Association of Cytokeratin 5 and Claudin 3 expression with BRCA1 and BRCA2 germline mutations in women with early breast cancer

Background It is important to identify biomarkers associated with BRCA mutation in women with early breast cancer (BC) to improve early identification of mutation carriers. Thus, in this study, we examined the protein expression of claudin (CLDN) 3, CLDN4, CLDN7, and E-cadherin. Moreover, we analyzed additional histopathological variables and their associations in familial BC. Methods Immunohistochemical analysis for CLDNs and E-cadherin was performed on 237 BC cases of three different subsets of BC tumors: 62 from BRCA1 mutation carriers, 59 from BRCA2 mutation carriers, and 116 tumors from patients with BRCA wild type (WT) as controls. Histopathological data were also analyzed in the different subgroups. Logistic regression and receiver operation characteristic (ROC) curve were conducted to investigate factors associated with BRCA tumors. Results Expression of CLDN3 positively correlated with BRCA-mutated BC. CLDN3 was expressed in 58% of BRCA1-mutated tumors compared to only 7% in BRCA2-mutated tumors (p < 0.001) and 1% in WT tumors (p < 0.001). CK5 and CK14 expression were also more likely to arise in BRCA1 tumors (44 and 16%, respectively) than in the control group (8 and 4%) (p < 0.001, p = 0.012, respectively). We also found a significantly higher proportion of CK5+ among BRCA1 tumors (44%) in comparison with BRCA2-related BC (8%) (p < 0.001). In addition, there was a significant difference between both groups regarding CK14: positive expression in 16 and 5%, respectively (p = 0.030). CK5 and CK14 did not differ between the BRCA2 group and the WT tumors significantly. In a multivariate regression model, expression of CK5 (Odds ratio (OR): 6.46; 95% confidence interval (CI): 1.52–27.43; p = 0.011), and CLDN3 (OR: 200.48; 95% CI: 21.52–1867.61; p < 0.001) were associated with BRCA1 mutation status. Conclusions Our data suggests that CLDN3, CK5, and CK14 in combination with ER, PR and HER2 are associated with BRCA1 mutation status. Electronic supplementary material The online version of this article (10.1186/s12885-019-5908-6) contains supplementary material, which is available to authorized users.


Background
Breast cancer (BC) is the leading cancer type among women in the world [1]. Familial BC, representing 5-7% of all BC, are hereditary and are associated with inherited gene mutations [2]. Approximately 25% of familial BC are due to germline mutations in the BRCA1 and BRCA2 genes, which are located on chromosome 17 and 13, respectively [2][3][4]. The average cumulative BC risk in BRCA1 mutation carriers by age 70 is 57-65%, whereas the cumulative BC risk in patients with BRCA2 mutation is 45-49% [5,6].
Claudin-low is another subtype of triple-negative BC, and can be characterized by low expression of claudin (CLDN) 3, CLDN4, CLDN7, and E-cadherin. The majority of claudin-low tumors have a poor prognosis [29][30][31]. CLDNs are structural and functional components of tight junctions which provide cell-cell adhesion in epithelial to endothelial cells [32]. There are at least 24 different CLDNs existing in humans, the expression of each seems to be tissue specific [33]. E-cadherin is one of the most important molecules in cell-cell adhesion in epithelial tissues [34]. Loss of intercellular adhesion by E-cadherin correlates with increased invasiveness and metastasis of tumors [34][35][36][37][38][39].
In this study, we analyzed clinicohistopathological features which are already associated with BRCA1/2 tumors (ER, PR, HER2, CK5 and 14, EGFR, among others). In addition, we selected three important CLDNs in BC, ie. CLDN3/4/7, and E-cadherin, which are used for characterization of the claudin-low subtype. We aim to define the expression profiles of these biomarkers in BRCA1 BC and compare these with BRCA2 and BRCA WT patients to improve early identification of mutation carriers. We

