2.1 Patients, sample collection and storage

In total, 90 individuals with positive H. pylori infection who met the following criteria were included in this study: men and women aged 40 to 64 years; self-reported alcohol consumption 2–3 times a month or less; no history of colon or rectum polyps since age 20, gallstones, gastric cancer, gastric resection, alarm symptoms for digestive or any other diseases, type 2 diabetes, ulcerative colitis, Crohn’s disease, coeliac disease, biliary cirrhosis, thyroid diseases, hepatitis B viral infection or serious psychiatric disorders. A ¹³C-Urea breath test (Euroisotop, Germany) was performed to diagnose H. pylori infection. A detailed questionnaire was collected from each study participant, which contained information such as age, BMI, medical history (e.g., gastrointestinal diseases, viral infections, autoimmune diseases, and cancer), and lifestyle habits. Samples from 60 individuals collected over two years were used for the ESBL panel experimental group to explore long term variation in the abundance of ESBL coding genes. Samples from the remaining 30 individuals collected over one year were used for validation of the ESBL panel to account for the abundance of ESBL-encoding genes over a shorter time period (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Study design.

First line H. pylori eradication therapy consisted of esomeprazolum 40 mg, clarithromycinum 500 mg and amoxicillinum 1000 mg, each twice per day for ten days.

https://doi.org/10.1371/journal.pone.0289879.g001

Fecal samples were acquired from each recruited patient before starting the standard first-line H. pylori eradication (HPE) therapy and one–two years after. Each patient was prescribed the following medications twice a day for ten days: Esomeprazolum 40 mg, clarithromycinum 500 mg, and amoxicillinum 1000 mg. All procedures conformed to institutional ethical standards, and written consent was obtained from all patients prior to enrollment in the study.

In total, 180 samples were obtained within 30 minutes of defecation. Samples were transferred to an OC-Sensor tube (approx. 0.06–0.1 g) (Eiken Chemical Co, Japan), immediately homogenized, and stored at -80°C until further processing according to previously validated storage conditions [45].

2.2 DNA extraction

Fecal samples from the OC-Sensor tubes were extracted using a disposable syringe and transferred to pre-labelled 5 mL tubes. Excess OC-Sensor tube solvent was removed from the samples by lyophilization for approximately 15 h in a Christ Alpha 1–2 LD Freeze Dryer (SciQuip Ltd., UK). DNA from the dry remnants of OC-Sensor samples was isolated using the FastDNA SPIN Kit for Soil (MP Biomedicals, USA) according to the manufacturer’s guidelines (S1 File).

2.3 ESBL gene cluster primer design

Known BL nucleotide sequences were obtained from the NCBI GenBank (accession date: 02.01.2018). Primer design was performed separately on TEM BLs, as they formed a homogenous group that was highly dissimilar from the rest of the BL sequences. Multiple alignments of the BL sequences were obtained using MAFFT v.7.392 [46]. The number of pairwise non-gapped mismatches between the BL sequences was calculated, and subsequent hierarchical clustering with complete linkage was performed on the distance matrix calculated from the alignment, using the number of non-gapped mismatches divided by the length of the alignment as the distance metric. Clusters were defined at a distance cut-off of 0.1. For each cluster, we identified contiguous conserved regions in the alignment using the Shannon index, for which at most 10% of the nucleotides would differ in each position. Regions with length 18 bases or longer were used in further experiments.

For each cluster, all possible pairs of conserved regions were computed such that the interval between regions in a pair was not longer than 500 bp. Each conserved region pair was scored by summing their Shannon indices at each position and sorted by their scores in ascending order. Primers were then designed on these region pairs with PRIMER3 v.2.4.0, by specifying PCR product sizes in the 200–500 bp range [47]. The best primer pairs for each cluster were evaluated by identifying the potential binding sites in the original list of BL sequences. A site was regarded as a potential binding site if there were up to three mismatches between the primer sequence and template. Additionally, no mismatches with the template were allowed for the last five nucleotides at the 3’ end of the primer. The binding site algorithm was implemented using the SeqAn library [48]. Regions bound by the primers were then aligned against the BL sequence database to evaluate potential off-target PCR products, that is, sequences that mapped against multiple BL clusters. Primer pairs that were overlapping (forming heterodimers estimated with primer3-py) or forming PCR products shorter than 50 bp were identified and pooled into separate sequencing batches. The designed primers were synthesized by MetaBion (Metabion International AG Ltd., Germany).

