Claudin-18 expression in oesophagogastric adenocarcinomas: a tissue microarray study of 523 molecularly profiled cases

Background Claudin-18 (CLDN18) is a highly specific tight junction protein of the gastric mucosa. An isoform of CLDN18, the Claudin 18.2, has recently emerged as an innovative drug target for metastatic gastric cancer. Methods We investigated the immunohistochemical profile of CLDN18, p53, p16, E-cadherin, MSH2, MSH6, MLH1, PSM2, HER2, and PDL-1 in a large series of 523 primary gastric carcinomas (GCs; n = 408) and gastro-oesophageal carcinomas (GECs; n = 115) and 135 matched and synchronous nodal metastases. The status of HER2 and EBER by means of chromogenic in situ hybridisation (CISH) was also evaluated. Results High membranous CLDN18 expression was present in 150/510 (29.4%) primary cases and in 45/132 (34.1%) metastases. An abnormal expression (i.e. nuclear and/or cytoplasmic) was observed in 115 (22.5%) primary cases and in 33 (25.0%) metastases. A 38.8% of the cases showed significant CLDN18 intratumoural variability among the different tissue microarray cores obtained from the same tumour. Positive membrane CLDN18 expression was statistically associated with non-antral GCs (p = 0.016), Lauren diffuse type (p = 0.009), and with EBV-associated cancers (p < 0.001). Conclusions CLDN18 is frequently expressed in gastric and gastro-oesophageal cancers; further studies should investigate the prognostic significance of CLDN18 heterogeneity in order to implement its test into clinical practice.

GCs and tumours with CIN have no specific clinicopathological features, even if the first mostly shows a diffuse pattern and the second is often located at the gastro-oesophageal junction.
Claudins (CDLNs) are a family of at least 27 transmembrane proteins, first described by Tsukita et al. 6 They are the major component of the tight junctions (TJ). 7,8 TJ are the main cell-cell contacts among epithelium and endothelium. 9 Particularly, CDLNs are composed by an extracellular loop, four4 transmembrane domains (including the N-terminus) and a cytoplasmic domain (including the C-terminus). 8 The C-terminal domain exhibits binding domains for a complex of proteins such as the scaffolding zonula occludens proteins (ZO) ZO-1, ZO-2, ZO-3 and multi-PDZ domain protein 1 (MUPP1), which are involved in signalling pathways. 10 CDLNs are mainly localised in the apical regions of the cellular membrane and play a critical role in cell-cell adhesion, maintenance of cell polarity and in selective paracellular permeability. [7][8][9] Different CLDNs are expressed in various tissues 8 and can be altered during carcinogenesis. 11 For example, Claudin-6 is involved in breast cancer and in cervical cancer development, [12][13][14] Claudin-11 in cutaneous melanoma, 15 Claudin-1, -4, -7 in colorectal cancer 16 and Claudin-8 in prostate cancer. 17,18 Claudin-18 (CLDN18) is a highly specific TJ component of the stomach. It is expressed in foetal and in adult normal gastric mucosa. 11,19 CLDN18 has two isoforms: Claudin-18.1 and Claudin-18.2, specific for pulmonary and gastric tissue, respectively. GCs and their metastases can retain the expression of this TJ protein. 8 Targeted treatment of GC is in continuous evolution. 8,20-22 Claudin 18.2 (CDLN18.2) represents an ideal therapeutic target due to its trans-membranous localisation. Claudiximab (IMAB362) is a monoclonal recombinant chimeric antibody (IgG1) specific for the CLDN18.2. 23 The antibody can bind CLDN18.2 on the cellular surface and the binding induces the activation of antibody and complement-dependent cytotoxicity. IMAB362 is currently tested in several clinical trials for treatment of advanced gastric carcinomas alone or in combination with standard chemotherapy, showing a favourable safety profile and promising preliminary results in terms of clinical efficacy. 8,20 With the present study, we investigated CDLN18 expression in 523 GCs and GECs, focusing on its association with the clinicopathological and molecular parameters.

