Enterobacter hormaechei replaces virulence with carbapenem resistance via porin loss

Significance

Enterobacter hormaechei is an understudied, emergent, multidrug-resistant pathogen. We deployed comparative genomics and in vivo infections to identify two porins necessary for lethal intraperitoneal infection and subsequent bacteremia of immunocompetent animals, but dispensable for infection of immunosuppressed animals. Porin loss increased fitness during antibiotic exposure, suggesting a trade-off between virulence and multidrug resistance.

Abstract

Pathogenic Enterobacter species are of increasing clinical concern due to the multidrug-resistant nature of these bacteria, including resistance to carbapenem antibiotics. Our understanding of Enterobacter virulence is limited, hindering the development of new prophylactics and therapeutics targeting infections caused by Enterobacter species. In this study, we assessed the virulence of contemporary clinical Enterobacter hormaechei isolates in a mouse model of intraperitoneal infection and used comparative genomics to identify genes promoting virulence. Through mutagenesis and complementation studies, we found two porin-encoding genes, ompC and ompD, to be required for E. hormaechei virulence. These porins imported clinically relevant carbapenems into the bacteria, and thus loss of OmpC and OmpD desensitized E. hormaechei to the antibiotics. Our genomic analyses suggest porin-related genes are frequently mutated in E. hormaechei, perhaps due to the selective pressure of antibiotic therapy during infection. Despite the importance of OmpC and OmpD during infection of immunocompetent hosts, we found the two porins to be dispensable for virulence in a neutropenic mouse model. Moreover, porin loss provided a fitness advantage during carbapenem treatment in an ex vivo human whole blood model of bacteremia. Our data provide experimental evidence of pathogenic Enterobacter species gaining antibiotic resistance via loss of porins and argue antibiotic therapy during infection of immunocompromised patients is a conducive environment for the selection of porin mutations enhancing the multidrug-resistant profile of these pathogens.

The Enterobacter cloacae complex (Ecc) is a group of seven genetically related bacterial species of clinical concern (1). Ecc bacteria are normal residents of the human intestinal microbiota, and pathogenic species, most often Enterobacter hormaechei and Enterobacter cloacae (1), can cause opportunistic infections in immunocompromised individuals and pediatric patients (2–7). Despite globally being one of the leading causes of mortality due to bacterial infection (8), surprisingly little is known about the molecular mechanisms of Ecc pathogenesis, with capsular polysaccharide being the sole virulence factor identified during mammalian infection (9). This lack of knowledge is especially troubling given the intrinsic and high rates of acquired antimicrobial resistance among Ecc bacteria.

Multidrug-resistant (MDR) Ecc bacteria are members of the ESKAPE pathogen group (Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacter spp.) (10). Intrinsic resistance to penicillins and first- and second-generation cephalosporins in Enterobacter spp. can be exacerbated by mutations promoting AmpC hyperproduction during antimicrobial therapy, conferring resistance to third-generation cephalosporins and aztreonam, and classifying these bacteria as extended-spectrum β-lactamase (ESBL)-producing organisms (11, 12). Plasmid-encoded aminoglycoside resistance and quinolone resistance often coincide with ESBL and/or carbapenemase production, and ESBL-producing Ecc and carbapenem-resistant Ecc (CREC) are of grave concern clinically (12). Carbapenemase-encoding plasmids are promoting the worldwide dissemination of certain Ecc lineages (13), such as the CREC sequence types 171 (ST171) and 78 (ST78) (14), though Ecc bacteria can gain carbapenem resistance via other mechanisms, such as impaired membrane permeability. Last, antibiotic treatment failure during Ecc infections is also associated with the formation of heteroresistant subpopulations (15, 16).

Porins are multimeric, β-barrel proteins in the outer membrane of Gram-negative bacteria that serve as initial entryways into the bacterial cell (17). Porins compose a major proportion of outer membrane proteins, with an estimated 10⁵ porin copies present in Enterobacteriaceae cells (18). While they act as nonspecific transporters, the size and charge of a given substrate dictates entry and passage through the porin (17), and certain antimicrobials are known for their use of porins for cellular entry, such as β-lactams and fluoroquinolones (19, 20). Loss of porin proteins and/or porin activity during antibiotherapy has been reported in clinical isolates of several Gram-negative pathogens, including the loss of porins OmpC and OmpF in Escherichia coli (21, 22), as well as their orthologs in Enterobacter species (23), K. pneumoniae (24, 25), and Salmonella enterica (26, 27). However, given the role of porins in nutrient uptake, antibiotic-induced loss of porin activity can result in fitness defects and potentially impair virulence (28). Although loss of the porin OprD in P. aeruginosa has been implicated in MDR as well as enhanced virulence (29), infection models have revealed an importance for porins in the virulence of E. coli (30), K. pneumoniae (31), A. baumannii (32), and Vibrio cholerae (33).

In this study, we combined infection models and comparative genomics with the aim of identifying virulence traits in a contemporary collection of MDR Ecc clinical isolates (14). By comparing the genomes of phylogenetically clustered E. hormaechei isolates, we identified two porins as important virulence determinants during infection of immunocompetent hosts. We refer to these porins as OmpC and OmpD, which are their homologs in E. coli and S. enterica, respectively. These porins were dispensable for virulence in immunocompromised hosts, and porin loss bestowed carbapenem resistance both in vitro and in an ex vivo bacteremia model. Our findings provide experimental evidence of the role of porin loss in MDR Ecc, as well as support a model where porin loss confers an overall fitness benefit in the context of antibiotherapy and immunosuppression.

Results

E. hormaechei ST78 Is Virulent in a Mouse Model of Peritonitis.

Genomic analyses have identified E. hormaechei ST171 and ST78 as the predominant lineages causing human infections in the United States and Europe (14, 34–37). Using a collection of contemporary ST171 and ST78 clinical isolates (14), we sought to assess the virulence potential of the two lineages in an established, preclinical murine model of peritonitis (9, 15). We observed that the isolates belonging to ST171 were essentially avirulent in this model, while the ST78 isolates were virulent, with approximately 50% of mice succumbing to infection within 96 h postinfection (h.p.i.) (Fig. 1A).