2.4 Sample preparation and targeted sequencing

Pools of primers with equal molarities and volumes were prepared for targeting the ESBL-coding genes and primers for the normalization of ESBL counts- primer pair Probio_Uni-F/Probio_Uni-R targeting 16S rRNA gene V3 region [49]. PCR amplification of ESBL coding gene regions was performed using a 10 μM custom designed primer pool (S1 Table), Phusion U Multiplex PCR Master Mix (Thermo Fisher Scientific, USA), and GeneAmp® PCR System 9700 (Thermo Fisher Scientific, USA). The reaction mixture was prepared according to the manufacturer’s recommendations, and the thermal conditions were set as follows: 98°C for 30 s; 35 cycles of 98°C for 10 s, 55°C for 30 s, 72°C for 15 s; with a final extension at 72°C for 7 m. The success of the reaction was then assessed by 1.2% agarose gel electrophoresis. 100 ng of the acquired amplicons were used for library generation using the Ion Plus Fragment Library Kit (Thermo Fisher Scientific, USA) and the NucleoMag® NGS Clean-Up and Size Select kit (Macherey-Nagel, Germany) purification module. The quality and quantity of the amplicons were assessed using an Agilent High Sensitivity DNA kit on an Agilent 2100 BioAnalyzer (Agilent Technologies, USA).

Prior to emulsion PCR, each library was diluted to 12 pM and pooled for up to 18 libraries per sequencing run. The Ion PGM^TM Hi-Q^TM View OT2 kit (Life Technologies, USA) and Ion OneTouch DL instrument (Life Technologies, USA) were used for template generation. Sequencing was performed on an Ion 318 v2 chip and Ion Torrent PGM machine using the Ion PGM^TM Hi-Q^TM View Sequencing kit (Life Technologies, USA). All procedures were performed according to the manufacturer’s instructions, and each run was expected to produce at least 80’000 reads per sample.

2.5 Sample preparation and shotgun sequencing

DNA samples for the shotgun metagenome analyses were normalized to an initial library input of 500 ng and sheared using a Covaris S220 Focused-ultrasonicator (Covaris, USA) to reach an average size of fragments 300 bp. Libraries with average insert sizes of 280 bp were prepared using MGIEasy Universal DNA Library Prep Set V1.0, (MGI Tech Co., China) according to the manufacturer’s recommendations. Quality control of the libraries was assessed using the Qubit High Sensitivity dsDNA assay kit on a Qubit 2.0 instrument (Thermo Fisher Scientific, USA) and the Agilent High Sensitivity DNA kit on an Agilent 2100 Bioanalyzer.

Sequencing depth was calculated to achieve at least 20 million reads per sample (paired end, read length 100 bp). Libraries were sequenced using the DNBSEQ-G400 sequencer and a DNBSEQ-G400RS High-Throughput Sequencing Set PE100 (MGI Tech Co., China) according to the standard workflow.

2.6 Data analysis of targeted ESBL coding genes

A sequencing platform-specific adapter clipping of the obtained raw reads was performed with Cutadapt v.1.16 [50]. Targeted sequencing data were then aligned against the curated BL sequence database using Bowtie2 v.2.3.5.1, pre-set at very sensitive [51]. Host reads from shotgun metagenomic sequencing data were filtered using Bowtie2 prior to mapping against BL sequences. 16S rRNA was quantified with SortMeRNA v.2.1 [52] using 16S rRNA sequences obtained from RNA central v10 [53] with search query ’rna_type: "rRNA" AND TAXONOMY: "9606" AND length: [19 TO 2000000000]’.