MATERIALS AND METHODS
Case selection A total of 523 archival cases of surgically treated GCs (n = 408) and GECs (n = 115) were retrieved from the archives of the Surgical Pathology Unit (Padua University). GECs were defined according to the American Joint Committee on Cancer (AJCC) TNM Classification Eighth edition, which define GECs as cancers with epicentre distant no more than 2 cm from the gastric cardia. 24 All tissue samples were processed according to standard protocols, with formalin fixation time <48 h. All cases were jointly reassessed by 3 pathologists (R.C., I.C. and M.F.), and representative, non-necrotic cancer specimens were selected for tissue microarray (TMA) construction. When available (n = 135), metastatic nodes were also included in the analysis. For metastatic samples, only metastatic foci >2 mm with >40% neoplastic component were selected for TMA preparation.
Only material that was not required for diagnosis was used and all patients signed an informed consent approved by the University Hospital of Padua Review Board, which allows researchers to use excess material for research purposes. The study was approved by the local Ethic Committee.
TMA construction Two neoplastic areas from two separate formalin-fixed, paraffinembedded (FFPE) blocks were selected and tissue cores (1 mm diameter) were punched out of these areas using the Tissue ArrayerMinicore 3 (Alphelys, Plaisir, France), as previously described. 21 In 520 of the 523 primary GCs/GECs, 2 tissue cores were obtained from each selected area (i.e., 4 samples per tumour); in 3 of the 523 cases, only 1 cancer tissue core was available from each neoplastic area (i.e., 2 samples per tumour). From the 135 synchronous nodal metastases, 2 tissue cores were obtained in 115 cases, whereas a single tissue core was obtained from 20 metastatic nodes.
As a result, a total of 2086 tissue cores of primary GC/GECs and 250 tissue cores of metastatic nodes were collected in 34  For the evaluation of CLDN18, the membrane immunoreaction was assessed using a semi-quantitative pathology H-score, defined as the aggregate of total percentage of tumour cells expressing CLDN18 at each particular intensity level from 0, +1 (weak intensity), +2 (moderate intensity) or +3 (strong intensity). In brief, the H-score was defined as: (Percentage of CLDN18 1+ tumour cells multiplied by intensity of 1) + (Percentage of CLDN18 2+ tumour cells multiplied by intensity of 2) + (Percentage of CLDN18 3+ tumour cells multiplied by intensity of 3). Thus this composite score can range from 0 (a tumour which is completely negative) to a maximum of 300 (a tumour in which all the cells feature a 3+ staining). Scores were categorised in negative/low (0 = 0-50) and positive/high (1 = 51-300). Where present, nuclear and/or cytoplasmic CLDN18 expression was noted but not retained for scoring.
Deficient mismatch repair (MMRd) status was assessed by testing MSH2, MSH6, MLH1 and PSM2, and samples were defined as MMRd when one or both proteins resulted negative. 25 p53 was considered as aberrant in the presence of complete loss or diffuse and strong nuclear immunostaining in neoplastic cells. 26 For the evaluation of p16, immunoreaction was assessed using a four-tier classification: 0, complete absence of p16 staining in all neoplastic cells; 1, staining only in isolated and dispersed neoplastic cells; 2, staining in patchy and scattered clusters of neoplastic cells; and 3, dense and continuous cytoplasmic/nuclear staining in all neoplastic cells (1). The resulting values were combined in 2-point total scale, characterised by negative/low (0 and 1) and positive/high (2 and 3) expression.
For the evaluation of HER2, the four-tier modified Herceptest score for biopsies was used. Score 0/1+ (no membranous immunostaining in any neoplastic cells/presence of tumour cell cluster with barely perceptible membranous reactivity irrespective of percentage of tumour cells stained) as negative; 2+ (presence of tumour cell cluster with weak or moderate basolateral-lateral membranous reactivity irrespective of the percentage of tumour cells stained) as equivocal; and 3+ (presence of tumour cell cluster with strong complete, basolateral or lateral membranous reactivity irrespective of the percentage of tumour cells stained) as positive. 21 E-cadherin expression was considered altered in the presence of complete loss or markedly reduced membranous staining (>30%), regardless of nuclear/cytoplasmic staining. 26 Only tumour PD-L1 expression was retained for scoring, and a 1% cut-off was used in the analysis.
HER2 chromogenic in situ hybridisation (CISH) CISH was performed according to the manufacturer's protocol (Dako Her2 CISH pharmDx Kit; Dako, Glostrup, Denmark). Areas with the highest HER2 counts with non-overlapping nuclei were analysed by counting HER2 and centromeric probe 17 (CEP17) signals in at least 40 nuclei. The ratio HER2/CEP17 was calculated. A case was considered HER2 amplified when the signal ratio was ≥2.0 or when HER2 signal cluster was observed.

EBER in situ hybridisation (ISH)
The Bond ready-to-use ISH EBER Probe was used in a Leica Bond-Max automation system according to the manufacturer's instructions (Leica Biosystems) to detect EBV infection.

Statistical analysis
Differences and associations between CLDN18 and clinicopathological variables or other IHC markers were tested by applying the χ 2 test and Fisher Exact test. A p value <0.05 was considered significant.

