Host adaptation and convergent evolution increases antibiotic resistance without loss of virulence in a major human pathogen

As human population density and antibiotic exposure increase, specialised bacterial subtypes have begun to emerge. Arising among species that are common commensals and infrequent pathogens, antibiotic-resistant ‘high-risk clones’ have evolved to better survive in the modern human. Here, we show that the major matrix porin (OmpK35) of Klebsiella pneumoniae is not required in the mammalian host for colonisation, pathogenesis, nor for antibiotic resistance, and that it is commonly absent in pathogenic isolates. This is found in association with, but apparently independent of, a highly specific change in the co-regulated partner porin, the osmoporin (OmpK36), which provides enhanced antibiotic resistance without significant loss of fitness in the mammalian host. These features are common in well-described ‘high-risk clones’ of K. pneumoniae, as well as in unrelated members of this species and similar adaptations are found in other members of the Enterobacteriaceae that share this lifestyle. Available sequence data indicate evolutionary convergence, with implications for the spread of lethal antibiotic-resistant pathogens in humans.

Introduction molecule diffusion but which may exclude more bulky anionic carbapenem and cephalosporin antibiotics [20].
Highly antibiotic-resistant K. pneumoniae is both a critical threat pathogen and a model of adaptation in a world with increasing human density and antibiotic exposure. The aim of this study was therefore to understand the pathogenesis and antimicrobial resistance implications of common changes in major porins that diminish membrane permeability.

Bacterial strains, plasmids, primers and growth conditions
The bacterial strains, plasmids and primers used in this study are listed in Table 1 and S1 Table. Porin mutants were constructed in three antibiotic-susceptible K. pneumoniae strains (ATCC 13883, and clinical isolates 10.85 and 11.76 from our laboratory). Bacterial isolates were stored at -80˚C in Nutrient broth (NB) with 20% glycerol and recovered on LB agar plates. Unless otherwise indicated, strains were routinely grown in Mueller-Hinton broth (MHB, BD Diagnostics, Franklin lakes, NJ, USA) or Luria-Bertani (LB, Life Technologies, Carlsbad, CA, USA). E. coli and K. pneumoniae strains carrying the chloramphenicol-resistant plasmids pKM200 and pCACtus were grown at 30˚C on LB agar or in LB broth supplemented with 20 μg/ml chloramphenicol (Sigma-Aldrich, St. Louis, MO, USA). The growth of bacterial cells was determined by measuring the optical density at 600 nm (OD 600 ) in an Eppendorf Biophotometer (Eppendorf AG, Hamburg, Germany).

Construction of porin mutants
Chemical transformation, conjugation and electroporation were carried out using standard protocols. Platinum pfx DNA polymerase (Invitrogen, USA) was used to amplify blunt-ended PCR products. All PCR products were purified (PureLink Quick PCR Purification Kit; Invitrogen, USA). PCR and Sanger sequencing were used to confirm all constructs. Genomic DNA extractions were performed using a DNeasy Blood and Tissue kit (Qiagen, Valencia, CA, USA) and plasmid DNA using a PureLink Quick Plasmid Miniprep kit (Life Technologies, Carlsbad, CA, USA) or a HiSpeed Plasmid Midi Kit (Qiagen, Valencia, CA, USA).
Porin deletions mutants of K. pneumoniae ATCC 13883, 10.85 and 11.76 were created by introduction of tetA (tetracycline-resistance) or aphA-3 (kanamycin-resistance) into unique sites in ompK35 and ompK36 (HincII and StuI, respectively) which had been previously cloned into pGEM-T easy (Promega, Madison, WI, USA). The disrupted porin genes were then cloned into the pCACtus temperature-sensitive suicide vector (pJIAF-7 to pJIAF-12) to replace the respective chromosomal genes by homologous recombination [25]. Confirmation of correct single-copy chromosomal mutations were finally verified by PCR (S1 Table).
OmpK36GD mutants were obtained by amplification of OmpK36 from each parental strain using K36GD1 / K36GD2 and K36GD3 / K36GD4 primers (S1 Table). The amplicon, containing a GD duplication in L3, was cloned first in pGEM-T easy and after digestion with SphI and SacI (New England Biolabs, MA, USA) was introduced into pCACtus. The pCACtusbased constructs (pJIAF-13 to pJIAF-18) were transformed into S17λpir and conjugated into K. pneumoniae ΔOmpK36 (kanamycin-resistance mutant) in which the interrupted gene was replaced by OmpK36 porin with GD duplication in L3 by homologous recombination. Mutants were selected by loss of kanamycin resistance and confirmed by PCR and sequencing.
Double mutants (ΔOmpK35ΔOmpK36 and ΔOmpK35OmpK36GD) were constructed using Lambda Red-mediated recombineering as described previously [26,27], with some modifications. A tetracycline cassette flanked by OmpK35 deletion (~2.5 kb in size) was PCR amplified from an OmpK35 deletion mutant (ΔOmpK35; tetracycline resistant-previously obtained) using primers ompK35X-F and ompK35X-R (S1 Table), and the PCR products were purified. The Red helper plasmid pKM200 was electroporated into ΔOmpK36 or OmpK36GD single mutants. ompK35:tetA fragments were electroporated into ΔOmpK36 or OmpK36GD clones carrying pKM200. Bacteria were grown at 30˚C for 2 h with agitation (225 rpm) followed by overnight incubation at 37˚C. Different dilutions of the electroporated cells were spread on LB agar plates containing 10 μg/ml tetracycline to select for transformants at 37˚C. The correct structure was confirmed by sequencing of PCR amplicons (primers ompK35F1 and ompK35R2, S1 Table).
All the engineered strains were verified by whole genome sequencing.

