A genome‐wide study of gastric intramucosal neoplasia based on somatic copy number alterations, gene mutations, and mRNA expression patterns

Abstract

We performed comprehensive analyses of somatic copy number alterations (SCNAs) and gene expression profiles of gastric intramucosal neoplasia (IMN) using array‐based methods in 97 intestinal‐type IMNs, including 39 low‐grade dysplasias (LGDs), 37 high‐grade dysplasias (HGDs), and 26 intramucosal carcinomas (IMCs) with stromal invasion of the lamina propria to identify the molecular mechanism of IMN. In addition, we examined gene mutations using gene panel analyses. We used cluster analyses for exclusion of arbitrariness to identify SCNA patterns and expression profiles. IMNs were classified into two distinct subgroups (subgroups 1 and 2) based on SCNA patterns. Subgroup 1 showed a genomic stable pattern due to the low frequency of SCNAs, whereas subgroup 2 exhibited a chromosomal instability pattern due to the high frequencies of SCNAs and TP53 mutations. Interestingly, although the frequencies of LGD and HGD were significantly higher in subgroup 1 than in subgroup 2, IMC was commonly found in both types. Although the expression profiles of specific mRNAs could be used to categorise subgroups 1 and 2, no clinicopathological findings correlated with either subgroup. We examined signalling pathways specific to subgroups 1 and 2 to identify the association of each subgroup with signalling pathways based on gene ontology tree visualisation: subgroups 1 and 2 were associated with haem metabolism and chromosomal instability, respectively. These findings reveal a comprehensive genomic landscape that highlights the molecular complexity of IMNs and provide a road map to facilitate our understanding of gastric IMNs.

Introduction

Gastric cancer (GC) is one of the most important causes of cancer‐related death worldwide [1]. Although most GCs are not diagnosed until advanced stages [2], many cases of this disease are discovered as intramucosal neoplasia (IMN), and the molecular alterations of these IMNs can be helpful for evaluating tumorigenesis in early GC [3, 4, 5]. GC can be classified as intestinal or diffuse based on morphological differences [6]. Intestinal‐type GC is typically associated with Helicobacter pylori infection and is common in routine practice [7]. In contrast, diffuse‐type GC is uniformly distributed and exhibits aggressive clinical behaviour, similar to typically scirrhous carcinoma, which is associated with a poor prognosis at advanced stages [8]. Intestinal‐type GC obtained from endoscopic submucosal resection can be useful for identification of early tumorigenesis of GC [3, 5, 9]. Despite the cumulative molecular evidence of intestinal‐type GC, detailed genomic‐scale data for IMNs, the earliest phase of GC, are limited because IMNs can exhibit various histological types [3, 10].

Somatic copy number alterations (SCNAs) are genomic structural alterations that result in abnormal gene copy numbers, including gene amplifications, gains, losses, and deletions [11]. SCNAs may influence expression of protein‐coding and non‐coding genes, affecting the activity of various signalling pathways [11]. Additionally, SCNAs are closely associated with neoplastic progression in various cancers, including oesophageal, gastric, colorectal, ovarian, and lung cancers [11, 12]. SCNAs arise as a result of preferential selection that favours cancer development; thus, activation of enhanced ‘driver genes’ in GC may enable novel opportunities for personalised therapy and also evaluation of the molecular mechanism of early carcinogenesis [11, 12, 13].

Genome‐wide differential expression patterns of mRNAs may facilitate the evaluation of carcinogenesis [14, 15, 16]. Recent data on gene expression profiling of various cancers, including gastric, colorectal, and ovarian cancers, improve our understanding of the hypothesis that cancer cells arise through distinct signalling pathways [14, 15, 16]. Although several large‐scale gene expression studies with cDNAs or oligonucleotide arrays have been performed in GC, these studies employed platforms that varied in the number and identity of the arrayed genes [15].

