The participants were recruited at Chung-Ang University Hospital between 2017 and 2022. The inclusion criteria for the neoplasm group were patients newly diagnosed with histologically confirmed GC or adenoma, whereas the control group consisted of patients with gastritis only without evidence of GC or adenoma on endoscopic examination. Patients with a history of malignancy, those who underwent prior gastric surgery, those associated with a recent use of antibiotics or probiotics within the last 3 months, or those under 20 years of age were excluded. The final analysis group of 141 participants comprised 132 patients and nine controls. This case-control study was approved by the Institutional Review Board of the Chung-Ang University Hospital (IRB No. C2016047[1790]) and was conducted in accordance with the Declaration of Helsinki. Informed consent was obtained from all participants.

The Vienna Classification System was used to diagnose and categorize gastric neoplasms [8]. Category 3 (non-invasive low-grade adenoma/dysplasia) was classified as the low-grade dysplasia (LGD) group. Category 4.1 (high-grade adenoma/dysplasia) and category 4.2 (non-invasive carcinoma [carcinoma in situ]) were classified as the high-grade dysplasia (HGD) group. Category 4.3 (suspicion of invasive carcinoma) and 5 (invasive neoplasia) were classified as the GC group.

A minimum of 10 mL of saliva was collected by allowing it to accumulate on the floor of the mouth, followed by spitting it into a specimen tube. Samples were stored at −80°C until use. Gastric juice samples were obtained according to the following protocol: Patients fasted for more than 8 h before sample collection. For patients who underwent endoscopic examination or resection, a trap tube was connected between the endoscope and the suction tube before the procedure. Gastric juice (7–30 mL) was collected by suction through the endoscope at the beginning of the procedure. Gastric tissues were collected from the same gastric sites in both the patient and control groups, following the same protocol. Two samples were obtained from the antrum and two from the body, resulting in a total of four tissue samples, which were then stored at −80°C. The endoscopes were washed and disinfected to avoid contamination by immersion in a detergent solution containing 7% proteolytic enzymes and 2% glutaraldehyde before use. Following the collection of gastric juice, the samples were kept at −20°C and immediately transferred to the nearby laboratory, without the use of preservative reagents. The collected gastric juice was diluted to 40 mM by the addition of 1 M Tris base and centrifuged at 600 × g for 10 minutes at 4°C to collect the supernatant. After transfer to a new 15 mL tube, it was again centrifuged at 1,500 × g for 10 minutes at 4°C. The supernatant was subsequently harvested and stored at −80°C.

All samples were diluted in 10 mL of PBS and incubated for 24 hours prior to DNA extraction. After the centrifugation step at 10,000 × g for 10 minutes at 4°C, the pellet containing the bacteria was diluted with 200 μL of PBS to create a suspension. Microbial genomic DNA was extracted using a DNeasy PowerSoil kit (QIAGEN, Hilden, Germany) according to the standard protocol provided in the manufacturer’s instructions. The isolated genomic DNA was amplified by targeting the V3–V4 hypervariable regions of the 16S rRNA gene. This amplification was carried out using specific primers (16S_V3_F: 5′-TCGTCGGCAGCGTCAGATGTGTATAA GAGACAGCCTACGGGNGGCWGCAG- 3′ and 16S_V4_ R: 5′-GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGACTACHVGGGTATCTAATCC- 3′). Subsequently, amplicon libraries were prepared from the amplified DNA fragments. The MiSeq Reagent Kit v3 (600-cycle) (Illumina, San Diego, CA, USA) was used to sequence the library reagents. The Nextra XT Index Kit v2 Set A (96 indices, 384 samples) (Illumina) was used for barcodes and adapters. Library preparation for sequencing followed the 16S Metagenomic Sequencing Library Preparation (Part # 15,044,223 Rev. B). All amplicons were sequenced using MiSeq (Illumina) according to the manufacturer’s instructions. The bacterial DNA in each sample was quantified using a QIAxpert system (QIAGEN).

Paired-end 16S rRNA gene sequences were input into the Quantitative Insights into Microbial Ecology software (QIIME2 v2021. 4) [9]. Adapter sequences were removed using cutadapt. Reads were filtered for quality and chimeric reads using dada2 with manual parameters (trim-left-f 0, trim-left-r 0, trunc-len-f 260, trunc-len-r 200, trunc-q 2, max-ee-f 3, max-ee-r 3) [10]. Taxonomic classification was performed using a Naïve Bayes classifier trained on the extracted V3–V4 region from the SILVA 138 database. All sequences classified as either chloroplasts or mitochondria were removed.

