Cytokeratin 5 and cytokeratin 20 inversely correlate with tumour grading in Ta non‐muscle‐invasive bladder cancer

Abstract Cytokeratin 5 is a marker of basal molecular subtypes of muscle‐invasive bladder cancer (MIBC), which correlates with worse overall survival compared to luminal subtypes. Our observations have not confirmed CK5 as a marker of high‐grade (HG) disease in Ta non‐muscle‐invasive bladder cancer (NMIBC). Therefore, to understand the basal‐luminal immunohistochemistry profile in Ta NMIBC, we performed immunohistochemistry for CK5, P40, P63 (basal), GATA3 and CK20 (luminal) and studied the correlation with HG and clinical outcome in 109 patients with Ta NMIBC. HG and low‐grade (LG) diseases were scored in each patient. Four different CK5 patterns were evaluated: absent (median 41.3%), normal (72.5%), rising (84.4%) and full thickness (23.9%). The median percentage of GATA3 was 100%. HG disease and CK5 expression and rising CK5 pattern had a significant inverse correlation, whereas HG disease and CK20 expression had a significant positive correlation. We also found a significant inverse correlation between CK5 expression and CK20 expression. Quantitative PCR confirmed that the presence of CK5 correlated with up‐regulation of CK5 RNA. None of the markers could differentiate patients with regard to clinical outcome. Our results suggest a role for CK5 and CK20 in differentiating between LG and HG disease in Ta NMIBC.

become BCG-refractory, which significantly impacts their prognosis. 4 Treatment of BCG-unresponsive NMIBC and MIBC is aggressive with either surgical resection of the bladder (cystectomy) or trimodal therapy (maximal transurethral resection, chemotherapy and radiotherapy).
Transcriptional analysis of MIBC has revealed several molecular subtypes, with luminal and basal as the most distinct. 5,6 Interestingly, patients with tumours of the basal subtype have worse overall survival (OS) compared to patients with the luminal subtype. In MIBC, CK5 has been suggested as a surrogate marker for the basal subtype, whereas CK20 and GATA3 have been linked to the luminal subtype. 5,7,8 Based on these findings, immunohistochemistry (IHC) using CK5 and GATA3 has been used to stratify patients in MIBC between the luminal and basal molecular subtypes. 9 Basal MIBC is recognized by its CK5+ and GATA3-profile, whereas luminal MIBC has a CK5-and GATA3+ profile. 9 Other commonly used basal markers are P63 and its isoform P40 (ΔNp63). P63 and P40 may act as either an oncogene or a tumour suppressor gene. 10,11 In MIBC, P40 correlates with high-grade (HG) disease and higher European Organisation for Research and Treatment of Cancer (EORTC) risk score. 12 Furthermore, patients with loss of P40 have a higher risk of NMIBC relapse and progression. 12 In contrast to the findings in MIBC, transcriptional analysis of NMIBC Ta and T1 disease identified a basal subtype with improved progression-free survival (PFS) compared to the luminal subtype. 13 Interestingly, this basal NMIBC subtype has increased RNA expression of CK5, whereas luminal NMIBC subtypes have higher RNA expression of CK20, a marker of apical umbrella cells. 13 The luminal marker GATA3 was expressed at even higher levels in the basal subtype in NMIBC than the other subtypes. 13 Interestingly, sub-classification of T1 NMIBC using the RNA expression of CK5 and CK20 can be used for the successful stratification of patients, demonstrating worse recurrence-free survival (RFS) and PFS in patients with high CK20 and low CK5 expression. 14 Similarly, our routine pathological observations did not confirm the potential of CK5 IHC as a marker of HG Ta NMIBC. Furthermore, we observed significant inter-tumour and intra-tumour heterogeneity of CK5 in Ta NMIBC. Given the contradictory findings reported in the RNA and IHC studies between NMIBC and MIBC, we studied the

IHC of the basal and luminal markers in a large group of patients with
Ta NMIBC for the first time. 15 To fully capture the heterogeneity of CK5 expression at the IHC level, we applied a novel approach using different CK5 expression patterns and correlated them with the percentages of HG and low-grade (LG) disease within one patient.
Lastly, we aimed to identify correlations between these CK5 IHC patterns and CK5 RNA expression, tumour grading (%), the other basal/luminal IHC markers and RFS.