Complementation of porin mutations
The ompK36 gene with its predicted ribosomal binding site and transcriptional terminator was PCR amplified using 5'-GACAAGCTTTAAAAGGCATATAACAAACAG-3' (forward) and 5'-CTGGGATCCAGCGAGGTTAAACCGG-3' (reverse, S1 Table). Genomic DNA from K. pneumoniae ATCC 13883 wild-type strain (Table 1) was used as template. To generate ompK36 PCR product with the L3 GD mutation, the ATCC OmpK36GD strain was used as template DNA (Table 1). DNA inserts containing ompK36 and ompK36GD were cloned into the low-copy number pACYC184 vector [28] at the HindIII/BamHI restriction sites to generate pJIQQ-1 (pACYC184-OmpK36) and pJIQQ-2 (pACYC184-OmpK36GD) plasmids (S1 Table), respectively. Two K. pneumoniae strains, 10.85ΔOmpK35ΔOmpK36 and JIE2771 (a clinical strain with naturally-occurring lesions in ompK35 and ompK36), were grown overnight in LB broth. On the next day, the strains were centrifuged and washed three times with ice-cold 10% glycerol. pACYC184, pJIQQ-1 and pJIQQ-2 were electroporated into the electrocompetent cells to generate the strains shown in Table 1. Sanger sequencing was performed to verify the absence of unintended non-synonymous mutations in the coding regions of ompK36 and ompK36GD. . All MICs were determined in triplicate at least on three separate occasions to obtain at least 9 discrete data points and compared For the in trans complemented strains, the susceptibilities of plasmid-bearing mutants of 10.85ΔOmpK35ΔOmpK36 to CEF, CFZ and FOX, as well as the susceptibilities of the plasmid-bearing mutants of JIE2771 to ETP, IPM and MEM, were performed in cation-adjusted Mueller-Hinton broth with chloramphenicol at a concentration of 25 μg/ml. Subsequent procedures follow those used for all other bacterial strains in this study.

Real-time reverse transcription-PCR
The expression levels of the different porins were measured by real-time RT-PCR. Cells were harvested in logarithmic phase at an OD 600 of 0.5-0.6. Total RNA was isolated using RNeasy system (Qiagen). RNA was treated with DNase (TURBO DNA-free Kit, Ambion). cDNA was synthesized by high-capacity cDNA reverse transcriptase kit (Applied Biosystems). One microgram of the initially isolated RNA was used in each reverse transcription reaction. cDNA was diluted 1:10 and 2 μl were used for the real-time reaction. Three biological replicates, each with three technical replicates, were used in each of the assays. The relative levels of expression were calculated using the threshold cycle (2 −ΔΔCT ) method [41]. The expression of rpoD was used to normalize the results. The primers used are listed in S1 Table.

