two samples from the same malignant focus, intrafocal heterogeneity was detected in 55% (106 of 193) of patients for PTEN loss and in 89% (33 of 37) for a high Ki67 expression (Figure 1D).Among patients with samples from at least two different malignant foci, interfocal heterogeneity was identified in 46% (68 of 149) of patients for PTEN loss and 6% (nine of 149) of patients for a Morphological and molecular biomarkers in PCa 857 high Ki67 expression (Figure 1E).Among the patients where interfocal heterogeneity was identified for PTEN expression, PTEN loss was exclusively identified in a non-index malignant focus in 40% (27 of 68) of patients (Figure 1E).Among the patients where interfocal heterogeneity was identified for Ki67 expression, a high Ki67 expression was exclusively identified in a non-index malignant focus in 22% (two of nine) of patients (Figure 1E).All the assessed biomarkers were significantly associated with a higher risk of BCR and CR (Figures 2A-E, 3A-E and Table 2).
The relative proportion of explained variation for the multivariable models were calculated to compare and identify the most important features in the prediction of BCR and CR.CAPRA-S contributed most to the explained variation for predicting BCR (77%) and CR (61%).A cribriform pattern was identified as the most important prognostic factor compared to the other biomarkers in the prediction of BCR (11% compared to ≤ 8%; Figure 2F) and high Ki67 as the most important prognostic factor in the prediction of CR (21% compared to ≤ 10%; Figure 3F).

H I G H K I 6 7 E X P R E S S I O N T O G E T H E R W I T H A C R I B R I F O R M P A T T E R N I D E N T I F I E S P A T I E N T S W I T H A H I G H R I S K O F B C R A N D C R
Although a cribriform pattern was identified as an independent prognostic factor in predicting BCR and CR, not all patients with a cribriform pattern will experience BCR nor CR.It was therefore explored whether PTEN, Ki67 or ERG in combination with a cribriform pattern could help to identify patients with a higher risk of relapse.The difference in prognosis based on PTEN, Ki67 or ERG expression was most pronounced for Ki67 (Figure 4 and Table 3), whereas differences were not as evident for PTEN or ERG (Supporting information, Figure S3 and Table 3).It was investigated whether interaction terms between a cribriform pattern, PTEN, Ki67 and ERG should be included; however, statistically significant interactions were not identified (all P > 0.2).

Morphological and molecular biomarkers in PCa 859
In multivariable analyses adjusting for CAPRA-S, the patient subgroup with a cribriform pattern and concurrent high Ki67 were more than two times as likely to experience BCR and four times as likely to experience CR as patients with a cribriform pattern, but low Ki67 (Table 3).Similar results were identified for ERG, where patients with a cribriform pattern and ERG expression were 1.8 times as likely to experience BCR and more than two times as likely to experience CR.For PTEN expression, patients with a cribriform pattern and PTEN loss were more than 1.5 times as likely to experience BCR, whereas a significant association was not identified for CR (Table 3).Clinical recurrence-free survival in patients treated with radical prostatectomy.Kaplan-Meier curves of clinical recurrence-free survival stratified (A) whether a cribriform pattern was present, defined as either non-cribriform or cribriform, (B) reactive stroma absent (À) or present (+), (C) PTEN status in tumour defined as either 'intact' or 'loss', (D) Ki67 status in tumour defined as either 'low' or 'high' and (E) ERG status in tumour defined as either negative (À) or positive (+).Kaplan-Meier curves were truncated when the number at risk within a particular group was fewer than five.F, The relative proportion of explained variation in clinical recurrence-free survival for CAPRA-S and each morphological and molecular feature included in the multivariable analysis.CAPRA-S, Cancer of the Prostate Risk Assessment Postsurgical score; CR, clinical recurrence.

Morphological and molecular biomarkers in PCa 861
As we identified that patients with a cribriform pattern and a high Ki67 had a particularly poor prognosis, we further investigated whether there were differences between this subgroup and the patient subgroup with a cribriform pattern and low Ki67 who were identified to have a better prognosis.The patient subgroup with a cribriform pattern and high Ki67 were more often classified in a higher CAPRA-S risk group, had ERG expression and developed mCRPC compared to patients with a cribriform pattern, but low Ki67 expression (Table 4).
We also assessed whether a high Ki67, PTEN loss and ERG expression was more often identified in TMA samples with a cribriform pattern compared to TMA samples with a non-cribriform Gleason pattern 4. A high Ki67 was identified in 10% (32 of 336) of cribriform pattern samples compared to 3% (21 of 723) of non-cribriform Gleason pattern 4 samples (P < 0.001).PTEN loss was identified in 57% (193 of 336) of cribriform pattern compared to 40% (287 of 723) of non-cribriform Gleason pattern 4 samples (P < 0.001).ERG expression was identified in 53% (179 of 226) of cribriform pattern compared to 43% (310 of 723) of non-cribriform Gleason pattern 4 samples (P = 0.002).

