Identification of genetic determinants of antibiotic resistance in Helicobacter pylori isolates in Vietnam by high-throughput sequencing

Abstract

The aim of this study was to identify genetic factors responsible for antibiotic resistance in Helicobacter pylori, a bacterium that can cause long-term gastroduodenal disease. The primary resistance of H. pylori to commonly used antibiotics was studied, and high-throughput next-generation sequencing (NGS) was employed to discover genetic determinants of resistance using a reference-based approach. A total of 123 H. pylori strains were cultured and tested for antibiotic susceptibility using an E test. Genotypic analysis was performed using NGS data with ARIBA v2.14.7 and PlasmidSeeker v1.3 for plasmid detection. Statistical correlations between resistant genotypes and phenotypes were evaluated. In addition, a genome-wide association study (GWAS) and linear mixed model were used to identify genetic variants associated with antimicrobial resistance phenotypes while adjusting for covariates such as population structure. Our results showed that 78.2% of the strains were resistant to metronidazole (MTZ), 22.5% to levofloxacin (LVX), 43.5% to clarithromycin (CLR) and 13.7% to amoxicillin (AMX). Resistance to tetracycline was not detected. Multi-drug resistance was detected in 48.8% of the strains. While plasmids were not detected, chromosomal genetic determinants of resistance to CLR, LVX, and AMX were identified, including mutations in 23S rRNA (A2142G and A2143G), gyrA (N87K/Y and D91Y/N/G), and pbp1 A (F366L, S414R, F473V, G595_V596insE, as well as the mutations T558S and T593A/G/P/S). Additionally, missense, frameshift, and nonsense mutations in rdxA were identified as genetic determinants of resistance to MTZ. No genetic determinants associated with tetracycline resistance were detected. A strong correlation was observed between resistance genotypes and phenotypes for CLR, LVX, AMX, and MTZ. In addition, we found that missense, frameshift and nonsense mutations in rdxA were genetic determinants of resistance to MTZ. We did not detect any genetic determinants associated with tetracycline resistance. There was a strong correlation between resistance genotypes and phenotypes for CLR, LVX, AMX, and MTZ. Furthermore, unitig-based GWAS revealed that AMX, LVX, and CLR resistance in H. pylori was mainly caused by chromosomal mutations that affected the targets of these antibiotics (pbp1 A, gyrA, and 23S rRNA, respectively). Our results highlight the need for regular evaluation and alternative therapies in Vietnam, given the high rates of H. pylori resistance to CLR, MTZ, and LVX. Our study also demonstrated the high capacity of NGS to detect genetic resistance determinants and its potential for implementation in local treatment policies.

Introduction

Long-term Helicobacter pylori infection in the human stomach is associated with gastroduodenal disease and an increased risk of developing gastric cancer [1]. H. pylori can cause dysbiosis in the gastric microbiota and activate internal signaling in the carcinogenesis pathway [2]. Early treatment of H. pylori infection can restore homeostasis, which is crucial for preventing severe clinical outcomes [2]. H. pylori antimicrobial resistance is increasing globally, particularly in the Asia–Pacific region, leading to treatment failure and burdening the healthcare system and patients [3]. The most recently reported primary resistance prevalences in the Asia–Pacific region were 30% for clarithromycin (CLR), 61% for metronidazole (MTZ), 35% for levofloxacin (LVX), 4% for tetracycline (TET), and 6% for amoxicillin (AMX) [3]. Routinely performing antibiotic susceptibility testing (AST) on each isolate in clinical laboratories, is recommended in certain situations to guide treatment, especially when patients have failed initial treatment, have allergies to first-line antibiotics, or live in areas with high resistance rates.

Traditional culture-based methods for detecting H. pylori infection are limited by long wait times and the need for well-equipped laboratories. Furthermore, interpreting the results is subjective and difficult to replicate. However, PCR-based techniques offer faster results and can be used directly on various samples. Additionally, molecular-based test kits that detect H. pylori and mutations that confer resistance to CLR and LVX are available, providing results within hours. However, detecting resistance to MTZ and AMX remains challenging because of the diverse molecular mechanisms involved [4]. High-throughput sequencing technologies, including whole-genome sequencing (WGS), offer promising alternatives for monitoring H. pylori antibiotic resistance, as well as infection with antimicrobial resistant strains, in clinical microbiology laboratories [5].