Here, we aimed to determine whether SCNAs could contribute to the development of intestinal‐type IMNs using a single nucleotide polymorphism (SNP) array suitable for genome‐wide evaluation. Additionally, we examined gene mutations in IMNs using a gene panel. Finally, we assessed the differential expression profiles of mRNAs in IMNs using an expression array to comprehensively elucidate the mechanisms of early gastric carcinogenesis.

Materials and methods

Patients

Gastric IMNs, including low‐grade dysplasia (LGD), high‐grade dysplasia (HGD), and intramucosal cancer (IMC), were examined at the Department of Molecular Diagnostic Pathology, Iwate Medical University from 2019 to 2022 [3]. Pathological diagnoses were made according to World Health Organisation criteria with slight modifications [17]. The pathological factors analysed included sex, age, tumour location, size, tumour subtype (based on tumour grade), and differentiation grade according to the Japanese Gastric Cancer Association seventh edition [18]. No patients had undergone chemo‐ or radiotherapy before endoscopic submucosal resection. Patients with a family history or medical history of GC were excluded. We excluded IMNs with a microsatellite instability (MSI)‐ or Epstein–Barr virus (EBV)‐related phenotype owing to the different pathogenesis of intestinal IMNs. Detailed clinicopathological findings are shown in Table 1. Informed consent was obtained from each patient according to institutional guidelines, and the research protocols were approved by the ethics committee of Iwate Medical University Hospital (reference number: HG2019‐149).

Table 1.

Clinicopathological findings of the intramucosal neoplasia cases

	Total
Total	97
Sex
Male	83
Female	14
Age, median [range] (years)	72 [44–87]
Locus
U/M/L	19/30/48
Size	16 [4–60]
Subtype
Low‐grade dysplasia	39
High‐grade dysplasia	32
Intramucosal cancer	26
Differentiation
Well differentiated	79
Moderately differentiated	17
Papillary type	1
Mucin type
Gastric	18
Large intestinal	7
Small intestinal	29
Mixed	43

U, upper portion; M, middle portion; L, lower portion.

Lesion sampling and DNA extraction

Tumour samples were obtained from resected stomach tissues using biopsy forceps within 10–20 min of resection. Normal gastric mucosa distant from the tumour was removed from the mucosa using biopsy forceps; as a control (germline data), gastric biopsy samples from patients with IMN with chronic gastritis were included. Tumour tissues for pathological analysis were obtained from a region of the resected stomach adjacent to the site used for molecular analysis (one sample was obtained as a representative sample). In addition, the tumour tissue samples examined were surrounded by tumour tissue on all sides to ensure the presence of tumour tissue in the sample (Figure 1). In this section, the proportion of tumour cells accounted for at least 50% of the tissue. If other histological components were contained in the histological section, genomic DNA and RNA were extracted from whole tumour tissues. In brief, microdissected tissues were incubated at 56 °C for 12–18 h in 50 μL buffer containing 0.5% Tween‐20 (Boehringer Mannheim, Mannheim, Germany), 20 μg proteinase K (Boehringer Mannheim), 50 mm Trizma base (pH 8.9), and 2 mm ethylenediaminetetraacetic acid. Proteinase K was inactivated by incubating the samples at 100 °C for 10 min. A representative example is shown in Figure 1.

Figure 1.

Representative images of macroscopic and histological features. (A) Macroscopic image, (B) loupe images of slices corresponding to lines a–d in (A), (C) low‐grade dysplasia, (D) high‐grade dysplasia, and (E) intramucosal carcinoma. Note that the tumour samples obtained for molecular analysis were surrounded by tumour tissues on all sides, ensuring the quality and quantity of the examined tumour tissue samples.

RNA extraction

Total RNA was isolated with RNeasy Mini kits (Qiagen, Valencia, CA, USA) per the manufacturer's instructions using the same site of sample collection for DNA analysis. The nucleic acid concentration was determined with a Nanodrop1000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA), and purity was verified on 1.5% agarose denaturing gels.