Determination of growth rate
Growth rates were determined as previously described [42]. Overnight broth cultures were diluted 1:1000. Six aliquots of 200 μl per dilution were transferred into 96-well microtiter plates (Corning Incorporated, Durham, NC, USA). Samples were incubated at 37˚C and shaken before measurement of OD 600 in a Vmax Kinetic microplate reader (Molecular Devices, Sunnyvale, CA, USA). Growth rates and generation times were calculated on OD 600 values between 0.02-0.09. The relative growth rate was calculated by dividing the generation time of each mutant by the generation time of the parental strain (K. pneumoniae ATCC 13883, 10.85 or 11.76), which was included in every experiment. Experiments were performed in six technical replicates in three independent cultures on three different occasions. Results are expressed as means ± standard errors of the means.
For complemented strains, K. pneumoniae plasmid-bearing mutants of 10.85 ΔOmp-K35ΔOmpK36 and JIE 2771 were streaked on Mueller-Hinton agar with 25 μg/ml chloramphenicol and incubated overnight at 37˚C. On the next day, bacterial cells were resuspended in 0.85% saline to a turbidity of 0.5 McFarland. The inoculum was diluted 1:400 in cationadjusted Mueller-Hinton broth, which were then transferred in six aliquots of 200 μl into 96-well microtiter plates (Corning Incorporated, Durham, NC, USA) and incubated with continuous gentle orbital shaking at 37˚C in a SpectraMax iD5 Hybrid Multi-Mode Microplate Reader (Molecular Devices, San Jose, CA, USA). Measurements of OD 600 were obtained every 5 minutes. Subsequent procedures follow those used for all other bacterial strains in this study.

In vitro competition experiments
Competition experiments were carried out as described previously [43]. Viable cell counts were obtained by plating every 24 h on antibiotic-free LB agar and on LB agar supplemented with antibiotic (kanamycin 20 μg/ml or tetracycline 10 μg/ml) to distinguish between mutants and wild-type cells. PCR (with primer pair K36GD4 / K36GD11 or K36GD12 / K36GD13 primers, S1 Table) was performed for the calculation of the competition results between the parental strain and OmpK36GD mutant (in this particular experiments, bacteria were diluted in fresh media every 24 h and PCR on 100 viable colonies of each replicate was performed every 48 h). All experiments were carried out in triplicate with three independent cultures. Mean values of three independent experiments ± standard deviation were plotted.

Mouse model of gastrointestinal tract colonization (GI) and competition experiments
Five to six week-old female BALB/c mice (Animal Resources Centre (ARC), Sydney, Australia) were used for GI colonization [44,45,46] and competition experiments. Mice were caged in groups of three and had unrestricted access to food and drinking water. Faecal samples were collected and screened for the presence of indigenous K. pneumoniae before inoculation. For the colonization study, three mice were inoculated with the parental strain or a porin mutant (1 x 10 10 CFU / mouse), suspended in 20% (w/v) sucrose. For individual colonization, ampicillin was added to drinking water on day 4 (0.5 g / L) after an inoculation [47]. For the competition experiment, equal volumes of the parental strain and each mutant or equal volume of different mutants (1x 10 10 CFU / mouse) were mixed and suspended in 20% (w/v) sucrose. Colonization was maintained with ampicillin 0.5 g / L throughout the experiment [48, 49,50]. Faeces samples were collected every second day, emulsified in 0.9% NaCl and appropriate serial dilutions plated on MacConkey-inositol-carbenicillin agar, which selectively recovers K. pneumoniae [51]. Animal experiments were approved by the Western Sydney Local Health District Animal Ethics Committee (AEC Protocol no. 4205.06.13).

Mouse model of virulence: Intranasal infection
Five-six week-old female BALB/c mice [Animal Resources Centre (ARC), Sydney, Australia] used in the inhalation (pneumonia) model [52,53,54] were exposed to ATCC 13883 and 10.85 and their isogenic ΔOmpK35OmpK36GD mutants. Overnight bacterial cultures were harvested, washed and resuspended at 10 9 CFU in 20 μl of saline and inoculated into the nasal passages. A control group of mice was inoculated with saline. Following infection, survival studies were performed (10 mice per strain). At the same time and using the same inoculum, organ (lung and spleen) and blood infection burdens were also assessed at various points throughout the infection period, by plating out blood and homogenised tissue onto LB agar, and counting CFU (5 mice per strain, per time point). Animal experiments were approved by the Western Sydney Local Health District Animal Ethics Committee (AEC Protocol no. 4275.06.17).

Structural modelling of OmpK36 variants
Tri-dimensional structural models of ATCC 13883 OmpK36 and its mutated variant OmpK36GD were computed with ProMod3 Version 1.1.0 on the SWISS-MODEL online server [55] using the target-template alignment method. The best scoring model used as a template was 5nupA (93.84% sequence identity, with a QMEAN equal to -2.29 and -2.14, respectively for both sequences). For comparison purposes, models were also computed using the second best OmpK36 structure available in PDB (1osmA). All predicted models were evaluated using MolProbity [56,57] and Verify3D [58,59], with Ramachandran plots generated by MolProbity indicating for all computed models that at least >98% of residues were in allowed regions. Predicted structures were displayed by PyMol software (version 2.1.1) [60].
Additionally, the specific impact of two amino-acids insertions was also investigated by altering the OmpK36 structure under PDB accession 5nupA, adding either the amino-acids GD-, TD-or SD-, after position G113 and modeling the resulting variant sequences in the same manner as mentioned above.