The mechanism by which H. pylori eliminates antibiotics has been extensively studied. Three resistance profiles have been identified, namely: single drug resistance, multidrug resistance, and heteroresistance [6]. These resistance profiles have overlapping fundamental mechanisms and clinical implications. The most common molecular mechanisms that have been studied involve chromosomal mutations that alter the structure of antibiotic targets [4]. Advances in understanding the fundamental molecular aspects of drug resistance in H. pylori have facilitated the development of molecular techniques for the rapid detection of antibiotic resistance in clinical specimens [7]. Specific mutations in domain V of 23S rRNA, which is a component of the bacterial protein translation machinery, cause CLR resistance [8]. The two most common mutations in 23S rRNA, A2142G and A2143G, are the main causes of CLR resistance in H. pylori [4]. Additionally, resistance to LVX is mainly caused by the specific mutations N87 K/I/Y/H and D91G/N/Y/A in the quinolone-resistance-determining region (QRDR) of the gyrA gene, which encodes DNA gyrase [9]. PCR-based techniques are used to detect these mutations in H. pylori clinical specimens [7]. Mutations in the gene encoding pbp1 A, which possesses both transglycosylase and transpeptidase activity, confer resistance to AMX [4]. The mechanism of MTZ resistance primarily involves the inactivation of both nitroreductase enzymes, frxA and rdxA. However, the mechanisms underlying resistance to MTZ and AMX are complex and still not well understood [4].

Bacterial genome-wide association studies (GWASs) are a new and exciting research technique in which human GWAS methods are used to understand how variations in microbial genomes affect host or pathogen phenotypes, such as drug resistance, virulence, host specificity, and prognosis [10]. However, this approach is prone to false positives due to correlations among genetic variations. H. pylori, which has a high rate of homologous recombination, does not exhibit this correlation, thus eliminating the problem of false positives [11, 12]. Although chromosomal mutations are widely accepted genetic determinants of antibiotic resistance in H. pylori, the H. pylori genome contains many genes that help the bacterium adapt to environmental changes such as antibiotics, host interactions, and the gastric environment [13].

The aim of this study was to determine how H. pylori develops antibiotic resistance by analyzing WGS data. Both reference-based and panGWAS approaches were used to identify genetic variations linked to antimicrobial resistance in H. pylori.

Methods

Bacterial isolation, whole-genome sequencing, and assembly

Using conventional culture techniques, H. pylori strains were isolated from patients who underwent upper endoscopy at Cho Ray Hospital in Ho Chi Minh City, Vietnam and 108 Military Hospital in Hanoi, Vietnam from November 2012 to May 2014 as described in our previous studies [14]. Exclusion criteria included a history of gastrectomy; previous H. pylori eradication therapy; treatment with bismuth-containing compounds, H2-receptor blockers, or proton pump inhibitors within 2 weeks of the start of the study; and patients with suspected recurrent Billroth anastomosis cancer, entire stomach tumor, or cardiac tumor. Briefly, antral biopsy specimens were homogenized, inoculated onto H. pylori–selective plates (Nissui Pharmaceutical Co., Ltd., Tokyo, Japan), and incubated for 3 to 10 days at 37 °C under microaerophilic conditions. The resulting purple colonies were subcultured in Brucella broth (Becton, Dickinson and Company, Franklin Lakes, NJ, USA) supplemented with 7% horse blood. Bacterial DNA was extracted in 2014 using the DNeasy Blood and Tissue Kit (Qiagen Inc., Hilden, Germany) for whole-genome sequencing. DNA concentrations were quantified using a Quantus Fluorometer (Promega Corporation, Madison, WI, USA) to confirm a minimum of 10 ng double-stranded DNA (dsDNA) per sample required for library preparation. Library preparation was performed using the TruSeq Nano DNA Library Prep Kit according to the manufacturer's protocol. Bacterial genomes were sequenced on either the HiSeq 2500 platform (Illumina Inc., San Diego, CA, USA) with 2 × 100 or 2 × 150 bp paired-end reads, or the MiSeq platform with 2 × 300 bp paired-end reads, in accordance with the manufacturer's instructions.