To classify sequencing reads into a specific ESBL cluster, we first created a classification scheme as follows: for each ESBL cluster, we identified regions where we expected the PCR product to form. Each putative product was aligned against all BL sequences to evaluate whether the product was specific to a particular cluster. We regarded putative products as specific if they mapped against sequences from only one cluster. Clusters were merged when a putative PCR product was specific to a set of clusters and if such clusters were not discernible. If multiple PCR products were mapped against the same position within a cluster, we established the alignment score threshold as the minimum score from a set of true positive alignments. An annotation table with cluster reference sequence IDs, start and end coordinates of the corresponding primer product regions, and alignment score thresholds were generated. Sequencing reads were assigned to clusters if overlapped with the coordinates in the annotation table and exceeded the alignment score threshold. Putative PCR products were identified using SeqAn, Pandas and Bowtie2. An annotation table and Python script reading sample binary alignment map files with the Pysam package (https://github.com/pysam-developers/pysam) were used to quantify the read count in each cluster for each sample. The read counts of each BL cluster were normalized to the read counts of the 16S rRNA gene of a particular sample.

2.7 Metagenome data analysis

Quality control and quality trimming of the obtained paired-end reads were performed using FastQC and Trimmomatic v0.39 [54] with a quality threshold of 20 and a minimum read length of 36. Quality filtered sequences were then aligned to the human genome reference GRCh37 (hg19, UID:2758) and sequences matching the human genome were removed using Bowtie2 v.2.3.5.1. The taxonomical profile of the metagenomic dataset was assigned using Kraken2 v.2.0.8 [55] and RefSeq database release 98 [56]. De novo read assembling into contigs was performed using the IDBA_UD [57] assembler with the k-mer length of at least 50. Generated assembly was evaluated using metaQuast [58]. The assembly database and the local alignment of input reads to assembly was performed using Bowtie2. Open reading frame detection and subsequent annotation was performed using PROKKA v.1.14.6 [59] with the manually curated Swiss-Prot UniProtKB [60] database (accessed 08.02.2021.). During the annotation, predictions of rRNA and tRNA, as well as contigs below 250 nt were excluded. Coordinates of predicted protein-coding features (CDS) were used for quantification against the assembly database using HTSeq [61] and the intersection-nonempty resolution mode. According to previous studies [62–64], metagenomic read counts were standardized using the Transcripts Per Million method [65] with an in-house built python script. Subsequently, from annotation files, CRISPR annotation was removed while contig IDs with the respective product information were retained using in-house built sed and awk scripts. Read counts were joined with the filtered annotation by contig ID column for each sample separately. Next, all samples were merged into a single dataset by annotation column using the Pandas [66] library within the Python environment.

Contigs were used to predict the resistome profile of the study subjects using The Resistance Gene Identifier (RGI) v.5.1.1 along with the Comprehensive Antibiotic Resistance Database (CARD) [67] and the DIAMOND [68] alignment tool. Results were gathered for each sample obtained using the heat map function of RGI, by organizing resistance genes based on the resistance mechanism and gene family. Additionally, hierarchical clustering was performed to cluster samples based on their similarity.

2.8 Statistical analysis of taxonomical data

Kraken reports were uploaded to the Pavian v.1.0.0 [69] package and taxonomic entries were filtered out if the sum of the assigned sequences for the taxonomic clade across the samples was below 200. Then, SIAMCAT v.1.9.0 [70] was used to evaluate the association of microbial species between pre- and post-eradication states. Briefly, the dataset of taxonomical entities was separated into two groups: pre-eradication (designated as case) and post-eradication (designated as control). The cut-off for the relative abundance of the species was set to 0.001. The association of microbial species between pre-and post-eradication states was determined using the Wilcoxon test at a significance level of p<0.05, with the False Discovery Rate multiple hypothesis correction method. In addition, log-transformed normalization was applied to the abundance matrix of microbial species and the Area Under the Receiver Operating Characteristics Curve (AU-ROC) was used as a non-parametric measure of the enrichment. All the acquired measures of association between the pre- and post-eradication groups were visualized in the SIAMCAT association plot.

Furthermore, the dataset was divided into four groups: F-post-erad denoted subjects of the post-eradication group with ineffective HPE; F-pre-erad denoted subjects of pre-eradication group with ineffective HPE; S-post-erad denoted subjects of post-eradication group with successful HPE and S-pre-erad denoted subjects of pre-eradication group with successful HPE. Alpha diversity metrics (Shannon, Chao1 and Observed) were calculated and visualized using Phyloseq v.1.30.0 [71]. Pairwise comparisons of alpha diversity metrics between the treatment states using the Wilcoxon rank sum test and the Holm P-value adjustment method were performed using the Vegan v.2.5–7 package. Non-metric multidimensional scaling was performed using Phyloseq.

