Enhanced Fitness of a Helicobacter pylori babA Mutant in a Murine Model

ABSTRACT

Helicobacter pylori genomes encode over 60 predicted outer membrane proteins (OMPs). Several OMPs in the Hop family act as adhesins, but the functions of most Hop proteins are unknown. To identify hop mutant strains exhibiting differential fitness in vivo compared to in vitro, we used a genetic barcoding method that allowed us to track changes in the proportional abundance of H. pylori strains within a mixed population. We generated a library of hop mutant strains, each containing a unique nucleotide barcode, as well as a library of control strains, each containing a nucleotide barcode in an intergenic region predicted to be a neutral locus unrelated to bacterial fitness. We orogastrically inoculated each of the libraries into mice and analyzed compositional changes in the populations over time in vivo compared to changes detected in the populations during library passage in vitro. The control library proliferated as a relatively stable community in vitro, but there was a reduction in the population diversity of this library in vivo and marked variation in the dominant strains recovered from individual animals, consistent with the existence of a nonselective bottleneck in vivo. We did not identify any OMP mutants exhibiting fitness defects exclusively in vivo without corresponding fitness defects in vitro. Conversely, a babA mutant exhibited a strong fitness advantage in vivo but not in vitro. These findings, when taken together with results of other studies, suggest that production of BabA may have differential effects on H. pylori fitness depending on the environmental conditions.

INTRODUCTION

Helicobacter pylori is a Gram-negative bacterium that colonizes the gastric mucosa in about 50% of the world’s population (1). H. pylori colonization of the stomach is a strong risk factor for development of peptic ulcer disease and gastric cancer (2, 3). Gastric cancer is the third leading cause of cancer-related death worldwide, accounting for about 0.8 million annual fatalities (4). The incidence of H. pylori-related gastric cancer varies markedly throughout the world. The risk of gastric cancer is influenced by multiple factors, including characteristics of the H. pylori strains, host genetic variation, and environmental exposures (5). For example, consumption of a high-salt diet is an important dietary risk factor for gastric cancer (6–8).

Approximately 4% of genes in the H. pylori genome (64 genes in the prototype strain 26695) are predicted to encode outer membrane proteins (OMPs) (9, 10). These OMPs have been classified into several families, including the Hop (H. pylori OMP), Hor (Hop-related), Hof (Helicobacter OMP), and Hom (Helicobacter outer membrane) families, as well as iron-regulated OMPs and efflux pump OMPs (9). In H. pylori strain 26695, 21 OMPs have been classified in the Hop family (9). Proteins in the Hop family are characterized by an interrupted β-barrel composed of 7 β-strands at the C terminus and one at the N terminus (11). The extracellular domains located between these N-terminal and C-terminal β-strands are presumed to be determinants of each Hop protein’s distinct functions. Several proteins in the Hop family (AlpA/B, OipA, BabA, SabA, HopQ, HopD [LabA], and HopZ) are reported to have adhesin activity, mediating H. pylori adherence to receptors on host cells (12–21). The functions of most other Hop proteins remain unknown.

Previous studies have analyzed adhesive properties of specific Hop proteins, but there has been relatively little effort to systematically assess the contributions of Hop proteins to H. pylori fitness. In the current study, we evaluated potential contributions of hop genes to H. pylori fitness during growth in vitro and in colonization of the mouse stomach and tested the hypothesis that specific members of this gene family contribute to H. pylori fitness in vivo but not in vitro. To facilitate the analysis, we developed methodology that allowed us to track changes in a population of H. pylori mutant strains over time. Specifically, we labeled individual hop mutant strains with unique nucleotide barcodes and monitored the composition of populations using either next-generation sequencing (NGS) or quantitative PCR (qPCR) methods. Using this approach, we show that mutagenesis of babA results in a marked increase in H. pylori fitness in the mouse model without a corresponding increase in fitness in vitro.

RESULTS

Nucleotide barcoding of H. pylori strains.

To evaluate the fitness of H. pylori strains under different conditions, we used a genetic barcoding approach (Fig. 1). We first generated a “control library” composed of 6 H. pylori strains (LH-1 to LH-6), each containing a unique nucleotide barcode inserted in the mdaB-hydA intergenic region, predicted to be a neutral locus (unrelated to bacterial fitness) (22, 23) (Table S1). In the initial development of methodology for tracking changes in a population of barcoded strains, we conducted experiments to assess the validity and accuracy of the approach. We pooled the six strains in equal proportions, and we also created pools in which one control strain (LH-1) was reduced in proportional abundance by 100-fold or omitted from the pool. As expected, the experimental detection of barcodes matched the known composition of the pools (Fig. S1). Specifically, in an experiment in which the pool was composed of equal proportions of strains, barcodes corresponding to LH-1 accounted for 16.7% of total counts (Fig. S1A). When the initial proportion of LH-1 was reduced by 100-fold, barcodes corresponding to LH-1 accounted for 0.22% of the total counts, with barcodes for other strains increasing in proportional abundance (Fig. S1B). Finally, when LH-1 was excluded from the pool, we no longer detected barcodes from this mutant (Fig. S1C). These data indicate that genetic barcoding combined with next-generation sequencing (NGS) is a sensitive and quantitative approach that can be used to track changes in the proportional abundance of individual mutants within a pool.

FIG 1.

