What's known on the subject? and What does the study add?
Only little and partly contradictory data are currently published about the prognostic role of immunohistochemically detectable proliferation-associated biomarkers in surgically treated squamous cell carcinoma of the penis (SCCP), and no data are available at present about their usefulness for refining the delineation between different Broders' grading categories (e.g. still G2 or just G3 SCCP?). Moreover, the accuracy of various conventional histopathological parameters for predicting cancer-specific survival (CSS) in surgically treated SCCP has not been systematically evaluated yet.
Based on the so far largest study cohort encompassing 158 consecutive patients with surgically treated PSCCs characterised by means of a central histopathological review, our data add the following to the currently available literature: (i) Ki-67, mini-chromosome maintenance 2 protein (MCM2), and geminin indicate a more aggressive behaviour in SCPP but do not represent independent prognostic parameters in the multivariable analysis in terms of CSS, (ii) these three biomarkers are not helpful for refining the delineation between different Broders' grading categories at the immunohistochemical level, and (iii) the conventional histopathological parameters staging, grading, nodal involvement, and lymphovascular invasion are independent prognostic parameters that together achieve a predictive accuracy of 82% for CSS.
To assess the role of cell proliferation-associated biomarkers to predict cancer-specific survival (CSS) in patients with surgically treated squamous cell carcinoma of the penis (SCCP).
Patients and Methods
A multicentre study enrolling 158 consecutive patients with surgically treated SCCP was performed.
After conducting a central histopathological review, the staining profiles of Ki-67, mini-chromosome maintenance 2 protein (MCM2) and geminin were evaluated for their correlation with conventional histopathological criteria and their prognostic relevance for predicting CSS in a multivariable Cox proportional hazards regression model (median [interquartile range] follow-up 33 [6–63] months).
Staining evaluation showed high interobserver agreement (92–96%). Ki-67 and MCM2 displayed a significant positive correlation with histological tumour grade, lymphovascular invasion (LVI) and nodal status, whereas geminin expression only correlated with tumour grade.
The 5-year CSS for the entire study cohort was 62%.
Univariable analysis showed a significant prognostic impact of Ki-67 (P = 0.026), MCM2 (P = 0.007), and geminin (P = 0.036).
In multivariable analysis, only pT (hazard ratio [HR] 1.67; P = 0.003) and pN stage (HR 2.62; P = 0.015) as well as tumour grade (HR 1.89; P = 0.036) and LVI (HR 2.66; P = 0.028) were identified as independent prognostic parameters for CSS.
The accuracy of the Cox model for CSS prediction was 0.820 (95% confidence interval 0.741–0.898).
At present, conventional histopathological criteria remain the most powerful predictors of CSS in surgically treated SCCP. Due to overlapping staining profiles, Ki-67, MCM2 and geminin, either singly or in various combinations, failed to immunohistochemically refine the boundaries between Broders' grading categories.
Ki-67, MCM2 and geminin do not represent independent prognostic parameters but reflect a more aggressive behaviour in surgically treated SCCP.
Further studies are needed to clarify the currently contradictory predictive role of proliferation-associated biomarkers in terms of predicting nodal involvement in SCCPs.
Squamous cell carcinoma of the penis (SCCP) is a rare tumour entity in Caucasian populations, with an estimated incidence of less than 1 in 100 000 in Europe . Ablative surgery aiming at negative surgical margins currently represents the mainstay of therapy, although organ-sparing procedures might also be indicated . In addition to histopathological stage and grade, outcome is essentially determined by tumour spread to locoregional lymph nodes (LNs) [1-3]. Although conventional histopathological criteria, e.g. depth of invasion, lymphovascular invasion (LVI) and koilocytosis, give prognostic information, the identification of biomarkers predictive of LN spread would substantially facilitate individual risk stratification [4-6]. Moreover, based on the assumption that dysregulation of replication licensing factors might be implicated in tumour progression, Kayes et al.  recently suggested that the identification of an individual ‘proliferative signature’ reflected by the immunohistochemical staining profile(s) of selected biomarkers involved in cell proliferation might help to refine prognostic evaluation.
In the present multicentre study, we looked at the prognostic role of the standard proliferation-associated marker Ki-67 as well as two other DNA replication markers, mini-chromosome maintenance 2 protein (MCM2) and geminin, in terms of cancer-specific survival (CSS) in the so far largest reported study cohort encompassing 158 consecutive patients with surgically treated SCCP. Moreover, we evaluated the ability of these biomarkers in terms of delineating the different Broders' grading categories at the immunohistochemical level.