2.9 Statistical analysis of amplicon data

The relative abundances of the ESBL clusters between the treatment states were compared using the two-tailed paired t-test in the Vegan v.2.5–7 package. The Kruskal-Wallis test was used to assess the significance in the abundance of individual ESBL clusters between the pre- and post-eradication states with the same package. To explore the clusters overlapping between datasets, cluster IDs which appeared at least once in a sample were extracted from all three datasets. A Venn diagram was constructed using the ggVennDiagram [72] v0.1.9 within the R environment.

2.10 Statistical analysis of functional data

UniprotKB entry IDs of the summarized annotation dataset were converted into Gene Ontology (GO) IDs using the UniProt online Retrieve/ID mapping tool (https://www.uniprot.org/uploadlists/). The UniProtKB entries which did not match any corresponding GO ID were removed from the dataset. Next, MaAsLin2 v1.8.0 [73] was used to determine the association of the microbiome functional profile with the treatment state. For the MaAsLin2 analysis, the q-value threshold for significance was set to 0.05, the minimum abundance for each GO term was set to 50, the minimum percentage of samples for which a GO term was detected at a minimum abundance was 25%, the random effect for the model was set to the patient ID, and the fixed effect for the model was treatment state. The significance of the association was controlled using the Benjamini-Hochberg multiple testing correction method.

2.11 Ethics Approval Statement

This study was approved by the Biomedical Ethics Committee of the Riga East University Hospital Support Foundation, approval No. 13-A/13 from October 3, 2013.

3.1 Experimental setting: Targeted ESBL analysis

The ESBL panel was designed to reflect the prevalence and abundance of the most common ESBL types. A total of 245 gene clusters encoding evolutionarily related ESBLs were targeted by designed primer pool, and primers targeting V3 region of the 16S rRNA gene were added to the pool to normalize the number of ESBL gene cluster counts (cluster names and respective primer sequences are shown in S1 Table). Analysis of the obtained raw sequence data resulted in 1’787 annotated microbial sources containing strain or clinical isolate IDs with respective BL gene groups, classes and gene names (S2 Table, sheet full annotation). Clusters with identical IDs were merged, and the lowest common ancestor was indicated as the microbial source (S2 Table, sheet merged annotation). Thus, from the 245 ESBL clusters targeted, results were obtained for 265 ESBL clusters, of which twenty clusters most likely represented sequence homology with similar BL gene clusters and thus were assigned to multiple clusters. In the pre-eradication subgroup, 89 ESBL clusters, while in the post-eradication subgroup 106 ESBL gene clusters were not detected in any of the study subject’s samples.

Assessing the presence of ESBL gene clusters in each sample in both treatment states, most clusters were absent, while the presence of a particular ESBL cluster in each sample was mostly dispersed. Examples of disperse clusters include bla_PER, bla_MSI, bla_CMH, bla_L1, bla_GES, bla_TMB, bla_VIM, bla_E, bla_PDC, and others. As a result, each sample was represented by 23 different ESBL gene clusters in the pre-eradication study group and 22 different ESBL gene clusters in the post-eradication study group.

Comparing pre- and post-eradication subgroups, the most prevalent clusters (Fig 2A) were the bla_EC gene group for class C ESBL with an annotated source of Escherichia coli (pre = 87.98%, post = 91.11%), cbl_A gene group for class A BL with annotated source of Bacteroides uniformis (pre = 3.56%, post = 5.44%), bla_MIR gene group for class C BL with annotated source of Enterobacteriaceae (pre = 2.52%, post = 0.64%), and bla_ACT gene group for class C BL with an annotated source of Enterobacteriaceae (pre = 2.35%, post = 0.12%). The normalized averaged relative abundance of ESBL genes was not significantly abundant between pre- and post-eradication (p = 0.5467). Next, we assessed if non-averaged relative abundance of any particular BL cluster differed significantly across treatment states. Thus, we identified eight clusters that were significantly different between pre- and post-eradication subgroups (Table 2, p<0.05). Among these dominated the cluster of class A bla_OXY group ESBL gene with the annotated origin from 12 different Klebsiella oxytoca strains, which had a higher relative abundance in the pre-eradication subgroup than in the post-eradication subgroup (p = 0.0076). Similarly, abundance of another strain-rich (n = 4) cluster of the B1 subclass bla_IMP group metallo-BL genes with annotated origins from Pseudomonas aeruginosa, P. fluorescens, P. putida and Acinetobacter baumannii was identified as a significantly different between the treatment states. However, the relative abundance of this one was noticeably lower than that of the previous one. The relative abundance of the B1 subclass bla_IMP group metallo-BL genes was significantly higher (p = 0.0424) in the post-eradication subgroup than in the pre-eradication subgroup.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Relative abundance and proportion of common ESBL coding genes between the experimental, validation-targeted, and validation-shotgun groups.