Fig. 1.

E. hormaechei ST78 is virulent in a mouse model of peritonitis. (A) Survival curves of mice infected i.p. with 10⁸ colony-forming units (CFU) of indicated E. hormaechei ST171 and ST78 isolates. Number of isolates tested per ST indicated in parentheses. n = 5 to 10 mice per isolate, and 5 mice for PBS control. P < 0.0001, log-rank (Mantel-Cox) test. (B and C) ST78 survival curve data shown in (A), separated by ST78 clade (B) and isolates belonging to ST78 clade D (C). In (B), number of isolates tested per clade indicated in parentheses. In (C), n = 10 mice per isolate, and 5 mice for PBS control. (D) Bacterial burdens of NR3072 and NR3033 in the peritoneal lavage fluid, blood, spleen, and lungs of mice, 16 h after i.p. injection with 10⁸ CFU. The horizontal dotted line indicates the limit of detection. n = 5 mice per isolate. Error bars indicate geometric SD. **P < 0.01, *P < 0.05, Mann–Whitney test.

We next determined whether specific ST78 sublineages were more virulent than others. We separated the isolates by ST78 clade (14) (Fig. 2A) and found that, on average, clade D isolates were the most virulent (Fig. 1B). Of the six ST78 clade D isolates in our collection, five were virulent, killing 60 to 80% of mice; however, one of the clade D isolates, NR3033, was avirulent (Fig. 1C).

Fig. 2.

Comparative genomics of E. hormaechei isolates. (A) Maximum-likelihood phylogeny of ST78 isolates. Clade designations have been previously described (14). (B) The genomes of six ST78 clade D isolates were compared via maximum-likelihood phylogeny, core-genome single-nucleotide substitution (SNV) counts, and clustering of orthologous genes. The first heatmap to the right of the phylogeny shows counts of isolate-pair SNVs, while the second heatmap shows the number of genes from ortholog clustering that are not shared between a pair of isolates. Through these comparisons, we uncovered that NR3033 features a 2,384-nucleotide deletion that disrupts yddG and ompD and a deletion of a guanine in nucleotide position 410 of ompC, resulting in a frameshift. (C) Neighbor-joining phylogeny of 6,833 E. hormaechei assemblies from NCBI. Pie charts display the proportions of porin genes ompD and ompC in an intact state (present) or with disruptive mutations: presumed gene deletion (not found), disruption by insertion sequence (IS), or frameshift/truncation. The large pie charts on the left depict porin status proportions for all isolates in the phylogeny. Smaller pie charts show porin gene status of subsets with at least 50 isolates that descend from an at least 500-SNVs-long phylogeny branch (highlighted in yellow). Additionally, porin gene status was tabulated for the 12 most frequent sequence types (ST), indicated with a red branch coloring and a label. ompD pie charts have a thicker border than ompC pie charts. *ompC was duplicated in two isolates.

To determine whether the NR3033 burden in tissues of infected mice was lower than those of the related virulent isolates, we infected mice intraperitoneally (i.p.) either with NR3033 or NR3072. Sixteen h.p.i., we humanely killed the mice and collected peritoneal lavage fluid, as well as harvested blood, spleens, and lungs to homogenize and plate for CFU. Across sampled tissues, the burden of NR3033 was ~10-fold less than that of the virulent isolate NR3072 (Fig. 1D).

Comparative Genomics of E. hormaechei Isolates.

The close genomic identity of the clade D isolates afforded an opportunity to perform comparative genomics to associate genetic differences with the avirulent phenotype of NR3033, and thus identify traits associated with E. hormaechei virulence. The six ST78 clade D isolates in our collection were at most 65 single-nucleotide variants (SNVs) apart and shared the great majority of gene content, with a maximum of 265 unmapped orthologs between isolate pairs, among a total of 4,736 distinct orthologs (Fig. 2B). Out of all the genetic differences between the avirulent NR3033 and its five virulent ST78 clade D relatives, we were most drawn to loss-of-function mutations in two porin-encoding genes in the NR3033 genome – ompC, which has a frameshift mutation approximately one-third of the way into the gene (deletion of a guanine at nucleotide position 410), and ompD, of which the majority of the coding sequence has been lost due to a 2,384-nucleotide deletion in the NR3033 chromosome (Fig. 2B). We next sought to quantify the presence of disruptive mutations in ompC and ompD among 6,833 publicly available E. hormaechei assemblies (Fig. 2C and Dataset S1). We found 311 different ST in this phylogeny (Fig. 2C), with ST78 (514/6,833 or ~8%) and ST171 (396/6,833 or ~6%) being the most frequent ones. Across all assemblies, ompC was most often found intact (~94%), though it was also not found (~3%), found truncated or with a frame shift (~2%), and found with an IS immediately upstream or disrupting the gene (<1%). ompD was also most frequently found intact (~95%) and, in a minority of cases, not found (~4%), truncated or with a frame shift (~1%), or affected by IS (<1%). In some lineages, porin genes appeared to be more frequently disrupted, especially ompC (Fig. 2C); in ST78, ompC (~98%) and ompD (~95%) were most frequently found intact, with few instances of disruption by truncation/frame shift or IS (<1%), and ompD not being found in some assemblies (~5%). While ompC showed similar percentages in ST171 (Dataset S1), ompD was found intact in a lower proportion of assemblies (~85%), not being found in some cases (~13%) or found disrupted by IS, frame shift, or truncation (~2%).