H. pylori barcoded mutant library design and workflow. (A) Target genes were subjected to insertional mutagenesis with a cassette comprised of a chloramphenicol acetyltransferase (cat) gene and unique 21-nucleotide barcodes. Flanking segments (∼500 bp) on either side of the cat cassette are derived from regions upstream and downstream of the desired site of insertion in the H. pylori chromosome. Conserved regions flanking each barcode allow unbiased parallel PCR amplification of barcodes from a pool of mutants. (B) The cassettes illustrated in panel A were introduced into selected sites in the H. pylori chromosome. Barcoded H. pylori mutants were pooled, and the resulting libraries were either passaged in vitro or administered to mice by oral gavage. At the end of the experiments, the relative abundance of barcoded strains in the populations was analyzed and compared to the abundance of barcoded strains in the input libraries. Input and output pools were analyzed by extracting H. pylori genomic DNA, amplifying barcodes via PCR, prepping barcoded amplicons with NGS adapters, and performing next-generation sequencing.

Tracking changes in composition of the control library in vitro.

We next evaluated the stability of the control library during growth in vitro. We prepared three independent preparations of the control mutant library on three separate days, each containing equal proportions of the component strains, and we passaged the libraries on blood agar plates (tryptic soy agar [TSA] with 5% sheep blood). Following 10 serial passages of the libraries over 21 days, we compared the composition of the passaged libraries (outputs) with the composition of the starting libraries (inputs) using PCR and NGS as described in Materials and Methods. As expected, at the start of the experiment, each of the input libraries contained similar proportions of the six component strains (Fig. 2A). At the end of the experiment, all six strains remained present in the output libraries. When comparing the relative abundance of various strains in the output libraries, there were statistically significant differences in the numbers of barcode counts detected (Table S2), but the magnitude of changes was relatively small (less than 10-fold differences in the numbers of barcode counts) (Fig. 2B). These data provide a frame of reference for interpreting subsequent experiments investigating the contribution of specific hop genes to H. pylori fitness.

FIG 2.

The control library proliferated as a stable community in vitro. We generated control libraries composed of 6 strains, each containing a unique nucleotide barcode at the mdaB-hydA intergenic locus. In three independent experiments conducted on separate days, a single input pool was inoculated to 3 separate plates (p1 to p3) which were passaged in vitro for 21 days. The composition of input (input 1 to 3) and output pools (p1 to p3) was analyzed as described in Materials and Methods. The barcode counts of input and output pools were normalized to standardized total counts. Panel A shows relative proportions of barcode counts and panel B shows log₁₀ ratios comparing output counts to the input counts (see details in Materials and Methods). Statistical analysis is shown in Table S2. The control library proliferated as a relatively stable community, without substantial changes in the proportional abundance of individual strains compared to that of the input pool.

Tracking changes in composition of the OMP mutant library in vitro.

To evaluate possible contributions of hop genes to H. pylori fitness, we generated an “OMP mutant library” containing a panel of barcoded strains with mutated hop genes (Table S1). This library also contained several barcoded control strains for comparison, including a strain containing the barcode in the mdaB-hydA intergenic region (LH-7) and two strains containing mutations in non-hop genes predicted to be required for H. pylori fitness in vivo but not required for fitness in vitro (ureA and flaA, encoding a subunit of urease and a flagellar component, respectively) (Table S1) (24–28). We prepared independent preparations of the OMP mutant library on three separate days, each containing equal proportions of the component strains. Following 21 days of serial passaging in vitro, we compared the composition of the passaged libraries with the composition of the input libraries. At the start of the experiment, each of the input libraries contained similar proportions of the component strains (Fig. 3A). There were marked changes in the composition of the libraries over the course of the experiment, and the observed changes in abundance of several mutants was much greater than what was observed in the previous experiment with control strains (compare Fig. 3B with Fig. 2B). We used multiple criteria to identify mutants that exhibited significant differences in fitness compared to that of control strains, as described in Materials and Methods. Neither the control strain nor the ureA mutant exhibited a fitness defect in vitro, but unexpectedly, the flaA mutant exhibited a fitness defect. Based on the combined results of two statistical analyses (t tests and analysis of confidence intervals), hopE and oipA mutants (designated LH-13_a and LH-16_a) exhibited significant fitness defects and a hopF mutant (LH-14_a) exhibited a fitness advantage (Fig. 3C and Table S2). Growth curve analyses of the hopE, oipA, and flaA mutant strains indicated that these mutant strains grew at rates similar to those of the J166 wild-type strain and several other mutants (Fig. S6).

FIG 3.

Fitness of a library of hop mutant strains during passage in vitro. We generated OMP mutant libraries composed of 16 hop mutants and 3 non-hop control mutants. In three independent experiments conducted on separate days, a single input pool was inoculated to one or more separate plates (p1 to p6) which were passaged in vitro for 21 days. The composition of input (inputs 1 to 3) and output pools (p1 to p6) was analyzed as described in Materials and Methods. (A, B) Panel A shows relative proportions of barcode counts and panel B shows log₁₀ ratios comparing the output pools with the input. (C) We estimated the variability observed in the control library (Fig. 2) by analyzing all pairwise differences of mean log₁₀ ratios within the control library (see Table S2). For each of the 15 pairwise differences, 99% confidence intervals were computed. Among these, the furthest boundary from 0 is 1.075 (LH-1 versus LH-4). Thus, we used the −1.075 to 1.075 range as the boundary for “no difference.” In other words, each mutant in the OMP mutant library needs to be further from the control strain (LH-7) than this distance to be considered a significant change from the input. We computed 99% confidence intervals of the difference between each mutant to the control (LH-7). All 15 pairwise comparisons within the control library are depicted (orange). If a comparison to the control strain (LH-7) within the OMP mutant library (blue) clears this threshold (99% confidence is wholly above or below the threshold) (*), we concluded that the mutant is significantly different from the control. Additional statistical analysis is shown in Table S2. The hopE (LH-13_a), oipA (LH16_a), and flaA (LH-9_a) mutants had significantly reduced barcode counts based on both these criteria, whereas hopF (LH-14_a) mutant had significantly increased barcode counts.