Section A-C: relative ESBL coding gene prevalence profile in the pre- and post-eradication samples of the analyzed datasets. For each dataset ESBL counts were normalized against 16S rRNA gene V3 region bacterial counts and summarized by treatment state. For visualization purposes, for the experimental (A) and targeted validation (B) groups, the abundance threshold was set to 0.1%, and for the shotgun validation (C) group it was 0.0001%. Section D-F: Venn diagram of the ESBL gene cluster counts observed in the experimental group and the validation group of shotgun and targeted sequencing. Only gene clusters that appear at least once in the respective sample group were included in the analysis. Abbreviations: ESBL–extended spectrum beta lactamases, BL–beta lactamases, Exp–experimental group, Vtr–targeted validation group, Vsh–shotgun validation group, Exp-pre–pre-eradication sample set of experimental group, Exp-post–post-eradication sample set of experimental group, Vtr-pre–pre-eradication sample set of targeted validation group, Vtr-post–post-eradication sample set of targeted validation group, Vsh-pre–pre-eradication sample set of shotgun validation group, Vsh-post–post-eradication sample set of shotgun validation group.

https://doi.org/10.1371/journal.pone.0289879.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Significant distribution of BL genes among annotated bacterial sources from patients comparing the pre- and post-eradication states.

https://doi.org/10.1371/journal.pone.0289879.t002

3.2 ESBL coding gene quantification using the validation group

The efficiency of ESBL panel was validated through targeted and shotgun metagenomic sequencing of DNA from another independent sample group (54 samples from 27 individuals), which also included pre-and post-eradication subgroups to validate the ESBL gene quantification experimental group. To provide results that are comparable to those of experimental group, we used the same ESBL reference database as before. As in the previous group, in the case of targeted sequencing data analysis, we used the number of reads of the V3 region of 16S rRNA gene to normalize ESBL quantification data between samples, while in the case of the shotgun metagenomic sequencing, the total number of 16S rRNA sequences were used for normalization.

Overall, sequencing and data analysis results for the validation group by targeted sequencing were similar to those of the experimental group. A total of 254 ESBL clusters were targeted, but data analysis resulted in identification of 265 ESBL clusters, of which twenty showed sequence homology and thus were assigned to multiple clusters. In total, we identified reads from 101 ESBL gene clusters in the pre-eradication subgroup samples and from 95 ESBL gene clusters in the post-eradication subgroup samples. Considering the arithmetic mean of each ESBL gene cluster within the respective eradication subgroup (Fig 2B), there were only seven clusters that exceeded the 0.05% abundance threshold, from which the most abundant ones were the C ESBL class bla_EC group genes (pre = 40.16%; post = 68.3%), the A BL class CblA family cbl_A group genes (pre = 32.86%; post = 27.94%), and the A ESBL class bla_OXY group genes (pre = 25.76%; post = 3.24%). The normalized average relative abundance of ESBL gene groups was not significantly different between pre- and post-eradication subgroups (p = 0.32464).