The complete genome sequence of H. pylori was obtained using a long-read sequencing platform (Pacific Biosciences of California, Inc., Menlo Park, CA, USA) and used as a reference. The SMRTbell Template Prep Kit 1.0 (Pacific Biosciences of California, Inc., Menlo Park, CA, USA) was used to prepare a SMRTbell library. DNA fragments greater than 17 kbp were selected using the BluePippin system (Sage Science, Inc., Beverly, MA, USA). Each H. pylori strain was sequenced on one SMRT cell using the PacBio RS II System with P6/C4 or P6/C4v2 chemistry and 360-min movies. The SMRT sequencing data were analyzed using SMRT Analysis v2.3.0 through the SMRT Portal. RS_HGAP_Assembly.2 was used to assemble the reads. The overlapping ends were trimmed, and the chromosomal contig was started from the ori-sequence. The contig was then resequenced with RS_Resequencing.1 to generate consensus sequences.

Determination of antibiotic resistance phenotypes and genotypes

A total of 123 clinical H. pylori isolates, previously obtained in a study conducted in Vietnam [14], were analyzed. From 2012–2014, the antimicrobial susceptibility testing (AST) was assessed using Epsilometer-test (Etest®, bioMerieux) for five antibiotics (amoxicillin, clarithromycin, levofloxacin, tetracycline, and metronidazole) following European Committee on Antimicrobial Susceptibility Testing (EUCAST) v.21.01.01 protocols on Müller–Hinton agar plates supplemented with 5% horse blood. The minimum inhibitory concentrations (MICs) of the antibiotics were monitored daily after incubation for 3–6 days. H. pylori strain 26,695 was used as a control strain. The clinical breakpoints distinguishing between resistant and susceptible strains were determined in accordance with the EUCAST guidelines, accessible at http://www.eucast.org/.

Bioinformatics analysis

The integrity of the raw sequence data obtained from high-throughput sequencing was assessed using FastQC v0.12.1. Reads of substandard quality (Q < 15), along with adapter sequences and sequences shorter than 36 residues, were eliminated using Trimmomatic v0.32 [15]. Three bases were truncated from both the beginning and end of a read if the sequence fell below the quality threshold, using the following function: TruSeq3-PE.fa:2:30:10:2:True LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:36. The raw sequence reads from each strain were assembled into contigs using Unicycler v0.5.0 [16]. Contigs shorter than 200 base pairs were excluded. To identify genome features within a collection of genomic DNA sequences, the assembled H. pylori genomes were annotated using Prokka v1.14.5, a rapid prokaryotic genome annotation pipeline [17]. The core genome alignment was extracted from the pangenome analysis of 123 strains using Panaroo v1.3.4, based on the GFF files generated by Prokka [18]. The maximum-likelihood (ML) phylogenetic tree based on core genome alignment was generated by IQTREE v2.2.2.7 using the generalized time-reversible model, 1000 bootstrap replicates, and ancestral state [19]. The ML phylogenetic tree, the distribution of susceptible/resistant phenotypes, and the distribution of genetic determinants of antibiotic resistance were visualized using Phandango [20].

Analysis of drug susceptibility genotypes encoded by plasmids in H. pylori isolates

Short-read sequences were used to identify plasmids that encode genes related to antibiotic resistance. This was achieved using PlasmidSeeker v1.3, with H. pylori 26,695 (NC_000915.1) serving as the reference genome [21]. The read coverage of short DNA fragments of length k (k-mers) was analyzed against a publicly available plasmid reference database (http://bioinfo.ut.ee/plasmidseeker/). The abundance of k-mers was used to differentiate between plasmid and bacterial sequences. A plasmid was deemed to be present if at least 80% of the k-mers were detected.

Analysis of drug susceptibility genotypes encoded by the genomic DNA of H. pylori isolates

H. pylori genomic sequences associated with resistance phenotypes to AMX, LVX, CLR, TET, and MTZ were retrieved and used to construct a variant-calling database. Reference sequences were obtained from H. pylori strain 26,695 (CP003904.1), including pbp1 A (AMX), 23S rRNA (CLR), gyrA (LVX), gyrB (LVX), 16S rRNA (TET), and rdxA (MTZ). ARIBA v2.14.7, a local-assembly tool for identifying antibiotic resistance genes [22], was used to detect relevant genetic variants. For variant calling, ARIBA used two inputs: a FASTA file of reference sequences, and paired-sequencing reads. Detailed information, including variants and amino acid changes in both coding and noncoding sequences between the sequencing reads and the reference sequences, is reported.