Study population
A total of 242 BC tissue microarrays (TMA) were obtained from the Kathleen Cuningham Foundation Consortium for research into Familial Breast cancer (kConFab) [http://www.kconfab.org]. We evaluated one case per patient, and one core per case. The core diameter was 0.6 mm. Three BC cases were excluded due to TP53 (n = 1) and PALP2 (n = 2) mutation status. We also eliminated two cases of ductal carcinoma in situ. Our analysis was therefore based on 237 BC tumors where 62 tumors originated from BRCA1 carriers, 59 tumors were from BRCA2 carriers, and we obtained the remaining 116 from BRCA WT patients. The BRCA WT subgroup served as controls in our study. The control group consists of tumors from non BRCA1/2 mutation carriers. We used these tumors from consecutive BC patients with familial history. BRCA testing and analysis are described in the Supplemental (Additional file 1). Clinicopathological information collected included age at diagnosis, tumor size, tumor morphology, tumor grade, ER, PR, HER2, CK5 and 14, and EGFR.
Immunohistochemical staining of the samples was performed as described in our previous study [45].  [46]. HER2-positivity was defined by staining of > 10% of tumor cells as proposed by the update of the American Society of Clinical Oncology/College of American Pathologists (ASCO/ CAP) clinical practice guideline. Staining of HER2 should also be strong and circumferentially membranous [47]. CK5, CK14, and EGFR were regarded as positive if any cytoplasmic and/or membranous staining was seen in the tumor cells [33]. Immunohistochemical analysis for CLDN3/4/7 and Ecadherin.
We used hydrogen peroxide and 3,3'-diaminobenzidine-tetrahydrochloride for visualization of the reaction. The slides were counterstained with haemalaun and exposed to a bluing reagent for different times. We used colon (CLDN3, E-cadherin) and breast tissue (CLDN7) as positive controls. CLDN4 was overexpressed in ovarian cancer.
Immunohistochemical staining was centrally reviewed by an experienced pathologist (MR) to ensure comparable results. The stained sections were visualized on an Olympus BX50 microscope (Olympus Corporation, Tokyo, Japan). The imaging software cell^P (Olympus Corporation) was used for taking pictures of the slides. The tumor samples were evaluated by differentiation between positive and negative staining. Only membranous staining was classified as positive. There is no standard for assessing the CLDN expression [48]. Positive expression of CLDN3, CLDN4, and CLDN7 was defined by any detectable staining in the membrane of the tumor cell. E-cadherin was regarded positive if any staining was observed. Thus, complete absence of any membranous E-cadherin immunoreactivity was considered as Ecadherin-negative [49].

Statistical methods
Descriptive statistics were performed to determine the characteristics of our study sample, which comprise three groups: BRCA1, BRCA2, and BRCA WT. ANOVA test and student's t-test were used to compare the mean age at diagnosis between two groups. Chi-square and Fisher's Exact (for smaller sample size) tests were used to compare the proportions of clinicohistopathological (categorical) parameters. To further determine the relationship between clinicohistopathological parameters with mutation status, Spearman's correlation analysis was performed. Logistic regression was conducted to identify independent factors associated with BRCA1 (and BRCA2) mutation status. Associations were summarized using the odds ratio (OR) and corresponding 95% confidence interval (CI) derived from the model estimates. The receiver operation characteristic (ROC) curves were constructed for the prediction of BRCA1 mutation status. The predictive ability of each model was summarized by the area under the curve (AUC), and optimal models were classified as those that yielded the highest AUC in the ROC analysis. We excluded all unknowns/undetermined values from analysis. Statistical significance was considered as p < 0.05 (2-tailed). We performed all statistical analyses using SPSS (v. 23.0 (SPSS Inc., Chicago, IL, USA)).