In contrast, analysis of shotgun metagenomic sequencing data from the validation group samples revealed that only 15 ESBL gene clusters were found in all 54 samples taken together. These included bla_R1, bla_CMY, cbl_A, bla_ACI, bla_PLA, bla_CKO, bla_EC, bla_SHV, bla_OXY, three of bla_ACT, and three of bla_OXA. However, their distribution was mostly dispersed among the samples. In the shotgun dataset, each sample was represented by 13 different ESBL gene clusters in the pre-eradication subgroup and by 8 different ESBL gene clusters in the post-eradication subgroup. Considering the arithmetic mean of all identified ESBL gene clusters in the pre- and post-eradication study subgroups (Fig 2C), 11 in the pre-eradication study subgroup and only five in the post-eradication study subgroup exceeded the 0.1% abundance threshold and the most abundant ones were the A BL class CblA family cbl_A group genes (pre = 56.04%, post = 53.0%), A ESBL class bla_ACI group genes (pre = 20.35%, post = 1.75%), D BL class bla_OXA group genes (pre = 12.02%, post = 32.13%), and C ESBL class bla_EC group genes (pre = 9.62%, post = 12.46%). The normalized average relative abundances of ESBL gene groups did not differ significantly between the pre- and post-eradication subgroups (p = 0.20919).

Further we assessed if the relative abundance of individual BL gene clusters differed significantly across the treatment states. Thus, within the targeted dataset, we identified 10 BL gene clusters that were significantly different between the pre- and post-eradication subgroups (Table 2, p<0.05). These included a cluster of various A ESBL class bla_OXA group genes with the annotated origin of Achromobacter xylosoxidans, a subclass B3 metallo-BL L1 family bla_L1 group gene with the annotated origin of Stenotrophomonas maltophilia, and an oxacillin- hydrolyzing D BL OXA-443 class bla_OXA group gene with the annotated origin of the Ralstonia picketii strain PIC-1. However, within the metagenomic dataset, we were unable to identify any specific BL cluster that differed significantly between treatment states and the most likely reason for that was the low number and low abundance of identified ESBL gene clusters (p>0.05).

We also quantified the number of ESBL gene clusters that overlapped between the experimental, targeted validation, and shotgun validation groups (Fig 2D–2F). A global comparison, for example, without considering the pre- and post-eradication subgroups, revealed that all three groups contained 13 clusters (Fig 2D) (cbl_A, bla_PLA, bla_EC, bla_ACI, bla_OXY, bla_CMY, bla_R1, three of bla_OXA, and three of bla_ACT). The highest number of shared ESBL gene groups was found between the experimental and targeted validation datasets (Fig 2E). Conversely, the shotgun dataset contained a notably lower amount of ESBL gene groups, resulting in lower coverage compared to the experimental dataset (Fig 2F). Thus, 69 ESBL gene groups were found between the experimental and targeted validation datasets, and only 6 shared ESBL gene groups were found between the experimental and shotgun validation datasets.

3.3 Taxonomical characterization

As our analyses included metagenomic sequencing, we also explored the taxonomic and functional composition of patient samples before and after H. pylori eradication therapy. A detailed taxonomic analysis of samples from the experimental group (n = 120) is described in our previous publication (Gudra et al., 2020). However, to be brief, we found that the dominant genera in the pre-eradication study subgroup were Bacteroides (10.3%), Oribacterium (9.08%), Prevotella (6.16%) and Parasutterella (4.87%), whereas in the post-eradication study subgroup–Bacteroides (9.75%), Streptomyces (7.75%), Oribacterium (7.41%) and Prevotella (5.81%). Alpha and beta diversities did not differ significantly between the pre-and post-eradication subgroups (p>0.05). Moreover, despite receiving eradication therapy, eight of the 60 individuals continued to test positive for H.pylori, and the available data show that there were no instances of recurrence.

In the current validation group (n = 27), 98.68±1.24% of taxonomic entries belonged to bacteria, 0.33±0.48% to viruses and 0.0026±0.003% to fungi and protozoa combined. A detailed summary of the taxonomy is provided in S1 File, but the most prevalent bacterial genera in the pre-eradication subgroup were Bacteroides (37.73%), Faecalibacterium (11.86%) and Bifidobacterium (6.67%), whereas in the post-eradication subgroup dominated Bacteroides (44.95%), Faecalibacterium (12.43%), and Alistipes (4.73%). We also found that at the species level, the most prevalent bacteria in the pre-eradication subgroup were Faecalibacterium prausnitzii (13.45%), Bacteroides vulgatus (10.38%), Bifidobacterium adolescentis (4.89%) and Bacteroides uniformis (4.53%), whereas in the post-eradication subgroup Faecalibacterium prausnitzii (14.21%), Bacteroides vulgatus (13.35%), Bacteroides dorei (6.12%) and Bacteroides uniformis (6.11%). Furthermore, we also tested if there was an association between microbial species and treatment state, and according to the acquired results, 12 microbial species were found to be differentially abundant between the pre- and post-eradication subgroups (Fig 3). Of these, two had increased relative abundance (E. bolteae, p_adj. = 0.049; E. lenta, p_adj. = 0.003), and 10 had decreased relative abundance (S. faecalis, p_adj. = 0.042; A. intestini, p_adj. = 0.048; Olsenella sp. GAM18, p_adj. = 0.01; A. fermentans, p_adj. = 0.015; E. hirae, p_adj. = 0.003; B. angulatum, p_adj. = 0.013; C. aerofaciens, p_adj. = 0.042; T. succinifaciens, p_adj. = 0.013; M. funiformis, p_adj. = 0.009; S. dextrinosolvens, p_adj. = 0.022) in the post-eradication subgroup compared to the pre-eradication subgroup.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Significantly associated microbial species between the pre- and post-eradication states (at a significance level p<0.05).

