Molecular and pathological analyses of gastric stump cancer by next-generation sequencing and immunohistochemistry

Gastric stump cancer (GSC) has distinct clinicopathological characteristics from primary gastric cancer. However, the detailed molecular and pathological characteristics of GSC remain to be clarified because of its rarity. In this study, a set of tissue microarrays from 89 GSC patients was analysed by immunohistochemistry and in situ hybridisation. Programmed death ligand 1 (PD-L1) was expressed in 98.9% of tumour-infiltrating immune cells (TIICs) and 6.7% of tumour cells (TCs). Epstein–Barr virus (EBV) was detected in 18 patients (20.2%). Overexpression of human epidermal growth factor receptor 2 and deficiency of mismatch repair (MMR) protein expression were observed in 5.6% and 1.1% of cases, respectively. Moreover, we used next-generation sequencing to determine the gene mutation profiles of a subset of the 50 most recent patients. The most frequently mutated genes were TP53 (42.0%) followed by SMAD4 (18.0%) and PTEN (16.0%), all of which are tumour suppressor genes. A high frequency of PD-L1 expression in TIICs and a high EBV infection rate suggest immune checkpoint inhibitors for treatment of GSC despite a relatively low frequency of deficient MMR gene expression. Other molecular characteristics such as PTEN and SMAD4 mutations might be considered to develop new treatment strategies.

IHC and ISH. In the tissues of all patients, the presence of dense tumour-infiltrating immune cells (TIICs) was observed. Moreover, we detected anti-programmed death ligand 1 (PD-L1) expression in TIICs of the tissues in almost all patients (88/89, 98.9%), while PD-L1 expression in tumour cells (TCs) was observed in a limited number of cases (6/89, 6.7%) ( Table 2 and Supplementary Table 1). EBV was detected in 18 (20.2%) samples by in situ hybridisation. Expression of epidermal growth factor (EGFR) and human epidermal growth factor receptor 2 (HER2) was detected in two (2.2%) and five (5.6%) samples, respectively (Supplementary Table 1). Mismatch repair (MMR) gene deficiency was observed in only one case (1.1%, Supplementary Table 1) that showed loss of MLH1 and PMS2 proteins. Table 1. Clinicopathological characteristics. The pathological stage as well as T and N numbers were defined using the 8th TNM classification. Histopathological types were classified in accordance with the Japanese classification of gastric carcinoma. www.nature.com/scientificreports/ Gene mutation profiles. Gene mutation profiles of the 50 most recent patients are summarised in Table 3.

Subgroup analysis.
We compared the results on the basis of the tumour location (Anastomosis group vs non-Anastomosis group) and EBV infection status (EBV positive vs. EBV negative). Comparison of the clinicopathological factors of the two groups is shown in Table 4. Tumours arising at the anastomosis site were more frequently observed at the initial gastrectomy performed for benign disease, reconstructed with the Billroth-II method, and a longer interval from initial surgery. The size of a tumour arising at the anastomosis site tended to be larger than that of a tumour arising at a non-anastomosis site.
All EBV-positive patients were male and the median age of EBV-positive patients was younger than that of EBV-negative patients with a significant difference. For other factors, there were no significant differences between the two groups.

