Gastric microbiota and predicted gene functions are altered after subtotal gastrectomy in patients with gastric cancer

Subtotal gastrectomy (i.e., partial removal of the stomach), a surgical treatment for early-stage distal gastric cancer, is usually accompanied by highly selective vagotomy and Billroth II reconstruction, leading to dramatic changes in the gastric environment. Based on accumulating evidence of a strong link between human gut microbiota and host health, a 2-year follow-up study was conducted to characterize the effects of subtotal gastrectomy. Gastric microbiota and predicted gene functions inferred from 16S rRNA gene sequencing were analyzed before and after surgery. The results demonstrated that gastric microbiota is significantly more diverse after surgery. Ralstonia and Helicobacter were the top two genera of discriminant abundance in the cancerous stomach before surgery, while Streptococcus and Prevotella were the two most abundant genera after tumor excision. Furthermore, N-nitrosation genes were prevalent before surgery, whereas bile salt hydrolase, NO and N2O reductase were prevalent afterward. To our knowledge, this is the first report to document changes in gastric microbiota before and after surgical treatment of stomach cancer.

gastrectomy for gastric cancer are often subjected to other surgical procedures, which alter the gastric environment. For example, highly selective vagotomy affects gastric secretion of gastric acid 11 ; cholecystectomy elevates gastric pH value 12 ; Billroth II reconstruction reduces pancreatic polypeptide secretion 13 . Following subtotal gastrectomy, there are several common side effects, including marginal ulcers, bile reflux, and stump cancer. Bile reflux after subtotal gastrectomy has been associated with the presence of Streptococcus and Veillonella in gastric aspirates 14 and Escherichia, Klebsiella, and Clostridium in the intestine 15 . Although gastric microbiota is altered after subtotal gastrectomy 14 , changes in diversity have not been well characterized. Despite characterization of gastric microbiota by culture-independent approaches, changes following subtotal gastrectomy in patients with gastric cancer are not completely understood. Therefore, there are many knowledge gaps, leading to a number of questions. For example, are different anatomic sites inhabited by different microbes? What is the compositional variation in gastric microbiota after subtotal gastrectomy? What is the biodiversity pattern before and after subtotal gastrectomy? Do metabolic functions embedded in gastric microbiota correspond to changes caused by subtotal gastrectomy? In this study, we aimed to address these questions by deep sequencing of microbial 16S ribosomal RNA (rRNA) genes in gastric tissues.
Gastric microbiota in gastric cancer patients (at various anatomic sites and before and after subtotal gastrectomy) was characterized by 16S rRNA gene sequencing. Within the 2-year timeframe of this follow-up study, 24 gastric biopsies were collected from 6 patients subjected to subtotal gastrectomy. Variations in gastric microbiota and predicted gene functions before and after tumor excision (subtotal gastrectomy) were determined.

Results
Statistical summaries of sequencing results. To characterize stomach bacterial microbiota and potential variations associated with subtotal gastrectomy, we collected tumor (abbreviated as T in figures) and non-tumor (N) tissues before surgery, as well as gastric stump (S) and high body (B) tissues after surgery, from 6 gastric cancer patients. A total of 4.6 million pair-end reads were generated, of which 3.2 million reads passed quality filtering and were non-chimeric. To determine bacterial community diversity and composition, reads were aligned to the Greengenes database and non-bacterial sequences removed. On average, 85% of reads in a sample were retained. In total, 2.7 million reads (on average, 113 ± 42 thousand reads per sample) were used for subsequent analyses.
Bacterial diversity. Based on 16S rRNA gene sequencing data, bacterial richness (number of operational taxonomic units (OTUs), richness value, and Chao 1 index) increased after surgery (P < 0.01, one-tailed Student's t-test; Table 1). Similarly, there was greater diversity of bacterial communities after surgery (Shannon index) than before surgery (P < 0.01, one-tailed Student's t-test; Table 1). However, within patients, there were no differences (P > 0.05, two-tailed paired t-test of Shannon indices) between tumor and non-tumor tissues (before surgery) or between gastric stump and high body tissues (after surgery). Before-surgery communities were characterized by rarefaction curves approaching an asymptote, whereas after-surgery communities were characterized by curves with steeper slopes ( Figure S1), indicating the potential for greater diversity to be discovered with more sequencing efforts. Good's coverage estimates sampling completeness by calculating the probability that a randomly selected read from a sample has been sequenced. At 97% sequence similarity level, Good's coverage values for all sampled bacterial communities ranged from 0.964 to 0.996 (except T594S, with 0.893 Good's coverage) when estimated using all reads (Table S1).