Section A–differentially abundant microbial species between pre- and post-eradication states; Section B–the significance of the enrichment calculated by a Wilcoxon test with FDR multiple hypothesis correction; Section C–generalized fold change of each significantly associated microbial species; Section D–the prevalence shift of each significantly associated microbial species; Section E–the Area Under the Receiver Operating Characteristics Curve (AU-ROC) as a non-parametric measure of the enrichment.

https://doi.org/10.1371/journal.pone.0289879.g003

Given that the majority of ESBLs detected during this work originated from the Enterobacteriaceae family, we decided to compare the abundance of enterobacteria between treatment states in the metagenomic dataset. Our ESBL targeted sequencing data showed that the bla_EC gene group with an annotated source of E. coli predominated in both treatment states, but in our metagenomic data, the incidence of E. coli was only 0.129% before and 0.338% after the eradication therapy. However, although this change in abundance is indeed low, it represents 2.6-fold difference (p>0.05). The relative abundance of entire Enterobacteriaceae family was 1.29% in the pre-eradication state and 2.44% in the post-eradication state, representing a 1.9-fold difference in favor of the post-eradication state (p>0.05). Further on assessing BL genes with significant distribution (Table 2) among the treatment states, Klebsiella sp. was one of the dominant annotated sources of ESBLs (Fig 3A, 3B). In metagenomic data, relative abundance of Klebsiella sp. was 0.062% in the pre-eradication state samples and 0.022% in the post-eradication state samples, which represents a 2.9-fold difference and agrees well with the targeted sequencing results. Bacterial species such as Pseudomonas aeruginosa (pre = 0.005%, post = 0.006%), Ralstonia picketii (pre = 0.014%, post = 0.004%) and Serratia marescens (pre = 0.0001%, post = 0.0011%) were at very low abundance in all samples, whereas A. xylosoxidans, S. maltophilia, C. piscinae, R. mannitolilytica, N. farcinica, B. clausii and A. baumannii were not found in the metagenomic dataset. When comparing the outcomes of metagenomic sequencing with targeted sequencing, the observed variance may be due to differences in applied wet-lab and data analysis methodologies, as well as to insufficient sequencing depth to detect very low abundance microorganisms.

Assessing Shannon and Chao1 indices and the observed OTUs (S1 File), only the Shannon index differed significantly between the pre- and post-eradication subgroups (p_adj. = 0.019). Furthering, non-metric multidimensional scaling was applied to assess sample-to-sample variability (S1 File). The analysis revealed non-specific clustering of samples at the species level, suggesting mutual sample similarity irrespective of gender and treatment state. In addition, four of the 27 study subjects continued to test positive for H.pylori after receiving the eradication medication, and the data that are available indicate that there were no incidences of recurrence.