Discussion
Understanding the molecular characteristics of GSC may enable selection of effective treatments and the development of new therapeutics. However, molecular characteristics of GSC remained to be investigated, partly because of their low incidence. Therefore, in the present study, we investigated protein expression and the mutation profiles of selected patients with GSC.
We found that PD-L1 expression in TCs (6.7%) of GSC was lower compared with that in TCs of PGC as reported previously (22.8%) 23 and that PD-L1 was expressed more frequently in TIICs of GSC compared with those of PGC 23 . These lymphocytes, which may reflect excess inflammatory stress caused by exposure to bile reflux, induce autoinhibition to prevent excess immune reactions against bile reflux-induced inflammation 24 and www.nature.com/scientificreports/ simultaneously contribute to immune escape of developing tumour cells. It is conceivable that PD-L1 expression not only in TCs but also in TIICs may predict the efficacy of immune checkpoint inhibitors for treatment of GSC. In fact, a combined positive score has been proven to be a significant indicator to predict the effect of immune checkpoint inhibitors on gastric cancers 25 . Therefore, we believe that a high frequency of PD-L1 expression in TIICs is a clinically significant characteristic of GSC. Because other clinical factors are known to be related to inflammation in the stomach and may induce PD-L1 expression in TC sand/or TIICs, we investigated the smoking status, drinking habits, helicobacter pylori infection, and adjuvant chemotherapy following initial gastrectomy (Supplementary Table 3). Therefore, no significant correlations with PD-L1 expression on TIICs were observed for any of the factors, although the number of available cases for the comparisons was limited.
We next focused on GSCs arising at the site of anastomosis because this type of GSC can be strongly affected by bile reflux-induced inflammation 8 . We found that GSCs at anastomosis sites expressed higher levels of PD-L1 in TCs than those not involving this site, although there was no significant difference. Expression of PD-L1 in TCs may be involved in the development of remnant gastric cancer, and surgeons may need to avoid Billroth-II anastomosis or add Braun anastomosis to Billroth-II anastomosis to lessen the reflex.
Deficient MMR gene expression was observed in one patient (1.1%), which is inconsistent with previous reports 17,19,20 . Although the reason is unknown, one possibility is the small number of patients studied in the previous report. Thus, the present study indicates that MMR deficiency is not a major contributor to the development and progression of PGC.
We expected to detect frequent mutations of KRAS, which frequently (50-60%) occur in gallbladder and biliary tract cancers of patients with pancreaticobiliary maljunction because of regurgitation of bile and pancreatic Table 4. Comparison of clinicopathological factors between patients with gastric stump cancer (GSC) in the anastomotic area (Anastomosis) or other areas (non-Anastomosis) and GSCs with EBV positivity or negativity. The pathological stage as well as T and N numbers were defined using the 8th TNM classification. Histopathological types were classified in accordance with the Japanese classification of gastric carcinoma. The TCGA classifies gastric cancer into four subtypes, one of which is EBV 21 . Here, we found that patients with EBV-positive GSC represented 20% of the cases, which is consistent with previous reports (20-40%) and higher compared with PGC patients (5-10%) [14][15][16] . The reason why GSC has high relevance with EBV is unclear, but male and Billroth-II reconstruction appear to be preferable factors for EBV. Nishikawa et al. suggested that an atrophic change of remnant gastritis in Billroth-II reconstruction is associated with EBV-positive GSC 16 . Furthermore, we detected higher frequencies of PD-L1-positive TCs and PIK3CA mutations, which is consistent with the TCGA classification. EBV-positive GSC is similar to EBV-positive PGC, and EBV-positive GSC accounts for a larger proportion of GSCs than PGCs. Therefore, establishing new therapeutics against EBV-positive GC may be an approach for more effective treatment of advanced GSC. Table 5. Comparison of immunohistochemistry and in situ hybridisation between gastric stump cancer (GSC) located in the anastomotic site (Anastomosis) or others (non-Anastomosis) and GSC with EBV positivity or negativity. TCs tumour cells, TIICs tumour-infiltrating immune cells.   [33][34][35] . The treatment strategy for GSC is based on the results of clinical trials of PGC, but it has been gradually demonstrated that the molecular features of PGC and GSC differ [17][18][19][20] as confirmed by the present study. As one example, the high expression rate of PD-L1 in the present study suggests that immune checkpoint inhibitors would be effective for treatment of unresectable GSC, especially GSC at the anastomotic area or EBV-positive GSC, by restoring an effective lymphocyte response against TC 36,37 . Moreover, molecularly targeted therapies against PI3K and the TGF-β axis may be candidates for treatment of advanced GSC, although therapeutic strategies for targeting mutations in PTEN and SMAD4 have not been established for gastric cancer. Although further research is needed, the present results may contribute to the development of therapies specific for GSC in the future.
Our study has several limitations. First, we examined GSC, but not PGC. Therefore, we employed the molecular profiles of gastric adenocarcinomas from TCGA, by which the recent consensus of gastric cancer molecular subtypes was established. Second, we performed IHC using TMAs and were unable to exclude the possibility that the data reflected tumour heterogeneity. Third, not only bile reflux-induced inflammation, but also many other factors may affect the development of GSC. Therefore, our results may not simply reflect the influence of bile reflux-induced inflammation. Despite these limitations, we believe that this is the first report to elucidate molecular characteristics of GSC with a relatively large number of patients and contributes to revealing the molecular characteristics of GSC.
In conclusion, the present study revealed the molecular and pathological characteristics of GSC, especially a high frequency of PD-L1 expression and EBV positivity as well as PTEN and SMAD4 mutations. GSC can be categorised as a specific entity of gastric cancer and therapeutic strategies for GSC may be developed in accordance with the molecular and pathological characteristics as suggested in the present study.