Discussion
The need for additional biomarkers in PCa risk stratification is evident, as a substantial number of patients will experience relapse despite treatment with RP. 31 Both morphological and molecular biomarkers have shown promise in improving PCa risk stratification. 1,3hrough a comparative analysis of the prognostic value of the biomarkers cribriform pattern, RS, PTEN, Ki67 and ERG, we demonstrate here that a cribriform pattern and Ki67 outperform RS, PTEN and ERG as prognostic biomarkers.Moreover, we show that for patients with a cribriform pattern, the concurrent presence of high Ki67 identifies a subgroup of PCa patients with a particularly poor prognosis, which we propose to name 'high proliferative cribriform prostate cancer'.
3][34] A cribriform pattern has been emphasised as an adverse feature in PCa, 35 and is now recommended to report in relation to PCa grading. 1 Furthermore, a cribriform pattern has been demonstrated to be a better prognostic biomarker than PTEN, 17 and also minor high-grade pattern 5 and percentage of Gleason pattern 4 in patients with grade group 2 PCa. 36,37However, to the best of our knowledge, a cribriform pattern has not been comprehensively compared to the other promising prognostic biomarkers: RS, Ki67 or ERG.In agreement with previous studies, 24,34,38 we show here that a cribriform pattern is an independent prognostic factor for predicting BCR and CR, but in contrast to these studies we show that a cribriform pattern is a robust prognostic biomarker even when considering RS, PTEN, Ki67 and ERG status.Interestingly, we found that a cribriform pattern was a particularly important factor  Morphological and molecular biomarkers in PCa 863 for predicting BCR (Figure 2F), and provided more information than the other assessed biomarkers.
Another study has also assessed a cribriform pattern in relation to CAPRA-S and demonstrated that a cribriform pattern has prognostic value in PCa. 39In the study by Jeyapala et al. 39 they demonstrated that a cribriform pattern has additional prognostic value and, together with CAPRA-S, could more clearly identify patients at risk of BCR.Beyond its value as a prognostic biomarker, we also identified that the presence of a cribriform pattern was associated with the finding of a post-RP persistent PSA and development of mCRPC, further supporting that a cribriform pattern is a marker of an aggressive phenotype.Taken together, these results support the recommendation to report the presence of a cribriform pattern in pathology reports.
Ki67 was also identified as a strong prognostic biomarker in PCa.Ki67 has consistently demonstrated prognostic value in PCa, 4 but is still not implemented in routine practice.Previous studies have used different thresholds for a high Ki67, ranging from 1% to 11%, with a mean threshold of about 6%. 4 In the current study, we applied a threshold of 5% and identified that fewer than 10% of patients had a high Ki67, which is slightly lower than previous studies using the same threshold. 5,28Discrepancies in the prevalence of a high Ki67 could be due to spatial heterogeneity, as we and others have demonstrated that the expression of Ki67 is highly heterogeneous both within and between different malignant foci. 5,40vertheless, we still identified that Ki67 was an independent predictor for BCR and CR, even when adjusting for a cribriform pattern, RS, PTEN and ERG status.
The prognostic value of Ki67 has been compared to both PTEN and ERG in a study by Leapman et al. 41 and to PTEN in a study by L eon et al. 42 Leapman et al. demonstrated that only PTEN was independently associated with metastasis, while none of the assessed biomarkers were associated with BCR. 41Leon et al. demonstrated that Ki67, and not PTEN, was associated with BCR, although not independently of other clinicopathological parameters. 42Differing results could be due to the above-mentioned studies including cohorts with a markedly lower proportion of high-grade and locally advanced PCas compared to our cohort, 41,42 interobserver interpretation of PTEN expression 43 or heterogeneity of PTEN, 44 Ki67 5,40 and ERG expression. 14Interestingly, although the limited follow-up time should be noted, we show here that besides CAPRA-S, Ki67 was identified as the most important factor in predicting CR (Figure 3F).We speculate that Ki67 is a strong determinant of clinical progression, and that these patients could be candidates for adjuvant treatment to improve their prognosis.Collectively, our results suggest that Ki67 could be of value beyond standard parameters included in routine pathology reports, although additional and larger validation studies are needed.The increasing use of digital pathology and artificial intelligence-based methods will hopefully, in the future, allow for simpler incorporation and more standardised reporting of Ki67 expression in daily practice.7][8] In our study, we identified that PTEN was strongly associated with a cribriform pattern and RS, which has also been demonstrated previously, [45][46][47] and could contribute to our finding of PTEN not being an independent prognostic factor, as we considered both morphological features in time-to-event analyses.Further, in agreement with previous reports, 44 we identified substantial heterogeneity in PTEN expression.This substantial heterogeneity could be an explanation for differences in its prognostic relevance, and also why we detected PTEN loss in a higher percentage of patients than in previous reports. 11Importantly, we found that PTEN loss was exclusively identified in a non-index malignant focus in a significant proportion of patients, which underscores the necessity of sampling all malignant foci for accurate assessment of PTEN expression.
RS has been highlighted as a marker of poor prognosis in both biopsy 48 and RP cohorts, 49 and its evaluation has been suggested as a novel grading approach in PCa. 2 However, in our study RS was not identified as an independent prognostic factor.
Differences could be that our evaluation differed from the proposed semi-quantitative method 50 and recently proposed quantitative binary scoring system. 51Although not an independent prognostic factor, we found that patients with RS were classified more frequently in a higher CAPRA-S risk group and also had a cribriform pattern, suggesting that it is associated with more aggressive PCa.The relationship between the tumour microenvironment and cribriform PCa is still largely unexplored. 52However, an association between IDC and RS has been demonstrated in needle biopsies, and it was hypothesised that RS facilitates the invasion of ducts and/or acini by malignant cells. 53Another study demonstrated that a distinct group of cancer-associated fibroblasts associated with poorer prognosis were enriched in cribriform PCa. 54Collectively, these results suggest that RS may contribute to the poorer prognosis of cribriform PCa.
The molecular mechanisms contributing to the unfavourable prognosis of a cribriform pattern remains to be fully elucidated.Moreover, although a cribriform pattern is associated with a poorer prognosis, 24,33,38 not all patients will experience relapse.Previous studies have also demonstrated that areas with a cribriform pattern have a higher proliferation rate compared to other growth patterns. 55,56e also demonstrated that samples with a cribriform pattern more often had a high Ki67 compared to non-cribriform Gleason pattern 4. Furthermore, we identified, that for patients with a cribriform pattern present, a concurrent high Ki67 identifies patients with an aggressive phenotype who are even more likely to experience disease progression (Table 3) which, to the best of our knowledge, has not previously been demonstrated.These results were not as evident for PTEN nor ERG, and we only found a weak association for PTEN.A study by Spieker et al. demonstrated that PTEN loss was not associated with increased risk of BCR for patients with IDC and/or ICC, and speculated that other molecular aberrations contribute to the poorer prognosis for these patients. 17We speculate that Ki67 is one of the molecular alterations that contribute to the poorer prognosis of patients with a cribriform pattern.The high relapse rate in patients with high Ki67 and cribriform pattern also transfers clinically with a significantly higher percentage of patients developing mCRPC compared to patients with cribriform pattern and low Ki67.We would welcome further studies which elaborate on the mechanisms underlying the poorer prognosis of the combination of a cribriform pattern and high Ki67.