Patients and Methods
Our study group consisted of 158 consecutive patients surgically treated for histologically confirmed primary invasive SCCP. All patients were free of distant metastases (cM0) at the time of surgery, and underwent radical (39 patients) or partial (119; amongst these, 23 excisional biopsies) penectomy, which was performed in eight different German hospitals between 1992 and 2010. None of the study patients received (neo-)adjuvant therapy. The patients' histories including pertinent diagnostic and treatment data were retrospectively obtained from the retrieved archived records. After our study was approved by the Medical Ethics Committee of Brandenburg State Medical Association (MEC No.: 37588/11), CSS was recorded for each study patient.
Briefly, patients with histologically confirmed poorly differentiated SCCP with invasion of the subepithelial connective tissue or beyond (≥pT1b stage according to seventh edition of the TNM classification system) and/or with inguinal LNs clinically palpable beyond the sixth week after surgery, were advised to undergo bilateral inguinal LN dissection (LND), which was subsequently followed by ipsilateral pelvic LND in case of histologically involved inguinal LNs. Despite strictly adhering to these criteria, only 49 (31%) patients underwent an inguinal LND. Patients with histologically involved LNs and/or corpora invasion (≥pT2 stage) had contrast-enhanced CT of the pelvis and abdomen, as well as a chest radiograph. Cranial CT and bone scintigraphy were only performed in clinically symptomatic patients. Surveillance after surgery followed similar guidelines . The median (interquartile range, IQR) follow-up period for the entire study cohort was 33 (6–63) months.
Central Histopathological Review
Data on maximum tumour size (in cm), anatomical tumour localisation, and pN stage (histologically re-classified according to the seventh edition of the TNM classification system, and dichotomized as pN0/pNx vs pN+) were collected from the original pathology reports. The haematoxylin and eosin (H&E)-stained histology slides and the archived paraffin wax-embedded tissue blocks were retrieved from the files of the eight participating hospitals, and were centrally reviewed by one clinical pathologist (S.G.) who was ‘blinded’ with respect to patient identity, as well as to the original pathology reports. During this central review, all slides were histologically re-examined, and all study cases were re-staged according to the seventh edition of the TNM classification system, and re-graded according to the Broders' grading system [8, 9]. Moreover, during the central histopathological review, the LN status was dichotomized as pN0/pNx vs pN+, and histological growth patterns, surgical margin status, perineural invasion and LVI were carefully examined and recorded . LVI was defined as the unequivocal conventional histological presence of tumour cells within endothelium-lined spaces. The depth of vertical tumour infiltration was dichotomized (≤5 vs >5 mm). After the central histopathological review had been completed, H&E stained-sections representative of histopathological stage and grade were forwarded to an independent second clinical pathologist (D.M.) who re-evaluated histological stage and grade, while being ‘blinded’ to the original pathology reports, as well as to clinical data and to the results previously documented during the central histopathological review.
Not including the differences with respect to pT1a/b stage definition (attributable to the integration of tumour grade and LVI), three study cases (1.9%) were assigned different pT categories by both study pathologists (S.G. and D.M.). In all, 36 cases (22.8%) were allocated different grading categories (no more than one grading point difference). All discrepant study cases were subsequently histologically re-examined and discussed by both study pathologists using a multi-head microscope, and consensus decisions in terms of histological staging and grading were found.
Representative serial sections (4-μm thick) were cut from the retrieved paraffin wax-embedded tissue blocks. Immunohistochemical staining was performed with the Ventana BenchMark® system (Ventana, Tucson, USA; Roche Diagnostics GmbH, Mannheim, Germany). Primary antibodies directed against Ki67 (monoclonal; clone MIB-1, M7240; DAKO Cytomation, Glostrup, Denmark; dilution 1:50), MCM2 (polyclonal; AB4461, Abcam, Cambridge, UK, dilution 1:100) and geminin (polyclonal; AB12147, Abcam, Cambridge, UK, dilution 1:200) were used. After incubation for 30 min, the staining kit UltraView Universal DAB Detection Kit (Ventana) was used, and counterstaining was performed using haematoxylin and bluing reagent (Ventana). Adequate positive and negative controls were run.
By following these standards, all study cases showed distinct nuclear biomarker expression (Fig. 1A–C).