PanGWAS approach to identifying genetic determinants of antibiotic resistance

The PanGWAS analysis was performed on a total of 255 H. pylori strains, comprising 159 WGS strains from the present study and 96 WGS strains obtained from two previous studies (Table S8) [5, 23]. To test associations between genetic variants and antibiotic resistance phenotypes, a single pan-genome-wide association study (panGWAS) was conducted using the Fast-LMM (factored spectrally transformed linear mixed model) algorithm implemented in pyseer v.1.3.11 [24]. PySeer estimates the effect of genetic variation in a bacterial population on a phenotype of interest while accounting for potentially very strong confounding population structure. The binary phenotype file for each antibiotic is supplied, with the sample names in the first column and the phenotype (1: resistant, 0: susceptible) in the last column. To reduce both computational and statistical complexity, unitigs identical to k-mers were used as input via the –kmers option. Unitigs are nodes in a compressed de Bruijn graph, and using them eliminates some of the redundancy present in k-mer counting, as well as requiring fewer tests [25]. The unitigs, which accounted for all sequence elements in the bacterial sequence data, were counted using the unitig-caller v.1.3.0 package (https://github.com/bacpop/unitig-caller). To reduce the false discovery rate, unitigs with a minor allele frequency greater than 5% were extracted. A unitig-based GWAS analysis was performed following the pyseer manual, utilizing a linear mixed regression model to control for population structure based on an n x n relatedness matrix calculated froma core-genome phylogenetic tree of H. pylori strains (https://pyseer.readthedocs.io/en/master/). A Q‒Q plot was created using the qq_plot.py script provided by PySeer to assess the number and magnitude of observed associations between the number of unitigs and antibiotic resistance compared with the association statistics expected under the null hypothesis of no association.

Results

Primary H. pylori antibiotic susceptibility in Vietnam

Table 1 shows the antibiograms of 123 isolates from Vietnam. MTZ resistance was the most common—it was observed in 78.2% (95% CI, 69.7; 84.9) of the isolates. This was followed by CLR resistance in 43.5% (95% CI, 34.7; 52.7), LVX resistance in 22.5% (95% CI, 15.7; 31.1), and AMX resistance in 13.7% (95% CI, 8.4; 21.3). No isolates were found to be resistant to TET. Only 14 isolates (11.3%, 95% CI: 6.5; 18.5) were found to be susceptible to all antibiotics tested (Table S7).

Table 1 Antibiotic resistance of 123 H. pylori isolates from Vietnam

Notably, 39.5% (95% CI, 30.9; 48.7) of the isolates exhibited single drug resistance, 32.2% (95% CI, 24.3; 41.3) were MTZ-resistant, and 5.6% (95% CI, 2.5; 11.7) were CLR-resistant. Only one isolate each exhibited single drug resistance to AMX and to LVX. The rate of multidrug resistance was high, at 49.2% (95% CI, 40.1; 58.2). Dual drug resistance to CLR + MTZ and LVX + MTZ was found in 18.5% (95% CI, 12.3; 26.7) and 8.8% (95% CI, 4.7; 15.6) of the isolates, respectively and dual drug resistance to AMX + MTZ and AMX + CLR groups was found in 1.6% (95% CI, 0.2; 6.2) and 3.2% (95% CI, 1.0; 8.5) of the isolates, respectively. Triple drug resistance to CLR + LVX + MTZ, AMX + CLA + MTZ, and AMX + LVX + MTZ accounted was observed in 8.8% (95% CI, 4.7; 15.6), 4.0% (95% CI, 1.4; 4.6), and 0.8% (95% CI, 0.0; 5.0) of isolates, respectively. Finally, 3.2% (95% CI, 1.0; 8.5) of the isolates were found to be resistant to AMX + CLR + LVX + MTZ (Table S7).

Relationship between antibiotic resistance patterns and strain relationships in Vietnamese H. pylori isolates

Reconstruction of a maximum likelihood tree, which pertains to core single nucleotide polymorphisms concerning varying resistance patterns (Fig. 1) showed no association between strain relatedness and resistance to any individual antibiotic or combination of drugs. In contrast, the isolates exhibited unique susceptibility profiles, irrespective of their genetic relatedness.

Fig. 1

The maximum likelihood phylogenetic tree based on whole-genome sequences, resistance patterns, and their corresponding genetic determinants for each antibiotic in the 123 H. pylori genomes sequenced in this study. The antibiotic sensitivity and resistance patterns are represented by dark blue and red rectangles, respectively. The presence and absence of mutations are indicated by green and pink rectangles, respectively

Plasmid detection in H. pylori isolates from Vietnam

A plasmid reference database including a total of 19,782 known plasmids from published species, 69 of which are specific to H. pylori, was used to determine whether any of the isolates carried one or more plasmids. No plasmids encoding antibiotic resistance genes were found among the 123 analyzed H. pylori genomes.