Bacterial community structure. Bacterial abundance (i.e., read count) was normalized by corresponding copy number of 16S rRNA genes (prior to community compositional analysis) to reduce potential bias of abundance estimation due to copy number variation 16 . Before surgery, stomach microbiota in tumor and non-tumor tissues was dominated by the phyla Proteobacteria (80% in N and 67% in T) and Actinobacteria (15% in N and 24% in T), followed by Firmicutes and Bacteroidetes (< 4% in both N and T; Fig. 1A). However, after surgery, Firmicutes and Bacteroidetes dramatically increased in both gastric stump and high body tissues (32 and 20% respectively, on average), while Proteobacteria and Actinobacteria decreased (Fig. 1A).
At the class level, there was obvious personal variation in stomach microbiota. For example, in 3 before-surgery samples Epsilonproteobacteria was dominant (> 95%), whereas 5 samples harbored Actinobacteria (at least 25%; Fig. 1B). After surgery, Bacilli (of Firmicutes) and Bacteroidia (of Bacteroidetes) significantly increased in the stomach of most patients, representing (on average) 25 and 19% of the microbiota, respectively.
Principal component analysis (PCA) with surgery status as instrumental variable revealed significant differences in bacterial genera abundance before and after surgery (P = 0.0001, Monte-Carlo simulation; Fig. 1C). Furthermore, from the results of linear discriminant analysis (LDA) effect size (LEfSe) analysis 17 , there were a total of 63 known genera with differential abundance before and after surgery. Among 19 genera known to be more abundant before surgery, Ralstonia and Helicobacter were the two with the largest LDA effect size (Fig. 1D), whereas Streptococcus and Prevotella represented the top two genera (of 43) after surgery.
Group differences (i.e., before and after surgery) in bacterial community structure were also observed using the Bray-Curtis measure of beta diversity 18 . The within-group distance was significantly lower than the between-group distance when samples were divided into before-and after-surgery groups (P < 0.001; Fig. 2A). However, comparisons between tumor and non-tumor tissues before surgery (Fig. 2B) and between gastric stump and high body tissues after surgery (Fig. 2C) showed insignificant or only marginally significant differences in distances. Interestingly, community dissimilarity within patients was lower than that between patients (regardless of surgery status), although only after-surgery communities demonstrated a significant difference (Fig. 2D).
Bacterial gene functions. Based on the community structure derived from 16S rRNA gene sequencing data, predicted gene functions of each sample were inferred from referenced bacterial genomes 19 . Enrichment Scientific RepoRts | 6:20701 | DOI: 10.1038/srep20701 analysis was performed (two-group comparison 20 ) to identify gene functions of differential abundances that were significantly enriched (Benjamini-Hochberg adjusted P < 0.05) in bacterial microbiota before and after surgery. In agreement with the dramatically changed community composition, there were 1879 and 2394 Clusters of Orthologous Groups (COGs) enriched in samples before (Table S2) and after surgery (Table S3), respectively. Given that there were so many genes of differential abundance, and based on the understanding of metabolic diversity of gut microbiota 21,22 , functional genes related to gastric disease and carcinogenesis were prioritized for further investigation.
Metabolic enzymes involved in denitrification (e.g., nitrate and nitrite reductases) were more abundant in the gastric microbiota before surgery (Fig. 3A). Furthermore, genes related to nitrosation (conversion of organic compounds to nitroso derivatives) were also more abundant before surgery, including the gene families of phenol degradation ( Fig. 3A; COG4313) and cytochrome (COG3474 and COG3258). From pathway-centric comparisons, biosynthesis of various vitamins, including biotin (vitamin H), riboflavin (vitamin B12), pyridoxal phosphate (vitamin B6), cobalamin (vitamin B12), thiamine (vitamin B1), and menaquinone (vitamin K), was more prevalent in the gastric microbiota before surgery than after surgery (Table S4). Conversely, after-surgery gastric microbiota encoded more bile salt hydrolase ( Fig. 3B; COG3049 and K01442) and genes related to reduction of nitric oxide (NO) and nitrous oxide (N 2 O). There were 36 COG Pathways enriched in the gastric microbiota after surgery (Table S5).

Discussion
This is the first follow-up study in which deep sequencing is applied to investigate the stomach microbiota in gastric cancer patients before and after subtotal gastrectomy. This study is also the first to predict gene functions associated with variations in gastric microbiota. There was an obvious shift in community composition of the gastric microbiota after surgery. Decreases in Proteobacteria and Actinobacteria and increases in Firmicutes and Bacteroidetes defined the after-surgery microbiota shift (at phylum level). Ralstonia and Helicobacter were the top two genera of discriminant abundance in the stomach of patients with gastric cancer, while Streptococcus and Prevotella were the top two genera after surgery. In contrast, there was relatively low divergence of gastric microbiota among various sites in the stomach. Corresponding to the community shift, the gastric microbiota also exhibited differential predicted gene functions. For example, denitrification and nitrosation genes were prevalent  Table S1). c Evenness was defined as Shannon/ln(# OTU). d Richness was defined as (# singleton OTU-1)/log 10 N. The maximum value was (N − 1)/log 10 N. e Good's coverage was defined as 1-(# singleton OTU)/N. * The values of richness indices increased in samples after surgery (P < 0.01, one-tailed Student's t-test).