Clinicopathological findings
Overall, the mean age of the patients was 69.3 ± 12.4 years (median 60; range 25-95). The male-to-female ratio was 1.98 (Table 1). All patients were Caucasian.
High CLDN18 was associated with tumour site (antrum vs corpus), with a higher prevalence of positive cases among proximal (i.e. corpus) GCs (p = 0.016), whereas the distribution between GCs and GECs showed no statistically significance (29.7 vs 28.3%). High CLDN18 expression was also associated with the Lauren Classification, with higher prevalence of positive cases in the diffuse pattern (p = 0.019) (Fig. 2a-d). No association emerged between CLDN18 and age/sex, grading, staging and Ming Classification. We tested the prognostic impact of CLDN18 on the series of 45 stage IV cases, in which 10 (22.2%) tumours showed high CLDN18 expression, but the association between CLDN18 status and patients' prognosis was not significant in this small series of advanced cancers.
Of note, a strong nuclear and/or cytoplasmic positivity was observed in 115 primary cases (22.5%): 95 with a nuclear positivity, 13 with a cytoplasmic positivity and 7 with a mixed nuclear/ cytoplasmic positivity. In 47/95 (49.5%) cases with nuclear staining, a concurrent membranous positivity was observed and only 10 of them were categorised as high CLDN18 tumours. Similarly, among metastatic samples, 27 (20.5%) had a nuclear positivity, 6 (4.5%) a cytoplasmic positivity and 5 (3.8%) a mixed positivity. The nuclear and/or cytoplasmic positivity showed no statistically significant associations with tumour location. Furthermore, no significant associations emerged comparing nuclear/cytoplasmic with membranous CLDN18 expression and with pathological staging. Intratumoural variability of membranous CLDN18 expression was investigated, considering CLDN18 expression among the multiple TMA cores collected from different areas of the same tumour. A tumour was considered as CLDN18 heterogeneous in case of concomitant presence of high-CLDN18 and low-CLDN18 TMA cores. Among primary tumours, 160 GCs (40.3%) and 38 GECs (33.6%) showed intratumoural variability within the analysed cores. Similar results were observed for metastatic samples, with a total of 38 cases (28.8%) with heterogeneous CLDN18 status. Focussing on CLDN18 expression, as assessed in matched primary and metastatic samples, this analysis was possible for 128 couples. CLDN18 status was consistent between the two matched samples in 111 cases (86.7%; 83 negative and 28 positive cases), with only 17 cases showing CLDN18 discordant status (Fig. 2e).
A total of 20/523 (3.8%) tumours were positive for EBER ISH. Of this group, 15 were GCs and 5 GECs. A significant association was observed between EBER status and CLDN18 expression (p < 0.001). In fact, 14/20 (70%) of EBER-positive tumours showed high expression of CLDN18. Moreover, 4 further EBER-positive cases showed low (i.e. >1 and <51) CLDN18 expression. Only two EBERpositive cases did not show any CLDN18 expression. The mean age of EBER positive tumours was 64.6 years, and the male-tofemale ratio was 3:1. Among the 15 EBV-associated GCs, 12 were located in the gastric corpus, 2 in the antrum, and 1 in an oesophago-jejunal anastomosis.
Immunohistochemical profiling has recently emerged as a suitable alternative for molecular classification of GC. 26,27 We focussed on the work of Ahn and collaborators 26