Allele‐specific copy number analysis of tumours (ASCAT)

We used a unique bioinformatics approach, ASCAT, to determine the allele‐specific copy number in solid tumours, simultaneously estimating and adjusting for both tumour ploidy and non‐aberrant cell mixture [19]. A given tumour sample can be classified into nMjor and nMinor according to a previous study [19]. This classification allows determination of genome‐wide allele‐specific copy number profiles, from which gains, losses, loss of heterozygosity (LOH), and copy neutral LOH (CN‐LOH) can accurately be determined. In the present study, measurements from individual SNP probes were aggregated to segments of unchanged allele‐specific copy numbers using ASCAT v3.1.0 via R v4.2.2 to perform deconvolution to remove the influence of DNA/RNA derived from the stromal component using ASCAT and DeMixT.

SNP array analysis

We examined SCNAs using the CytoScan 750K platform (Thermo Fisher Scientific), which contains more than 550,000 non‐polymorphic markers and over 200,000 SNP markers, with an average intragenic marker spacing of 1,737 bp and intergenic marker spacing of 6,145 bp. These platforms are composed of microarrays containing non‐polymorphic probes for copy number variations from coding and non‐coding regions of the human genome as well as polymorphic SNP probes. All procedures were performed according to the manufacturer's instructions. Slides were analysed with a GeneChip Scanner 3000 7G (Thermo Fisher Scientific). CytoScan raw data files were pre‐processed using Affymetrix Power Tools (Thermo Fisher Scientific). The detailed method is described elsewhere [20].

Classification of SCNAs

We classified SCNA patterns into three categories: gain, LOH, and CN‐LOH. LOH was considered a gross chromosomal change resulting in loss of the entire gene and surrounding region; gain was defined as a gross chromosomal change caused by a gain of the entire gene and surrounding region. CN‐LOH was defined as LOH without a copy number change (CN = 2). Detailed classification criteria are described elsewhere [3].

Clariom S Human Array and gene expression analysis

For each array experiment, 500 ng total RNA was used for labelling with the Clariom S Human Array (Thermo Fisher Scientific), which contains 21,453 mRNAs. Probe labelling, chip hybridisation, and scanning were performed according to the manufacturer's instructions. A Probe Set (gene‐exon) was considered expressed if at least 50% of the samples were detected above background (DABG) values below the DABG threshold (DABG < 0.05). Array data were processed using oligo v1.62.2, and transcriptome deconvolution was performed using DeMixT v1.14.0 via R v4.2.2. The detailed method is described previously [21].

Library preparation for next‐generation sequencing (NGS)

NGS libraries were prepared using a custom panel (Illumina, San Diego, CA, USA) containing 753 amplicons covering 82 exonic regions across 28 genes that may play a role in the development of intestinal type IMN, including LGD, HGD, and IMC (APC, BRAF, TP53, CDKN2A, MET, ATM, MLH‐1, PMS2, HRAS, AXIN2, BAX, DCC, MSH2, POLE, RNF43, PTEN, EPCAM, MSH6, BUB1B, RHOA, KRAS, NRAS, SMAD4, CDK4, PIK3CA, STK11, TGFBR2, and EGFR).

Analysis, quality metrics, and variant detection

Integrated analysis software (Illumina) was used for image analysis, base calling, and assignment of quality scores, which were automatically performed for primary analysis. The sequencing method and analyses are described in Supplementary materials and methods.

Detected variants that did not satisfy our previously described criteria or exhibited strand bias were further assessed during interpretation. A pathologist (NU or TS) familiar with the gene sequencing technology evaluated variants by identifying missense, frameshift, stop‐gained, or in‐frame insertion‐ or deletion‐affected sequences. Variant classification was performed using both the ClinVar (http://www.ncbi.nlm.nih.gov/clinvar) and COSMIC (http://cancer.sanger.ac.uk/cosmic) databases. Pathogenic and likely pathogenic variants were determined according to the above databases.

Additional methodological detail is provided in Supplementary materials and methods.

Statistical analysis

Differences among groups in clinicopathological variables were analysed using Fisher's exact test with statistical software (JMP Pro 16.1 software package for Windows; SAS, Tokyo, Japan). Differences in age and tumour size distributions were evaluated using Mann–Whitney U tests in JMP Pro 16.1 (SAS).