We then conducted further experiments to evaluate if similar changes in the library composition would be observed after passage on TSA with 5% sheep blood compared to that on two alternate media (sodium bisulfite-free Brucella [BSFB]-cholesterol agar plates containing either routine levels of supplemental sodium chloride [0.5%] or increased levels of sodium chloride [0.9%]). Barcode counts from multiple hop mutant strains (including hopE, oipA, and hopF mutants) were significantly different compared to barcode counts from the control strain under all three conditions (Fig. S2 and Table S3).

Further evaluation of mutants in vitro.

To address the possibility of unrecognized secondary mutations in mutant strains, we repeated the mutagenesis of mouse-adapted wild-type strain J166, thereby generating independently constructed replicates of the original mutants. Specifically, we generated newly constructed hopE, oipA, and hopF mutants (designated LH-13_b, LH-16_b, and LH-14_b) since the original hopE, oipA, and hopF mutants exhibited fitness defects in the initial experiments (Fig. 3 and Fig. S2). We also generated a second flaA mutant strain (LH-9_b) to determine if the unexpected fitness disadvantage of the flaA mutant observed in Fig. 3 and Fig. S2 would be recapitulated. For comparison, we also generated a newly constructed alpA mutant (LH-11_b) and babA mutant (LH-24_b). In the initial experiments (Fig. 3), the alpA mutant exhibited a nonsignificant trend toward a fitness defect and the babA mutant did not exhibit any detectable fitness defect.

We evaluated the fitness of the new hop mutants, along with that of the new flaA mutant, using small-scale competitions and pairwise competition assays. Small-scale competitions were conducted using libraries composed of 12 barcoded mutant strains (7 control strains [LH-1 to LH-7] and the new alpA, hopE, oipA, hopF, and babA mutants) and the populations were analyzed via NGS, whereas pairwise assays were carried out with 1:1 mixtures of mutants competed with LH-7 and analyzed via qPCR. Consistent with results of the initial experiment (Fig. 3), we observed fitness defects of hopE and oipA mutants (LH-13_a,b and LH-16_a,b) when testing independently constructed mutant strains (Fig. S2 to 4 and Tables S3 to S5). The significance of the fitness defect observed for the alpA mutant compared to LH-7 was variable depending on the type of statistical analysis performed (Fig. S2 to S4 and Tables S3 to S5). The properties of the original hopF mutant (LH-14_a) were not recapitulated by the newly constructed hopF mutant strain (LH-14_b). As expected, there were no significant differences when comparing the babA mutants (LH-24_a and LH-24_b) with the LH-7 mutant, confirming that babA does not contribute to H. pylori fitness in vitro (Fig. S2 to S4 and Tables S3 to S5).

Among the 5 genes analyzed in the experiments described above, several are monocistronic (for example, oipA and babA), whereas alpA and hopE are each predicted to be transcribed within operons. This raises the possibility that mutagenesis of the latter genes might alter transcription of downstream genes (29). Therefore, we assessed if the fitness of alpA and hopE mutants could be restored via complementation. In parallel, we undertook complementation of the flaA mutant, which had an unexpected fitness defect. Complemented mutants were generated as described in Materials and Methods. Quantitative real-time PCR (qRT-PCR) experiments showed that alpA and hopE were transcribed as expected in the complemented mutants, and motility of the flaA mutant was successfully restored by the complementation (data not shown). To evaluate the fitness of the complemented strains compared to that of the mutants from which they were derived, we did competition experiments in which three strains (LH-7 control strain, a mutant strain, and the corresponding complemented mutant strain) were combined at equal proportions and passaged on blood agar plates for 21 days. Subsequently, the relative abundance of each strain compared to that of the control strain was analyzed by qPCR as described in Materials and Methods. As expected, the relative abundance of the alpA, hopE, and flaA mutants was reduced compared to the relative abundance of the control strain (Fig. S5). Unexpectedly, complementation did not rescue the fitness of these three mutant strains (Fig. S5). Potential explanations for why the fitness of these mutants was not restored by complementation are considered in the Discussion.

Tracking changes in composition of the control library in vivo.

We next undertook experiments to analyze the fitness of H. pylori strains in vivo. To assess changes in a population of control strains in vivo, we infected mice with the control library for 21 days (corresponding to the same time period analyzed for the in vitro experiments). Animals were euthanized, H. pylori strains were cultured from the stomachs, and the composition of the resulting populations was compared to the composition of the input libraries (Fig. 1B). As seen in Fig. 4A, we detected barcodes corresponding to each of the strains in the population. These data confirm that multiple H. pylori strains can cocolonize the same stomach. Notably, the distribution of barcode counts in individual mice was highly variable. In each case, the composition of the H. pylori population recovered from individual mice was markedly different from that of the input population, and populations from individual mice were markedly different from populations of other mice (Fig. 4A). This variation occurred in a random manner, with no statistically significant differences detected when comparing individual control strains with other control strains (Fig. 4 and Table S7). Therefore, in contrast to the relatively stable composition of the control library observed during passage in vitro, we consistently observed marked changes in the composition of the control library when it was introduced into murine stomachs (compare Fig. 4B with Fig. 2B). These findings are consistent with the existence of a nonselective bottleneck in vivo.

FIG 4.

Multiple strains of H. pylori can colonize the same stomach. C57BL/6J mice were infected with control libraries for 21 days in two independent experiments. H. pylori was cultured from the stomachs, and the composition of input and output pools was analyzed as described in Materials and Methods. The labels m1 to m6 indicate results for individual mice in separate experiments. (A, B) Panel A shows relative proportions of barcode counts, and panel B shows log₁₀ ratios comparing the output pools with the input pools. Statistical analysis is shown in Table S7. All control strains (LH-1 through LH-6) within these libraries could be detected after 21 days, but the numbers of barcode counts corresponding to each mutant were highly variable from mouse to mouse.

