Helicobacter pylori infection associates with fecal microbiota composition and diversity

Helicobacter (H.) pylori is the most important cause for peptic ulcer disease and a risk factor for gastric carcinoma. How colonization with H. pylori affects the intestinal microbiota composition in humans is unknown. We investigated the association of H. pylori infection with intestinal microbiota composition in the population-based cohort Study-of-Health-in-Pomerania (SHIP)-TREND. Anti-H. pylori serology and H. pylori stool antigen tests were used to determine the H. pylori infection status. The fecal microbiota composition of 212 H. pylori positive subjects and 212 matched negative control individuals was assessed using 16S rRNA gene sequencing. H. pylori infection was found to be significantly associated with fecal microbiota alterations and a general increase in fecal microbial diversity. In infected individuals, the H. pylori stool antigen load determined a larger portion of the microbial variation than age or sex. The highest H. pylori stool antigen loads were associated with a putatively harmful microbiota composition. This study demonstrates profound alterations in human fecal microbiota of H. pylori infected individuals. While the increased microbiota diversity associated with H. pylori infection as well as changes in abundance of specific genera could be considered to be beneficial, others may be associated with adverse health effects, reflecting the complex relationship between H. pylori and its human host.

only in the stomach, but also in cecal and colonic samples compared to non-infected controls 8 . However, no specific taxa with significantly altered abundance could be identified in their colon. In humans, only small studies analyzed changes in intestinal microbiota during and after eradication therapy of H. pylori [9][10][11] . Due to the dramatic effect of the antibiotics used for H. pylori eradication in those studies, changes in the microbiome cannot unambiguously be attributed to the absence of the pathogen. Older studies relied on culturing techniques 12 which are inherently inappropriate to investigate the predominantly anaerobic gut milieu. In the present study we analyzed fecal microbiota profiles generated by 16S rRNA gene sequencing of 212 H. pylori infected and 212 phenotypically matched control individuals from the population-based Study-of-Health-in-Pomerania-TREND (SHIP-TREND) 13 .

Results
Phenotypic matching of the 212 H. pylori infected and 212 H. pylori negative subjects was performed to control for putative confounders known to influence intestinal microbiota such as age, sex, body mass index (BMI), alcohol consumption, smoking, proton pump inhibitor (PPI) intake, and diet [14][15][16][17][18] . A history of peptic ulcer disease was also considered for matching because affected individuals were more likely to have been subjected to eradication therapy which is assumed to influence gut microbiota 10 . After matching H. pylori cases and controls exhibited similar distribution patterns for all accounted phenotypic variables (Table 1 and Supplementary Table S1). None of the selected individuals were under antibiotic therapy at the time of sample collection.

Beta diversity analysis of H. pylori infected individuals as compared to non-infected controls.
Beta diversity analysis estimates how samples differ from each other. We used the commonly applied Bray-Curtis dissimilarity which is calculated based on the minimal shared abundance of each taxon. Thus, dual absence of taxa is not treated as similarity. Figure 1 shows the result of a principal coordinate analysis (PCoA) based on Bray-Curtis dissimilarity including all 424 microbiota samples. H. pylori infection was associated with a clear shift mainly along the first principal coordinate axis. Permutational analysis of variance (PERMANOVA) based on Bray-Curtis dissimilarity confirmed a significant shift of H. pylori cases compared to controls (r² = 0.011, p < 0.001). Association of continuous H. pylori antigen levels with beta-diversity explained even more variation (r² = 0.023, p < 0.001). The association with continuous H. pylori antibody levels was much weaker (r² = 0.006, p = 0.014). In a next step, we investigated whether H. pylori antibody or H. pylori stool antigen levels showed a significant association with beta-diversity only within the group of controls or H. pylori cases, respectively. No significant association was found within the group of controls. Within the group of H. pylori cases, H. pylori antibody levels did not show an association with beta-diversity. In contrast, continuous H. pylori antigen levels were associated with distinct changes in beta-diversity (r² = 0.029, p < 0.001). This effect size was even larger than that of age (r² = 0.015, p = 0.005) or sex (r² = 0.018, p = 0.002). No significant associations with beta-diversity were found for BMI, alcohol, smoking, use of PPI, or history of peptic ulcer disease in the group of H. pylori cases.