For alpha diversity analysis, samples were rarefied to the minimum read number (1,969) to normalize the read counts. Alpha diversity was further examined by adjusting for age and sex using analysis of variance (ANOVA). Principal coordinate analysis (PCoA) was conducted to determine the individual taxa-level clustering of groups based on the Bray–Curtis dissimilarity distance and weighted UniFrac distance. The p value for the PCoA was calculated by permutational multivariate analysis of variance (PERMANOVA) using dissimilarity matrices. Additionally, we verified the PCoA after adjusting for age and sex using PERMANOVA. To analyze the differences in microbiome composition and functional profiles between groups, the Wilcoxon rank-sum test was performed. Linear discriminant analysis (LDA) Effect Size (LEfSe) was also used to determine the differentially abundant taxa between the groups for the selection of biomarkers with statistical and biological significance. When converting the feature table to LEfSe format, 100,000 was used as the normalization value. The LEfSe algorithm utilized the Kruskal–Wallis test with a cutoff LDA score (log10) of 3.5 for bacteria and 0.25 for functional profiles. During the LEfSe analysis, the false discovery rate (FDR) was applied to further validate the p values. This value represents the q-value and was examined to determine the significance of the results and assess false positives. Additionally, an analysis was conducted using thse multivariate association with linear models (MaAsLin2) package in R, adjusting for age and sex as covariates. A q-value of less than 0.25, adjusted for the FDR using the Benjamini–Hochberg method, was considered statistically significant. This threshold is the default significance threshold in MaAsLin2, and is commonly used in microbiome studies [11]. To reduce noise, taxa present in less than 10% of all samples with a mean relative abundance of less than 0.01% were removed before conducting LEfSe and MaAsLin2 analyses. For the analysis of the relative abundance and functional profile prediction, the Total-Sum Scaling method was used for normalization. Spearman’s rank correlation coefficient was used to analyze the correlation between the microbiota compositions of all samples. Spearman’s rho and p values were calculated using the cor.test in R [12]. The analyzed correlation coefficient value (rs) was found to be statistically significant when the p value was < 0.05, indicating a meaningful relationship. We performed multiple testing corrections for the Wilcoxon rank-sum test using the Benjamini–Hochberg procedure to control for the FDR. The functional profiles of the gastric and oral microbiomes of the control and each disease stage were predicted using Tax4Fun2 [7]. All analyses were conducted using R Statistical Software (version 3.6.1; R Foundation for Statistical Computing, Vienna, Austria).

Samples were collected from 141 patients (nine controls, 58 with LGD, 33 with HGD, and 41 with GC). The study participants had a mean age of 64.4 ± 11.9 years, with a male proportion of 70.9% (100/141). The clinical characteristics of the enrolled patients according to each sample type are summarized in Table 1.

We analyzed the microbial diversity in the microbiota of the gastric juice among the four groups. In the alpha diversity analysis of the gastric juice, the Simpson indices were significantly lower in the HGD and GC groups than in the LGD and control groups (p < 0.05; Fig. 1A). In contrast, there was no significant difference in the alpha diversity of the microbial composition within the gastric tissue (both antrum and body) and oral samples among the four groups.

The beta diversity was assessed between the different stages of GC in the gastric juice, tissue (antrum, body), and saliva samples (Fig. 1B). A significant difference in microbiota composition was confirmed among the four groups of gastric juice samples, and the clustering distinction of the control group was particularly prominent (p < 0.001; Fig. 1B). Regarding the beta diversity, there was also a significant difference in the microbial composition of the gastric tissue among the four groups (antrum, p < 0.05; body, p < 0.001; Fig. 1B). Moreover, no significant differences were found in the beta diversity of the oral microbiota among the four groups (p = 0.463, Fig. 1B).

After adjusting for age and sex, the alpha and beta diversity analyses exhibited numerical fluctuations in certain p-values; however, the overall significance remained consistent both before and after the adjustment. Additionally, to clearly demonstrate the differences in beta diversity based on location corresponding to each stage of gastric carcinogenesis, schematic diagrams were constructed (Fig. 1C). Significant differences were observed in microbiome composition among the different sample types at all stages of gastric carcinogenesis. In particular, the difference in microbial composition between the saliva and gastric juice was more pronounced in the control group than in the disease groups (LGD, HGD, and GC groups).

A difference was observed in the gastric juice microbial community between the control and disease groups (Fig. 2A). The abundance of Pseudomonadota (formerly Proteobacteria) decreased, while that of Bacillota (formerly Firmicutes) and Actinomycetota (formerly Actinobacteria) increased in the gastric juice of the disease group compared to that of the control group. Moreover, a difference in the microbial composition of the tissues between the control and disease groups (both the antrum and the body) was observed. In the tissues, the abundance of Campylobacterota increased in the disease groups compared to that in the control group. In contrast, no significant difference in the oral microbial composition between the control and disease groups was observed. In the saliva, Bacillota, Actinomycetota, and Bacteroidota predominated in both the control and disease groups, and the composition fractions of each group were very similar, with few differences between the groups.