After a few weeks had elapsed since the performance of the central histopathological review, the immunostained slides were evaluated in succession by both independent study pathologists who were ‘blinded’ to the grading categories previously allocated by themselves and by the other pathologist as well as to patients identity and clinical outcome. At least 500 tumour cell nuclei were evaluated in each study case. Briefly, expression of the three biomarkers was scored according to the following approach recently described: score 1+, marker expression predominantly in basal cell nuclei with no more than occasional extension to suprabasal cell nuclei (Fig. 1A); score 2+, nuclear expression in basal and many suprabasal cell nuclei, but no full-thickness marker expression (Fig. 1B); score 3+, nuclear marker expression throughout the entire neoplastic epithelium corresponding to full-thickness marker expression (Fig. 1C) [4, 5]. This scoring system was used as evaluation of the staining architecture throughout the neoplastic epithelium based upon these defining criteria was felt to be more reproducible compared with estimating the percentage of stained nuclei. However, to show that this selected approach correlates well with the frequently used method of estimating the percentage of stained nuclei, a subset of 45 randomly selected study cases were re-examined by estimating the percentage of stained nuclei a few weeks later by both independent raters ‘blinded’ to the staining categories previously allocated to these study cases by themselves and by the other pathologist. Study cases assigned discrepant scores by both independent raters were subsequently discussed on a multi-head microscope, and consensus decisions in terms of biomarker scoring were recorded.
CSS rates, defined as the period of time between primary surgery and cancer-related death according to death certificates or the end of follow-up after surgery (at which point data were censored), were defined as to represent the study endpoint in our investigation.
Correlations between the different immunostaining patterns and other study criteria were evaluated using Spearman's rank coefficients (based on Spearman's ρ/R). Univariable survival analyses were carried out using the Kaplan–Meier method (with differences quantified using the log-rank test), and with a Cox-proportional hazards regression analysis. Multivariable analyses were calculated using a Cox regression analysis model (reverse elimination according to likelihood ratio, step by step exclusion of the least significant factors with P > 0.10). The predictive accuracy of the final multivariable model was calculated according to the Frank Harrell method . Briefly, with this method, ‘1’ reflects a perfect discriminatory model, and ‘0.5’ is equivalent to coin flipping. The level of statistical significance was set at P < 0.05, and all statistical tests were two-sided.
Pertinent clinical and histopathological characteristics of the study cohort including biomarker expression are summarised in Table 1.
Table 1. Pertinent clinical and pathological characteristics of the study patients (n = 158) with surgically treated penile squamous cell carcinoma.
Inter-observer concordance for staining evaluation of nuclear Ki-67, MCM2 and geminin expression was found in 146 (92.4%), 152 (96.2%) and 150 (94.7%) study cases, respectively. Hence, 12, six and eight consensus decisions for scoring evaluation were reached by both raters during their discussion of discrepant study cases.
Both staining scores (staining architecture throughout the epithelium vs estimated percentage of stained nuclei) allocated to a subset of 45 randomly selected study cases by two independent raters showed a significant positive correlation (rater A: R = 0.85, rater B: R = 0.90; P < 0.01), indicating that both staining scores are equivalent.
Ki-67 and MCM2 highly correlated with each other (R = 0.65; P < 0.001). Geminin also correlated positively with Ki-67 and MCM2 (R = 0.32; P < 0.001 and R = 0.29; P < 0.001). A statistical comparison of the three biomarkers with the recorded conventional histopathological criteria showed a positive correlation between Ki-67 expression and histological tumour grade (R = 0.44; P < 0.001, Fig. 1A–C), LVI (R = 0.27; P < 0.001) as well as with dichotomized LN status (pN0 vs pN+, R = 0.20; P = 0.011). MCM2 also correlated positively with tumour grade (R = 0.45; P < 0.001), LVI (R = 0.27; P = 0.001) and with the dichotomized LN status (R = 0.18; P = 0.026). By contrast, nuclear geminin expression only showed a positive correlation with tumour grade (R = 0.30; P < 0.001). However, due to partly overlapping staining profiles between different grading categories, the three biomarkers failed to clearly delineate the different Broders'grading categories at the immunohistochemical level.
The distribution of Ki-67 scores (three categories with scores from 1+ to 3+, all averages given with their corresponding 95% CIs in brackets) for G1-G2-G3 tumours was 1.90 (1.75–2.05), 2.30 (2.12–2.48) and 2.67 (2.48–2.86). The distribution of MCM2 scores for G1-G2-G3 tumours was 1.87 (1.73–2.02), 2.28 (2.10–2.45) and 2.65 (2.44–2.86). The distribution of geminin scores for G1-G2-G3 tumours was 2.04 (1.90–2.18), 2.29 (2.05–2.53) and 2.59 (2.39–2.79) in our study.