Differences in SCNA numbers, including gains, LOH, and CN‐LOH, were evaluated using the Kruskal–Wallis test in JMP Pro 16.1 (SAS). Differences in SCNAs among IMNs, LGDs, HGDs, and IMCs were analysed using Fisher's exact tests. Results with p values less than 0.05 were considered significant. p values were adjusted with the Benjamini–Hochberg false discovery rate (FDR) for multiple comparisons.

Differences in mRNA expression levels between subgroups 1 and 2 or among LGDs, HGDs, and IMCs were analysed using limma v3.54.1 via R v4.2.2 (FDR p value <0.05 and logFold change ≥|1|).

Results

All tumours were categorised as the microsatellite stable (MSS) phenotype. Eleven patients with the MSI phenotype were excluded from the study. No tumours with an EBV phenotype were found. Figure 2 shows the workflow of the study.

Figure 2.

Workflow in the current study.

Unsupervised hierarchical clustering analysis based on the copy number alteration (CNA) patterns in IMNs with the MSS phenotype

We sub‐classified IMNs with the MSS phenotype into two subgroups according to SCNA patterns (Figure 3A; subgroup 1, 71 tumours; subgroup 2, 26 tumours) using unsupervised hierarchical clustering analysis. Clinicopathological analyses showed that the frequencies of LGD and HGD were higher in tumours in subgroup 1 (LGD, 89.7%, 35/39 tumours; HGD, 75%, 24/32 tumours) than in subgroup 2 (LGD, 10.3%; 4/39 tumours; HGD, 25%, 8/32 tumours) (Table 2). However, there were no significant differences in the frequencies of IMC between subgroup 1 (46.2%; 12/26 tumours) and subgroup 2 (53.8%, 14/26 tumours) (Table 2). Detailed data are shown in Table 2.

Figure 3.

(A) Hierarchical clustering analysis based on SCNAs. wel, well differentiated; mod, moderately differentiated; pap, papillary; LGD, low‐grade dysplasia; HGD, high‐grade dysplasia; IMC, intramucosal cancer; U, upper portion; M, middle portion; L, lower portion. (B) Comparison of the numbers of genes with any SCNA, gains, CN‐LOH, and LOH in each subgroup of 97 gastric IMNs with the MSS phenotype. (a) Comparison of the number of genes with gains between subgroups 1 and 2. (b) Comparison of the number of genes with LOH between subgroups 1 and 2. (c) Comparison of the number of genes with CN‐LOH between subgroups 1 and 2. (d) Comparison of the number of total genes with any SCNA between subgroups 1 and 2.

Table 2.

Comparison of clinicopathological findings between subgroups based on the SCNA patterns produced by hierarchical clustering analysis

	Total	Subgroup 1 (%)	Subgroup 2 (%)	p value
Total	97	71 (73.2)	26 (26.8)
Sex				0.3399
Male	83	59 (71.1)	24 (28.9)
Female	14	12 (85.7)	2 (14.3)
Age, median [range] (years)	72 [44–87]	72 [44–87]	74 [61–85]	0.4078
Locus				0.2899
U/M/L	19/30/48	14/25/32	5/5/16
Size	16 [4–60]	16 [5–60]	18.5 [4–50]	0.3298
Subtype				0.0006
LGD	39	35 (89.7)*	4 (10.3)*
HGD	32	24 (75) †	8 (25) †
IMC	26	12 (46.2)	14 (53.8)
Differentiation				0.0275
Well	79	62 (78.5)*	17 (21.5)*
Moderately	17	8 (47.1)	9 (52.9)
Papillary	1	1 (100)	0 (0)
Mucin type				0.1735
Gastric	18	14 (77.8)	4 (22.2)
Large intestinal	7	3 (42.9)	4 (57.1)
Small intestinal	29	24 (82.8)	5 (17.2)
Mixed	43	30 (69.8)	13 (30.2)

U, upper portion; M, middle portion; L, lower portion; LGD, low‐grade dysplasia; HGD, high‐grade dysplasia; IMC, intramucosal cancer; Well, well differentiated; Moderately, moderately differentiated.