To test for phylogenetic clustering of porin mutants, we identified for each isolate in our phylogeny the same-lineage nearest neighbor (NN) in terms of the number of SNVs; on average, isolates had a distance of 3.9 SNVs to their NN. Some assemblies with “quasi-disrupted” (short for disrupted or missing) ompD had a tendency to cluster in our phylogeny, with 184 / 361 isolates having NNs with identically (same mutations or same gene absence) quasi-disrupted ompD; their NN SNV distance was significantly lower than among all NNs (< 1 SNVs on average; P < 0.0001; Dataset S1). While assemblies with quasi-disrupted ompC had a higher proportion of NNs with quasi-disrupted ompC (255 / 399) and with identical ompC quasi-disruption (250/399), their NN SNV distance was higher than among all NNs (4.2 SNVs on average, P = 0.3; Dataset S1), indicating a higher degree of de novo ompC disruption and repair as opposed to vertical descent.

Curious about the prevalence of porin disruption in Enterobacter species other than E. hormaechei, we next inspected ompC and ompD in 3,850 publicly available Enterobacter assemblies. There were 41 different species in this dataset and the most frequent ones were Enterobacter cloacae (n = 695), Enterobacter roggenkampii (n = 688), and Enterobacter kobei (n = 593; Dataset S1). In most cases, ompC was found intact (~94%) and, less frequently, truncated or with a frame shift (~3%), not found (~2%), or affected by IS (~1%; Dataset S1). In contrast, ompD was found intact less frequently than ompC (~64%), missing in many assemblies (~35%) and sometimes truncated or with a frame shift (~1%), or affected by IS (< 1%; Dataset S1). ompC and ompD disruption through point mutations or IS was comparable in E. hormaechei and other Enterobacter species, but ompD was completely or frequently missing in some Enterobacter species, consistent with its limited occurrence among Enterobacteriaceae (17).

Porins Are Required for the Virulence of ST78 Clade D Isolates.

To assess whether loss of OmpC and OmpD impairs the virulence of E. hormaechei E. hormaechei ST78 clade D strains, we generated unmarked, in-frame deletion mutations in both ompC and ompD in two virulent isolates – NR3072 and NR2339. While single porin gene deletions (∆ompC or ∆ompD) in NR3072 reduced virulence in the murine peritonitis model, only the double deletion mutant (NR3072 ∆ompC ∆ompD) was completely avirulent (Fig. 3A). To ensure the avirulent phenotype was due to porin loss, we complemented the NR3072 double deletion mutant with each porin-encoding gene (and their native promoters) at the attTn7 site of the chromosome (NR3072 ∆ompC ∆ompD attTn7::ompC+ompD). Virulence was restored to wild-type (WT) NR3072 levels in the complemented mutant, whereas the double deletion mutant with an antibiotic resistance cassette alone at the attTn7 site was nearly avirulent (Fig. 3B). Importantly, this porin-dependent virulence was not specific to NR3072, as NR2339 ∆ompC ∆ompD was also avirulent, and the complemented NR2339 double mutant was similarly virulent to the parental NR2339 isolate (Fig. 3C). Although additional genetic differences were detected in the NR3033 genome in reference to the genomes of the five virulent strains (Fig. 2B), deleting genes representing these variants from the NR3072 genome did not affect the virulence of this strain (SI Appendix, Fig. S1). Therefore, the simultaneous loss of ompC and ompD was the only genetic manipulation that abrogated virulence of ST78 clade D isolates.

Fig. 3.

Porins are required for the virulence of E. hormaechei ST78 clade D isolates. (A) Survival curves of mice i.p. infected with 10⁸ CFU of NR3072, NR3072 ∆ompC, NR3072 ∆ompD, and NR3072 ∆ompC ∆ompD. n = 10 mice per strain. ***P < 0.001, log-rank (Mantel-Cox) test. (B) Survival curves of mice i.p. infected with 10⁸ CFU of NR3072 ∆ompC ∆ompD attTn7::Cm and NR3072 ∆ompC ∆ompD attTn7::ompC+ompD. n = 10 mice per strain. **P < 0.01, log-rank (Mantel-Cox) test. (C) Survival curves of mice i.p. infected with 10⁸ CFU of NR2339, NR2339 ∆ompC ∆ompD, and NR2339 ∆ompC ∆ompD attTn7::ompC+ompD. n = 10 mice per strain. **P < 0.01, log-rank (Mantel-Cox) test. (D) Bacterial burdens of NR3072, NR3072 ∆ompC ∆ompD, and NR3072 ∆ompC ∆ompD attTn7::ompC+ompD in peritoneal lavage fluid, blood, spleen, and lungs 16 hours after i.p. infection. The horizontal dotted line indicates the limit of detection. n = 15 mice per strain. Error bars indicate geometric SD. ****P < 0.0001, ***P < 0.001, **P < 0.01, *P < 0.05, or not significant (ns), Mann–Whitney test.

Given porins are known importers of nutrients, we hypothesized the attenuated virulence of OmpC- and OmpD-deficient strains may be due to fitness defects resulting from impaired growth within the host. To test this, we grew NR3033, NR3072, and their porin gene-manipulated derivatives in nutrient-rich and nutrient-poor media and plated for viable bacteria over time (SI Appendix, Fig. S2). While all strains grew equally well in nutrient-rich medium, porin-deficient strains (NR3033 and NR3072 ∆ompC ∆ompD) exhibited statistically significant growth defects in comparison to their respective porin-intact relatives (NR3033 attTn7::ompC+ompD, or NR3072 and NR3072 ∆ompC ∆ompD attTn7::ompC+ompD) when nutrients were limiting. Thus, during murine infection, when nutrient availability is likely diminished compared to rich culture broth, nutrient import via porins may be required for E. hormaechei to cause lethal infection.

Porin Loss Leads to Decreased Infectivity in the Murine Peritonitis Model.