Testing the fitness of hop mutant strains in vivo.

We next infected mice with the OMP mutant library and analyzed output pools cultured from the murine stomachs at 21 days postinfection. Similar to the results of in vivo experiments with the control library, we observed a high variability in the distribution of barcode counts in individual mice, and in each case, the composition of the population recovered from individual mice was markedly different from the composition of the input population (Fig. 5). Consistent with previous reports that ureA and flaA are required for gastric colonization (24–28), we detected very low numbers of barcode counts for the ureA and flaA mutants. Interestingly, we did detect barcode counts specific for ureA or flaA mutants at levels substantially above the limit of detection (defined in Materials and Methods) in several of the animals (Fig. 5B and Table S7). The babA mutant was the dominant strain (accounting for >50% of total barcode counts) in output pools from the majority of the mice (Fig. 5 and Table S7). Similar results were observed in mice infected with independent input pools. A dominance of the babA mutant was also observed in mice infected for 90 days (Fig. S7). At this time point (90 days postinfection), barcode counts corresponding to the babA mutant accounted for over 95% of counts from 4 out of 5 mice (Fig. S7). Statistical analysis was complicated by a very high level of variability in results for individual animals, which is presumably attributable to the bottleneck phenomenon observed in Fig. 4. In a statistical analysis using the approach employed for Fig. 3, none of the OMP mutants met the criteria for statistically significant changes. Nevertheless, we were able to identify hop mutants that exhibited marked changes in numbers of barcode counts compared to those of the control strain (LH-7). Specifically, there was a >100-fold reduction in the ratios of output/input barcode counts for alpA and hopE mutants compared to those of the control strain (Fig. 5 and Table S7). The numbers of barcode counts corresponding to alpA and hopE mutants were similar to the numbers of barcode counts corresponding to ureA and flaA mutants. The babA mutant exhibited an increased abundance of similar magnitude (>100-fold increase).

FIG 5.

Fitness advantage of a babA mutant in vivo. C57BL/6J male mice were infected with OMP mutant libraries for 21 days in three independent experiments. H. pylori was cultured from the stomachs, and the composition of input and output pools was analyzed as described in Materials and Methods. The labels m1 to m18 indicate results for individual mice in separate experiments. (A, B) Panel A shows relative proportions of barcode counts, and panel B shows log₁₀ ratios relative to the input. At the end of the experiment, half of the infected mice had >50% of total counts originating from the babA (LH-24_a) mutant barcode. (C) If the 99% confidence interval was wholly above 2 or below −2 in the log₁₀ scale, then the OMP mutant strain was considered significantly different from the control strain (LH-7). Statistical analysis is shown in Table S7. Using this approach, we determined that the relative abundance of alpA (LH-11_a), hopE (LH-13_a), ureA (LH-8), and flaA (LH-9_a) mutants was decreased, whereas the relative abundance of the babA mutant was increased relative to LH-7. An asterisk (*) indicates >100-fold difference in ratio of output compared to input.

When comparing the results of experiments conducted in vivo with results of experiments conducted in vitro (compare Fig. 3 with Fig. 5), the ureA mutant exhibited a fitness defect in vivo but not in vitro, as expected. We did not identify any hop mutants that exhibited unequivocal fitness defects exclusively in vivo without corresponding fitness defects in vitro. Interestingly, the babA mutant exhibited a strong fitness advantage in vivo compared to the fitness of control strains and other mutants, but the babA mutant did not exhibit a corresponding fitness advantage in vitro. The babA mutant accounted for greater than 50% of nucleotide barcode counts in bacterial populations recovered from 26/40 (65%) mice but did not exhibit a similar fitness advantage in 10 experiments conducted in vitro (P = 0.0002). Therefore, the babA mutant exhibited a fitness advantage in vivo but not in vitro.

To further evaluate the in vivo fitness advantage observed with the babA mutant, we mixed the LH-7 control strain with either the original babA mutant (LH-24_a) or the subsequently constructed babA mutant (LH-24_b) in equal proportions and passaged these mixtures on blood agar plates for 21 days. In parallel, the mixtures were introduced into mouse stomachs via oral gavage, and the gastric H. pylori populations were assessed at 21 days postinfection. We then analyzed the relative abundance of the babA mutants compared to that of the control strain by qPCR as described in Materials and Methods. As expected, there were no significant differences when comparing the relative abundance of the babA mutants with that of the control strain in vitro (Fig. 6). In mice, however, we observed that both babA mutants outcompeted the control strain (Fig. 6), confirming that the babA mutant exhibited enhanced fitness in vivo but not in vitro.

FIG 6.

Fitness of babA mutants in vitro and in vivo evaluated by qPCR. Each babA mutant [original (LH-24_a) and replicate (LH-24_b) mutants] was mixed with the control strain (LH-7) in a 1:1 ratio. The mixtures were passaged on blood agar plates for 21 days and also used for infection of mice, which were euthanized at 21 days postinfection. We used qPCR to determine the relative abundance of each strain in the input and output populations, based on comparison to a standard curve generated for each target. The log₁₀ ratio (output/input) was calculated based on the relative abundance of each mutant compared to that of the control strain in the output and input populations. At the end of the experiment, the relative abundance of the original (LH-24_a) and independent (LH-24_b) babA mutants remained similar to that of LH-7 in vitro but was increased in mice. The changes in abundance of the babA mutants in mice were statistically significant (an asterisk indicates a P value of <0.001). Statistical analysis is shown in Table S8.