3.4 Functional analysis of the metagenomic profile

The functional profiles of the samples were very similar across treatment states, with the predominant gene products being tyrosine recombinase XerC (in both pre- and post-eradication subgroups: 0.59%), TonB-dependent receptor P3 (pre = 0.53%, post = 0.48%), adaptive response sensory kinase SasA (pre = 0.4%, post = 0.45%), the sensor histidine kinase RscC (pre = 0.32, post = 0.37%) and the TonB-dependent receptor SusC (pre = 0.32%, post = 0.25%) (Fig 4). Furthermore, the resulting UniProt IDs were converted to Gene Ontology (GO) annotations, and the original annotations that did not have the corresponding GO IDs were removed from the dataset. Thus, in total 1’993 unique GO IDs were obtained. To assess the relationship between functional profile and treatment state, association analysis was performed revealing 18 GO ID entries that were significantly different between pre- and post-eradication subgroups (S1 File). Significant associations between GO IDs and treatment state were found for entities related to molecule (GO:0035442, q-value = 0.012; GO:0034219, q-value = 0.012) and ion transport (GO:0006811, q-value = 0.005; GO0042777, q-value = 0.012; GO:0015693, q-value = 0.013; GO:0006817, q-value = 0.043), several biosynthetic processes (GO:0019242, q-value = 0.032; GO:0045226, q-value = 0.043), including cobalamin (GO:0009236, q-value = 0.032) and dTMP (GO:0006231, q-value = 0.044) biosynthesis, DNA restriction-modification system (GO:0009307, q-value = 0.012), metabolic processes (GO:0019568, q-value = 0.044; GO:0006541, q-value = 0.044; GO:0019243, q-value = 0.032), DNA-templated transcription and initiation (GO:0001123, q-value = 0.032), ribosomal small subunit biogenesis (GO:0042274, q-value = 0.012), adenine salvage (GO:0006168, q-value = 0.044) and cellular phosphate ion homeostasis (GO:0030643, q-value = 0.044). Notably, most of the abundances of significant GO IDs entities were increased in the post-eradication subgroup (n = 15), while only a few were increased in the pre-eradication subgroup (n = 3). GO entities that exhibited increased abundance in the pre-eradication subgroup and decreased abundance in the post-eradication subgroup were related to ribosomal small subunit biogenesis, DNA-templated transcription and initiation, and adenine salvage.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Top 50 most abundant gene products obtained by shotgun metagenomic sequencing between the pre- and post-eradication states.

https://doi.org/10.1371/journal.pone.0289879.g004

The diversity of antimicrobial resistance (AMR) genes found in the metagenomic sample set was revealed by additional mapping to the CARD database. Thus, we identified 54 gene families associated with AMR, conferring five different mechanisms of resistance (S3 Table). A total of 1’080 AMR genes were detected in the pre-eradication subgroup, of which 125 had 100% sequence similarity and 955 had strictly significant (≥95% identity) sequence similarity to the CARD reference database. We detected 1’265 AMR genes in the post-eradication subgroup, of which 179 had 100% sequence similarity and 1’086 had strict (≥95% identity) sequence similarity to the CARD reference database. AMR genes present in at least 70% of samples in each treatment state (n = 20) were the antibiotic target alteration genes ErmB and ErmF; antibiotic inactivation genes chloramphenicol acetyltransferase, aadS and lnuC; antibiotic efflux genes Mef(En2), tet(40), and adeF; antibiotic target alteration gene rpoB; antibiotic target protection genes tet(W/N/W), tet32, tetM, tetO, tetQ, and tetW; and the antibiotic target replacement gene dfrF. AMR genes present in all patient samples in the pre-eradication subgroup were antibiotic efflux gene adeF, antibiotic target protection gene tetQ, and antibiotic target replacement gene dfrF. In contrast, AMR genes present in all patient samples of the post-eradication study subgroup were the antibiotic target alteration gene ErmF; antibiotic efflux genes tet(40) and adeF; antibiotic target protection genes tetO, tetQ, and tetW; and the antibiotic target replacement gene dfrF. After applying a hierarchical clustering analysis, 12 samples collected from different individuals and within a sample cluster showed a similar reservoir of detected AMR genes. Only two study subjects had pre- and post-eradication subgroup samples, respectively. The identified gene groups included rifamycin-resistant beta-subunit of RNA polymerase (rpoB) genes, resistance-nodulation-cell division antibiotic efflux pump genes, pmr phosphoethanolamine transferase genes, multidrug and toxic compound extrusion transporter genes, major facilitator superfamily antibiotic efflux pump genes, macrolide phosphotransferase genes, ampC-type BL genes, general bacterial porins with reduced permeability to BL genes, ATP-binding cassette antibiotic efflux pump genes, and others.