Genome sequencing and comparative analysis
All isolates used in final experiments were subjected to whole genome sequencing to verify their altered sequence and ensure that no additional mutations had arisen. Genomic DNA was extracted from 2 ml overnight cultures using the DNeasy Blood and Tissue kit (Qiagen). Paired-end multiplex libraries were prepared using the Illumina Nextera kit in accordance with the manufacturer's instructions. Whole genome sequencing was performed on Illumina NextSeq 500 (150bp paired-end) at the Australian Genome Research Facility (AGRF) and at Professor Vitali Sintchenko's laboratory (Translational Public Health Bacterial Genomics Group, Centre for Infectious Diseases and Microbiology (CIDM) Public Health, Westmead Hospital, NSW, Australia). Raw sequence reads are available on NCBI under Bioproject accession number PRJNA430457. Reads were quality-checked, trimmed and assembled using the Nullarbor pipeline v.1.20 (available at: https://github.com/tseemann/nullarbor), as previously described [61], but with the exception of the assembly step which was performed using Shovill (available at: https://github.com/tseemann/shovill), a genome assembler pipeline wrapped around SPAdes v.3.9.0 [62] which includes post-assembly correction. Assemblies were also reordered against reference strain K. pneumoniae 30660/NJST258_1 (accession number CP006923) using progressive Mauve v.2.4.0 [63] prior to annotation with Prokka [64] and screened for antibiotic resistance genes using Abricate v.0.6 (available at: https://github.com/ tseemann/abricate).

Population analysis
To investigate the significance of OmpK35 and OmpK36 mutations in a wider population, we collected a total of 1,557 draft and complete K. pneumoniae genomes publicly available in Genbank (Feb 2017, S2 Table). Sequences were typed using Kleborate v0.1.0 [65] to identify MLST (S3 Table) and minimum spanning trees were generated using Bionumerics v.7.60. Presence and absence of porins were assessed in the pangenome using Roary v3.6.0 [66] with default parameters, and mutations in loop 3 (L3) identified using BLAST. The 2,253,033 bp core genes alignment predicted by Roary was used to build a maximum-likelihood tree using IQ-TREE v1.6.1 [67], with a GTR+G+I nucleotide substitution model and branch supports assessed with ultrafast bootstrap approximation (1,000 replicates). Trees were visualized alongside contextual information with Phandango [68].
Statistical analysis was performed using Chi-squared test and Wilcoxon test, to determine associations between ST, porin defects and antibiotic resistance genes. Extended mosaic plots were used to assess the distribution of OmpK35 and OmpK36 with or without GD/TD insertion across i) ST, ii) country of origin and iii) year of isolation. Extended mosaic plots offer a convenient way to visualize the relative frequencies of a set of categorical data using proportional areas, as well as the fit of a log-linear model (assuming independence). Areas are thus colored according to the direction and magnitude of standardized deviation from the expected frequency (Pearson residual). Cut-offs of +/-2 and 4 are defined heuristically on the assumption that the Pearson residuals approximate a standard normal distribution, and can be approximated to the statistical significance alpha = 0.05 and and alpha = 0.001 levels, respectively [69]. All statistical analyses were performed in R version 3.5.1 and "vcd" package. Relevant R scripts were also made available at https://github.com/nbenzakour/Klebsiella_ antibiotics_paper.

Statistical analysis
For doubling time and Real Time RT-PCR, the results were analysed using the Student t test to determine their significance. For survival studies the results were analysed using Long-rank (Mantel-Cox) test and Gehan-Breslow-Wilcoxon test. To compare bacterial load in organs during lung infection, results were compared using Mann-Whitney unpaired t test. The analyses were performed using Prism7 (GraphPad Software).