^*Bonferroni adjusted p < 0.01.

^† Bonferroni adjusted p < 0.05.

The median total number of genes with CNAs per patient was 6,856, with a median of 2,340 gains (range: 0–40,191), 0 LOHs (range: 0–9,031), and 1,762 CN‐LOHs (range: 0–9,119) among all patients. The median total number of genes with CNAs per patient was 3,725, with a median of 1,046 gains (range: 0–16,465), 0 LOHs (range: 0–9,031), and 1,835 CN‐LOHs (range: 0–7,552) in subgroup 1. In addition, the median total number of genes with CNAs per patient was 40,110, with a median of 40,010.5 gains (range: 10,025–40,191), 0 LOHs (range: 0–6,495), and 0 CN‐LOHs (range: 0–9,119) in subgroup 2. There were significant differences in the total numbers of SCNAs between subgroups 1 and 2 (Figure 3B; p < 0.001). Moreover, significant differences in the average number of copy number gains among subgroups 1 and 2 were found (Figure 3B; p < 0.001). Although LOH was common in the two subgroups, the average number of CN‐LOHs was significantly higher in subgroup 1 than in subgroup 2 (Figure 3B; p = 0.023).

There were significant differences in the total numbers of SCNAs between LGD and HGD and between LGD and IMC (supplementary material, Figure S1; p = 0.0030, p = 0.0051, respectively). Additionally, the total number of copy number gains was significantly higher in LGD than in IMC or HGD (supplementary material, Figure S1; p = 0.0006, p = 0.0015, respectively). CN‐LOH and LOH were common events among LGD, HGD, and IMC (supplementary material, Figure S1).

Regions of SCNAs detected in more than 30% of cases are summarised in supplementary material, Tables S1 and S2. Allelic loci specific to each subgroup were identified; there were many characteristic regions of SCNAs specific to each subgroup. Detailed data are summarised in supplementary material, Tables S1 and S2. Differences in regions of SCNA with more than 30% cases between each subgroup are listed in supplementary material, Table S3. Regions of SCNAs detected in more than 30% of cases in LGD, HGD, and MC are shown in supplementary material, Tables S4–S6.

Gene mutation frequencies (APC, KRAS, and SMAD4) were not different between tumours in subgroups 1 (TP53: 12/71, 16.9%; APC: 14/71, 19.7%; KRAS: 6/71, 8.5%; SMAD4: 0/71, 0%) and 2 (TP53: 10/26, 38.5%; APC: 2/26, 7.7%; KRAS: 0/26, 0%; SMAD4: 1/26, 3.8%). However, there was a significant difference in TP53 mutations between subgroups 1 and 2. Detailed data for mutations are depicted in supplementary material, Table S7.

Associations of significant SCNAs with LGD, HGD, and IMC

We examined significant differences in the frequencies of SCNA regions among LGD, HGD, and IMC. Significant differences in the frequency of gains among LGD, HGD and IMC were found (supplementary material, Table S8). There were only two regions with a significantly different frequency of gains between LGD and HGD and between LGD and IMC: 6p12.3–q27 and 7q11.21–q36.3. However, no significant difference in the frequency of SCNAs between HGD and IMC was observed. We also investigated candidate oncogenes mapping to the chromosomal regions, presented in supplementary material, Table S9. Although no candidate oncogene was found at 6p12.3–q27, three candidate oncogenes, MET proto‐oncogene, receptor tyrosine kinase (MET), smoothened, frizzled class receptor (SMO), and transformation/transcription domain associated protein (TRRAP), were detected at 7q11.21–q36.3. Finally, no common SCNAs were found among LGD, HGD, and IMC.

