1Department of Oncology and Radiotherapy, Medical Research Center Oulu, Oulu University Hospital and University of Oulu, P.O. Box 22, 90029 Oulu, Finland

2Department of Pathology, Medical Research Center, Oulu University Hospital, Oulu, Finland

2Department of Pathology, Medical Research Center, Oulu University Hospital, Oulu, Finland

3Department of Pathology, University of Eastern Finland, Kuopio, Finland

1Department of Oncology and Radiotherapy, Medical Research Center Oulu, Oulu University Hospital and University of Oulu, P.O. Box 22, 90029 Oulu, Finland

4Department of Oncology, and Cancer Center, Kuopio University Hospital, and Institute of Clinical Medicine, University of Eastern Finland, Kuopio, Finland

2Department of Pathology, Medical Research Center, Oulu University Hospital, Oulu, Finland

1Department of Oncology and Radiotherapy, Medical Research Center Oulu, Oulu University Hospital and University of Oulu, P.O. Box 22, 90029 Oulu, Finland

The datasets analyzed during the current study are available from the corresponding author on reasonable request.

The subtype of claudin-low breast cancer can be reliably determined only by gene-expression profiling. Attempts have been made to develop immunohistochemical surrogates, which nearly always focus on membranous claudin expression.

We assessed the immunohistochemical expression of both membranous and cytoplasmic claudins 3, 4 and 7 in a series of 197 non-metastatic breast cancers, enriched with triple-negative breast cancers (TNBCs; 60%). The expression of epithelial-to-mesenchymal transition-regulating transcription factors Sip1, Zeb1 and vimentin had previously been determined in the same material.

In multivariate analysis, strong cytoplasmic claudin 3 expression was associated with poor relapse-free survival (RFS), disease-free survival, distant disease-free survival, breast cancer-specific survival and overall survival among TNBC patients (for RFS, RR 5.202, 95% CI 1.210–22.369, p = 0.027, vs. T-class, RR 0.663, 95% CI 0.168–2.623, p = 0.558, and N-class, RR 3.940, 95% CI 0.933–16.631, p = 0.062). Cytoplasmic claudin 3 expression was also associated with strong nuclear Sip1 expression (p = 0.000053), TNBC phenotype (p = 0.012) and within them, non-basal-like phenotype (p = 0.026). Cytoplasmic claudin 7 was associated with dismal RFS (RR 6.328, 95% CI 1.401–28.593, p = 0.016, vs. T-class, RR 0.692, 95% CI 0.242–1.982, p = 0.493, and N-class, RR 2.981, 95% CI 1.1016–8.749, p = 0.047). Low cytoplasmic expression of claudins 3, 4 and 7 together also predicted poor RFS (RR 6.070, 95% CI 1.347–27.363, p = 0.019, vs. T-class, RR 0.677, 95% CI 0.237–1.934, p = 0.467, and N-class, RR 3.167, 95% CI 1.079–9.290, p = 0.036).

Immunohistochemical expression levels of cytoplasmic claudins 3 and 7 appear to be novel prognostic factors in TNBC.

Claudins are tight junctional proteins of 20–24 kDa that are present on the apicolateral membranes of epithelial, endothelial and mesothelial cells [1, 2]. They have barrier and fence functions, and they take part in signal transduction [2]. They have four transmembranous domains and the molecules form two extracellular loops that harbor sites for functions of claudins such as paracellular permeability and attachment sites for Clostridium perfringens toxin or hepatitis C virus [1, 2]. The intracytoplasmic carboxyterminal part of the molecule has PDZ domains for attachment to ZO1–3, by which claudins may influence cellular signaling [3]. There are 27 different claudins known [4].

In malignancies, the expression of claudins varies, depending on the site and type of the tumor [5] and claudins can be used in the differential diagnosis of tumors in some cases. As an example, claudins 3 and 4 are very likely to be expressed in metastatic carcinomas of the pleura, while mesotheliomas are usually negative, and the expression of claudins 4 and 7 has been suggested to differentiate cholangiocarcinoma and hepatocellular carcinoma [6, 7]. Claudins 3 and 4 are particularly overexpressed in several carcinomas, including breast cancer [2].