We infected mice i.p. with NR3072, NR3072 ∆ompC ∆ompD, or NR3072 ∆ompC ∆ompD attTn7::ompC+ompD, humanely killed mice 16 h.p.i., and harvested tissues to determine the infectivity of the three constructs. Like the comparison between NR3072 and NR3033 (Fig. 1D), burdens of the porin-deficient NR3072 mutant were approximately 10-fold lower than WT NR3072 (Fig. 3D). At the site of infection, NR3072 ∆ompC ∆ompD exhibited a burden defect in the peritoneal lavage, and dissemination from and/or survival outside the peritoneum was impaired for this mutant, with lower burdens in the bloodstream and disseminated organs. At this time point, the burdens of NR3072 ∆ompC ∆ompD attTn7::ompC+ompD often trended toward, though were not restored to, WT NR3072 burdens. Although the genetically complemented mutant exhibited equal virulence to WT NR3072 (Fig. 3 A and B), the decreased burdens of the genetically complemented strain may be a result of the unnatural complementation strategy (ompC and ompD fused together and delivered to the attTn7 site of the chromosome). Overall, loss of OmpC and OmpD in the NR3072 background phenocopied the infectivity defect seen in the naturally porin-deficient NR3033 isolate.

Both OmpC and OmpD Contribute to Carbapenem Sensitivity in ST78 Clade D Isolates.

Given the role porins play in antibiotic uptake, as well as the selective mutations that occur in porin-encoding genes during antibiotic therapy, we sought to determine whether the loss of OmpC and OmpD in the E. hormaechei ST78 clade D isolates promotes carbapenem resistance. Notably, the natural OmpC- and OmpD-deficient NR3033 isolate was originally classified as carbapenem-resistant, while its clade D siblings were classified as carbapenem-sensitive [SI Appendix, Table S1 and (14)]. When the ST78 clade D isolates and their genetically manipulated strains were grown in rich medium (TSB) supplemented with 0.5 µg/mL meropenem, we noticed an ompC- and ompD-dependent meropenem sensitivity for NR3033, NR3072, and NR2339 (Fig. 4A). While genetically complementing NR3033 with either ompC or ompD alone did not sensitize the isolate to meropenem, simultaneously complementing with both porin-encoding genes did, indicating both porins are involved in the import of the antibiotic. Conversely, the WT strains of NR3072 and NR2339 were sensitive to meropenem, as both strains produce OmpC and OmpD. Deletion of both ompC and ompD rendered the isolates resistant to meropenem, while complementing the double deletion mutants with both ompC and ompD restored the sensitive phenotype. In NR3072, deletion of either ompC or ompD showed partial desensitization to meropenem, whereas both genes needed to be deleted from the NR2339 chromosome to provide any level of resistance to this isolate.

Fig. 4.

Both OmpC and OmpD contribute to carbapenem sensitivities of E. hormaechei ST78 clade D isolates. (A) Growth curves of NR3033, NR3072, NR2339, and their porin gene-manipulated derivatives in 0.5 µg/mL meropenem. n = 6 biological replicates per strain. Error bars indicate SE of the mean. (B–D) Minimum inhibitory concentrations (MICs) of meropenem (B), ertapenem (C), and imipenem (D) against NR3033, NR3072, NR2339, and their porin gene-manipulated derivatives. n = 3 to 6 biological replicates per strain for each antibiotic. Error bars indicate SE of the mean. ****P < 0.0001, ***P < 0.001, or not significant (ns), one-way ANOVA with Sidák’s multiple comparison test.

Next, we used E-test strips impregnated with three clinically relevant carbapenems to measure the MICs of the antibiotics against the ST78 clade D isolates and their genetically manipulated strains. Meropenem (Fig. 4B), ertapenem (Fig. 4C), and imipenem (Fig. 4D) all showed porin-dependent antibacterial activity. The MICs of the three carbapenems were all statistically significantly higher against ompC and ompD mutants (NR3033, NR3072 ∆ompC ∆ompD, and NR2339 ∆ompC ∆ompD) than they were for the porin-intact strains (NR3072, NR2339, and the genetically complemented strains). The E-test strips showed NR3072 was inherently less sensitive than NR2339 to the carbapenems, potentially explaining why meropenem was more inhibitory toward NR2339 than NR3072 during growth in broth culture (Fig. 4A). In addition to the tested carbapenems, NR3072 ∆ompC ∆ompD was more resistant to the beta-lactam/beta-lactamase inhibitor combination piperacillin/tazobactam than were NR3072 and NR3072 ∆ompC ∆ompD attTn7::ompC+ompD, suggesting these porins also import penicillin antibiotics (SI Appendix, Fig. S3).

OmpC- and OmpD-Deficient ST78 Isolates Are Virulent in Immunosuppressed Mice.

Since NR3033 was originally isolated from a patient with pneumonia [SI Appendix, Table S1 and (14)], we reasoned that porin loss did not abrogate virulence of the pathogen in this patient. As clinical Enterobacter infections typically occur in immunocompromised individuals, such as those in neonatal intensive care units, on mechanical ventilation, or undergoing chemotherapy (12, 38–40), we hypothesized that Enterobacter porin mutants may retain virulence in immunosuppressed hosts, and that such mutations may be selected for during antibiotic therapy. To assess the impact of immunosuppression on the virulence of porin mutants, we dosed mice with the immunosuppressive drug cyclophosphamide (41, 42) prior to infection with E. hormaechei isolates to render the animals neutropenic (Fig. 5A). OmpC and OmpD were dispensable for virulence in this model, as the natural ompC and ompD mutant NR3033, as well as laboratory-generated NR3072 and NR2339 mutants, were similarly virulent to WT NR3072 (Fig. 5B). Importantly, i.p. inoculation with the E. coli DH5α λpir laboratory strain (or PBS control) did not cause mortality in cyclophosphamide-treated mice, indicating this model did not result in lethal infection by otherwise nonpathogenic Gram-negative bacteria. Thus, OmpC and OmpD are dispensable for E. hormaechei virulence in an immunosuppressed host.

Fig. 5.