Alpha diversity analysis of H. pylori infected individuals compared with non-infected controls.
Alpha diversity estimators characterize the diversity of an ecological community within a sample. We determined several alpha diversity indices with different emphases. The 'Shannon diversity index' (H) and the 'Simpson diversity number' (N2) include information about the number of different operational taxonomic units (OTUs), as well as the abundance of each taxon in the respective samples. The 'Phylogenetic diversity' tries to predict the genetic relatedness of the taxa in each sample. We compared the alpha diversity scores between H. pylori cases and controls using the parametric two-tailed t-test in case of the normally distributed 'Shannon diversity index' (Supplementary Fig. S1). In case of the non-normal distributed 'Simpson diversity number' and 'Phylogenetic diversity' (Supplementary Fig. S1) the two-tailed Mann-Whitney test was applied. Alpha diversity estimations revealed significantly higher scores in H. pylori cases compared to controls for 'Shannon diversity index' and 'Simpson diversity number' (Table 2 and Fig. 2). Although the median and mean 'Phylogenetic diversity' scores of H. pylori cases were also higher compared to controls, this was not significant.  19 . To investigate whether the H. pylori carrier status affects the enterotype distribution, we performed enterotype clustering similar to the approach described by Arumugam et al. We found only two enterotypes in our dataset either dominated by Bacteroides or Prevotella. Ruminoccous or unclassified Ruminococcaceae were present in both groups and did not constitute a unique cluster. In controls, the enterotype 1 was present in 66.5% and enterotype 2 in 33.5% of cases. Analysis of H. pylori carriers revealed a shift from enterotype 1 (56.1%) towards enterotype 2 (43.9%) compared to controls (p = 0.036; Fisher's exact test), (Fig. 3b).

Association of intestinal microbiota with H. pylori stool antigen load in H. pylori infected individuals.
The results of the beta-diversity analysis revealed a significant microbiota variation even within the group of the 212 H. pylori cases with respect to the individual H. pylori load determined by stool antigen test. Hence, we performed linear regression analysis to identify the genera that are associated with the H. pylori stool antigen load. We analyzed all genera present in at least ten percent of all samples. Due to the reduced statistical   www.nature.com/scientificreports www.nature.com/scientificreports/ power in this smaller data subset we focused on more prominent taxa by additionally excluding all genera with a mean abundance of less than or equal to 0.1% and all unclassified taxa at genus level. This analysis identified four genera that were all negatively associated with H. pylori stool antigen load, namely Bacteroides (q = 0.003),   Table S3). Performing a similar analysis for the individual H. pylori serology level instead of the stool antigen load did not yield a significant association.