Driver genes associated with gastric carcinogenesis exhibiting gains

We examined previous comprehensive data [22] to identify oncogenic driver genes associated with gastric carcinogenesis exhibiting significant gene gains. According to those data, we found previously known (TP53 [17p13.1], ARID1A [1p36.11], and CDH1 [16q22.1]) and novel (MUC6 [11p15.5], CTNNA2 [2p12], GLI3 [7p14.1], RNF43 [17q22], and others) mutated driver genes. Among these candidate genes, MUC6, CTNNA2, and GLI3 were located in the regions found to have gains.

Transcriptome analysis of gastric IMN

We compared differentially expressed mRNAs from tumour samples with those of normal mucosa samples in each subgroup separately. We identified 1,158 differentially expressed mRNAs (1,104 upregulated and 54 downregulated) in subgroup 1 tumour samples (Volcano plot; supplementary material, Figure S2A). In contrast, we found 1,137 differentially expressed mRNAs (1,030 upregulated and 107 downregulated) in subgroup 2 tumour samples (Volcano plot; supplementary material, Figure S2B). Additionally, we examined mRNA expression patterns for significant mRNA expression (subgroup 1: 1,158; subgroup 2: 1,137) obtained from Volcano plot analysis using unsupervised hierarchical clustering analysis. Cluster analysis identified two distinct subgroups based on mRNA expression patterns (Figure 4). No significant differences in clinicopathological findings were identified between these subgroups (supplementary material, Table S10).

Figure 4.

Hierarchical clustering analysis based on the expression patterns of mRNAs. wel, well differentiated; mod, moderately differentiated; pap, papillary; LGD, low‐grade dysplasia; HGD, high‐grade dysplasia; IMC, intramucosal cancer; U, upper portion; M, middle portion; L, lower portion.

Next, we investigated the expression of mRNAs closely associated with activities relevant to cancer using gene ontology (GO) analysis. We found 79 and 27 GO terms from subgroups 1 and 2, respectively. Using these GO terms, we created tree views associated with each subgroup (Figure 5). Finally, we obtained GO trees for visualisation using GO terms specific to the expression of mRNAs from subgroup 1 (subgroup 1 GO tree; Figure 5A) and subgroup 2 (subgroup 2 GO tree; Figure 5B). The subgroup 1 GO tree could be divided into several functions, including cellular iron ion homeostasis, RNA stabilisation, ubiquitin‐dependent endoplasmic‐reticulum‐associated protein degradation pathway, and maturation of large subunit rRNA. Using a similar method, the subgroup 2 GO tree could also be divided into several functions, most of which were associated with chromosomal instability.

Figure 5.

GO tree visualisation using GO terms. (A) GO tree visualisation using GO terms specific to subgroup 1. (B) GO tree visualisation using GO terms specific to subgroup 2.

Discussion

SCNA detection in chromosomal DNA from tissue samples may facilitate histological diagnosis and prediction of prognosis [3, 4, 16]. Moreover, SCNAs may be the focus of targeted therapies for GC [11]. In this study, we show that SCNA patterns in IMNs can be classified into two subgroups: subgroup 1 was characterised by a low frequency of SCNAs, whereas subgroup 2 was associated with a high frequency of SCNAs. Thus, subgroup 2 was related to chromosomal instability and may contribute to neoplastic progression, and subgroup 1 was related to genomic stability, which was assumed to have different pathogenic molecular alterations compared with chromosomal instability type alterations, such as epigenetic alterations [16]. This hypothesis is supported by the finding that SCNA changes (total number of SCNAs, total number of CN gains, and total number of CN‐LOHs) were significantly higher in subgroup 2 than in subgroup 1 [16]. Approximately 70% and 30% of cases were assigned to subgroups 1 and 2, respectively. Thus, approximately 30% of IMNs may have an aggressive type. Although the frequencies of LGD and HGD differed significantly between subgroups 1 and 2 (subgroup 1 > 2), there were no significant differences in the frequency of IMC. This finding may be interesting given that tumours in subgroup 1 were a mixture of various histological grades, whereas tumours in subgroup 2 were primarily composed of HGD and IMC. Accordingly, although the low SCNA frequency was associated with the indolent type, tumours in subgroup 1 were not all benign [3, 10].