| Patients
Patients with Ta NMIBC diagnosed after transurethral resection of the bladder tumour (TURBT) between May 2016 and February 2019 were included in the study cohort. Ethical approval was obtained from the local ethics committee (Ethics Committee AZ Klina no. OG 146; approval no. 086/200/019). Patients with invasive disease (≥T1), clinical node-positive disease or metastatic disease at the time of TURBT were excluded from the analysis. Histopathology was reassessed in all cases by an expert uro-pathologist (TG) blinded to clinical outcome. The following parameters were recorded: tumour stage, tumour grade, presence of carcinoma in situ (CIS) and the presence or absence of detrusor smooth muscle. As heterogeneity of tumour grading within one patient is possible, we scored the tumour grade traditionally as either LG or HG according to the WHO 2004 classification, but also scored the percentage of LG and HG disease within each patient. This scoring is a novel method of assessing HG disease burden that can be used to assess the heterogeneity of tumour grading within a patient. To promote reproducibility, grades were scored in increments of 10%.
A retrospective chart review was performed for the following clinico-pathological data: age, gender, number of lesions, lesion size, primary vs. recurrent NMIBC and prior intravesical treatment.
The following survival outcomes were collected: RFS, PFS, cancerspecific survival (CSS) and OS. All survival end-points were defined as the time from date of TURBT; for RFS, the end-point was the date of first recurrence of disease at cystoscopy, for PFS the date of first diagnosis of ≥T2 disease, for CSS the date of death due to BC, and for OS the date of death due to any cause.

| Immunohistochemistry
Depending on the extent of the TURBT, 1 or >1 formalin-fixed paraffin-embedded (FFPE) blocks were available per patient. In the case of the availability of >1 FFPE block, we selected a representative FFPE block after the presence of a relevant tumour was assessed on all haematoxylin-and-eosin-stained slides. Sections (4 μm) were cut and mounted on poly-L-lysine-coated glass slides. IHC was performed using the automated Benchmark Ultra IHC system (Roche Diagnostics). The automated procedure consisted of blocking endogenous peroxidase activity using 0.3% H 2 O 2 in methanol, with heat-induced epitope retrieval (HIER) in Tris-EDTA buffer (pH 7.8) at 95°C for 44 minutes (standard CC1). Slides were incubated with primary antibodies for 15 minutes, a peroxidase-labelled polymer for 30 minutes and then a substrate chromogen (mixed DAB Refine) for 10 minutes. Nuclear counterstaining was performed with haematoxylin.
Antibody clones were selected for their epitope selectivity (see Table 1 for details); all have been extensively validated for clinical diagnostic practice (www.nordi qc.com). Prior to enrolment in the study, antibodies were validated on control tissue (tissue with known endogenous expression) for staining specificity and reliability. The primary antibodies used were anti-CK5, anti-GATA3, anti-P40, anti-P63 and anti-CK20 (Table 1).
Cytokeratin 5 is a marker of undifferentiated basal cells in the urothelium and has been correlated with the basal molecular subtype in RNA expression studies in both MIBC and NMIBC. Based on the clinical observation by our uro-pathologists (TG and LL) that CK5 expression can be heterogeneous within each patient, we developed a scoring system for CK5 expression using different patterns to fully capture the heterogeneity of CK5 IHC within each patient. The CK5 expression pattern in IHC could be either 'absent' (ie no CK5 expression present), or 'normal', 'rising' or 'full-thickness' when CK5 expression was present. These four CK5 expression patterns are illustrated in Figure 1. The normal pattern is CK5 in the basal layer, as is expected from a basal marker, whereas the rising pattern comprises CK5 in the basal layer with increasing CK5 expression towards the luminal side. Comparable expression patterns have been described in both NMIBC and MIBC, which strengthens our methodological approach. 16 To capture heterogeneity, we scored the presence of each pattern as the percentage of each pattern per patient using a visual estimation ( Figure 2). In addition, we scored the presence of the other basal (P40, P63) and luminal (GATA3, CK20) markers per CK5 pattern present in each patient ( Figure S1). Finally, an additional assessment was made of the apical expression pattern of CK20; in normal urothelium, CK20 is present in umbrella cells only. Based on TA B L E 1 Properties of the antibody clones used apical negative, and apical attenuated), and the percentages of these patterns were scored per CK5 pattern ( Figure S2).
Our workflow of assessing heterogeneity in tumour grading, CK5 IHC patterns and other IHC markers is illustrated in Figure 2. IHC staining was evaluated and scored by a specialized uro-pathologist (TG).