In addition to barrier and fence functions, individual claudins harbor different properties affecting tumor growth and spread. Claudin 4 has been shown to induce angiogenesis, the spread and proliferation of MCF-7 cells, while abrogating apoptosis [8, 9]. Claudin 4 appears to be overexpressed frequently in metastatic breast cancer tissues compared with primary sites [10]. Inhibiting claudin 3 overexpression in MCF-7 cells has resulted in decreased tumor cell migration [11]. Claudins may also influence the prognosis of tumors. Low-level claudin 4 expression is associated with poor prognosis in esophageal and pancreatic carcinoma [12, 13]. On the other hand, claudin 3 overexpression is an indicator of poor prognosis in serous ovarian carcinoma, while its downregulation predicts poor survival in squamous cell lung carcinoma [14, 15]. Low-level claudin 7 expression is associated with better prognosis of patients with oral squamous cell carcinoma [16], and in prostate carcinoma, with increased tumor grade [17].

Triple-negative breast cancer (TNBC) is a subtype with lack of expression of ER, PR and HER2 and it constitutes of about 15–20% of cancer cases [18]. TNBCs are enriched in basal-like (BLBC) and claudin-low breast cancer molecular subtypes, the former expressing basal cell markers and the latter, in addition to low claudin 3, 4, 7 and E-cadherin expression, showing induced expression of EMT (epithelial-to-mesenchymal transition)-related genes, immune system-related genes and stem-cell features [18, 19]. The estimated incidence of claudin-low breast cancer is 7–14% and long-term prognosis is relatively poor [19–21].

The clinical research on claudins in cancers is rapidly growing and monoclonal claudin antibodies have also shown promising results in a phase II trial in cases of gastric cancer [22]. The clinical benefit of finding this subgroup in breast cancer is still limited, since identifying a tumor as a claudin-low subtype requires gene expression profiling from fresh frozen tumor material. Different approaches to define claudin-low subtypes by immunohistochemistry (IHC) have been proposed, but none have been validated in independent cohorts.

Claudins thus have various biological and pathological properties, depending on their specific subtypes and localization. Previous claudin protein expression studies in breast cancer have mainly been concentrated on membranous claudin expression and/or have not involved the expression of separate claudins. We aimed to clarify if the expression of claudins 3, 4 and 7, in membranes and cytoplasm, could be associated with the outcome of the disease. Since claudins are overexpressed in TNBCs, we used TNBC-enriched material, previously assessed for expression of major EMT regulators.

There was a total of 197 women with non-metastatic breast cancer in the research material (Table 1). Of these, 119 were TNBC cases (60.4%) and 78 non-TNBC. Of 99 evaluable TNBC cases, 87 (73.1%) showed a basal-like phenotype as they expressed either CK5/6 or EGFR-1. The median follow-up time was 100.0 months (mean 94.0 months).

Patient material

The specimen fixation, storing and staging was performed as previously described [23]. Tumor differentiation was classified according to the WHO Classification of Tumors [24].

Claudin primary antibodies, designed for formalin-fixed paraffin-embedded tissue sections, were purchased from Zymed Laboratories Inc. (San Francisco, CA, USA). The antibodies used were polyclonal rabbit anti-claudin 3 (Z23.JM), monoclonal mouse anti-claudin 4 (clone 3E2C1), and polyclonal rabbit anti-claudin 7 (ZMD.241). Sections of 5 μm were deparaffinized and rehydrated. They were first heated in a microwave oven in tris-EDTA for 10 min and then incubated with the primary antibody for 60 min. The dilution was 1:50 for all anti-claudins and DAKO EnVision kits were used according to the manufacturer’s instructions for the detection of primary antibody. Color was developed by using diaminobenzidine, the sections were counterstained with hematoxylin and mounted with Pertex (Leica Microsystems, Germany). Negative controls were handled as previously described but with the primary antibody replaced by serum or PBS. Positive controls included tumor samples previously known to be positive for the claudins.

Tumors exhibiting nuclear estrogen/progesterone receptor (ER or PR) expression in more than 9% of invasive tumor cells were considered as steroid receptor-positive. The TNBC group did not show any ER or PR positivity. In other words, tumors expressing ER or PR in 1–9% of invasive cells were excluded from the study. If a specimen exhibited a membranous HER2-positive result (1+ to 3+ on a scale of 0 to 3+) in IHC, HER2 gene amplification status was determined by means of chromogenic in situ hybridization. Breast cancers with six or more gene copies of HER2 in cells were considered HER2-positive. Expression of Ki-67 was studied by means of IHC as described previously [25]. The methods and results concerning cytokeratin 5/6, epidermal growth factor receptor and EMT marker immunostaining and assessment in this material have also been reported earlier [23]. The triple-negative tumors that also expressed either EGFR and/or CK5/6 were classified as basal-like breast cancers [25–27].