in patient stomachs before surgery, whereas bile salt hydrolase and NO and N 2 O reductases were prevalent after surgery. After gastrectomy, the stomach was dominated by only four phyla (Proteobacteria, Firmicutes, Bacteroidetes, and Actinobacteria), which are in abundance in the gastric microbiota of healthy individuals 3,5 . The relative abundance of these phyla covaries with H. pylori status 9 . Composition of the gastric microbiota gradually changes along with progression of gastric diseases (from gastritis to intestinal metaplasia and ultimately gastric cancer 7,23 ). This study extended observations corresponding to surgical removal of gastric cancer tissue, resulting in increases in Firmicutes and Bacteroidetes and decreases in Actinobacteria and Proteobacteria.
In the present study, gastric cancer tissue and neighboring normal tissue demonstrated similar microbiota. After subtotal gastrectomy, the within-patient microbiota revealed even greater similarity between gastric stump and high body tissues, similar to the findings of a previous report in which there was little difference in gastric microbiota between antrum and body biopsy specimens 4 . Based on cultures of biopsy specimens, 62% of gastric microbiota exist in both the antrum and body of the stomach 24 , which is as expected as these two sites are distinct niches for microbial colonization, given their differential ability to secrete gastric acid 25 . Based on the present results and those of previous studies, we inferred that the gastric environment as a whole is more critical than the individual anatomical sites in determining microbial composition.
Gastric acidity is a barrier to microbial overgrowth 26,27 . However, the secretion of gastric acid is reduced after vagotomy 11 (i.e., cutting branches of the vagus nerve), resulting in a mildly acidic stomach that allows for more bacterial colonization 26 (mostly ingested). Similarly, acid-reducing drugs reportedly increase bacterial colonization in stomach 28 . This phenomenon correlated with our results that the richness and diversity (Shannon index) of gastric microbiota increased in after-surgery samples. Another potential side effect of vagotomy is cobalamin deficiency, due to decreased gastric intrinsic factor 29 . Consistent with this prediction, our pathway-centric comparison identified decreased abundance of genes for biosynthesis of cobalamin (and other vitamins) after surgery. Therefore, growth of bacteria lacking cobalamin synthesis genes might be the microbial response following vagotomy that contributes to cobalamin deficiency, although it has been proposed that gastrointestinal bacteria in humans are competitors for cobalamin rather than contributors 30 .
In addition to vitamin biosynthesis, other predicted gene functions of differential abundances were analyzed before and after surgery. For example, nitrate and nitrite reductase, phenol degradation gene, and cytochrome were differentially abundant before surgery, all of which are related to bacteria-mediated N-nitrosation 22 . It is known that N-nitroso compounds are causative factors in carcinogenesis. Therefore, the enrichment of genes functionally associated with N-nitrosation before surgery was in agreement with the findings of the cancerous stomach in this study. Since cytochrome commonly presents in respiratory chains and nitrate reductase is used by many Proteobacteria for anaerobic respiration, the differential abundance of these genes was likely a reflection of the high level of Proteobacteria before surgery. As for nitrite reductase, COG1251 (nirB) and COG2146 (nirD) are functionally associated with nitrate assimilation in various bacteria 31 , suggesting that these two genes are not involved in gastric NO production. To confirm the phenomenon, further experiments are required. After surgery, the gastric microbiota demonstrated increases in bile salt hydrolase (COG3049 and K01442), NO reductase (COG3256 and COG3901), and N 2 O reductase (COG4263). After Billroth II reconstruction and cholecystectomy (i.e., gallbladder removal), bile salts continuously pass through the stomach (bile reflux) due to anatomical change. This phenomenon speculatively changes the gastric environment in patients after surgery, making the stomach a potential niche for microbes capable of degrading bile salts. This plausibly explains the enrichment of bile salt hydrolase in gastric microbiota after subtotal gastrectomy. Furthermore, NO can be produced enzymatically by activated leukocytes and bacteria 32 and its bactericidal effect has been suggested to protect the stomach from pathogenic colonization 33 . However, bacterial NO reductase participates in the defense against NO toxicity 34 , implying that the gastric microbiota has a higher capability of NO detoxification after surgery than before surgery. Abundant N 2 O reductase after surgery corresponds to the reported increase in N 2 O concentration after partial gastrectomy 35 , although the microbial effect of N 2 O in stomach is not fully understood.