Porins are dispensable for virulence in immunocompromised hosts. (A) Schematic of E. hormaechei infection in the cyclophosphamide model of immunosuppression. (B) Survival curves of cyclophosphamide-treated mice i.p. infected with 10⁸ CFU of E. coli DH5α λpir or E. hormaechei ST78 strains NR3033, NR3072, NR3072 ∆ompC ∆ompD, or NR2339 ∆ompC ∆ompD. n = 10 mice per E. hormaechei strain and PBS control, n = 5 mice for DH5α λpir. (C) Schematic of meropenem-treated ex vivo human whole blood bacteremia model. (D) Burdens over time of NR3072 attTn7::Hyg (“WT”) and NR3072 ∆ompC ∆ompD attTn7::Cm (“∆ompC ∆ompD”) during coinfection of human whole blood, with or without meropenem treatment. The horizontal dotted line indicates the limit of detection. n = 4 biological replicates (4 different blood donors). Error bars indicate geometric SD. ****P < 0.0001, one-way ANOVA with Sidák’s multiple comparison test.

Porin Loss Provides a Fitness Advantage in the Context of Meropenem Treatment During Bacteremia.

To investigate whether porin loss provides a fitness advantage during antibiotic therapy, we used an ex vivo human whole blood bacteremia model (9). Human blood was freshly drawn into anticoagulant-coated tubes, treated with 0.5 µg/mL meropenem (or buffer control), and inoculated with an equal mixture of NR3072 attTn7::Hyg and NR3072 ∆ompC ∆ompD attTn7::Cm at a multiplicity of infection of 1 (~10⁶ CFU/mL total) (Fig. 5C). Plating for CFU over time onto agar containing either 200 µg/mL hygromycin or 50 µg/mL chloramphenicol allowed us to determine the population of the individual strains in this coinfection experiment. While there were decreases in burdens for both strains at early time points, regardless of meropenem, we saw significant porin-dependent phenotypes 24 h.p.i. (Fig. 5D). In blood lacking meropenem, the burden of the WT NR3072 strain was approximately five times (one-half-log) greater than that of NR3072 ∆ompC ∆ompD 24 h.p.i. However, in blood containing meropenem, the burden of NR3072 ∆ompC ∆ompD was approximately six-logs greater than that of WT NR3072 at the 24-h time point. Therefore, porin loss provides a strong competitive advantage during bacteremia and ensuing antibiotic therapy.

Discussion

Despite the striking global burden of Enterobacter infections and the MDR propensity of these pathogens, few investigations into the molecular mechanisms of Enterobacter pathogenesis exist. Such studies include virulence screens in invertebrate hosts and the identification of capsular polysaccharide as a critical virulence factor and potential vaccine target (9, 43). To improve our prophylactic and therapeutic interventions targeting Enterobacter infections, our knowledge on Enterobacter pathogenesis must improve. In this study, we leveraged comparative genomics on a collection of contemporary Enterobacter clinical isolates to associate genetic signatures with virulence in a mouse model of peritonitis, and we identified two porins, OmpC and OmpD, as determinants of lethal infection in the ST78 clade of E. hormaechei.

How OmpC and OmpD, as well as other porins, promote lethal Enterobacter infection remains to be determined. Given the known role of porins in nutrient uptake (17), loss of these proteins may solely decrease bacterial fitness during murine infection, which is in line with the infectivity defects seen in porin-deficient strains in our model (Figs. 1 D and 3 D), as well as the growth defects seen in minimal medium (SI Appendix, Fig. S2). Moreover, porins may provide a nutrient-independent benefit during infection, such as regulating interactions with host epithelial and/or immune cells. In contrast to our data on the importance of OmpC and OmpD for lethal E. hormaechei infection, the porin OprD has been shown to negatively impact the virulence of P. aeruginosa, and oprD mutant P. aeruginosa displayed dramatic changes in transcriptional landscape compared to a WT strain (29). Thus, Enterobacter porins may indirectly promote virulence by supporting a cellular environment in which global transcription, or transcription of select genes, enhances pathogenesis.

Although OmpC and OmpD were required for virulence in our murine model of peritonitis (Fig. 3), our genomic analyses of 6,833 publicly available E. hormaechei genomes revealed that putative disruptive mutations in porin-encoding genes are common in some lineages (Fig. 2C and Dataset S1), apparently as a result of multiple selection and reversion events. Notably, these disruptive mutations are uncommon in the high-risk lineages ST78 and ST171. Additionally, we detected frequent disruptive mutations in other porin genes, such as ompF. Therefore, genomic evidence suggests one or more selective pressures are leading to the mutational landscape in E. hormaechei porin-related genes. Given the known role of porin loss in antibiotic resistance [Fig. 4 and (21–27)], antibiotic therapy likely is promoting the prevalence of mutations in porin-related genes. However, our murine infection data indicate the E. hormaechei porins OmpC and OmpD (and possible additional, untested porins) are required for full virulence of this pathogen. This discrepancy may be explained by the fact that our standard murine peritonitis model used immunocompetent mice. Perhaps porins are necessary for Enterobacter pathogens to survive in and infect a host with a strong immune response, while porin-deficient Enterobacter can be virulent and even selected for in immunosuppressed hosts under antibiotic treatment, with the latter often being the victims of Enterobacter infections (2).

It is intriguing to note how much more prevalent disruptive mutations are in porin-related genes within E. hormaechei ST171 genomes compared to ST78 genomes (Dataset S1). It is possible the avirulent phenotype of ST171 isolates in the murine peritonitis model (Fig. 1A) is due to porin loss, especially compared to ST78 isolates, which are relatively more intact with regard to porin-related gene status (Fig. 2 and Dataset S1). That being said, porin deficiency did not impair E. hormaechei virulence in an immunosuppressed model of infection (Fig. 5B). Moreover, we showed that both OmpC and OmpD sensitized E. hormaechei to meropenem (Fig. 4A), and that porin loss provided a fitness advantage in a bacteremia model of meropenem treatment (Fig. 5D). Therefore, the naturally OmpC- and OmpD-deficient isolate NR3033 may be an example of an Enterobacter pathogen having undergone antibiotic-induced selective pressure to lose porin proteins. Considering data from the cyclophosphamide and ex vivo bacteremia models, we propose the immunocompromised host is a suitable environment for the selection of porin-abrogating mutations that enhance the MDR profile of Enterobacter pathogens. This scenario may hold true for other opportunistic pathogens as well, further stressing the need for proper antibiotic stewardship in the clinical setting.