DISCUSSION

Within the human stomach, H. pylori can be found attached to gastric epithelial cells or within the mucus layer (30–32). Adherence to gastric epithelial cells potentially facilitates nutrient acquisition and allows delivery of effector molecules into gastric cells (33–35). Several OMPs in the Hop family are known to function as adhesins, but the functions of most remain unknown. To facilitate the study of hop mutant strains, we used a genetic barcoding method that allowed us to track changes in the proportional abundance of barcoded H. pylori strains within a mixed population. We anticipated that this experimental approach would lead to the identification of specific hop genes that are required for H. pylori fitness in vivo but not in vitro. We did not identify any mutants with these properties. Instead, we observed that H. pylori babA mutants have enhanced fitness compared to that of other strains in vivo, without a corresponding change in fitness in vitro.

Previous studies have used signature tagged mutagenesis (STM) and microarray-based tracking of transposon mutants (MATT) to identify H. pylori genes essential for gastric colonization (26, 27). In a previous STM study, gerbils were infected with pools of 24 individually tagged H. pylori mutants for 21 days (26). Mutants present in the input pool but absent in the output pool were then identified using a tag-specific PCR method. Both flaA and ureA were identified as essential for gastric colonization (26). Among the genes in the hop family, only hopG was shown to be essential for gastric colonization (26). In a previous study that used MATT to identify genes that contribute to gastric colonization (27), mice were infected with pools of 48 H. pylori transposon mutants. For each experiment, the input pool DNA was labeled with Cy-3 (green) and the output pool DNA was labeled with Cy-5 (red). Microarray methodology was then used to quantify the red/green ratio for each gene. Numerous genes, including ureA, were essential for gastric colonization. Among the genes in the hop family, none were shown to be essential for gastric colonization (27). Notably, this study did not include analysis of oipA, hopF, sabA, or babA mutants, so the role of these OMPs in gastric colonization was not assessed. Since both previous studies relied on random transposon mutagenesis for generation of mutants, there is uncertainty about the comprehensiveness of the mutant pools analyzed, and a quantitative determination of relative levels of colonization among the pooled mutant strains was not undertaken in either study. The methods used in the current study allowed us to evaluate mutant pools with known composition and allowed us to quantify the proportional abundance of individual mutants within a mixed population, thereby assessing whether mutagenesis of target genes leads to a fitness advantage, disadvantage, or neutral phenotype relative to the fitness of barcoded control strains and other mutants.

As an initial step toward development of the methodology, we generated a barcoded control library in which all strains were genetically manipulated at a neutral locus, unrelated to bacterial fitness. Following 21 days of growth in vitro, we observed relatively low variability in composition of the control library, indicating that these strains were able to form a stable community and did not encounter any strong nonselective pressures in vitro.

Next, we assembled a library of hop mutant strains and analyzed changes in its composition during 21 days of passage in vitro. The hopE and oipA mutants exhibited a fitness defect that was detected using three types of solid culture medium, and the results were corroborated when independently generated hopE and oipA mutants were analyzed. Likewise, the alpA mutant exhibited a trend consistent with a fitness defect in vitro. Unexpectedly, we also observed that flaA mutants exhibited a fitness defect in vitro. To further evaluate the fitness defects of these mutants, we generated complemented mutant strains. The rdxA locus, which was used as the site for introduction of these complemented genes, was reported previously to have no effect on H. pylori fitness (36). Complementation restored expression of alpA and hopE, and restored motility in the flaA mutant, but unexpectedly, complementation failed to reverse the fitness defects of these mutants. There are multiple possible explanations for the failure of complementation to restore fitness of the mutants. One possibility (relevant for alpA and hopE) is that the insertion of barcoded antibiotic resistance cassettes in these genes might have altered the transcription of other genes in the operon. Another possibility is that cotranscription of these operon-transcribed genes is necessary for proper function; in this scenario, complementation into a heterologous locus would not rescue fitness. As another possibility, insertional mutagenesis of OMP-encoding genes might have resulted in the production of truncated proteins, causing a detrimental effect on H. pylori fitness. Finally, it is possible that the mutagenesis may have altered production of sRNAs. H. pylori is known to produce hundreds of different sRNAs (29), and we speculate that there may be strain-specific differences in sRNA production in the strain used for the current study compared to that in a prototype strain used for previous studies of H. pylori sRNA production (29). Since the complementation experiments did not provide evidence indicating a contribution of alpA or hopE to H. pylori fitness, we did not undertake further studies to investigate the mechanistic basis for the fitness defects of these mutants in vitro.

There were several challenges associated with analysis of the barcoded libraries in vivo. PCR amplification of the barcoded regions from the mouse stomachs was not consistently successful, so we cultured H. pylori from the mouse stomachs and analyzed the cultured bacteria. Consequently, there could be bias related to detection of strains that had properties favoring efficient culture from the mouse stomachs. Conversely, a strength of this approach was that it ensured that the barcodes detected originated from viable organisms. The second limitation was that the numbers of individual H. pylori colonies cultured from stomachs varied considerably among mice. To reduce bias, we restricted our analysis to mice from which at least 100 H. pylori colonies were cultured.

As an initial step in undertaking experiments in vivo, we orogastrically infected mice with the control library. Following 21 days of infection, we observed that all of the control library strains were detected in each stomach, indicating that multiple strains of H. pylori can colonize the same stomach. Notably, the proportional abundance of each strain varied from mouse to mouse, suggesting that strong nonselective pressures influenced the population colonizing each mouse. This is consistent with the results of previous studies, which suggested that H. pylori experiences a bottleneck early in infection (32, 37–39). Thus, the animal experiments with the control library provided insight into the nonselective environmental pressures encountered in vivo and established a foundation for understanding the range of experimental variability that can be expected in the absence of functional strain-specific differences.