Morphological and molecular biomarkers in PCa 865
In breast cancer, adjuvant treatment is recommended to selected patients based on morphological and molecular classification. 57The question arises as to whether a similar strategy is applicable in PCa, and specifically, whether patients with 'high proliferative cribriform prostate cancer' could benefit from adjuvant treatment after RP.
The current study has some limitations.One limitation is that cribriform pattern and RS were reviewed on whole-mount sections, whereas PTEN, ERG and Ki67 were reviewed on TMAs, which could lead to an under-representation of the true prevalence of the molecular biomarkers in our cohort, and specifically their expression in a cribriform pattern.To minimise the risk of bias as a result of using TMAs, we aimed to include multiple TMA samples from each patient; nevertheless, it is possible that some cases are classified incorrectly due to undersampling of an area with biomarker expression.Another consideration is the use of CAPRA-S score, rather than the individual clinicopathological variables.It is common practice to group patients based on individual variables into risk groups such as the European Association of Urology (EAU) risk group classification. 58CAPRA-S represents one type of risk classification tool, and although some information could be lost by combining information from different clinicopathological variables into a single score, it represents the combined impact of mostly categorical parameters.The CAPRA-S tool aims to offer a simpler and more straightforward method to calculate the risk of cancer recurrence after RP and to communicate results concerning the risk of recurrence to patients. 16,59In addition, the results presented in this study will need to be validated in other and larger cohorts.Also, the optimal tissue specimen and analysis method for assessing Ki67 in PCa still needs to be determined.One could foresee that assessment of Ki67 in prostate needle biopsies that contain areas with a cribriform pattern would lead to RP with extended pelvic lymph node dissection if a high Ki67 were demonstrated.In RP specimens with a cribriform pattern, a high Ki67 might be a factor that prompts the investigation of adjuvant therapies after RP, as these patients have a particularly high risk of PCa progression.Accordingly, the findings of a cribriform pattern together with a high Ki67 might be investigated as part of prognostic and treatment algorithms.
In conclusion, a cribriform pattern and Ki67 were identified as robust and important prognostic biomarkers.Our findings support the recommendation to include a cribriform pattern in pathology reports.Moreover, our findings suggest that Ki67 could aid beyond established parameters included in pathology reports, although additional validation studies and assessment of the optimal approach for Ki67 evaluation are needed.Furthermore, we identified that Ki67 identifies a patient subgroup with cribriform PCa with a particularly poor prognosis, which we propose to name 'high proliferative cribriform prostate cancer'.We speculate that this patient subgroup could be candidates for adjuvant treatment after RP to reduce the number of patients experiencing clinical progression.excluded leaving a total of 418 patients.PSA, prostate-specific antigen; RP, radical prostatectomy; TMA, tissue microarray.
Figure S2.Examples of 'normal' and 'reactive' stroma.A, Example of 'normal' prostatic stroma which is composed mainly of smooth muscle cells, fibroblasts and extracellular matrix.B, Example of 'reactive' prostatic stroma, where the stroma undergoes changes so that the normally smooth-muscle cell enriched-stroma is replaced with a myofibroblast/cancer-associated fibroblast-enriched stroma.Morphological changes as described by De Vivar et al. 1 Here, the glands have a more angled appearance and the stroma is more cellular and demonstrates reduced eosinophilia compared to normal stroma.Furthermore, the extracellular matrix is looser and fibrillary images at 20x magnification.Figure S3.Biochemical recurrence-free and clinical recurrence-free survival in patients treated with radical prostatectomy stratified by cribriform pattern and molecular biomarker status.Biochemical recurrencefree survival stratified by presence of a cribriform pattern and either (A) PTEN expression in tumour defined as either 'intact' or 'loss' or (B) ERG expression in tumour, where (+) indicates presence of ERG expression and (À) indicates absence of ERG expression.Clinical recurrence-free survival stratified by presence of a cribriform pattern and either (C) PTEN expression in tumour defined as either 'intact' or 'loss' or (D) ERG expression in tumour where (+) indicates 6 presence of ERG expression and (À) indicates absence of ERG expression.Kaplan-Meier curves were truncated when the number at risk within a particular group was fewer than five.Abbreviations: BCR, biochemical recurrence; CR, clinical recurrence.
Table S1.REMARK checklist.Table S2.Parameters included in the CAPRA-S score.
Table S3.Protein expression of PTEN, Ki67, and ERG in all evaluable samples.Table S4.Clinicopathological and molecular characteristics for all patients.
Table S5.Clinicopathological associations of the cohort stratified by presence or absence of a either a cribriform pattern or reactive stroma.
Table S6.Clinicopathological associations of the cohort stratified by presence or absence of PTEN loss, ERG expression, and a high Ki67.
Table S7.Univariable and multivariable Cox regression analyses assessing the prognostic value of a cribriform pattern and Grade Group.Morphological and molecular biomarkers in PCa 869 and graded as present or absent, irrespective of the amount of RS present.Example images of RS and normal stroma are shown Ó 2023 The Authors.Histopathology published by John Wiley & Sons Ltd., Histopathology, 83, 853-869.
C R I B R I F O R M P A T T E R N A N D K I 6 7 O U T P E R F O R M S R S , P T E N A N D E R G A S P R O G N O S T I C B I O M A R K E R S I N P R I M A R Y P C A In total, 21% (89 of 418) of the patients experienced BCR and 11% (47 of 418) experienced CR with a median time to BCR of 3.3 years (IQR = 1.9-6.0years) and CR of 5.2 years (IQR = 3.4-7.4years).