DISCUSSION
The main aim of this study was to investigate CLDN18 expression in a large mono-Institutional series of GCs and GECs using IHC. Taking into account the emergent role of the monoclonal antibody Claudiximab (IMAB362) targeted against the isoform Claudin-18.2, these results were compared to clinicopathological and molecular parameters, in order to detect possible distinctive features of CLDN18-positive GC/GECs. CLDN18 is a highly specific gastric claudin expressed in the normal adult gastric mucosa, as well in the gastric foetal tissue, with a tendency to be conserved in GCs. 11,19,28 Our results are consistent with previous findings, since the presence of CLDN18 was documented in 29.4% primary cases and in 34.1% nodal metastases. No significant differences emerged in relation to patients' age, sex, gastric vs gastro-oesophageal tumour localisation, grading, pathological staging, and Ming Classification. Among the GCs, we found a significant association between CLDN18 expression and gastric tumour localisation (i.e. tumour localised in the gastric corpus showed a higher prevalence of CLDN18-positive cases) and Lauren Classification. These data could be explained by both the role of CLDN18 in the paracellular ion transport, which is primarily associated with the gastric body 29 and by the "intestinalisation" observed during Correa's cascade in most antral GCs, which can lead in many instances to the loss of gastric-specific markers. 19 According to the Lauren Classification, a higher prevalence of positive CLDN18 cases had a diffuse type. Considering the TGCA GC classification, GS tumour category often shows a prevalent diffuse pattern. A typical molecular alteration of this subtype is an interchromosomal translocation between CLDN18 gene and ARHGAP26. ARHGAP26 is a GTPase-activating protein (GAP). Its role is to facilitate RHO GTPase to the GDP state and consequently to induce the cellular motility. 2 These types of translocations are common in diffuse GC and have been demonstrated to be negative prognostic factors. [30][31][32][33] Of note, Tanaka and colleagues demonstrated that cases with CLDN18 translocation are significantly characterised by a CLDN18 IHC overexpression. 32 An interesting observation in our study was the nuclear and/ or cytoplasmic CLDN18 immunoreactivity. This intracellular distribution was not significantly linked to a loss or weaker expression of membranous localisation. Previous studies have already focussed on the TJ nuclear/cytoplasmic positivity. 34,35 Somoracz and colleagues 34 demonstrated a nuclear positivity of tricellulin in a subset of hepatocellular carcinoma. The nuclear positivity was associated with a weaker membranous expression of this TJ, suggesting a possible disturbed intracellular trafficking of molecules. French et al. 35 stressed that the nuclear Claudin-1 (CLDN1) was linked to benign nevi and to early melanomas vs a high cytoplasmic and membranous CLDN1 expression mainly linked to the metastatic counterparts. Furthermore, these authors demonstrated that CLDN1 translocation from nucleus to cytoplasm was driven by protein kinase A (PKA) via phosphorylation.
Among the CLDN18-positive cases, another point was the high prevalence of its membranous intratumoural variability. This has been similarly demonstrated for HER2 in GC [36][37][38][39][40] and could affect (i) biomarker evaluation in biopsy specimens and (ii) any targeted therapeutic approach. Further "real world" studies should investigate the minimum number of GC/GEC biopsies to have an adequate CLDN18 evaluation.
The association between CLDN18 expression and p53, p16, E-cadherin, PD-L1 and HER2 was investigated. The only significant association emerged between CLDN18 and E-cadherin. Higher prevalence of positive E-cadherin was found among positive CLDN18 tumours. This association was independent of the Lauren Classification.
We found a significant association between CLDN18 and EBVpositive status. Shinozaki et al. 3 previously investigated the Claudins' expression in EBV-GCs compared to EBV-negative GCs and their results indicated a CLDN18-positive predominance in EBV-associated GCs. Since preserved expression of CLDN18 was described not only in mature but also in immature gastric epithelium, 19 our data could support the hypothesis that EBVassociated GCs could arise directly from immature proliferating cells. There were no significant differences between prevalence of EBV-associated GCs and EBV-positive GECs. Jovov et al. 41 demonstrated an almost absolute predominance of CLDN18 expression in Barrett's oesophagus compared to other Claudins. Shinozaki's study 3 underlined the lack of CLDN18 in intestinal metaplasia in the stomach. These previous results support the hypothesis that CLDN18 expression is mainly dependent on the molecular profile of immature tissue-specific cells regardless of the presence or not of an intestinal phenotype. Anyway, this point remains unclear and would require further investigation.
The prevalence of CLDN18-positive cases is relatively lower in our study in comparison to the recent published clinical trial 42 or a previous paper by Matsusaka and colleagues; 28 however, our series considered only a 8.6% of stage IV tumours and this can significantly affect the prevalence of CLDN18-positive cases. Moreover, different antibodies are available, and they are characterised by different sensitivities/specificities to CLDN18. In particular, by checking the Human Protein Atlas database (www.proteinatlas.org; 21 January 2018) three different antibodies are on the market and detect, on the same series of GC specimens, CLDN18 high expression in 36.4 to 70.0% of the cases, further supporting the need for a standardisation of testing.
In summary, this is the largest study investigating CLDN18 expression among GECs. One third of the cases retained CLDN18 expression and this was significantly associated with gastric corpus location, diffuse-type GC and with the presence of EBV infection. Our group has recently started a new prospectively conceived translational study on a selected series of stage IV GC, focusing on CLDN18.2 isoform and its prognostic role and predictive significance to current standard treatments. Ethics approval and consent to participate: Only material that was not required for diagnosis was used and all patients signed an informed consent approved by the University Hospital of Padua Review Board which allows researchers to use excess material for research purposes. The study was approved by the local Ethic Committee.

AUTHOR CONTRIBUTIONS
Funding: This work was partly supported by the grants from the Italian Association for Cancer Research (AIRC Regional grant 2008 N. 6421) and DOR grant from Padua