Discussion
We investigated changes in the intestinal microbial community structure of H. pylori infected individuals, finding significant alterations of the microbiota composition and diversity. Strikingly, in H. pylori cases, the alpha diversity estimators 'Shannon diversity index' and the 'Simpson diversity number' exhibited generally higher scores, indicating increased microbial diversity. High diversity is generally considered to be an indicator of a healthy gut microbiome, while a decrease is associated with poorer health or unhealthy habits. For example, a plant rich diet increases alpha diversity 20 , whereas a Western lifestyle or obesity are associated with reduced bacterial diversity 21,22 . In addition, conditions such as recurrent antibiotic-associated diarrhea 23 , Crohn's disease 24 , or ulcerative colitis 25 , have all been reported to be associated with reduced intestinal diversity. Considering these findings, the observed positive correlation of H. pylori infection with increased diversity suggests a putative beneficial effect of H. pylori as it may strengthen the host's resilience against microbiome perturbations or gastrointestinal infections. In more detail, a total of 13 taxa exhibited different abundance values between H. pylori infected cases and controls. We found Parasutterella to be decreased in case of H. pylori infection, a genus that has been reported to be increased in the ileal submucosa of Crohn's disease patients 26 . H. pylori infected individuals exhibited increased levels of the facultative pathogen Haemophilus and decreased levels of Pseudoflavonifractor. The latter has been reported to be involved in the production of short-chain fatty acids (SCFA) such as butyrate 27 . SCFA represent an important energy source for colonic epithelia 28 and have been shown to modulate the immune response by www.nature.com/scientificreports www.nature.com/scientificreports/ regulating colonic regulatory T-cell function and alleviate experimental colitis in rodents 29 . Thus, the decrease of the SCFA producer Pseudoflavonifractor may be disadvantageous for the human host as well.
In addition to the binary comparison of H. pylori infected individuals with controls we addressed putative differences within the group of H. pylori positive individuals and found that the H. pylori antigen load was associated with larger alterations in the fecal microbiota than age or sex. Of note, this association was not found for variations in the H. pylori antibody titer. The H. pylori antigen load was negatively associated with the four genera Bacteroides, Barnesiella, Fusicatenibacter, and Alistipes. Several of these taxa have previously been attributed health promoting features. The presence of Barnesiella was reported to be positively associated with higher eradication rates of antibiotic-resistant bacteria after fecal microbiota transplantation in mice 30 and humans 31 , whereas higher rates of chemotherapy-related blood stream infections were observed in individuals with reduced Barnesiella counts 32 . Fusicatenibacter is known to be involved in SCFA production and to produce lactic acid 33 . It may therefore exert anti-inflammatory properties as shown for other lactic acid producing bacteria 34 . Finally, Alistipes is a supposed producer of the SCFA butyrate, which can alleviate intestinal inflammation 35 . Consequently, the observed alterations in individuals with a particularly high H. pylori antigen load may together cause adverse consequences for the human host. The putative health benefits mediated by the increased microbial diversity associated with H. pylori infection could be counteracted when H. pylori load is high.
Arumugam et al. first reported three different fecal microbiome clusters they designated as enterotypes 19 . It was proposed that each enterotype is characterized by a marked occurrence of Bacteroides (enterotype-1), Prevotella (enterotype-2), or Ruminococcus (enterotype-3). These authors also reported that enterotype-1 generates energy primarily by fermentation of carbohydrates and proteins, since genes encoding for galactosidases, hexosaminidases and proteases were more common. In contrast, the dominating genus of enterotype-2, Prevotella, is supposed to be a mucin degrader 36 . Furthermore, Arumugam et al. did not find a significant correlation of any of these enterotypes with BMI, sex, or age. In a later study, only two clusters dominated by Bacteroides or Prevotella were confirmed 37 . Ruminococcus which did not form an individual cluster was subsequently fused with that of Bacteroides. Both remaining clusters were primarily separated by dietary effects. The Bacteroides dominated enterotype was associated with the intake of animal protein and saturated fats, whereas the Prevotella dominated enterotype was associated with the intake of carbohydrates and a plant-based diet. In the present study, using a similar approach to that of Arumugam et al., two clusters either dominated by Bacteroides or Prevotella could be identified. The Prevotella dominated enterotype 2 was more common in H. pylori infected individuals. As our dataset was matched with respect to diet, this observation could not merely be explained by differences in intake of animal proteins or plant based products. There is currently an ongoing debate about the general concept of enterotypes. It was proposed that the clustering results which revealed separated enterotypes were rather guided by the dominant abundance of Bacteroides or Prevotella, respectively, and did not result from the presence of consistent microbial communities in each cluster 38 . Furthermore, it was emphasized that Bacteroides and Prevotella may form continuous gradients rather than being clearly segregated into two groups 39,40 . While contributing to the debate about the adequacy of the term 'enterotype' is beyond the scope of this study, we found H. pylori infection to be associated with a shift of the intestinal microbiota towards a Prevotella dominated microbiome, irrespective of the 'enterotypes' concept.