Materials and Methods

Ethics Statement.

Freshly isolated human whole blood was collected and used for experiments under a protocol approved by the New York University Grossman School of Medicine Institutional Review Board for Human Subjects (IRB number i14-02129_CR6). All donors provided written consent before participating in the study.

Experiments using animals were conducted under a protocol approved by the Institutional Animal Care and Use Committee of New York University Langone Health (IA16-00050) and were performed according to guidelines from the NIH, the Animal Welfare Act, and US Federal Law.

Bacterial Strains, Plasmids, and Growth Conditions.

All bacterial strains and plasmids used in this study are listed in SI Appendix, Tables S2 and S3, respectively. The clinical isolates were reported previously (9, 14). Routine culture of Enterobacter isolates was performed in tryptic soy broth (TSB) with aeration at 37 ˚C or on tryptic soy agar (TSA) at 30 ˚C. Escherichia coli strains were routinely cultured in lysogeny broth (LB) with aeration at 37 ˚C or on LB agar at 37 ˚C. Media were supplemented with antibiotics, when necessary, as follows: 200 µg/mL hygromycin (Gold Biotechnology) for Enterobacter and E. coli, 50 µg/mL chloramphenicol (Gold Biotechnology) for Enterobacter (TSA/Cm₅₀ plates incubated at 37 ˚C), 34 µg/mL chloramphenicol for E. coli, and 100 µg/mL ampicillin (Fisher BioReagents) for E. coli. 2,6-diaminopimelic acid (DAP, Thermo Scientific) was added to LB or LB agar at a concentration of 200 µg/mL for culturing E. coli strain RHO3.

Animal Housing.

Experimental animals received PicoLab Rodent Diet 20 (LabDiet) and acidified water ad libitum. Animals were housed under a normal light cycle (12 h on/12 h off) and at a temperature of 21 ± 2 ˚C.

Murine Peritonitis Model.

8 wk-old female C57BL/6J mice (The Jackson Laboratory) were i.p. injected with 10⁸ CFU of Enterobacter isolates in 300 µL volume. For survival experiments, mice were monitored multiple times daily following infection and morbidity, mortality, and weight loss were assessed. Mice that exhibited severe morbidity and/or lost ≥ 20% of their starting body weight were humanely killed. Severe morbidity was determined by hunched posture, lack of movement, ruffled fur, and inability to eat or drink. Unless otherwise noted, experiments were conducted in duplicate, each replicate with five mice, for a total of 10 mice per experiment.

For experiments assessing bacterial tissue burdens following i.p. infection, mice were humanely killed 16 h.p.i., and peritoneal lavage fluid, blood, and organs were collected. Blood samples were incubated on ice in 0.1% saponin (MilliporeSigma) for 20 min before plating, and organs were homogenized before plating. Peritoneal lavage fluid, saponin-treated blood, and homogenized organs were serially diluted in PBS and plated on TSA for CFU enumeration.

Sequencing, Assembly, and Annotation of Bacterial Genomes.

The Enterobacter isolates investigated in this study were previously sequenced using Illumina short reads and assembled; isolate NR0013 was also sequenced with Nanopore long reads as previously described (NCBI BioProject PRJNA867733) (9). Isolates NR2329, NR2333, NR2339, NR3033, and NR3072 were additionally sequenced with Nanopore long reads, as follows. 2 µg of DNA in 50 µL volume was sheared to 30 kb with the Covaris g-tube (Covaris, #520079) following the manufacturer’s protocol in a tabletop microcentrifuge. Sheared DNA was repaired, and end prepped with the NEBNext FFPE DNA repair kit (M6630) and the NEB Ultra II End-prep kit (E7546) for 30 min @ 20 ˚C and 30 min @ 65 ˚C. Following incubation, the samples were cleaned with 0.5x volume of Ampure XP beads (30 µL) with a standard bead cleanup protocol to remove any small fragments of DNA. Clean sheared DNA was eluted in 25 µL of water and proceeded to the native barcoding ligation step. A unique barcode from kit EXP-NBD104 (Oxford Nanopore) was added to each sample and ligated using the NEB Blunt/TA Ligase Mix (M0367) for 1 h at room temperature. Following ligation, samples were cleaned with Ampure XP beads and eluted in dH₂O. Barcoded DNA was individually checked on Qubit using the high sensitivity DNA kit. Samples were combined at equimolar amounts to create a 1 µg pool to input into the adapter ligation step using the Adapter Mix II (Oxford Nanopore, SQK-LSK-109) and the NEBNext Quick-Ligation Reaction Buffer (B6058) plus the NEB T4 DNA Ligase 2M U/mL (M0202) for 1 h at room temperature. Following ligation of the pooled sample, Ampure XP beads were added and the sample was bound. Following binding, the sample was placed on a magnetic rack, residual supernatant was removed, and the beads were washed twice with Long Fragment Buffer (SQK-LSK-109) to remove smaller DNA fragments. Following the second wash and supernatant removal, the final pool was eluted with Elution Buffer for 1 h @ 37 ˚C. The final pool was collected and run on Qubit to verify the loading concentration and 1 uL was run on Genomic DNA Screentape with the Agilent Tapestation system to assess base pair size. The pool was loaded onto the PromethION flow cell (FLO-PRO002) following the protocol instructions and run for 72 h with active basecalling turned on.