| Real-time quantitative polymerase chain reaction
To

| Immunohistochemistry and CK5 expression patterns
All Ta NMIBC samples presented CK5 expression by IHC, but with variable intra-tumoural heterogeneity. We correlated the percentage of any CK5 IHC, regardless of the pattern, with the percentage of HG disease and identified a moderate inverse correlation (ρ) of −0.31 (P = .001). We also correlated each CK5 IHC pattern with the percentage of HG disease and identified an inverse correlation between HG disease and the percentage of rising pattern (ρ = −0.28; P = .003).
Using hierarchical clustering for the percentages of HG disease, CK5 patterns and CK20, we identified five distinct clusters ( Figure 3): CK5 absent/CK20+, CK5 normal, CK5 normal/CK20+, CK5 normal/rising and CK5 rising/full thickness. Interestingly, patients with a high percentage of HG disease mostly had an absence of CK5 or normal CK5 expression, whereas patients with a low percentage of HG disease had mostly rising and full-thickness CK5 expression ( Figure 3). The full-thickness CK5 expression pattern almost did not co-occur with the other CK5 expression patterns, as is visible in the first cluster in Figure 3. More detailed data regarding the Spearman correlation between the percentage of HG and CK5 patterns are given in Table S1.
Furthermore, we found a positive correlation between the percentage of CK20 and percentage of HG disease (ρ = 0.21; P = .04).
The findings regarding CK5, CK20 and tumour grading are further illustrated in Figure S3.
We also assessed the differences in IHC and CK5 expression patterns between patients with primary (n = 59) and recurrent disease (n = 48; Table S2). Any CK5 expression differed significantly (P = .03), as well as CK5 full-thickness expression (P = .02). We performed a similar analysis based on EAU risk categories for patients with low-risk (n = 8), intermediate-risk (n = 10) or high-risk disease (n = 90; Table S3). We found no significant differences between EAU risk categories in IHC and CK5 expression patterns. The CK5 fullthickness pattern was present in 8% of cases (n = 4) with recurrent disease. The other IHC markers (GATA3, P40, P63 and CK20) and CK5 patterns did not differ significantly between patients with primary and recurrent disease.
We examined the correlation between the relative CK5 RNA expression levels and CK5 IHC patterns and found positive correlations between CK5 RNA expression and the presence of any CK5 in IHC (ρ = 0.79; P = .0008), and between CK5 RNA expression and the presence of the rising CK5 pattern (ρ = 0.74; P = .002; Figure S4).

| Survival analysis
Recurrence-free survival at 1 and 2 years ( Figure 4A)

| D ISCUSS I ON
Here, we describe the expression patterns of luminal (GATA3, CK20) and basal (CK5, P40, P63) IHC markers in a group of patients Third, we observed a significant positive correlation between the percentage of HG disease and expression of the luminal IHC marker CK20. We also observed that the luminal marker GATA3 was omnipresent in Ta disease and did not correlate with any specific tumour feature. Fourth, CK5 inversely correlated with CK20.
Finally, we found no significant impact of any of the IHC variables on RFS.
In recent years, research on the molecule profile of BC has expanded quickly, with various genomic studies being published.
Most of these studies have focused on MIBC, and comprehensive classification schemes have been proposed by different groups. [6][7][8][17][18][19][20][21][22] Recently, a consensus paper proposed a set of six molecular classes, with the basal/squamous (Ba/Sq) group as the most prominent subgroup (35%). 6 Although less molecular data are available for NMIBC, three molecular subgroups have been defined by several groups. Hedegaard et al 13 reported three major tumour classes with basal-and luminal-like characteristics and different clinical outcomes in Ta and T1 NMIBC. 22 To the best of our knowledge, profiling studies with a unique focus on Ta NMIBC have not been published previously.
Several groups have tried to characterize these molecular subgroups by IHC, and GATA3 and CK5 have emerged as the most frequently used luminal and basal marker, respectively. 9,23 Our data suggest different biology underlying CK5/CK20 expression in Ta NMIBC and MIBC. 6 Loss of CK5 and gain of CK20 at the protein level in Ta NMIBC can be hypothesized to mark an evolution to a more aggressive Ta tumour type, whereas the opposite has been reported in MIBC. 6,7 In MIBC, CK5 expression is a marker of basal differentiation and CK20 of luminal differentiation, which is not the case in Ta NMIBC. The diffuse expression of the basal cell markers P40 and P63 in Ta NMIBC, regardless of the CK5 pattern, supports this thesis. As mentioned above, the role of CK5 Note: Percentages of CK20 patterns (umbrella cells positive, umbrella cells negative and accentuated) within the present CK20 as is illustrated in Figure S2.  NR). B, Kaplan-Meier plot of RFS of all patients with follow-up data stratified by administration of a single post-operative instillation of intravesical chemotherapy (n = 107). RFS in the not administered group at 1 year and 2 years was 55% (95%CI: 39-77) and 50% (95%CI: 35-73); and in the administered group 90% (95%CI: 83-97) and 60% (95%CI: 47-77), respectively. Median RFS was not reached in both groups findings suggest different applicability and relevance of the GATA3/ CK5/CK20 panel for the classification and grading of NMIBC compared to MIBC. The added value of these markers to clinicopathological grading has to be proven in larger prospectively designed trials with longer term data on clinical outcome. If validated, CK5 and CK20 may serve as additional markers for differentiating between HG and LG Ta NMIBC disease.

ACK N OWLED G EM ENTS
Steven Joniau is a senior clinical researcher at the Research Foundation of Flanders (FWO).

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflicts of interest related to this manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data available upon request due to privacy/ethical restrictions.