The underlying cause of the observed fecal microbiota changes associated with H. pylori infection is unknown. The altered gut microbiome in H. pylori-infected Mongolian gerbils has been explained by gastric hypochlorhydria 7 . It seems, indeed, plausible that a reduced production of gastric acid would promote the passage of acid-sensitive bacteria leading to an enriched diversity of the intestinal microbiome. During H. pylori infection there are generally two periods characterized by reduced gastric acid secretion: First, the initial infection phase can be followed by acute gastritis with temporarily impaired production of gastric acid 4 . Second, in a later stage of H. pylori infection many individuals develop pangastric, hypoacidic chronic gastritis caused by destruction of parietal cells 41 . However, this simple model would not be supported by the observation that gastric acid-suppressed PPI users were found to exhibit a reduced intestinal diversity 14,16 . Modulation of the distal gut microbiota diversity by H. pylori is therefore more complex. It has been proposed that the impact of H. pylori infection is not restricted to the gastrointestinal tract. In murine models, H. pylori demonstrated immunoregulatory features by preventing allergic asthma through the induction of regulatory T cells via IL-18 mediated tolerogenic reprogramming of dendritic cells, which then ensures the persistence of the pathogen 42,43 . In humans H. pylori could also regulate the intestinal microbial community composition in a similar fashion by modifying the host's immune response.
In the complete dataset including H. pylori cases and controls, we did not find any stool microbiota profile with presence of the genus Helicobacter. Although PCR based approaches have managed to detect H. pylori in fecal samples before, these used distinct sample preparation (e. g. immunomagnetic separation) and/or specific primers for H. pylori genes for its detection 44 . Yet, other investigations failed to detect fecal H. pylori DNA even when using specific primers 45 . As the gastric pathogen H. pylori may mostly not survive under the anaerobic conditions of the distal intestine its fecal DNA concentrations are likely very low compared to the majority of anaerobic bacteria. It may therefore escape detection in fecal samples when investigating with an untargeted approach such as 16S rRNA gene sequencing.
The strength of the present study includes the large study population, the high specificity of the H. pylori diagnosis based on the assessment of both serology and stool antigen testing, and the thorough phenotypic matching of factors known to influence intestinal microbiota in order to avoid bias. Its main limitations result from the design as an association study. It may be possible that the H. pylori stool antigen levels were influenced by the gut microbiota composition, rather than the H. pylori load determining the gut microbiota. However, given the high sensitivity and specificity of the H. pylori stool antigen test in comparison to histology, culture, rapid urease test, or urea breath test 46 , it seems unlikely that the antigen levels were affected by gut microbiota to a great extent. A (2019) 9:20100 | https://doi.org/10.1038/s41598-019-56631-4 www.nature.com/scientificreports www.nature.com/scientificreports/ poorer performance of the stool antigen test could be predicted if defined gut bacteria would specifically degrade the target antigen.
It has been assumed that humans have been infected with H. pylori for at least ~100,000 years 47 and distinctly more than half of the population in many developing and emerging countries are currently infected. In light of the sometimes life-threatening disorders that can arise in H. pylori infected individuals, this long lasting evolutionary relationship may be explainable by counterbalancing beneficial effects of the infection. Among them are the already mentioned protection against atopic diseases described in murine models 42 and, according to our data, possible suppression of other gastrointestinal pathogens by increasing microbial diversity. Therefore, H. pylori eradication may trigger unwanted detrimental gut microbiota alterations, namely a reduced microbiota diversity, in individuals with low-grade infection. This underlines the need for careful decision making regarding H. pylori eradication in each individual. However, the complex relationship between H. pylori and its human host is once again demonstrated by the probably adverse microbial alterations found in individuals with high H. pylori antigen load. The large proportion of H. pylori infected individuals in the global population emphasizes the importance of this finding. As the applied method of 16S rRNA gene sequencing only allows taxon identification, future whole-genome sequencing approaches will have to define the functional potential of the enriched taxa in H. pylori cases. Of the total dataset of 931 individuals, we excluded 12 participants due to missing phenotype data and further 6 individuals because of an intake of antibiotics at the time of sample collection. This resulted in 221 H. pylori infected individuals of whom stool DNA for microbiota analysis by 16S rRNA sequencing was available for 212 individuals. As controls, 212 samples from individuals with both negative H. pylori stool antigen test as well as H. pylori serology were selected. All control samples were matched with respect to age, sex, BMI, alcohol consumption, smoking, PPI usage, history of peptic ulcer disease, and dietary habits using the 'R' 49 package 'MatchIt' (option 'nearest') 50 .