Genetic determinants of antibiotic resistance in Vietnamese H. pylori isolates

CLR resistance

Our investigation of CLR resistance focused on genetic variations within the full-length 23S rRNA gene. A single point mutation in the peptidyl transferase region of domain V of this gene is known to be responsible for CLR resistance. Given the presence of two copies of the 23S rRNA gene in the H. pylori genome, the depth of reads mapped to each nucleotide variation position was analyzed to detect allelic variation. A total of 186 variants were identified among 53 CLR-resistant and 70 CLR-susceptible strains. Notably, only two mutations, A2142G (p = 4 × 10^–11) and A2143G (p = 5 × 10^–23), were significantly associated with CLR resistance (Table 2). Among the CLR-resistant strains, 85% (51/60) carried the A2143G mutation, 6.7% (4/60) carried the A2142G mutation, and 8.3% (5/60) did not have either mutation. Conversely, all but one of the CLR-susceptible strains (98.6%, or 76/77) carried neither of the mutations.

Table 2 Summary of the known and novel mutations in the 23S rRNA gene in H. pylori isolates

Furthermore, strains harboring either the A2142G or A2143G mutations exhibited significantly greater CLR minimum inhibitory concentrations (MICs) than those without mutations (all p < 0.05) (Fig. 2A). However, there was no significant difference in the average MICs among strains with the A2142G mutation and those with the A2143G mutation. Interestingly, all reads (100%) from the 54 CLR-resistant strains mapped to either A2142G or A2143G, indicating that the mutations occurred in both copies of the 23S rRNA gene. Furthermore, two of the CLR-resistant strains carried the A2143G mutations in only one copy of the 23S rRNA gene (Table S1). This finding underscores the uniformity of the mutations in the CLR-resistant strains.

Fig. 2

A Distribution of clarithromycin minimum inhibitory concentrations (MICs) and corresponding mutations for all observed combinations of relevant antibiotic resistance determinants. B Distribution of levofloxacin MICs for all observed combinations of relevant antibiotic resistance determinants. Dotted horizontal lines mark clinical breakpoints. *** indicates p < 0.001 compared with isolates without mutations, ** indicates p < 0.01, * indicates p < 0.05

LVX resistance

Point mutations in the QRDR region of gyrA confer resistance to LVX in H. pylori. Genetic variations in gyrA were examined in 95 LVX-susceptible and 28 LVX-resistant strains. Among 213 coding mutations identified, only N87 K/Y and D91 N/G/Y (p < 0.001) were significantly associated with LVX resistance (Table 3 and Table S2). Notably, 21.4% (6/28), 14.28% (4/28), 3.5% (1/28), and 51.7% (15/28) of the LVX-resistant strains exhibited Gly-91, Asn-91, Tyr-91, and Lys-87/Tyr-87 mutations, respectively. Among the 28 LVX-resistant strains, 26 (92.8%) carried mutations at codons 87 and 91. Conversely, four strains did not exhibit any mutations at these codons, yet were LVX-resistant (Table S2).

Table 3 Summary of known and novel gyrA mutations in H. pylori isolates

In general, strains with mutations at either codon 87 or 91 had significantly greater LVX MICs than those without mutations (all p < 0.05) (Fig. 2B). Interestingly, only one strain was mutated at both codon 87 and codon 91, and it had a MIC of 4 µg/mL (Fig. 2B). Furthermore, it has been observed that two known variants D481E and R484 K of gyrB were detected in two LEV-R strains, while they were absent in LEV-S strains (Fig. S1 and Table S5).

AMX resistance

A comprehensive examination of pbp1 A mutations was conducted in 17 AMX-resistant and 106 AMX-susceptible strains, given the known role of this gene in AMX resistance. A total of 176 distinct coding mutations in pbp1 A were identified (Table S3). Notably, isolates with four different mutations—F366L (p < 0.001), S414R (p < 0.001), F473 V (p < 0.001), and G595_V596 insE (p < 0.001)—also carried a T558S mutation. T593 A/G/P/S (p < 0.001) mutations were significantly associated with AMX resistance (Table 4). In conclusion, strains harboring mutations at codons 366, 414, 473, 558, 593, and 595 demonstrated significantly higher AMX MICs than those without mutations (all p < 0.05) (Fig. 3A).