Methods
Patients and clinical data. We enrolled 102 patients who underwent gastrectomy for GSC between 1998 and 2016 at the National Cancer Center Hospital East. No patients underwent systemic chemotherapy before surgery. We excluded 13 patients without tumour samples. A total of 89 patients were included in the present study (Fig. 2).
We retrospectively reviewed patients' clinicopathological factors included in their medical records. Pathological T and N factors were determined according to the 8th edition of the TNM classification 38 . Histopathological types were classified in accordance with the Japanese classification of gastric carcinoma 39 . The Institutional Review Board of the National Cancer Center, Japan, approved this study (IRB file no. 2017-114; approval date: October 16, 2017). All procedures conformed with the standards of the Declaration of Helsinki and current ethical guidelines. Informed consent was obtained from all participants included in the study.
Tissue microarrays. We reviewed haematoxylin and eosin (H&E)-stained sections prepared for pathological diagnosis before constructing tissue microarrays (TMAs). A set of TMAs from 89 patients were constructed as follows. Two representative tumour cores (2 mm in diameter) were obtained from formalin-fixed, paraffinembedded (FFPE) tissue blocks representative of the lesions. Then, the cores were embedded in paraffin and serial 4 µm-thick sections were prepared for H&E staining, IHC, and ISH.
The staining intensity of EGFR was graded on a scale from 0 to 3+ (0, no staining; 1, faint staining; 2, weak or moderate staining; 3, strong staining). An IHC score of 3+ was defined as positive and IHC scores of 0, 1+, and 2+ were defined as negative in accordance with a previous report 26 .
The HER2 score was evaluated in accordance with Hofmann's criteria 40 . For cases with equivocal (2+) HER2 staining, dual colour in situ hybridisation (DISH) was performed using an INFORM Dual ISH HER2 kit (Ventana). IHC scores of 3+ or 2+ with an HER2:CEP17 (centromeric probe 17) ratio of ≥ 2.0 were defined as positive.
Deficient MMR was defined as complete loss of any of the following MMR genes in TCs: MLH1, MSH2, PMS2, or MSH6.
All specimens were reviewed by M.W. and T.K. If expression scores differed between two cores, the higher score was selected. Representative images are shown in Fig. 3.  Table 4). Subsequently, amplicons were barcoded using an Ion Xpress Barcode Adaptors 1-16 Kit (ThermoFisher Scientific). Emulsion PCR was performed using an Ion OneTouch Dx with an Ion PGM Hi-Q View OT2 Kit (ThermoFisher Scientific). Template-positive ion www.nature.com/scientificreports/ spheres were concentrated using an Ion OneTouch ES Dx (ThermoFisher Scientific) and loaded onto an Ion PGM 318 Select Chip v2 (ThermoFisher Scientific). Sequencing was performed using an Ion torrent PGM Dx with an Ion PGM Hi-Q View Sequencing Kit. Data were analysed using Ion Reporter 5.4 (ThermoFisher Scientific). Mutations were filtered in accordance with the following criteria: (i) allele frequency of > 5% and (ii) minor allele frequency of < 5%. The minimum number of reads was 6 and the minimum allele frequency was 0.01. All mutations were manually curated by referring to mutation databases OncoKB 41 and ClinVar 42 . Briefly, if a mutation was described as (i) "pathological" or "likely pathological" in ClinVar or (ii) "gain-of-function (loss-of-function)" or "likely gain-of-function (likely loss-of-function)" in OncoKB, it was defined as pathological. If a mutation in a tumour suppressor gene caused a "frameshift" or "nonsense" mutation, it was defined as pathological.
The reference molecular profiles of PGC were obtained from the TGCA dataset 21 . Briefly, mutated gene profiles were obtained from the "Stomach Adenocarcinoma (TCGA, Nature 2014)" dataset. We excluded stage X and IV cases, and only profiles of stage I-III subsets were examined to calculate the mutation frequencies of corresponding genes. Statistical analysis. Statistical analyses were performed using JMP version 11 (SAS Institute, Cary, NC).
The Fisher's exact test was used to compare categorical valuables. P-values of < 0.05 are considered statistically significant.