CSS for the entire study cohort was 88% after 12 months, 70% after 36 months and 62% after 60 months. Univariable analysis of CSS showed a significant prognostic impact of Ki-67 and MCM2, whereas geminin expression failed to convey any significant impact on CSS (Table 2, P = 0.047, P = 0.011 and P = 0.054, respectively). In the multivariable analysis, only pT-stage (hazard ratio [HR] 1.67; P = 0.003), dichotomized LN status (HR 2.62; P = 0.015), tumour grade (HR 1.89; P = 0.036) and LVI (HR 2.66; P = 0.028, Table 2) were identified as independent prognostic parameters, and together achieved a predictive accuracy of 82% for predicting CSS. In contrast, the remaining conventional histopathological criteria recorded during the central histopathological review, as well as the expression patterns of Ki-67, MCM2 and geminin, failed to have any significant independent prognostic impact for CSS rates in our study cohort. The c-index of the Cox model for cancer-specific mortality was 0.820 (95% CI 0.741–0.898; P < 0.001).
Table 2. Univariable and multivariable analysis of carcinoma-specific mortality in the study patients (n = 158) with surgically treated SCCP.
Tumour localization on penile shaft beyond the sulcus (ref.glans/foreskin before sulcus)
1.48 (0.58–3.79), 0.416
Partial penectomy (ref. total penectomy)
0.37 (0.22–0.64), <0.001
Multiple tumours (ref. solitary tumour)
1.58 (0.38–6.57), 0.529
Tumour size (cont., per cm)
1.25 (1.05–1.50), 0.014
Tumour invasion >5 mm (ref. ≤5 mm)
4.37 (1.55–12.31), 0.005
pT-stage (5 categories)
2.13 (1.56–2.91), <0.001
1.67 (1.20–2.41), 0.003
Positive LNs (ref. pN0/pNx)
6.27 (3.27–11.99), <0.001
2.62 (1.21–5.67), 0.015
Lymphovascular invasion (ref. negative)
7.85 (3.87–15.90), <0.001
2.66 (1.11–6.37), 0.028
Perineural invasion (ref. negative)
3.15 (1.49–6.67), 0.003
Positive surgical margins (ref. negative)
1.67 (0.40–7.01), 0.482
Tumour grade (3 categories)
3.41 (1.95–5.99), <0.001
1.89 (1.04–3.43), 0.036
Ki-67 (3 categories)
1.47 (1.01–1.82), 0.047
MCM2 (3 categories)
1.67 (1.13–2.47), 0.011
Geminin (3 categories)
1.46 (0.99–2.15), 0.054
The present study suggests that the proliferation-related markers Ki-67, MCM2 and geminin can be reproducibly assessed by the pathologist, and appear to have an association with a more aggressive behaviour of SCCP. However, due to the positive correlations of the investigated biomarkers with histological tumour grade, it cannot be deduced from the present data whether or not the more aggressive behaviour of SCCP exhibiting increased proliferative activity is attributable to increased proliferation by itself or whether the latter just represents a reflection of other prognostically adverse parameters, e.g. poor histological differentiation and, with respect to Ki-67 and MCM2, of the associated increased frequency of LVI and LN involvement.
On the other hand, none of the three biomarkers had any independent predictive impact for CSS rates according to the present results. These findings are consistent with the advanced study recently conducted by Stankiewicz et al.  who also suggested that Ki-67 has no prognostic value for CSS and overall survival in patients with SCCP. Moreover, according to the present data, different staining profiles of the markers assessed, either singly or in various combinations (‘proliferative signature’), failed to clearly delineate different Broders' grading categories at the immunohistochemical level, due to partly overlapping staining profiles between different grading categories. Therefore, these markers seem to offer little help to the clinical pathologist in the grading of conventionally challenging SCCP (e.g. still G2 or just G3?), and obviously, do not represent useful immunohistochemical adjuncts to overcome the well-known inter-observer variability for histological grading of SCCP (23% in the present study) . These findings, at first glance, seem to call the prognostic and diagnostic usefulness of the selected biomarkers into question for surgically treated SCCPs. However, at present, data on the prognostic role of Ki-67 expression in surgically treated SCCP are limited and, at least in part, contradictory, especially in terms of predicting LN involvement [7, 13-17]. As inguinal LND is a procedure with considerable morbidity, markers determined during the evaluation of penectomy specimens, which would allow prediction of LN spread more accurately, would be helpful for clinical management . At present, this task might be accomplished by histopathologically assessing grade and LVI . However, conflicting data currently exist about the predictive role of immunohistochemistry in this regard. In particular, some studies reported locoregional penile cancer LN metastases to be linked to increased proliferation rates, whereas another study reported an inconsistent predictive role of Ki-67 expression with a labelling index of >10% encountered in 64% of patients with uninvolved vs 39% of patients with involved LNs (P = 0.009) [7, 13, 14]. These conflicting reports underline the uncertainty that currently surrounds the prognostic role of proliferation-associated markers for predicting LN involvement in patients with SCCP.