Table 4 Summary of pbp1 A mutations in H. pylori isolates

Fig. 3

A Distribution of amoxicillin minimum inhibitory concentrations (MICs) and corresponding mutations. B Distribution of metronidazole MICs for all observed combinations of relevant antibiotic resistance determinants. Dotted horizontal lines mark clinical breakpoints. *** indicates p < 0.001 compared with groups without mutations, ** indicates p < 0.01, * indicates p < 0.05

Metronidazole resistance

Loss-of-function (e.g., missense, nonsense, and frameshift) mutations in rdxA play a crucial role in MTZ resistance. All the MTZ-resistant strains had at least one missense/nonsense/frameshift mutation in the rdxA gene (Fig. 3B and Table S4 and S6). The MICs of most strains were greater than 8, indicating that mutation of this gene abolishes its function, leading to an increase in MTZ resistance. Several missense mutations were identified, including C19Y, K20 T, S43L, V55G, M56 V/L, N73 K, G145R, G162R, and G163D/S (Table S6). These missense mutations have been shown to encode functional codons in a previous study [26]. In addition to the nonsense mutations, frameshift and truncation mutations in rdxA were also identified (Table S6). As shown in Fig. 3B, the isolates carrying rdxA frameshift and truncation mutations were generally resistant to MTZ, indicating that these two types of mutations have a significant impact on rdxA function.

Correlations among antibiotic susceptibility genotypes and phenotypes

The effectiveness of WGS as a tool for predicting bacterial susceptibility to various antibiotics was assessed. To do this, the antibiotic resistance phenotypes were compared with the presence and absence of mutations. The results, as presented in Table 5, demonstrated a significant degree of concordance between genotypic and phenotypic drug resistance. This finding implies that the genetic makeup of bacteria (genotype) is a reliable predictor of their resistance or susceptibility to antibiotics (phenotype). More specifically, the concordance was found to be in perfect agreement for CLR (0.95), strong for AMX (0.90), moderate for MTZ (0.81), and in substantial agreement for LVX (0.60).

Table 5 Agreement between antibiotic resistance genotypes and phenotypes of in H. pylori strains isolated in Vietnam

Pangenome analysis of antibiotic resistance in H. pylori

A De Bruijn graph, also known as a pangenome graph, was constructed from a total of 255 H. pylori genomes. Accession numbers and assembly levels of all genomes used were provided in Supplementary Table S8. The pangenome analysis included a total of 3,928,855 unitigs. Filtering based on a minor allele frequency of at least 5%excluded 2,346,859 unitigs, and the remaining 1,581,996 unitigs were tested for their associations with antibiotic resistance. Q‒Q plots were created using the P values of the associations between unitigs and antibiotic resistance (Fig. 4) to assess the number and magnitude of the observed associations compared with the statistical values expected for the null hypothesis of no association. The pangenome GWAS showed that the P values of most unitigs were as expected under the null hypothesis for AMX, LVX, CLR, and MTZ resistance (Fig. 4A, 4B, 4 C, 4D). In addition, there was a strong association between LVX resistance, CLA resistance, and AMX resistance and the unitigs.

Fig. 4

Q‒Q plot constructed to assess the genome-wide association study results. Each dot in (A), (B), (C), and (D) indicates a unitig. Y-axis: observed − log10(P) of each unitig, where P is its P value. X-axis: expected − log10(P) under the null hypothesis of no association

The hits column indicates the number of unitigs mapping to the pangenome. -log10(P value): The -log10 of the Bonferroni P value of the unitigs. The average minor allele frequency is the second highest frequency of variants in the population. Average beta: effect size of variants on the antimicrobial resistance phenotype.

Mapping significant unitigs to our isolate dataset identified 42 and four genes related to AMX and LVX resistance, respectively (Fig. S2 and Table 6). In contrast, significant CLR-resistance unitigs only mapped to 23S rRNA, while no unitigs were found to be associated with MTZ-resistance—all P values were under the significance threshold. In brief, the pbp1 A, gyrA, and 23S rRNA genes exhibited the strongest correlations with AMX-resistance, LVX- resistance, and CLR- resistance, respectively (Table 6). Genetic variations in 23S rRNA were most strongly related to CLR resistance, with an effect size of 0.84, a -log10 (P value) of 77, and 77 hits. For AMX resistance and LVX resistance, pbp1 A and gyrA mutations were most common, with 15 and 21 hits, respectively, and -log10(P values) of 12.7 and 8.7, respectively (Table 6). Additionally, two genes associated with AMX resistance, encoding components of the type IV secretion system: virB11 and virD4—were identified. Both virB11 and virD4 play crucial roles in DNA processing during horizontal gene transfer. Furthermore, lpxD, which encodes UDP-3-O-acylglucosamine N-acyltransferase, a bacterial outer membrane biogenesis protein, and RluB, which encodes RNA pseudouridine synthase, an enzyme-directed rRNA binding gene, correlated with AMX resistance [27, 28]. In addition to gyrA, we identified another candidate gene, tlpC, that was strongly associated with LVX resistance. tlpC gene encodes a methyl-accepting chemotaxis transducer that plays a vital role in lactate utilization in H. pylori [11].