Methods
16S rRNA gene sequencing. Sequencing was performed as described before 51 . In brief, isolated DNA from fecal samples was used for amplification of the V1-V2 region of bacterial 16S rRNA genes on a MiSeq platform (Illumina, San Diego, USA). MiSeq Fast-Q files were created by CASAVA 1.8.2 (https://support.illumina.com/ sequencing/sequencing_software/casava). After quality trimming of sequences with Sickle (https://github.com/ najoshi/sickle), forward and reverse reads were merged and filtered using VSEARCH 52 . Subsequently all reads were quality filtered by FastX Toolkit (http://hannonlab.cshl.edu/fastx_toolkit). At this step, only reads with a quality score of at least 30 (error probability 1 in 1,000) per base in 95% of sequenced nucleotides were included. To reduce redundancy among sequences de-replication was performed. OTUs were clustered using VSEARCH demanding a minimum sequence similarity of 97%. After chimera filtering by USEARCH 53 each sample was normalized to 10,000 reads by random selection. Four of the 424 samples contained slightly less than 10,000 reads (9897, 9864, 9815, and 9408 reads). For assignment of taxonomy the SINTAX classifier was used 54 . A confidence of at least 80% for each taxonomic rank was ascertained. All taxa with a confidence below 80% were assigned to an arbitrary taxon as unclassified family, order, class, phylum, or bacteria, respectively. Data analysis. All statistical analyses were performed using 'R' . Bar plots were created with GraphPad Prism 5 (GraphPad Software, San Diego, USA). For calculation of 'Shannon diversity index' and 'Simpson diversity number' the R package 'vegan' 55 was used based on OTU counts. 'Phylogenetic diversity' was determined using the package 'picante' 56 . Q-Q plots for normality assessment were generated using the qqplot function of the 'stats' package. The Bray-Curtis dissimilarity was calculated based on genus level data using the 'vegan' function vegdist. PCoA was performed with the cmdscale function from the 'vegan' package. Square root transformation of the Bray-Curtis dissimilarity was performed prior to the ordination to avoid negative eigenvalues. To determine the contribution of phenotypic variables to the Bray-Curtis dissimilarity, PERMANOVA was done using the 'vegan' function adonis and the significance assessed by 10,000 permutations. For regression of phenotypic variables on the two major principal coordinate axes the 'vegan' function envfit was used. The two-tailed Mann-Whitney test was performed for assessment of significance in case of continuous data. A two-tailed t-test was applied for www.nature.com/scientificreports www.nature.com/scientificreports/ significance assessment of the normally distributed 'Shannon diversity index' . Fisher's exact test was utilized for categorical data. Comparison of the relative abundance of all taxa that were present in at least ten percent of all samples at genus level between H. pylori cases and control individuals was performed using the two-tailed Mann-Whitney test. The resulting p-values were adjusted for multiple testing following the Benjamini-Hochberg procedure and called 'q-values' . Association of the H. pylori antigen load with individual genera within the group of H. pylori infected individuals was examined as follows: All classified genera present in at least ten percent of all samples and with a mean abundance of greater than 0.1% were analyzed using a linear regression model ('R' function lm) based on log transformed abundance data. For this analysis zero-values were ignored, i.e. treated as missing values, to avoid biased linear regression estimates due to inflation of zeros. The putative confounder age and sex were included in the model and resulting p-values adjusted for multiple testing following the Benjamini-Hochberg procedure. P-values or q-values < 0.05 were considered significant. All p-and q-values were rounded to three significant digits.

Data availability
All microbiome and phenotype data were obtained from the Study-of-Health-in-Pomerania (SHIP/SHIP-TREND) data management unit and can be applied for online through a data access application form (https:// www.fvcm.med.uni-greifswald.de/dd_service/data_use_intro.php).