Results
The characteristics of the patients are summarized in Table 1. Immunohistochemical staining of CLDN3, CLDN4, and CLDN7, and E-cadherin is shown in Fig. 1 and Fig. 2. The tumor size of most of the tumors in all of the three groups (45% of BRCA1, 53% of BRCA2, and 63% of WT tumors) was < 2 cm. The histological type is dominated by invasive ductal carcinoma in all groups (72-90%). Grade 3 was the most common tumor grade in BRCA1 tumors (65%), followed by BRCA2 (44%) tumors, compared to only 27% of BRCA WT tumors (p < 0.001, p = 0.006, respectively). Negative expression of ER and PR was significantly more common among BRCA1 mutation carriers (76 and 79%, respetively) compared to the BRCA2 group (22 and 39%, respectively) (both p < 0.001) and in comparison with the WT subgroup (24 and 36%, respectively) (both p < 0.001). HER2 was positive in 54% of BRCA2 versus 19% of WT tumors (p < 0.001) and versus 21% of BRCA1-mutated tumors (p < 0.001). Positive HER2 was found to be associated with a BRCA2 mutation compared to the WT.
Tumors with expression of CK5 and CK14 were more likely to arise in BRCA1 (44 and 16%, respectively) than in the control group (8 and 4%) (p < 0.001, p = 0.012, respectively). We also found a significantly higher proportion of positive CK5 expression among BRCA1 tumors (44%) in comparison with BRCA2-related BC (8%) (p < 0.001). CK14 was positive in 16% of BRCA1 versus 5% of BRCA2 tumors (p = 0.030). Furthermore, EGFR negatively correlated with the BRCA2 mutation status in comparison with BRCA WT with statistical significance.
Thirty-six of 41 CLDN3-positive cases had a BRCA1 mutation, and 40 of 41 such cases have a mutation in either BRCA1 or BRCA2. CLDN3 was positively correlated with BRCA1-mutated BC. Positive CLDN3 was found in 58% of BRCA1-mutated tumors compared to only 7% of BRCA2-mutated tumors (p < 0.001) and 1% of WT tumors (p < 0.001). Positive CLDN3 was also significantly more frequent in the BRCA2 in comparison with the WT group (p = 0.039) ( Table 1). A positive correlation was observed between CLDN3 with tumor grade and CK5. In contrast, CLDN3 negatively correlated with ER and PR. E-cadherin expression was significantly different in proportion between tumors of BRCA WT and BRCA2 mutation (77% versus 68%, respectively, p = 0.002). With regard to CLDN4 and CLDN7, there were no significant differences observed among the groups. Positive CLDN4 was very common (86-98%), but CLDN7 was rarely positive (0-2%).
For BRCA2 BC versus WT, univariate analysis showed that tumor grade, the expression of HER2, and Ecadherin were independent markers of BRCA2 status. In  (Table 2).
For BRCA2 versus BRCA1 tumors, univariate analysis showed that tumor grade, ER, PR, HER2, CK5, CK14, and CLDN3 were independent parameters of BRCA2 status. However, in a multivariate model, all parameters were not significantly associated with the mutation status, except for CLDN3. Positive CLDN3 in BRCA2-related tumors has an OR of 0.05 (95% CI: 0.01-0.26; p < 0.001) with reference to BRCA1 tumors. Thus, negative CLDN3 was found in 85% of BRCA2-related tumors versus 39% of BRCA1 BC (p < 0.001).
Our receiver operation characteristic (ROC) analysis showed that the base model that correlates with BRCA1 status, which consisted of only ER, PR and HER2, yielded an area under the curve (AUC) of 0.792. The addition of CLDN3 to this model resulted in an AUC of 0.931. When CK5 was added to this model, the model yielded an AUC of 0.942. Further addition of CK14 resulted in an AUC of 0.946 in the ROC curve which is shown in Fig. 3.