Hybrid assemblies for NR0013, NR2329, NR2333, NR2339, NR3033, and NR3072 were generated as described before (9). In summary, Nanopore reads were quality filtered with Filtlong v. 0.2.0 (https://github.com/rrwick/Filtlong), assembled with Trycycler v. 0.4.1 (44), and polished with medaka v. 1.4.3 (https://github.com/nanoporetech/medaka) using the r941_prom_high_g4011 model. We used pilon v. 1.24 for additional polishing with Illumina short reads. The resulting assemblies were annotated with PGAP v. 2021-11-29.build5742 (45). Sequencing data and assemblies were added to NCBI BioProject PRJNA867733.

Comparative Genomics.

On March 26, 2024, we downloaded 6,106 Enterobacter assemblies from NCBI RefSeq (https://www.ncbi.nlm.nih.gov/refseq/) and complemented them with 4,580 Enterobacter assemblies from NCBI GenBank (https://www.ncbi.nlm.nih.gov/genbank/) that are excluded from RefSeq because they stem from large multi-isolate projects. After species identification with GTDB-Tk v.2.1.1 (46), we filtered for species E. hormaechei A and retained a set of 6,836 assemblies. ST for these assemblies were determined with mlst v. 2.19.0 using default parameters (https://github.com/tseemann/mlst).

The snippy v. 4.6.0 pipeline (https://github.com/tseemann/snippy) was used with default parameters for alignment, variant calling, and core-genome computation. For isolates in our collection, snippy was applied to the short-read sequencing data using the annotated NR3033 assembly (see the Sequencing, Assembly, and Annotation of Bacterial Genomes section of the Materials and Methods) as a reference. Core genomes were computed for 18 ST78 isolates (4,439,247 alignment positions of which 1,158 were variable) and a subset of six closely related ST78 clade D isolates (the same for which hybrid assemblies were generated; 4,876,976 alignment positions of which 85 were variable) and used to obtain two maximum likelihood phylogenies computed with IQ-TREE v. 1.6.10 (47) using automatic model selection and 1,000 ultrafast bootstrap replicates. For assemblies downloaded from NCBI, snippy was applied to the contigs, using RefSeq assembly GCF_001729745.1 as a reference. A core genome was computed (449,100 alignment positions of which 98,302 were variable) and SNV counts were used to create a neighbor-joining phylogeny using the ape package v. 5.5 (48) in R v.3.6.0 (49). We excluded assemblies GCF_029029395.1, GCF_008082005.1, and GCF_012975085.1 due to outlying branch lengths. We determined each assembly’s SNV NN by identifying the lowest SNV count when comparing to all other assemblies; ties were resolved by random selection. We computed the percentiles of the distribution of SNV counts for the NN set; SNV counts rose moderately with increasing percentiles up to the 90th percentile (25 SNVs) thereafter rising sharply to a maximum of 4,954 SNVs at the 100th percentile. To exclude NNs from different lineages, we disregarded those with SNV counts greater than 25.

Predicted proteomes from our hybrid assemblies were used for clustering of orthologous genes with orthofinder v. 2.5.2 (50) using the MSA method for gene tree inference.

For identifying porin-related genes in the assemblies downloaded from NCBI, we used a database of 27 nucleotide sequences (Dataset S1) in a blastn v. 2.11.0 (51) search with default parameters. Additionally, we searched for IS from ISfinder (52) with a blastn search with a minimum nucleotide identity of 70% and minimum query coverage per high-scoring pair of 70%. Using python v. 3.9.2 (https://www.python.org/), we identified the porin-gene hit with the highest bitscore whenever there were overlapping hits, requiring a minimum nucleotide identity and query coverage of 90% for E. hormaechei and of 80% and 70%, respectively, for other Enterobacter species. To identify IS interrupting porin genes, we inspected porin gene hits with a minimum nucleotide identity of 85% and cross-referenced them with ISfinder hits, allowing for a maximum of 30 nucleotides between porin-gene hits and IS hit. Additionally, we identified IS potentially located in the promoter region if an IS was up to 500 nucleotides upstream of a porin-gene hit. For E. hormaechei, nucleotide sequence matches for porin-gene hits were extracted using BioPython v. 1.78 (https://biopython.org/) and aligned to their corresponding porin-gene nucleotide sequence using nucmer v.3.23 (53) via pymummer (https://github.com/sanger-pathogens/pymummer), specifying a minimum identity of 90%, a minimum length of 20, a break length of 200, and the maxmatch and show_snps flags. We then used python code inspired on ariba v. 2.14.6 (54) to identify amino-acid mutations in these alignments. For other Enterobacter species, we extracted nucleotide sequence matches for ompC and ompD and identified frame shift and stop codon mutations, as follows. To identify relevant frame shifts, we used the BioPython Align module to align the extracted matches to their reference, scanned the alignment for frame shifts by keeping track of alignment gaps, and deemed a porin sequence frameshifted if more than 10% of its nucleotides occurred in a frameshifted region. To identify premature stop codons, we used the Align module with a BLOSUM62 matrix to identify the correct reading frame for translating the extracted nucleotide sequence; we identified the first occurring stop codon in the translated sequence and deemed it premature if the number of amino acids preceding it was less than 90% of the translated reference sequence length. To summarize porin-gene status for Fig. 2, we used phylobase v. 0.8.12 (https://github.com/fmichonneau/phylobase) to identify branches in our neighbor-joining phylogeny of length 500 or greater, from which at least 50 tips descend.

Phylogenies were plotted with iToL (55).

Genetic Manipulation of Enterobacter.