Claudin immunoreactivity was assessed semiquantitatively by dividing the immunoreactivity into five groups: 0–5%, 5–25%, 25–50%, 50–75% and over 75% positive. Membranous and cytoplasmic expression were assessed separately. For claudins 3 and 4 less than 50% positivity was considered to be low expression. Since claudin 7 expression was significantly weaker, less than 5% was considered as low expression. Claudin-low breast cancers were defined as those having low membranous expression of claudins 3, 4 and 7. Claudin assessments were performed by an experienced histopathologist (YS), who was blind to the clinical data at the time of the analysis.

Statistical analysis was performed using IBM SPSS Statistics software, v. 23.0.0.0 (IBM Corporation, Armonk, NY, USA). T-class was divided in statistical analyses to either T1 or T2–4, and nodal status to either positive or negative. Expression of Ki-67 was divided into 0–14% or > 14% and grade was either grade I–II or grade III in the analyses. The significance of associations was defined by using two-sided Pearson’s Chi-square tests. Kaplan–Meier curves with the log-rank test were applied in survival analysis. Disease-free survival (DFS), relapse-free (RFS), distant disease-free (DDFS), breast cancer-specific (BCSS) and overall (OS) survival were calculated from the time of diagnosis to disease recurrence at any site (DFS), in the ipsilateral breast, scar, or axilla (RFS), at distant sites (DDFS), to the time of confirmed breast cancer-related death (BCSS) or time of death from any cause (OS). Cox regression analysis was applied in multivariate analysis, where the most important traditional prognostic factors, T-class (T1 or T2–4) and N-class (N0 or N1–3), were included to the model. In all statistical analyses, p-values less than 0.05 were considered significant.

Among the total of 197 patients, claudin 3 was reliably assessable in 187 (94.9%) cases, claudin 4 in 191 (97.0%) and claudin 7 in 185 (93.9%) cases. Claudin expression is presented in Table 2 and examples of staining patterns are shown in Fig. 1.

Expression levels of claudins 3, 4 and 7 in the whole material, separately in TNBC and non-TNBC groups and significance in comparison of the two groups (p-value, 2-sided Pearson’s chi-square test)

Immunohistochemical expression of claudins 3 (a), 4 (b) and 7 (c) in breast cancer. Asterisks demonstrate cytoplasmic and arrows membranous immunostaining

Cytoplasmic claudin 3 was overexpressed in TNBC tumors (p = 0.012), and within them, in non-basal-like TNBCs (p = 0.026) (Table 3). Likewise, cytoplasmic claudin 4 was associated with the non-basal-like phenotype of TNBCs (p = 0.00090). Stronger membranous claudin 3 expression was associated with patients with only bone metastases as the first metastatic site (p = 0.032).

Significant 2-sided p-values of associations between claudin (cl) 3, 4 and 7 expression and traditional prognostic factors, EMT-regulating transcription factors and survival in univariate analysis

The directions of the associations are described in the Results section. Cyt, cytoplasmic; Membr, membranous; TNBC, triple-negative breast cancer; BLBC, basal-like breast cancer; DFS, disease-free survival; DDFS, distant disease-free survival; RFS, relapse-free survival; BCSS, breast cancer-specific survival; OS, overall survival

Cytoplasmic claudin 7 expression was associated with smaller tumor size (p = 0.0053), better differentiation (p = 0.043) and with sites other than bone as the first metastatic site (p = 0.016). Cytoplasmic claudin 7 was also overexpressed in TNBC tumors (p = 0.0026), and within them, in non-basal-like TNBCs (p = 0.023). Membranous claudin 7 expression showed an inverse association with proliferation rate (p = 0.0042) and it was overexpressed in TNBC tumors (p = 0.00023).

Low-level membranous expression of claudins 3, 4 and 7 together was associated with node negativity (p = 0.013). Low-level expression of claudins 3, 4 and 7 together in cytoplasm was connected with higher grade (p = 0.013), larger primary tumor (p = 0.0074), a non-TNBC phenotype (p = 0.0033), and within TNBCs it was strongly connected with basal-like breast cancers (p = 0.0050).