Table 6 Genes significantly related to antibiotic resistance and their functions

Discussion

The aims of this study were to identify genetic factors associated with antibiotic resistance and to investigate the practicality of using genomic NGS techniques to monitor antibiotic resistance in H. pylori clinical samples originating from Vietnam. As a preliminary step, the susceptibility of H. pylori strains to AMX, CLR, LVX, and MTZ were evaluated. These critical antimicrobials are traditionally used in both the first- and second-line treatment of H. pylori infection, typically in combination with proton pump inhibitors.

A large dataset of Illumina-based WGS data were examined and detected no plasmids in our isolates, confirming that all resistance determinants were encoded within the bacterial genome. Next, WGS-based approach were employed to identify mutations in genes related to antimicrobial resistance in each isolate. Our findings showed that phenotypic resistance to AMX was primarily associated with mutations in the pbp1 A gene located within penicillin-binding protein (PBP) motifs, specifically SAIK367-371, KTG555-557, and SNN559-561, as well as within the C-terminus at codons T593 and G595. Intriguingly, novel mutations were also identifed near or outside the PBP motifs, namely: F366L, S414R, F473 V, S543R, T558S, and N562H/Y. These mutations may confer AMX resistance, and thus necessitate further investigation in vitro. Notably, previous studies have also reported the presence of mutations in pbp1 A outside of the PBP motif and C-terminus that are relevant to AMX resistance, suggesting that other regions of this gene are potentially significant to the development of AMX resistance [6, 29,30,31]. Thus, the AMX resistance observed in our isolates could be due to reduced affinity of the antibiotic for the protein encoded by the pbp1 A gene. The CLR resistance observed in our H. pylori isolates stemmed from two point mutations in the peptidyl transferase–encoding region of the 23S rRNA that account for nearly 90% of primary CLR resistance in H. pylori isolates from Western countries. The predominance of A2143G in CLR resistance (92%) raises concerns because this is the main mutation thought to be responsible for failure of H. pylori eradication. Although mutations outside of domain V of the 23S rRNA gene (T2182 C, C2195 T) have been reported to be linked to CLR resistance, any such mutations were not detected in our isolates [32, 33]. The LVX resistance observed in our study was primarily attributed to mutations in the QRDR regions of the gyrA and gyrB genes. In the gyrA QRDR, mutations were detected at codons 87 (N to K, Y) and 91 (D to N, G, Y) in 87% of the resistant strains. Additionally, the A231 V mutation, which is located outside the gyrA QRDR, was seen more frequently in LVX-resistant isolates than in LVX-sensitive isolates (p < 0.001). The relevance of this mutation to LVX resistance requires further investigation.

Various mutations in rdxA and frxA, such as null mutations, frameshift mutations, and missense mutations, have been reported to confer resistance to MTZ. In our study 97 out of 123 samples (78.8%) exhibited MTZ resistance, and all of the resistant strains carried rdxA mutations, confirming that mutations in rdxA were responsible for the high rate of MTZ resistance. Some of the isolates with mutations exhibited an MTZ MIC ≤ 8 µg/ml, and all of these mutations were missense mutations. In contrast, the MTZ-resistant isolates predominantly carried frameshift and truncation mutations. In total, 24 out of 27 (88.8%) of the MTZ-resistant isolates analyzed in our study carried missense mutations. Our previous study, conducted in Cambodia, indicated a 96.4% rate of MTZ resistance, and loss of function due to nonsense or frameshift mutations in rdxA was shown to play an important role in MTZ resistance. Out of the 30 MTZ-resistant Cambodian isolates examined, 15 (50%) carried frameshift mutations, and 15 (50%) carried nonsense mutations. However, the kappa index results from the current study were similar to those from the study performed in Cambodia [5]. There are two factors contributing to the low Cohen's kappa values observed for LVX (0.6) and MTZ (0.81). First, the gyrA/gyrB and rdxA genes may have multiple mutations, some of which may not directly correlate with phenotypic resistance. This genetic variability can lead to discrepancies between genotype and phenotype [34]. The rdxA gene can have various mutations, including missense, nonsense, and frameshift mutations. Not all of these mutations result in a functional change that affects metronidazole activation [35]. Certain mutations in rdxA are related to the phylogenetic origin of the strain rather than directly causing resistance [36]. Second, bacteria may have compensatory mechanisms that mitigate the effects of certain mutations, leading to discrepancies between the presence of a mutation and the actual resistance phenotype [37].