The LEfSe results for gastric juice samples confirmed significant differences in the abundance of specific strains, including Ralstonia insidiosa, Schaalia odontolytica, Rothia mucilaginosa, Peptostreptococcus stomatis, Streptococcus salivarius, H. pylori, Streptococcus sanguinis, Gemella haemolysans, Streptococcus australis, Streptococcus cristatus, Haemophilus parainfluenza, and Prophyromonas pasteri, among others, within each group (Fig. 2B, q < 0.05). Among these characteristic bacteria, the abundance of R. insidiosa decreased in the disease groups compared to the control group (q < 0.01). However, the abundance of other strains increased in the disease groups than in the control group, with the highest proportion observed in the LGD group (q < 0.05). In the gastric fluid sample analysis, adjusting for sex and age resulted in a loss of significance for some bacterial strains. However, the strains S. odontolytica, P. stomatis, and S. cristatus generally maintained their significance even after adjustment for age and sex (Supplementary Table 1, 2). In the tissues (antrum), the abundance of Pseudomonas yamanorum, H. parainfluenza, and S. cristatus differed significantly between each disease group (q < 0.05). Among the three bacterial species, P. yamanorum and H. parainfluenza were more abundant in the HGD group than in the control group, whereas S. cristatus was the most abundant in the LGD group. In the body tissues, the abundance of Prevotella pallens significantly differed in each disease group (q < 0.05). The abundance increased in the disease group compared to the control group, with the highest observed in the LGD group. In the oral samples, the abundance of Prevotella melanionogenica, G. haemolysans, and Rothia aeria showed significant differences among the disease groups (q < 0.05), with G. haemolysans exhibiting a significantly lower proportion in the control group than in the other disease groups (q < 0.05). Moreover, the bacterial strains exhibited significant changes (q < 0.05) following the LEfSe analysis (Supplementary Table 3–8).

The degree of similarity in bacterial composition among the oral, gastric juice, and gastric tissue samples within each individual disease group was assessed using correlation analysis (Fig. 3, Supplementary Table 9). The similarity between the samples increased as the disease progressed. Moreover, the similarity in microbial composition was relatively low among the different sample types in the control group. The similarity in microbial composition between the saliva and gastric juice samples increased from 0.22 in the control group to 0.58, 0.60, and 0.58 in the LGD, HGD, and GC groups, respectively. Similar tendencies were also found between the gastric juice and tissues (antrum and body), as well as between the saliva and tissues (antrum and body), with higher similarity of microbial composition in the GC groups than in the control group (0.42–0.51 vs. 0.30–0.27 and 0.68–0.66 vs. 0.36–0.34, respectively). Meanwhile, the microbial similarity between the antrum and body tissues consistently displayed a stable level of similarity regardless of the disease status. Furthermore, there was a tendency for an increase in the similarity between samples with respect to the stages of gastric carcinogenesis. However, some samples showed a slight decrease in similarity as they progressed from the LGD group, which exhibited the highest similarity, to the HGD and GC groups (GJP-TIA, GJP-TIB, and ORWP-TIB).

When comparing the functional profiles based on microbiome clustering in four distinct samples across the disease stages, we anticipated that changes would primarily occur in signal transduction (Gastric juice: putative ABC transport system permease protein - K02004, two-component system, OmpR family, response regulator - K02483, PTS system, beta-glucoside-specific IIB component - K02756, PTS system, beta-glucoside-specific IIC component - K02757, PTS system, beta-glucoside-specific IIA component - K02755, two-component system, OmpR family, phosphate regulon sensor histidine kinase PhoR - K07636, two-component system, NarL family, response regulator DesR - K07693, and two-component system, chemotaxis family, chemotaxis protein CheY - K03413; Body tissue: ABC-2 type transport system ATP-binding protein - K01990 and ABC- 2 type transport system permease protein - K01992), DNA replication and expression (Gastric juice: glycine cleavage system transcriptional repressor - K03567; Antrum tissue: adenine-specific DNA-methyltransferase - K07316 and DNA (cytosine-5)-methyltransferase 1 - K00558; Body tissue: adenine-specific DNA-methyltransferase - K07316), and ATP-related pathways (Gastric juice: macrolide transport system ATP-binding/permease protein - K18230 and zinc/ manganese transport system ATP-binding protein - K02074; Body tissue: ATP-binding cassette, subfamily B, bacterial - K06147; Saliva: macrolide transport system ATP-binding/ permease protein - K05685) (Fig. 4, q < 0.05). The functional profiles exhibited significant changes (q < 0.05) in the LEfSe analysis, and this significance was maintained by the application of q-values (Supplementary Table 10–13).