Finally, we compared the fitness of hop mutants in vivo to the fitness of the corresponding mutants in vitro. We anticipated that we would identify hop mutants required for colonization of the stomach but not required for H. pylori fitness in vitro. A ureA mutant exhibited the expected behavior (essential for fitness in vivo but not in vitro), which indicated that the methodology was capable of identifying such mutants (24, 25, 40). Unexpectedly, we did not identify any hop mutants that exhibited fitness defects in vivo but not in vitro. Functional redundancy among Hop proteins might account for the failure to identify any mutants that exhibited these characteristics. Another possibility is that the colonization advantage of the babA mutant might have hindered efforts to detect mutants that have reduced fitness in vivo. In future studies, it might be informative to conduct experiments in which babA mutants are excluded.

Although we were not able to identify any mutants that exhibited a fitness defect in vivo but not in vitro, we found that the babA mutant exhibited a fitness advantage in vivo but not in vitro. Several previous studies observed that the BabA adhesin is selected against (via the loss of babA expression or loss in Lewis^b binding) during experimental H. pylori infection of mice, Mongolian gerbils, or rhesus macaques (41–44). The current results are consistent with such observations.

In H. pylori strains isolated from humans, babA may be either present or absent (45, 46), and in some instances when the gene is present, the BabA protein is not produced. We presume that there are selective pressures in humans that favor retention of babA, at least under some circumstances. Previous studies have shown that BabA is important for H. pylori attachment to the human gastric mucosa and allows for reversible attachment as gastric cells are shed off into the lumen (47). Therefore, BabA might confer a benefit to H. pylori at specific stages of infection in humans but might be deleterious at other stages of infection. Alternatively, BabA may confer a benefit in individuals who produce Le_b antigen (the BabA receptor) but not in individuals who do not produce Le_b. The results of the current study, along with those of previous studies, suggest that the presence of babA is deleterious in the mouse model (41–44). The mechanisms leading to selection against babA-positive strains in mice (in contrast to retention of babA-positive strains in humans, at least under some circumstances) are not yet known. One of the relevant differences between the human and mouse gastric environments is that mice lack Le_b (the BabA receptor) (48). We presume that the presence of babA confers both benefits and costs to H. pylori and that retention or loss of the gene is dependent on the balance of benefits and costs in various gastric environments.

In summary, this study provides new insights into the contributions of specific Hop OMPs toward H. pylori fitness in vitro and in vivo as well as the role of nonselective and selective pressures that shape the H. pylori community during gastric colonization. In addition, the quantitative monitoring of nucleotide barcodes described in this study offers numerous advantages compared to comparative analysis of strains labeled with different antibiotic markers. These methodologic approaches will be useful for future studies of H. pylori population dynamics, both in vitro and in vivo. For example, this approach will be useful in studying H. pylori fitness in various in vitro environments to study effects of pH, medium composition, and other variables. We anticipate that this approach will be generally useful for studying the roles of H. pylori genes in various phenotypes, as well as for analysis of the selective pressures that shape the evolution of H. pylori populations.

MATERIALS AND METHODS

Bacterial strains and culture conditions.

H. pylori strain J166 (GenBank accession CP007603) was originally isolated from a human with a duodenal ulcer (44, 49). This strain is motile, contains the cag pathogenicity island, and exhibits Cag T4SS activity. We chose strain J166 for the experiments in this study because it has been shown previously to colonize multiple animal species (mice, macaques, Mongolian gerbils, and guinea pigs) and achieves a colonization density in the mouse stomach lower than that of several other mouse-colonizing strains (e.g., SS1 and PMSS1), which might facilitate the detection of subtle differences in colonization among multiple mutants. To generate a mouse-adapted wild-type J166 strain, we infected a mouse with strain J166 for 3 months. A single colony isolate that retained Cag T4SS activity, VacA activity, and motility was selected for use in this study. Mouse-adapted wild-type J166 and barcoded mutant strains (described below) were grown on Trypticase soy agar plates supplemented with 5% sheep blood (TSA with 5% sheep blood) in room air supplemented with 5% CO₂ at 37°C. All mutant strains were generated from mouse-adapted wild-type strain J166 (a single colony output from a 3-month mouse infection) by selective growth on medium containing chloramphenicol (5 μg/ml). Escherichia coli strain DH5α, used for plasmid propagation, was grown on Luria-Bertani (LB) agar plates or in LB liquid medium supplemented with ampicillin (100 μg/ml) at 37°C.

Barcode generation.

The KAnalyze program was used to compile lists of overlapping 7-nucleotide sequences (7-mers), offset by a single nucleotide (50). A list of all possible 7-mers was compared to a set of 7-mers not present in the genome sequence of H. pylori strain J166 (nullomers). Three randomly selected nullomers then were concatenated to generate 21-mer barcodes that would differ from any 21-mer found in H. pylori J166 by at least 3 nucleotide positions.

Generation of barcoded H. pylori strains.

The mdaB-hydA intergenic region (corresponding to HP0630-HP0631 in H. pylori 26695) was identified in H. pylori strain J166 (GenBank accession CP007603) and was used as a neutral site for introducing nucleotide barcodes. A control library was composed of H. pylori strains (LH-1, LH-2, etc.) containing barcodes in the mdaB-hydA intergenic region. Similarly, we searched the J166 genome with known hop gene sequences from H. pylori strain 26695 to identify J166 orthologs with sequence identity greater than 90%. We identified 20 hop genes in J166 and successfully mutagenized 16 (Table S1). Two pairs of hop genes are duplicate copies (hopJ/K and hopM/N); one copy of each duplicate was mutagenized for this study. If no phenotype (such as a fitness advantage or disadvantage) was detected for these two mutants, it could be attributed to the presence of a duplicate gene. Additionally, we did not mutate alpB and we did not successfully isolate a hopZ mutant. Altogether, the first OMP mutant library is composed of 16 hop mutants, 1 neutral intergenic mutant (mdaB-hydA: a neutral locus), and 2 additional control mutants (ureA and flaA, previously reported to be essential for gastric colonization) (24–28), totaling 19 uniquely barcoded H. pylori mutants. Independently generated mutants were obtained by once again transforming the mouse-adapted wild-type strain J166 with the appropriate plasmids. Hop-encoding genes and control genes selected for mutagenesis are shown in Table S1.