We examined the driver genes responsible for gastric carcinogenesis identified in a previous comprehensive study [22]. Using those data, we found previously known (TP53, ARID1A, and CDH1) and novel (MUC6, CTNNA2, GLI3, and RNF43) mutated driver genes associated with gastric carcinogenesis. Among these candidate genes, MUC6, CTNNA2, and GLI3 mapped to the regions displaying gains identified in this study. However, high expression of these genes should be confirmed by RT‐PCR or immunohistochemistry to confirm that they are associated with the corresponding gains.

We examined correlations between the significant SCNAs and each grade of IMN (LGD, HGD, and IMC). We found two regions exhibiting a significantly different gain frequency between LGD and HGD and between LGD and IMC: 6p12.3–q27 and 7q11.21–q36.3. Three candidate oncogenes (MET, SMO, and TRRAP) were located at 7q11.21–q36.3. Among these three genes, MET and TRRAP are amplified [23, 24]. In a systematic review with meta‐analysis, higher MET amplification and expression in GC were reported to be an indicator of poor prognosis, although contradictory data have been found [23]. TRRAP is mutated or amplified in different cancer types, including melanoma, GCs, and uterine cancers, and such alterations have been implicated in oncogenic transformation [23, 24]. Those results suggest an important role of TRRAP and its cofactor KAT5, which potentially play roles in cancer initiation or progression by activating downstream mitotic genes in cancer cell growth [25]. The predictive role of MET and TRRAP amplification in gastric tumorigenesis should be further examined in clinicopathological studies.

In this study, no significant differences in the numbers of SCNAs between HGD and IMC were found. Therefore, assessment of SCNAs may not help to make a differential diagnosis between HGD and IMC, which is important in routine practice. This may be due to the small sample number in this study. Additional studies with more cases are needed. Furthermore, the current findings implied that HGD and IMC might need to be classified into the same category. However, more studies are needed to reach such a conclusion. Nevertheless, we suggest that analysis of SCNAs will improve understanding of the tumorigenesis of IMNs.

We investigated the expression pattern of mRNAs in IMNs using cluster analysis. Although the mRNA expression patterns of subgroups 1 and 2 differed, no significant differences in the frequencies of clinicopathological findings were found. Thus, distinct expression patterns may be necessary for the development of IMNs, even if there are no associations with clinicopathological findings. Additionally, analysis of the GO tree visualisation specific to subgroup 1 showed that cellular iron ion homeostasis was closely associated with subgroup 1, consistent with a recent study demonstrating that this function is closely associated with many biology processes, including cell proliferation, metabolism, and mitochondrial function [26, 27]. Disruption of iron homeostasis promotes production of reactive oxygen species (ROS), which enhance tumour development [26]. The mechanisms and regulation of iron homeostasis and iron‐mediated ROS in tumours may play important roles in the development of IMNs with a genome stable phenotype [26, 27]. In addition, a previous study showed that haem oxygenase 1, a key gene in the haem pathway, is closely associated with antioxidation, anti‐apoptosis, and anti‐proliferation in GC [28]. In contrast, pathways related to chromosomal instability were observed in subgroup 2. This hypothesis is supported by the finding that IMNs assigned to subgroup 2 were characterised by the chromosomal instability type with accumulation of SCNAs [3, 16, 29]. Thus, we suggest that there may be two pathways, i.e. iron ion homeostasis and chromosomal instability, involved in the development of IMNs.

Mutation of the TP53 gene was frequently observed in subgroup 2, which showed a high frequency of SCNAs [3, 5]. This finding suggests that TP53 mutations enhance the genomic instability occurring in tumour cells, as supported by a previous study [30]. APC mutations may also have a minor role in the development of IMNs, although findings of previous studies have been contradictory [31, 32, 33]. In the current study, the genes we evaluated that may cause driver mutations were scarce in GC. Although numerous gene mutations are observed in GCs [16], most mutated genes may exhibit the passenger effect [16]. To identify the roles of gene mutations in the development of IMNs, whole‐genome mutation analysis may be more suitable than panel sequencing.