Figure 1 .
Figure 1.Prevalence of morphological and molecular biomarkers and associations between clinicopathological, morphological and molecular characteristics.A, Images of the morphological biomarkers showing the prevalence, and examples of, a cribriform pattern (which included invasive cribriform carcinoma and intraductal carcinoma) and reactive stroma.Hematoxylin & eosin staining, images of invasive cribriform carcinoma and intraductal carcinoma at 109 magnification and image of reactive stroma at 209 magnification.B, Images of the molecular biomarkers showing the prevalence, and examples of, PTEN loss, high Ki67 and ERG expression.C, Table showing significant associations between biomarkers, with CAPRA-S score, and selected clinical endpoints.Different shades of grey used to indicate the level of statistical significance, with darker shades indicating a higher level of statistical significance.D, Percentage of patients with intrafocal heterogeneity for PTEN and Ki67 expression.E, Percentage of patients with interfocal heterogeneity for PTEN and Ki67 expression.In patients where interfocal heterogeneity was identified, it was further assessed whether PTEN loss or a high Ki67 was identified in the index focus or exclusively in a non-index focus.CAPRA-S, Cancer of the Prostate Risk Assessment Postsurgical score; mCRPC, metastatic castration-resistant prostate cancer; PCa, prostate cancer; PSA, prostate-specific antigen.