Barcoded mdaB-hydA control strains and hop mutant strains were generated by transforming mouse-adapted wild-type H. pylori strain J166 with a cassette comprised of a chloramphenicol acetyltransferase gene and unique 21-nucleotide barcode tags specific for each mutant, flanked by 500 bp homologous to the targeted insertion site to facilitate the insertion of this cassette via homologous recombination into the targeted site (Fig. 1). H. pylori Hop proteins have conserved sequences at the N and C termini and a region of diversity that is predicted to contribute to each Hop’s distinct functions (9). Simple insertions of the barcoded cassette were targeted to this region of diversity, approximately 500 bp downstream of the transcriptional start site. Following transformation, H. pylori transformants were plated on sodium bisulfite-free Brucella (BSFB) agar plates supplemented with cholesterol and containing chloramphenicol (5 μg/ml) (BB5ChlC) (51, 52). Transformants were validated via sequencing of PCR amplicons generated with primers flanking the predicted barcoded cassette insertion site (about 600 bp upstream and 600 bp downstream of the insertion site).

Growth curve analysis.

H. pylori strains J166 wild-type and barcoded mutant strains (alpA, hopE, oipA, babA, and flaA mutants) were grown overnight in sodium bisulfite-free Brucella broth supplemented with 10% fetal bovine serum (FBS) in room air supplemented with 5% CO₂ at 37°C. These were subcultured at a starting optical density at 600 nm (OD₆₀₀) of 0.1, and the optical densities were then measured every 6 h for 30 h, with a final measurement taken at 48 h.

Serial passaging of barcoded libraries in vitro.

H. pylori input pools were generated by combining barcoded mutant strains at equal proportions in bisulfite-free Brucella broth, resulting in a final concentration of the pooled mutants of 1.0 × 10⁹ cells/ml. To circumvent a potential loss of bacterial viability that is sometimes encountered with growth of H. pylori in liquid cultures, we passaged the pooled bacteria on solid media. The control library input pool was plated on blood agar plates and passaged every 2 days for 21 days, for a total of 10 passages. The OMP mutant library input pool was plated on blood agar plates supplemented with 10% FBS or BSFB agar plates (0.5% or 0.9% sodium chloride) supplemented with cholesterol and passaged every 2 days for 21 days, for a total of 10 passages. When bacteria were passaged from one plate to another, a zig-zagging streak was taken across the bacterial lawn to allow a comprehensive sampling.

Orogastric infection of mice with barcoded libraries.

Male conventional C57BL/6 mice, 6 to 8 weeks old, were used in all studies (The Jackson Laboratory). H. pylori input pools were generated by resuspending individual barcoded mutants in prewarmed BSFB broth. Each barcoded mutant was then combined at equal proportions; the final concentration of the combined mutants was 1.0 × 10⁹ cells/ml in BSFB broth. Mice were anesthetized and infected with 0.5 ml of H. pylori input pools via oral gavage. Mice were sacrificed at 21 or 90 days postinfection. After euthanasia, the stomachs were excised, the forestomach was removed, and the remaining stomach was opened along the minor curvature. Stomach contents were washed away with sterile phosphate-buffered saline (PBS) and stomach tissue was homogenized in 500 μl of Brucella broth for 10 min using a Bullet Blender (Next Advance). A total of 100 μl of homogenized stomach tissue was cultured on TSA plates supplemented with 5% sheep blood, nalidixic acid (10 μg/ml), vancomycin (50 μg/ml), amphotericin (20 μg/ml), and bacitracin (100 μg/ml). Samples were processed as described below.

Next-generation sequencing and sequence data analysis.

Genomic DNA was extracted from H. pylori input and output pools (bacterial lawns from in vitro passages or pooled single colonies grown from mouse stomach lysates, >100 colonies/mouse). Subsequently, barcodes were amplified via PCR with primers that bind to conserved regions flanking the barcodes. Specifically, we used 5′-ATCTACTGCCGATATTTACG-3′ as a forward primer and 5′-TAAATCCACTGTGATATCCAT-3′ as a reverse primer. Following a PCR cleanup step (Qiagen), amplicon libraries were generated using NEBNext Ultra II DNA library prep kit for Illumina. Sequencing was performed using paired-end (PE) 150 bp on the Illumina NovaSeq 6000 at approximately 10 million PE reads/sample. Flanking sequences were cleaved from barcodes in silico with the “Seqtk_Trimfq” Galaxy software tool (https://github.com/lh3/seqtk/). Barcodes were then sorted by barcode identity and quantified using “Barcode Splitter” Galaxy software tool (http://hannonlab.cshl.edu/fastx_toolkit/). Individual barcode counts for each mutant within a single sample were normalized to the same total standard count across all experiments (gene of interest [goi] normalized count). For instance, for a given gene of interest,

$$\mathrm{go}{\mathrm{i}}_{\mathrm{normalized\hspace{0.17em}count}}=\left(\frac{\mathrm{go}{\mathrm{i}}_{\mathrm{raw\hspace{0.17em}count}}}{\mathrm{tota}{\mathrm{l}}_{\mathrm{raw\hspace{0.17em}count}}}\right)\times \mathrm{tota}{\mathrm{l}}_{\mathrm{standard\hspace{0.17em}count}}$$