Outer membrane porins and resistance to beta-lactam antibiotics
Minimal inhibitory concentrations for commonly used carbapenems (ertapenem and meropenem), third-generation cephalosporins (ceftazidime, cefotaxime and ceftriaxone), cephamycins (also called 'second generation cephalosporins', cefoxitin and cefuroxime), first generation cephalosporins (cephalothin and cefazolin), and the semi-synthetic penicillin ampicillin were determined in three K. pneumoniae strains and their isogenic porin mutants, with representative results in Table 2 (for complete results, see S4 Table). The SHV enzyme  Table 2 and S4 Table) is associated with a minor increase in MIC for carbapenems and cephalosporins (Table 2 and S4 Table), with a lesser impact from OmpK36GD mutations, consistent with an important role for OmpK36 in the nutritious growth media (MHB) normally used for standardised MIC determinations (Table 2 and S4  Table).
OmpK35 loss (ΔK35 in Table 2 and S4 Table) has little impact alone but further increases MICs for most antibiotics in the presence of OmpK36 lesions (e.g. ΔK35ΔK36 and ΔK35ΔK36GD). In addition to ertapenem non-susceptibility, ΔOmpK35ΔOmpK36 and ΔOmpK35OmpK36GD strains are clinically resistant to first (e.g. cephalothin, CEF) and second generation cephalosporins/ cephamycins (e.g. cefoxitin, FOX) ( Table 2 and S4 Table).
Naturally occurring plasmids from other K. pneumoniae strains encoding a common ESBL (bla CTX-M-15 ) [33], a metallo-carbapenemase (bla IMP-4 ) [34] and a serine-carbapenemase (bla KPC-2 ) [22] were transferred into ATCC 13883 and its isogenic mutants by conjugation, with transfer verified by PCR (S1 Table) and S1/PFGE (S1 Fig). Even the common ESBL CTX-M-15 confers reduced susceptibility to ETP in the presence of an OmpK36 deletion or inner channel mutation (GD duplication), especially if accompanied by an OmpK35 defect ( Table 3). Expression of the specialised carbapenemases IMP and KPC from their naturally occurring plasmids resulted in greatly increased carbapenem MICs (Table 3), with the double porin mutants being highly resistant to all carbapenems tested.
K pneumoniae JIE2771 is a clinical isolate of K pneumoniae carrying bla KPC and a natural double mutant of ompK35 and ompK36 [22]. As expected, attenuation of the resistance phenotype was evident in this wild-type double mutant and susceptibility restored to the constructed 10.85 double mutant by in trans complementation with ompK36 but not ompK36GD (S5 Table).

Altered expression of other common porins in ΔOmpK35, ΔOmpK36, and OmpK36GD
Other porins may compensate for the loss of major outer membrane porins in K. pneumoniae [76,77,78,79]. Expression of ompK35, ompK36, ompK37, phoE, ompK26 and lamB was measured in isogenic porin mutants of ATCC 13883 and 10.85 K. pneumoniae strains (Fig 1,  Host adaptation and convergent evolution in a major human pathogen S6 Table). Neither the introduction of a GD duplication into the OmpK36 inner channel (OmpK36GD) nor the loss of OmpK35 (ΔOmpK35 and ΔOmpK35OmpK36GD) affected expression of OmpK36 in MH broth. Loss of OmpK36, however, was associated with increased OmpK35 expression in MH broth, in which OmpK36, but not OmpK35, is ordinarily expressed (S2 Fig). Restitution of OmpK36 by replacing the interrupted gene with ompK36GD directly in the chromosome restored normal porin regulation (Fig 1, S6 Table). Loss of both of these major porins (ΔOmpK35ΔOmpK36) resulted in increased expression of phoE and lamB. (Fig 1, S6 Table).

Relative fitness costs of major porin lesions
Exponential phase growth in MH broth was only affected when both major porins were absent (ΔOmpK35ΔOmpK36, S7A Table). In trans complementation with either ompK36 or ompK36GD resulted in amelioration of the growth defect in JIE2771 wild-type double mutant and the constructed 10.85ΔOmpK35ΔOmpK36 (S7B Table). The ability of ΔOmpK35 strains to directly compete against their intact isogenic parents in MH broth was little affected over seven-day growth (Fig 2A1 and 2A2 and S3 Fig). However, any ΔOmpK35 mutant was rapidly outcompeted by its isogenic parent in nutrient-limited conditions (S4A1 and S4B1 Fig). Furthermore, competition experiments clearly illustrate the importance of OmpK36 in high osmolarity highly nutritious media (Fig 2B1 and 2B2 and S3  Fig) but not in low nutrient conditions (S4B1 and S4B2 Fig). OmpK36GD strains are clearly much more able than ΔOmpK36 strains to compete with their isogenic parent strains (Fig 2C1  vs 2B1 and 2C2 vs 2B2). For ATCC 13883, at day 3, the OmpK36GD population was still 40% of the total combined population (Fig 2C1), while ΔOmpK36 fell to 20% in the same period ( Fig 2B1). This difference was more marked in the presence of an OmpK35 lesion but ΔOmp-K35OmpK36GD populations were still clearly more able than ΔOmpK35ΔOmpK36 to compete with the intact parent strain (Fig 2F1 vs 2E1). In fact, the introduction of an OmpK36GD mutation had no detectable cost at all in K. pneumoniae 10.85 (Fig 2C2 vs 2B2 and 2F2 vs 2E2), with ΔOmpK35OmpK36GD competing very successfully against the isogenic parent 10.85 (Fig 2F2: 37±4% and 26±15% of the total population represented by ΔOmpK35OmpK36GD on days 6 and 7 respectively). Finally, as expected, directly competing OmpK36GD with ΔOmpK36 (and ΔOmpK35OmpK36GD with ΔOmpK35ΔOmpK36) further illustrates the competitive advantage, with OmpK36GD strains quickly displacing isogenic ΔOmpK36 strains in MH broth (Fig 2D1 and 2G1).  Mouse gut colonizing studies yielded similar results (Fig 3). Mice were confirmed not to harbor indigenous K. pneumoniae on arrival [51], and stable colonisation at *10 9 CFU/g faeces was achieved (S5 Fig). OmpK35 deficient mutants (ΔOmpK35) were not disadvantaged ( Fig 3A3) and OmpK36GD strains strongly outperformed OmpK36 strains in competition with their isogenic parents (Fig 3A1 vs 3B1 and 3A2 vs 3B2). Similarly, direct in vivo competition confirmed a clear fitness advantage of OmpK36GD over ΔOmpK36 (Fig 3C1 and 3C2).