Zeb1 expression in cancer cells was associated inversely with membranous claudin 7 expression (p = 0.010), while relatively strong cytoplasmic claudin 7 expression was associated with increased cytoplasmic Sip1 expression (p = 0.0012). An association between cytoplasmic claudin 3 and nuclear Sip1 expression was extremely significant (p = 0.0000053). Low-level expression of claudins 3, 4 and 7 in cytoplasm was associated with low levels of cytoplasmic and nuclear Sip1 expression (p = 0.0029 for both).

Cytoplasmic claudin 3 was associated with poor DFS (p = 0.0009), DDFS (p = 0.006), RFS (p = 0.00001), BCSS (p = 0.001) and OS (p = 0.018) in univariate analysis in the whole material (Fig. 2). However, since cytoplasmic claudin 3 was associated strongly with TNBC, and there was only one patient in the non-TNBC group with strong claudin 3 expression, the association between survival and cytoplasmic claudin 3 expression was significant only within TNBC patients (DFS, p = 0.0012; DDFS, p = 0.007; RFS, p = 0.000028; BCSS, p = 0.005; OS, p = 0.016).

Kaplan–Meier curves of studied outcomes according to the expression of cytoplasmic claudin 3 (a–e), cytoplasmic claudin 7 (f) and claudins 3, 4 and 7 together (g). Crosses indicate censored cases

In multivariate analysis, when T-class (T1 or T2–4) and N-class (N0 or N1–3) were taken into account, claudin 3 still remained a significant prognostic factor (Table 4). Notably, in RFS analysis the prognostic role of cytoplasmic claudin 3 (RR 6.162, 95% CI 1.785–21.272, p = 0.004) exceeded that of T-class (RR 0.714, 95% CI 0.240–2.124, p = 0.544) and N-class (RR 3.076, 95% CI 1.032–9.170, p = 0.044). When the analysis was performed separately in TNBC and non-TNBC cases, in TNBC strong cytoplasmic claudin 3 expression was solely an independent predictor of worse RFS (RR 5.202, 95% CI 1.210–22.369, p = 0.027, vs. T-class, RR 0.663, 95% CI 0.168–2.623, p = 0.558, and N-class RR 3.940, 95% CI 0.933–16.631, p = 0.062), while in cases of non-TNBC none of these three variables was an independent predictor of RFS in this model.

Risk ratios, 95% confidence intervals and corresponding p-values concerning tumor size (T), nodal status (N) and cytoplasmic claudin 3 expression in multivariate analysis of disease-free survival (DFS), distant disease-free survival (DDFS), relapse-free survival (RFS), breast cancer-specific survival (BCSS) and overall survival (OS)

Strong cytoplasmic claudin 7 expression was associated with poor RFS (p = 0.0024), and it was also significant in multivariate analysis (RR 6.328, 95% CI 1.401–28.593, p = 0.016, vs. T-class, RR 0.692, 95% CI 0.242–1.982, p = 0.493, and N-class, RR 2.981, 95% CI 1.106–8.749, p = 0.047). Stronger cytoplasmic claudin 4 expression predicted worse DDFS in non-TNBC patients (p = 0.017), but this did not remain significant in Cox regression analysis.

Low-level expression of claudins 3, 4 and 7 together in cytoplasm predicted poor RFS (p = 0.0033), being similar in TNBCs (p = 0.036) and non-TNBCs (p = 0.038). In Cox regression analysis, this remained as a significant factor (RR 6.070, 95% CI 1.347–27.363, p = 0.019, vs. T-class, RR 0.677, 95% CI 0.237–1.934, p = 0.467, and N-class, RR 3.167, 95% CI 1.079–9.290, p = 0.036).

Our aim was to establish if separate IHC assessment of the expression levels of claudins 3, 4 and 7 could be associated with different outcomes in breast cancer. Previous IHC studies concerning claudins in breast cancer have rarely involved both cytoplasmic and membranous claudin expression, but we decided to evaluate them separately. In addition, we had previously characterized major EMT-regulating transcription factors in most samples in the current material [28], which allowed us to correlate Sip1, Zeb1 and vimentin expression to the expression levels of claudins 3, 4 and 7. Other strengths of the current study were sufficient follow-up and the use of appropriate definition of TNBC, i.e. ER and PR cut-off levels were set at