Discussion
Taken together, results from our study showed that CLDN3, CK5, and CK14 are associated with BRCA1 mutation status when used in a model where ER, PR and HER2 status are already known. Our research shows that staining of CLDN3, CK5, and CK14 in combination with ER, PR and HER2 indicate an association with BRCA1 mutation status.
Our findings show that tumors with expression of CK5 and CK14 were more likely to arise in BRCA1 than in the control group. This has been reported previously by Lakhani et al. where they investigated immunohistochemical staining for basal markers. They found that CK5/6, CK14, and EGFR were more frequent in BRCA1 tumors compared to non-mutation BC (58% versus 7, 61% versus 12, 67% versus 21%, p < 0.0001 in each case, respectively) [42]. This supports that most of BRCA1- related cancers belong to the basal-like subtype [2]. Foulkes and colleagues showed that the expression of CK5/6 was statistically significantly associated with BRCA1-related BC [18,50]. Furthermore, an association between positive CK5/6 and BRCA mutation status was shown by Murria Estal et al. [51] However, Mohanty et al. could not show a statistically significant difference of CK14 expression between BRCA-mutated and sporadic BC [52]. In addition, CK5/6, CK14, and E-cadherin were not associated with BRCA1 status in a study from Hassanein et al. [53] Eerola et al. reported that CK14 was significantly associated with BRCA1 tumors in univariate analysis [54]. In our study, the model which consisted of ER, PR, HER2, positive expression of CLDN3, CK5, and CK14 yielded an AUC of 0.946 in the ROC analysis to indicate an association with the BRCA1 mutation status. We also found a significantly higher proportion of CK5 expression in BRCA1 than BRCA2 BC. EGFR, an additional biomarker of the basal-like subtype, negatively correlated with the BRCA2 mutation status in comparison with BRCA WT in our study. Expression of CK5/6 and CK14 in BRCA2-related BC is rare and does not differ from sporadic tumors [7,[40][41][42]55]. Results from our study showed similar findings. Otherwise, Eerola et al. showed a positive expression of CK5/6 and CK14 in 7.7 and 26.9%, respectively, in BRCA2 tumors. Only regarding CK14, there was a significant difference between these tumors and the sporadic ones [54].
CLDN3 and CLDN4 are commonly expressed in BC [32,56,57]. Madaras et al. examined the expression of CLDN3, 4, and 7, among other variables, in BRCA-mutated and BRCA WT tumor tissues. In BRCA-mutated tumors, CLDN3, 4, and 7 were expressed at higher level compared to BRCA WT tumors [58]. Higher overexpression rates for CLDN3, 4, and 7 were found in BRCA1-related BC compared to sporadic BC [48]. We showed that CLDN3 was positively correlated with BRCA1-mutated BC. Positive CLDN3 was also significantly more frequent in the BRCA2 in comparison with the WT group. However, in our study there was no association between CLDN4 or CLDN7 and BRCA-mutated BC found. Positive CLDN4 was very common, but CLDN7 was rarely positive in all groups (BRCA1, BRCA2, WT).
In the analysis of CIMBA (Consortium of Investigators of Modifiers of BRCA1/2), Mavaddat et al. showed that 78% of BRCA1-related tumors were ER-negative, but in patients with BRCA2 mutation status ER was negative only in 23% of the tumors. The authors could demonstrate that in the BRCA1 group, 79% were PR-negative and 90% HER2-negative compared to 36 and 87% in BC patients with BRCA2 mutation status, respectively. These results allow to conclude that ER-positive tumors occur more likely in BRCA2 than in BRCA1 cases [44]. Several studies could also demonstrate that negative ER and PR were significantly more common among BRCA1-mutated tumors compared to sporadic cases, among others [11-13, 20, 48, 53]. Therefore, these results are similar to our findings. A low prevalence of positive HER2 (6.8%) expression was shown among patients with BRCA2-related BC by Evans et al., too [59]. BRCA2 tumors often show no or low expression of HER2 [7,13,17,40] However, Armes et al. showed a strong expression of HER2 in 44% (4 of 9) of BRCA2 tumors [60]. Accordingly in our study, positive HER2 was found to be associated with a BRCA2 mutation compared to the WT.
The limitations of this study are a small sample size resulting in a large OR as observed in the results for CLDN3, and should consequently be interpreted with caution. A larger sample size for future studies would be necessary to confirm these findings. Moreover, we did not analyze different molecular subtypes in this study. Future studies should therefore investigate the relationship between the molecular subtypes and the BRCA mutation status. Additionally, the problem of selection should be mentioned as a further limitation of our study. We selected patients with familial breast cancer based on the BRCA1/2 mutation status. Thus, the controls consist of predominantly luminal subtypes.
The dataset of this study should be considered as a training dataset, the analysis is hypothesis generating. This aspect is one of the limitations of our study which is exploratory. We do not have a sufficiently high number of independently collected samples to test the parameters in an independent dataset, too. Thus, further studies are necessary regarding this point.
The examination of TMA should be mentioned as a further limitation of this study. However, the staining was fairly homogenous across the tumor. There were no areas with stronger or weaker intensity observed. TMA is an acceptable and sufficient device, which has been developed and used in research settings [61][62][63][64][65].
In conclusion, findings from our study showed that CLDN3, CK5, and CK14 in combination with ER, PR and HER2 are associated with BRCA1 mutation status.

Conclusions
Our data suggests that CLDN3, CK5, and CK14 in combination with ER, PR and HER2 indicate an association with BRCA1 mutation status.