Pathogenicity is not attenuated in ΔOmpK35OmpK36GD strains
In a mouse pneumonia model [52,53,54], we showed no difference in lethality between a wild type strain and its isogenic mutant ΔOmpK35/OmpK36GD (Fig 4). Intranasal inoculation of mice with 10.85 and ATCC 13883 ΔOmpK35/OmpK36GD strains showed that these mutations had no significant impact on virulence, with equivalent mortality curves (Fig 4A1 and  4A2) and similar viable counts developing in lung, blood and spleen over the course of infection compared with their isogenic wild-type strains 10.85 and ATCC 13883, respectively ( Fig  4B to 4D). As OmpK36 deletion mutants have been clearly shown by other studies to be attenuated in vivo [80,81], and we also demonstrate this completely predictable virulence cost by in vitro and in vivo competition assays, experiments in the acute pneumonia model were confined to these two isolates and their key isogenic ompK36 variants in order to minimize the use of animals in experimentation.

Structural impact of OmpK36 loop L3 mutations
Two crystal structures of native OmpK36 available in the Protein Data Bank under accession number 5nup (2.9 Å, Xray) and 1osm (3.2 Å, Xray) were evaluated as templates for structural modelling of OmpK36 and OmpK36GD from ATCC 13883, with targets and templates sharing around 93% nucleotide sequence identity. While Ramachandran plots analysis for all predicted models show at least 98% of residues in allowed regions, other metrics such as QMEAN and Molprobity score were marginally better for ATCC 13883 OmpK36 and OmpK36GD models based on the 5nup structure (Fig 5, S8 Table). Although several differences can be observed in the final alignment (Fig 5D), the most prominent differences between the original structure ( Fig 5A) and the ATCC 13883 OmpK36 model lie within the loop L6, which can be seen in yellow, slightly obstructing the outmost channel of the porin (Fig 5B). Much more striking is the impact of single two amino-acid -GD insertion within loop L3, which is expected to further constrict the porin channel ( Fig 5C) and is likely responsible for the difference in phenotype between the two variants.