In conclusion, subtotal gastrectomy alters gastric microbiota in terms of diversity, community composition, and predicted gene functions. These changes in the microbial community of the stomach are closely associated with the altered gastric environment after subtotal gastrectomy.

Methods
Study subjects and gastric tissue specimen collection. Gastric tissue specimens were collected from early-stage gastric cancer patients before and after curative subtotal gastrectomy at Taichung Veterans General Hospital. Patients with previous malignancies or who had received chemotherapy, radiation therapy or prior gastric surgery were excluded. In addition, patients who had received proton pump inhibitors, H2 receptor antagonists, antibiotics, or probiotics within 1 month of tissue collection were excluded.
Gastric cancerous tissues and neighboring non-tumor tissues were collected before surgery. All patients underwent subtotal gastrectomy, which included excision of 40-50% stomach, vagotomy, cholecystectomy, and Billroth II reconstruction. Bile flow was through the stomach after surgery due to the method of Billroth II reconstruction. Approximately 2 years after subtotal gastrectomy, gastric tissue specimens were collected from the gastric stump (1 cm away from the anastomosis site) and high body of the lesser curvature of the stomach. Comprehensive oral explanations were given and signed informed consent was obtained from all study subjects. The experiments were carried out in accordance with the protocols approved by the Institute Review Board of Taichung Veterans General Hospital. 16S rRNA sequencing and analysis. The hypervariable region V1-V3 of bacterial 16S rRNA genes was amplified by PCR using bar-coded universal primers 27F (F, forward primer; 5′ -AGAG TTTGATCMTGGCTCAG-3′ ) and 534R (R, reverse primer; 5′ -GTATTACCGCGGCKGCTG-3′ ) 36 39 to remove reads shorter than 450 bp or longer than 550 bp, as well as reads with minimum average quality score < 27. In addition, reads containing an ambiguous base or homopolymer exceeding 8 bp were excluded. Chimera detection was performed using USEARCH (reference mode and 3% minimum divergence).
Quality-filtered and non-chimeric reads were analyzed (UPARSE 40 pipeline) to generate OTUs per sample (at 97% identity level). The OTU representative sequences were searched against the Greengenes 13_5 database using USEARCH global alignment to identify the corresponding taxonomy of the best hit. Any OTU without a hit or with only a weak hit, i.e. the function "(% sequence identity + % alignment coverage)/2" was < 93, was excluded from further analysis. The abundance of each taxon was counted and corrected with PICRUSt 19 , in which the pipeline divided the read count of each taxon by the corresponding 16S rRNA gene copy number. Diversity indices (e.g., Shannon, Simpson, Chao 1, and Good's coverage) were estimated with Mothur.
Statistical analysis of bacterial community. All statistical analyses were performed using R software (http://www.r-project.org/), unless otherwise specified. Gene copy number-corrected abundance of genera was total-sum scaled per sample. A pseudocount of 0.0001 was added to the relative abundance (in percentage) before logarithmic transformation 41 . PCA was performed on log-transformed data using the R package ade4 42 to analyze genera abundance before and after surgery. Between-group inertia percentages were tested (Monte-Carlo test with 10000 permutations) to determine the P-values of PCA results. To identify organismal features differentiating communities of stomach bacteria before and after surgery, LEfSe 17 was applied with α = 0.05 (Kruskal-Wallis and Wilcoxon tests) and effect size threshold of 2 on linear discriminant analysis (LDA) through the web site, http://huttenhower.sph.harvard.edu/galaxy. Community structure similarities within and between groups were assessed using the Bray-Curtis distance 18 via the R package vegan 43 , based on relative abundance of bacterial genera.
Prediction and analysis of gene functions of bacterial microbiota. Metabolic profiles of bacterial communities were predicted with PICRUSt, which forecasts abundance of genes of metabolic function based on the 16S copy number-corrected OTU composition. Functional genes were categorized (by PICRUSt) into COGs and KEGG Orthology (KO) gene families. To identify gene functions that differentiated bacterial communities before and after surgery, COG gene abundance was subjected to enrichment analysis of two-group comparison, using the R package ShotgunFunctionalizeR 20 . This analysis normalized gene abundances using a generalized linear model with Poisson canonical logarithmic link function and determined differential significance (P-value) via a binomial method with a Benjamini-Hochberg false discovery rate correction to adjust q-values for multiple testing. In addition to gene-centric analysis, ShotgunFunctionalizeR was used to perform pathway-centric analysis. The sets of functionally related COG families were grouped into COG Pathways and COG Categories based on JGI IMG/M 44 . To identify pathways of differential abundance before and after surgery, the two-group comparison of pathways was performed by ShotgunFunctionalizeR using Poisson model to determine differential significance (P-value) via a binomial method with a Benjamini-Hochberg false discovery rate correction to adjust q-values for multiple testing.