Additionally, the log₁₀ ratio was calculated by taking the log₁₀ of the output count divided by its respective input count for each strain. Due to the limitations imposed by log₁₀ fold calculations, all barcode counts that fell to 0 were replaced by 0.01 for quantification purposes. As such, the limit of detection corresponded approximately to a log₁₀ ratio of −8.0.

$$\mathrm{lo}{\mathrm{g}}_{\mathrm{10}}\mathrm{\hspace{0.17em}ratio=lo}{\mathrm{g}}_{\mathrm{10}}\left(\frac{\mathrm{go}{\mathrm{i}}_{\mathrm{output}}}{\mathrm{go}{\mathrm{i}}_{\mathrm{input}}}\right)$$

Small-scale competition assays.

We generated two libraries containing 12 barcoded mutant strains at equal proportions at a final concentration of 1.0 × 10⁹ cells/ml in BSFB. The first library was composed of the original control strains (LH-1 through LH-7) mixed with the original barcoded hop mutants (alpA, hopE, hopF, oipA, and babA mutants) (LH-11_a, LH-13_a, LH-14_a, LH16_a, LH-24_a). The second library was composed of the original control strains (LH-1 through LH-7) mixed with the new independent barcoded hop mutants (alpA, hopE, hopF, oipA, and babA mutants) (LH-11_b, LH-13_b, LH-14_b, LH-16_b, LH24_b). Each library was plated and serially passaged on blood agar plates every 2 days for 21 days. Genomic DNA was isolated from input and output pools and processed for next-generation sequencing as described above.

Pairwise competition assays.

The LH-7 control strain was combined 1:1 with original flaA, alpA, hopE, hopF, oipA, and babA mutants (LH-9_a, LH-11_a, LH-13_a, LH-14_a, LH-16_a, LH-24_a) or subsequently generated mutants (LH-9_b, LH-11_b, LH-13_b, LH-14_b, LH-16_b, LH-24_b) and passaged on blood agar plates every 2 days for 21 days. Mice were also infected with 1:1 mixtures of LH-7 competed with either the original or independent babA mutant (LH-24_a or LH24_b). Subsequently, we extracted genomic DNA from the inputs and outputs using phenol-chloroform-based methods (53). Real-time qPCR analysis of DNA samples was performed with SYBR green fluorophore (iQ SYBR green supermix; Bio-Rad) on an ABI StepOnePlus machine. Primer sequences can be found in Table S6. A standard curve of each DNA target was generated using 10-fold dilutions starting at 50 ng/well. The abundance of individual strains was calculated using the appropriate standard curve for each DNA target.

Generation of complemented mutant strains.

Plasmids for complementation contained the relevant gene with its endogenous promoter along with a unique 21-nucleotide barcode, flanked by 500 bp regions homologous to sequences flanking the rdxA gene. These plasmids were designed to allow insertion of the relevant gene into the rdxA locus, rendering these strains metronidazole resistant (54). Complemented mutant strains (LH-26, LH-27, LH-28) were generated by transforming J166 alpA (LH-11_a), hopE (LH-13_a), and flaA (LH-9_a) mutants with the appropriate plasmids. Following transformation, H. pylori transformants were plated on BSFB agar plates supplemented with cholesterol and containing metronidazole (7.5 μg/ml). Transformants were validated via sequencing of PCR amplicons generated with primers flanking the rdxA gene (about 600 bp upstream and 600 bp downstream of the insertion site).

Competition assays with complemented mutants.

The LH-7 control strain was mixed 1:1:1 with a hop mutant strain [alpA (LH-11_a), hopE (LH13_a), or flaA (LH-9_a) mutants] and the corresponding complemented mutant (LH-26, LH-27, or LH-28) and passaged on blood agar plates every 2 days for 21 days, for a total of 10 passages. Genomic DNA was extracted from the input and output populations using phenol-chloroform-based methods (53). We used real-time qPCR methodology described above to assess the abundance of each strain in the output and input populations. Primer sequences can be found in Table S6.

Statistical analysis.

For experiments analyzing OMP mutants in vitro, we used the following approach to identify mutants in the OMP mutant library that were significantly different from the control strain (LH-7) while accounting for the observed variability in the control library. All pairwise mean differences within the control library were estimated with 99% confidence intervals, and the furthest boundary (in either positive or negative direction) from 0 was selected as the boundary of no difference. Two-sample t test without assuming equal variance was used to compare a mutant in the OMP mutant library with the control strain (LH-7). A 99% confidence interval (corresponding to type I error rate of 1%) was computed for each comparison (Fig. 2 and 3 and Fig. S3E and F). Strains with a P value less than 0.01 (when comparing to LH-7) and also a 99% confidence interval (CI) that does not overlap the boundary of no difference (based on the control library) were considered significant.

For analysis of in vivo experiments, thresholds of −2 and 2 (corresponding to >100-fold differences) were used as a threshold for defining marked differences of mean log₁₀ ratios (Fig. 5C). Strains with a P value less than 0.01 (when comparing to LH-7) and also a 99% CI that does not overlap the ±2 threshold were considered significant. In experiments without a corresponding control library for comparison, we required a 99% confidence interval to be wholly above or below 0, which is equivalent to a multiplicity-controlled type I error rate of 1% (Fig. S2). Finally, for analysis of qPCR data, a one-sample t test was used to estimate log-transformed OMP/LH-7 ratios with a 99% confidence interval (Fig. 6, Fig. S4, and Tables S8 and S5). All confidence intervals and P values are reported in supplemental tables (Tables S2, S3, S4, S5, S7, and S8).