OmpK35 loss and convergent evolution of OmpK36GD
The successful antibiotic resistance, colonisation and pathogenicity phenotypes of ΔOmp-K35OmpK36GD strains should be reflected in their representation among strains causing human infection. Of 165 unique K. pneumoniae ompK36 sequences in GenBank, 16% varied from the consensus L3 inner channel motif (PEFGGD). The most common was the GD  [82,83], including of the pore eyelet region [84]). Using the native OmpK36 structure 5nup as a template, modelled structures of mutants, namely GD-, TD-and SD-insertions in L3, were computed as previously described, and showed similar restriction of the porin channel, slightly greater in the case of a bulkier amino-acid such as Threonine (  Host adaptation and convergent evolution in a major human pathogen To investigate OmpK36 among clinical isolates without specialised carbapenemases, we specifically analysed L3 variation in all such K. pneumoniae isolates with an Ertapenem MIC > 1 in our local clinical collection (Table 1) by PCR and sequencing (S1 Table). Of (n = 51), 17 strains (33%) were identified: all revealed either the previously described GD or TD mutation in the L3 loop of ompK36 on sequencing and these encoded up to 6 distinct betalactamases. These isolates were genetically diverse but belonged to major epidemic clones found elsewhere in the world: i.e. ST14, ST16, ST101, ST147, with as many as 6 distinct ompK35 mutations, all of which introduced disrupting frame-shifts and all of which were relatively lineage-specific (S10 Fig). Finally, all K. pneumoniae (complete and draft) genomes available from Genbank, i.e. 1,557 entries (as of February 2017) were examined: the two common (GD and TD) variants are shown in a minimum spanning tree built using MLST profiles (Fig 6) to be distributed across the whole spectrum of diversity of K. pneumoniae, including in most major epidemic clones, e.g. ST258 and its derivative ST512, ST11, ST101, ST147, ST14 and ST37.
A maximum likelihood phylogeny using a 2,253,033 bp core genome alignment of all 1,557 genomes was computed to contextualize variations in ompK36 and ompK35, with metadata relative to the population (year, source, geographical region of isolation, as well as major betalactamases genes) (Fig 7). Those genes most relevant to a carbapenem resistance phenotype are shown, and the expected clustering of some of these is as expected (e.g. bla CTX-M-15 with OXA-1 and TEM-1b ). Major associations with other genes not affected by porin changes are not shown (e.g. aminoglycoside resistance due to 16S methylase genes that are common companions of bla NDM , other class I integron cassettes from the array in which bla IMP-4 is found, etc). The predominance of ompK36 variations in L3 compared to its loss or disruption is evident at a glance, as is the common loss or disruption of ompK35 in unrelated strains. There is Host adaptation and convergent evolution in a major human pathogen Fig 7. Maximum likelihood tree of 1,557 K. pneumoniae strains. A phylogenetic tree was built using a 2,253,033 bp long core alignment. Contextual information relevant to the collection was visualized using Phandango and includes ST (of which the major ones are indicated on the tree); GD or TD insertion in the loop L3 of ompK36, in black and red, respectively; presence or absence of ompK36, in orange and purple, respectively; presence or absence of ompK35, in orange and purple, respectively. Additional metadata include year(date) of isolation, in a gradient from purple to yellow; source and geographical region of isolation in a rainbow gradient; and presence of major beta-lactamases (bla) alleles identified, in dark blue. https://doi.org/10.1371/journal.ppat.1007218.g007 Host adaptation and convergent evolution in a major human pathogen PLOS Pathogens | https://doi.org/10.1371/journal.ppat.1007218 March 15, 2019 no obvious relationship between ompK36 L3 variations and the presence of bla KPC but there is strong clustering of these variations in certain types (ST258, 512 etc).
As expected for a gene so clearly linked to fitness and virulence, ompK36 is highly conserved across the dataset (present in 1,499 out of 1,577), and we found no statistical evidence of STdependence (Chisq = 207.51, df = 227, p-value = 0.8188). Conversely, ompK35 (evidently dispensable in the host) is disrupted in nearly a third of all strains (Fig 6), with statistically significant association with ST (Chisq = 603.7, df = 227, p-value = 5.748e-36). Three-way comparison of the distributions of frequencies of presence/absence of ompK35, mutations in ompK36, and ST (considering only those STs harbouring ompK36 GD/TD variants) was performed using an extended mosaic plot (S11 Fig). Standard Pearson's residuals were calculated and displayed on the mosaic plot to identify over-represented categories (residuals [2,4] and >4) and under-represented categories (residuals [-2,-4] and <-4). We found statistically significant evidence (residual cut-off 2 and 4 equivalent to p < 0.05 and p < 0.001) that i) some STs prevalently have both ompK36 and ompK35 intact (mainly ST15, ST16, ST17); ii) others prevalently have intact ompK36 with ompK35 disrupted (ST129, ST258); and iii) some STs prevalently have ompK36 (GD/TD) variants combined with ompK35 disrupted (ST11, ST14, ST147, ST258 and ST37). Furthermore, we found statistically significant evidence for more disruptions of ompK35 in i) strains from the USA compared to other countries (S12A Fig) and ii) strains from 2011 and 2014 (S13A Fig). We also observed over-representation of ompK36 (GD/TD) variants in i) China, Greece, Germany, Italy and India (S12B Fig) and ii) in 2011 (S13B Fig). Finally, we looked at associations between the number of resistance genes and porin defects in major STs, and found that the presence of ompK36 GD/TD variants did not correlate with a higher number of resistance genes (with the exception of OmpK36GD in ST14). In fact, successful clones such as ST258 and ST11 harbouring OmpK36GD encoded significantly less resistance genes (p<0.001, Wilcoxon test) (S14 Fig). It should be noted that due to the inherent opportunistic nature of the sampling present in Genbank (e.g. USA), our conclusions are only applicable to this dataset. More sampling would be required to assess the significance of porin mutations in an unbiased K. pneumoniae population.