Until now, to our knowledge, the study conducted by Kayes et al.  is the only one to have addressed the predictive role of MCM2 and geminin in surgically treated SCCP, although the study endpoint defined in that investigation differed from ours (overall survival vs CSS). Kayes et al. found a significant prognostic impact of Ki-67 and MCM2 on overall survival in their univariable analysis (with geminin achieving P = 0.215). However, in that study enrolling 84 patients, all three biomarkers failed to show any significant independent prognostic value in the multivariable model. Moreover, in addition to their positive correlations with pN stage, all three biomarkers assessed displayed differing expression patterns for different tumour grades , a finding which is consistent with the present data. Taken together, the data reported by Kayes et al.  as well as the present findings suggest that the three biomarkers evaluated do not represent appropriate tools for refining the predictive power of currently established survival nomograms in SCCP.
In addition to its retrospective nature, there are three other important limitations to the present study that merit discussion. Firstly, only 31% of the present study patients underwent LND, which clearly precludes a reliable investigation of the predictive impact of the evaluated biomarkers in terms of LN involvement as well as the evaluation of the prognostic impact conveyed by LN spread for CSS, although our preliminary data suggest that pN0 and pNx study patients fail to differ significantly with respect to CSS (Fig. S1). In short, the regional LN status was dichotomised for statistical calculations because, according to the present data, dichotomisation (pN+ vs pN0/pNx) achieved improved accuracy of the Cox models used for predicting CSS compared with non-dichotomised inclusion of the LN status into the Cox models. Notably, only pN+ but not pNx status was linked to significantly shorter CSS rates compared with histologically confirmed LN-negative (pN0) study patients, whereas pN0 and pNx study patients failed to differ significantly in their CSS rates according to the present data (pN+ vs pN0: HR 8.82, 95% CI 2.57–30.31, P < 0.001; pNx vs pN0: HR 1.51, 95% CI 0.44–5.22, P = 0.509). However, this limitation also indicates that the frequencies of LNDs performed in the landscape of contemporary clinical and oncological practice appear to be clearly below the frequency that might be expected according to current guideline-based recommendations, a notion which may have contributed to poorer clinical outcome in a subset of the present study patients . This discrepancy might be attributable to the significant co-morbidity associated with LND, which might cause some degree of reluctance amongst some clinicians for using the procedure [20, 21]. Secondly, possible confounding factors that may have been brought about by evaluating a cohort of study patients collected from many centres over 18 years cannot be excluded. Thirdly, the non-inclusion of co-morbidity factors into the multivariable Cox model used represents another limitation, as co-morbidity also could have an effect on CSS.
In conclusion, conventional histopathological criteria currently remain the best predictors of outcome after surgery for SCCP. In particular, histologically confirmed staging, LN involvement, LVI, and grading together achieve a predictive accuracy of 82% for predicting CSS according to the present data. Staining evaluation of the proliferation-associated biomarkers Ki-67, MCM2 and geminin may be reproducibly accomplished by the pathologist by considering the staining architecture throughout the epithelium, which appears to correlate well with the frequently used method of estimating the percentage of stained nuclei. Although increased nuclear expression of these three markers seems to indicate a more aggressive behaviour of SCCP, none of them achieved an independent predictive contribution in terms of CSS. However, at present, our understanding of their predictive role in SCCP still seems to be preliminary and, at least in part, even contradictory. Therefore, further advanced clinical studies are clearly needed to clarify the presently questionable association of proliferation-associated biomarkers with histologically confirmed LN involvement in surgically treated SCCP.
In alphabetical order, we thank the following Department Directors for supporting the present study by providing archival wax-embedded tissues for our investigations: V. Henn, MD (Department of Pathology, Ruppiner Clinics GmbH, Germany), J. Jander, MD (Department of Pathology, Rhön Clinic Frankfurt/Oder, Germany), H. Lobeck, MD (Department of Pathology, Ernst-von-Bergmann Clinic Potsdam, Germany), R. Pauli, MD (Department of Pathology, Health Care Centre Brandenburg an der Havel GmbH, Germany), and M. Tuffaha (Carl-Thiem Clinic Cottbus, Germany).