Figure 2 .
Figure2. Biochemical recurrence-free survival in patients treated with radical prostatectomy.Kaplan-Meier curves of biochemical recurrence-free survival stratified by (A) whether a cribriform pattern was present, defined as either non-cribriform or cribriform, (B) reactive stroma absent (À) or present (+), (C) PTEN status in tumour defined as either 'intact' or 'loss', (D) Ki67 status in tumour defined as either 'low' or 'high' and (E) ERG status in tumour defined as either negative (À) or positive (+).Kaplan-Meier curves were truncated when the number at risk within a particular group was fewer than five.F, The relative proportion of explained variation in biochemical recurrencefree survival for CAPRA-S score and each morphological and molecular feature included in the multivariable analysis.BCR, biochemical recurrence; CAPRA-S, Cancer of the Prostate Risk Assessment Postsurgical score.

Figure 3 .
Figure3.Clinical recurrence-free survival in patients treated with radical prostatectomy.Kaplan-Meier curves of clinical recurrence-free survival stratified (A) whether a cribriform pattern was present, defined as either non-cribriform or cribriform, (B) reactive stroma absent (À) or present (+), (C) PTEN status in tumour defined as either 'intact' or 'loss', (D) Ki67 status in tumour defined as either 'low' or 'high' and (E) ERG status in tumour defined as either negative (À) or positive (+).Kaplan-Meier curves were truncated when the number at risk within a particular group was fewer than five.F, The relative proportion of explained variation in clinical recurrence-free survival for CAPRA-S and each morphological and molecular feature included in the multivariable analysis.CAPRA-S, Cancer of the Prostate Risk Assessment Postsurgical score; CR, clinical recurrence.

Figure 4 .
Figure 4. Biochemical recurrence-free and clinical recurrence-free survival in patients treated with radical prostatectomy stratified by cribriform pattern together with Ki67 status in tumour.A, Biochemical recurrence-free and (B) clinical recurrence-free survival, both stratified by presence of a cribriform pattern and Ki67 expression in tumour defined as either 'low' or 'high'.The Kaplan-Meier curves were truncated when the number at risk within a particular group was fewer than five.BCR, biochemical recurrence; CR, clinical recurrence.

Table 1 .
Clinicopathological characteristics of the cohort

Table 2 .
Biochemical recurrence (n = 418; 89 events) should be above both the column HR (95% CI) and P. Currently, it is only above HR (95% CI) †The proportional hazards assumption was met for all analyses except a cribriform pattern in univariable analysis assessing biochemical recurrence.*Statistical significance.Ó 2023 The Authors.Histopathology published by John Wiley & Sons Ltd., Histopathology, 83, 853-869.

Table 3 .
Univariable and multivariable Cox regression analyses for biochemical recurrence-free and clinical recurrence-free survival in patients treated with radical prostatectomy Ó 2023 The Authors.Histopathology published by John Wiley & Sons Ltd., Histopathology, 83, 853-869.

Table 3 .
(Continued)Cancer of the Prostate Risk Assessment Postsurgical score; CI, confidence interval; HR, hazard ratio.