In this study, genetic variations were explored in the H. pylori pangenome associated with resistance to first- and second-line antibiotics. Using a unitig-GWAS approach, mutations in pbp1 A, gyrA, and 23S rRNA were found to associate with resistance to AMX, LVX, and CLR, respectively, indicating that chromosomal mutations disrupt the interaction between antibiotics and their targets. pbp1 A gene had an effect size of 0.47 on AMX resistance. In H. pylori, pbp1 A encodes PBP1 A, which is involved in peptidoglycan biosynthesis, similar to Streptococcus pneumoniae, in which mutations in PBP1 A confer resistance to beta-lactams [38]. Previous studies have also detected antibiotic resistance–associated mutations in other genes participating in the peptidoglycan biosynthesis pathway, including pbp2x, pbp1 A, pbp2B, mraW, and mraY [38]. In addition to pbp1 A, our GWAS identified novel mutations in four other genes. The first two genes, virD4 and virB11, are involved in the type IV secretion system, which helps translocate effector proteins directly into neighboring cells [39]. The H. pylori type IV secretion system may confer antibiotic resistance by promoting biofilm formation, rendering microbial populations impervious to antibiotics, environmental insults, and host defenses [40]. The type IV secretion system is not present in all H. pylori isolates [41]; and indeed, our previously study found that the severity of H. pylori infection in patients with gastroduodenal disease depends on the presence of the functional type IV secretion system [42]. Additionally, there is a statistically significant correlation between the presence of a functional type IV secretion system and resistance to penicillin in Neisseria gonorrheae [43].

The third and fourth genes with novel mutations identified in this study include rluB and lpxD. lpxD is involved in lipid A biosynthesis in gram-negative bacteria, and mutations in this gene may confer resistance to antimicrobial peptides that target the outer membrane [44, 45]. RluB, which encodes RNA pseudouridine synthase, confers resistance to streptomycin and is an important factor affecting the bacterial response to low doses of antibiotics from different families [28, 46]. The tlpC gene encodes a methyl-accepting chemotaxis receptor protein involved in promoting H. pylori growth in the stomach and mediating responses to lactate attractants; however, its potential role in LVX resistance remains unexplored [47].

Finally, although our results provide valuable insights, they are based on hypothetical scenarios and computational predictions. Future work, in particular in vitro transformation analyse, is crucial to eliminate potential confounding biases, such as concomitant mutations, and to validate the hypotheses presented here [48].

Conclusions

Standard triple therapy for eradicating H. pylori in Vietnam may no longer be effective because of the high rates of resistance to CLR, MTZ, and LVX. Therefore, it is necessary to periodically evaluate antibiotic resistance in this area. Given the increasing resistance rates, clinicians should consider antimicrobial susceptibility testing before prescribing therapy to ensure effective treatment. Reliable genetic markers for the rapid and accurate prediction of H. pylori resistance to CLR, LVX, and AMX were identified: 23S rRNA (A2142G and A2143G), gyrA (N87 K/Y and D91Y/N/G), and pbp1 A (F366L, S414R, F473 V, G595_V596 insE, T558S, and T593 A/G/P/S). While mutations in the rdxA gene are associated with MTZ resistance, their significance varies, as not all mutations result in phenotypic resistance due to genetic variability, phylogenetic background, and potential compensatory mechanisms. We also showed that an NGS-based approach is a useful and reproducible method for detecting antibiotic resistance–related genetic determinants for rapid molecular diagnostics, enabling personalized treatment strategies that can improve eradication rates and reduce treatment failures. Our findings suggest that the mechanisms underlying AMX and MTZ resistance are polygenic and require further exploration. Through our bacterial genome-wide association study, we identified several novel antimicrobial resistance mutation candidates in H. pylori, and further research is necessary to elucidate the mechanism of drug resistance in H. pylori.