Previous studies have shown that the mucin phenotype plays important roles in early gastric carcinogenesis [34, 35, 36]. Moreover, there are several genetic pathways in GC based on the mucin phenotype [34, 35, 36]. However, our current findings show that the mucin phenotype plays a minor role in early gastric carcinogenesis and that the mucin phenotype is not significantly associated with SCNA patterns. Additionally, although we examined the mRNA expression patterns of the tumours using cluster analysis for exclusion of arbitrariness, significant differences in the frequency of the mucin phenotype were not found between the two subgroups, supporting that the mRNA expression pattern did not affect the mucin phenotype. These findings suggest that the mucin phenotype does not play a role in early development in terms of SCNAs and mRNA patterns occurring in IMNs.

There were some limitations to this study. First, the cohort size was small. However, the number of IMN cases was within the minimum number that we calculated before the study, and we believe that the sample number we used was sufficient to identify the molecular carcinogenesis of gastric IMNs. Second, the sample sites for DNA/RNA analyses and histological examinations were different. However, we carefully obtained samples from a region of the resected stomach adjacent to the site used for molecular analysis. Finally, we excluded IMNs with MSI‐ and EBV‐related cancer phenotypes owing to differences in the pathogenesis of intestinal‐type IMNs. GC is a heterogeneous disease [37], and this characteristic may impede identification of the underlying pathogenesis of IMNs. Therefore, we examined intestinal‐type IMNs to focus on evaluation of gastric carcinogenesis.

In conclusion, the current findings shed light on the molecular pathogenesis of GC, with a specific emphasis on SCNAs that display tumour progression in GC. Identification of SCNA patterns by cluster analysis may be helpful for evaluating the development of IMNs. There were two distinct subgroups in terms of SCNA patterns: subgroup 1 exhibited a genome stability type characterised by a low frequency of SCNAs, whereas subgroup 2 exhibited a chromosomal instability type associated with a high frequency of SCNAs. In addition, the expression pattern of mRNAs could also be categorised into two distinct patterns (subgroups 1 and 2). However, such subgroups did not affect the clinicopathological findings, including mucin phenotype and tumour grade. Further analysis of GO terms showed that subgroups 1 and 2 were related to haem metabolism and chromosomal instability, respectively. Additional studies are needed to fully elucidate the mechanisms mediating the molecular carcinogenesis of IMNs.

Author contributions statement

YK supported in the data preparation and interpretation. MO performed all data collection and statistical analyses. RS, NU and NY supported in the data preparation. TM provided clinical support during the preparation of the manuscript. HS assisted with the molecular analyses. TS contributed to the preparation of the manuscript, including all aspects of the data collection and analysis.

Supporting information

Supplementary materials and methods

Figure S1. Comparison of the numbers of genes with SCNAs (gains, CN‐LOH, or LOH) in 97 gastric intramucosal neoplasias with the MSS phenotype

Figure S2. Volcano plots of differential mRNA expression between tumour and normal samples

Table S1. Frequent SCNA regions in subgroup 1

Table S2. Frequent SCNA regions in subgroup 2

Table S3. Significant differences in the frequencies of SCNA regions between subgroups 1 and 2

Table S4. Frequent SCNA regions in LGD

Table S5. Frequent SCNA regions in HGD

Table S6. Frequent SCNAs regions in IMC

Table S7. Gene mutations in intramucosal gastric neoplasia

Table S8. Significant differences in the frequencies of SCNA regions among LGD, HGD, and IMC

Table S9. Oncogenic genes mapping to significant regions exhibiting gene gains based on the Cancer Gene Census (https://cancer.sanger.ac.uk/census)

Table S10. Comparison of clinicopathological findings between subgroups based on mRNA expression pattern in IMNs

Data availability statement

The data that support the findings of our study are available from the corresponding author upon reasonable request.