The pTOX5 allelic exchange vector was used to generate in-frame, unmarked deletion mutations in Enterobacter isolates following a previous protocol (56), with slight modifications. Approximately 500 base pairs (bp) of sequence upstream of the region to be deleted was fused to ~500 bp of sequence downstream of the region to be deleted via splicing by overhang extension PCR. Fusion products were ligated into pTOX5, resulting plasmids were conjugated into Enterobacter using E. coli RHO3 on agar containing DAP and 2% D-glucose for 6 h at 37 ˚C, and merodiploids were selected on agar containing 50 µg/mL chloramphenicol and 2% glucose. Resulting colonies were patched onto M9 agar containing 50 µg/mL chloramphenicol and 2% L-rhamnose to verify rhamnose sensitivity. Rhamnose-sensitive merodiploids were grown in LB+2% D-glucose for 2 to 3 h at 37 ˚C with aeration (one colony in 2 mL broth), cells were washed twice in M9 salts+2% L-rhamnose, and dilutions were plated on M9 agar+2% L-rhamnose to isolate single colonies. Deletion mutations were confirmed via PCR and Sanger sequencing.

To genetically complement ompC and ompD, the sequences of the two genes from NR3072, including 404 bp upstream of ompC and 352 bp upstream of ompD, were submitted to Twist Bioscience for DNA synthesis. The synthesized fusion product of the two genes and their upstream regions was PCR-amplified and ligated into the pUC18-R6K-mini-Tn7T-Cm vector (9). This complementation vector was introduced via triparental conjugation between an E. coli S17 λpir strain harboring the plasmid, an E. coli S17 λpir strain harboring the helper transposase plasmid pTNS2 (57), and an Enterobacter recipient. Conjugation was conducted on LB agar for 6 h at 37 ˚C, Enterobacter cells containing the complementation construct were selected on TSA containing 50 µg/mL chloramphenicol, and complemented clones were verified via PCR. The pUC18R6K-mini-Tn7T-Hyg plasmid was generated via replacing the gentamicin resistance-conferring aacC1 gene in pUC18R6K-mini-Tn7T-Gm (57) with the hygromycin resistance-conferring gene from plasmid pMQ310 (58). NR3072 attTn7::Hyg was generated via triparental conjugation with S17 λpir pUC18R6K-mini-Tn7T-Hyg and S17 λpir pTNS2, and NR3072 ∆ompC ∆ompD attTn7::Cm was generated via triparental conjugation with S17 λpir pUC18R6K-mini-Tn7T-Cm and S17 λpir pTNS2.

Growth Curves in Rich and Minimal Media.

Overnight cultures of Enterobacter strains were diluted in PBS, and diluted cells were used to inoculate media at starting bacterial concentrations of ~10⁵ CFU/mL. TSB was used as a rich medium and RPMI+01.% HSA+10mM HEPES was used as a minimal medium. Cultures were grown with aeration at 37 ˚C, and samples were taken at 0, 1, 3, 5, and 7 h postinoculation to serially dilute and plate onto TSA for CFU.

Meropenem Growth Curves.

Overnight cultures of Enterobacter strains were diluted 1:10 into PBS, and 1 µL of these diluted cells were added to 100 µL TSB containing 0.5 µg/mL meropenem trihydrate (TCI America) in the wells of a honeycomb plate. Plates were grown in a Bioscreen C Automated Plate Reader (Oy Growth Curves Ab, Ltd.) at 37 ˚C with agitation, and the OD₆₀₀ values of the wells were measured every 40 min for 24 h. Six biological replicates were performed per Enterobacter strain.

Antibiotic MIC Determination.

Liofilchem® paper strips containing varying concentrations of meropenem, ertapenem, imipenem, or piperacillin/tazobactam were used to measure MICs against Enterobacter strains. Bacteria from overnight cultures were sampled using sterile cotton-tipped applicators and spread on the surfaces of TSA plates. Once dry, antibiotic-impregnated strips were placed on the surfaces of the agar, and plates were incubated at 37 ˚C for 24 h, after which the MICs were measured. At least three biological replicates were performed to test each antibiotic against all indicated Enterobacter strains.

Cyclophosphamide Immunosuppression Model.

Cyclophosphamide monohydrate (MilliporeSigma), solubilized in water and filter-sterilized, was i.p. injected into 7-wk-old female C57BL/6J mice (The Jackson Laboratory) at a 150 mg/kg body weight dose. Four doses were administered on alternating days, and 2 d following the final dose, 10⁸ CFU of bacteria (or sterile PBS as a control) were i.p. injected into the mice. Mice were then followed for morbidity and mortality, and humanely killed when necessary.

Ex Vivo Bacteremia Model.

Consented volunteers provided blood, which was collected into BAPA (benzylsulfonyl-D-Arg-Pro4-amidinobenzylamide) anticoagulant tubes (Diapharma Group, Inc.). Meropenem trihydrate (TCI America) was added to blood at a concentration of 0.5 µg/mL, after which blood was inoculated with a 1:1 mixture of NR3072 attTn7::Hyg and NR3072 ∆ompC ∆ompD attTn7::Cm, at a starting concentration of 10⁶ CFU/mL. Cultures were incubated on a roller drum at 37 ˚C, and at 0, 3, 5, and 24 h.p.i., samples were taken and incubated in 0.1% saponin (MilliporeSigma) on ice for 20 min. Saponin-treated blood was then serially diluted in PBS and plated onto TSA containing either 200 µg/mL hygromycin or 50 µg/mL chloramphenicol to determine the CFU/mL for each strain. Four biological replicates were performed, each using blood from a different donor.

Statistics.

GraphPad Prism 10 was used for all statistical analyses, the results of which are detailed in the figure legends. To compute a P-value for the mean over a particular set of NN SNV counts, we constructed 10,000 equinumerous randomization sets by sampling with replacement from all NN SNV counts. A one-sided P-value was then calculated as the fraction of randomization sets with means larger (or smaller) than that of the set in question.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Data, Materials, and Software Availability

All bacterial genome sequencing data generated by this study are publicly available in the NCBI BioProject PRJNA867733 (https://www.ncbi.nlm.nih.gov/bioproject/?term=prjna867733) (59). All other data are included in the article and/or supporting information.