Discussion
β-lactam antibiotics are among the most commonly prescribed for severe infections [85,86] and the emergence of β-lactam resistance in K. pneumoniae has become a global health threat [87,88]. In general, E. coli and K. pneumoniae carrying transmissible β-lactam resistance genes have predictable and normally distributed β-lactam MICs [21] but carbapenem MICs in K. pneumoniae are bimodally distributed with higher MICs correlating with OmpK36 defects [21]. OmpK36 loss or mutation is not uncommonly reported in highly resistant clinical isolates producing KPC, ESBL and AmpC β-lactamases [20,89,90].
Diffusion of β-lactam antibiotics through non-specific porins such as OmpK35 and OmpK36 is dependent on size, charge and hydrophobicity [91,92], with bulky negatively charged compounds diffusing at a lower rate than small zwitterions of the same molecular weight [93]. OmpK35 is much less expressed in high osmolarity nutrient-rich conditions than OmpK36, which has the narrower porin channel of the two (S2 Fig) [9] and large negatively charged β-lactams such as third-generation cephalosporins and carbapenems diffuse more efficiently through OmpK35 than OmpK36 [80,94]. Here we confirm the significantly increased MICs, commonly attributed to mutations in these two major porins [10,95,96] in three K. pneumoniae strains (the widely-published ATCC strain 13883 and two locally isolated clinical strains (Table 2 and S4 Table) and unequivocally identify the primary role of OmpK36 in carbapenem resistance.
Comparable MIC changes in single (OmpK36GD and ΔOmpK36) and double (ΔOmp-K35OmpK36GD and ΔOmpK35ΔOmpK36) mutants indicate that duplication of a glycine aspartate (GD) pair in a critical position in the porin eyelet region (loop 3) is almost as effective as a complete deletion of the porin in excluding large anionic antibiotics. Both single and double porin mutants were susceptible to extended-spectrum cephalosporins (cefotaxime and ceftazidime) in the absence of acquired hydrolysing enzymes, demonstrating the impotence of the naturally occurring chromosomal SHV enzymes [70,71,72] against these compounds [95].
Differences relating to porin permeability in K pneumoniae are most striking and important in the presence of acquired carbapenemases and it is clear that these permeability changes greatly enhance the associated resistance phenotypes. The common Ambler Class A serine protease KPC-2 and Class B metalloenzyme IMP-4 expressed from their natural plasmids produce only borderline resistance against meropenem and the smaller zwitterionic imipenem in the presence of the 'wild type' OmpK36 osmoporin ( Table 3) but MICs that exceed therapeutic tissue levels [97,98] are the rule in strains of the commonly occurring ΔOmpK35OmpK36GD genotype.
We also show here that the OmpK35 matrix porin has little or no relevance in vivo or in vitro conditions that reliably predict antibiotic efficacy in the clinic (MICs and competitive fitness in Mueller-Hinton broth). Consistent with this, a high percentage of clinical isolates whose genomes have been lodged with GenBank appear to have lost their ability to express OmpK35 altogether (Fig 7). Increased production of the larger channel OmpK35 is expected under low-temperature, low-osmolarity and low nutrient conditions (S2 Fig). These favour survival outside the mammalian host and we show that ΔOmpK35 strains fail to compete successfully with their isogenic parents in nutrient-limited conditions (S4 Fig). We confirm that OmpK35 is not naturally expressed at significant levels in optimal growth conditions nor in the mammalian host, as previously described [78,80]. As expected, competition experiments, the most sensitive and direct measures of comparative fitness, evince no discernible disadvantage from the loss of OmpK35 in vivo [19,99].
Loss of OmpK36 trades off nutrient influx for antibiotic resistance [42,80], and we show that these more resistant bacteria cannot compete successfully with the antibiotic-susceptible populations from which they arise once antibiotic selection ceases to operate (Fig 2). Double porin mutants (ΔOmpK35ΔOmpK36) are the most antibiotic-resistant (Table 2 and S4 Table) but this resistance comes at the cost of a 10% relative growth reduction in nutritious media (S7 Table). OmpK36, the main porin normally expressed in vivo, is responsible for most of this fitness cost (Figs 2 and 3 and S3 Fig). The less permeable phosphoporin PhoE and maltodextrin channel LamB, most important in the usual compensatory response when OmpK35 is not available, are not efficient substitutes (Fig 1 and S6 Table). Defects in these porins have been implicated in carbapenem resistance in association with only an AmpC-type enzyme [42,76,79,100], but other defects are ill-defined and the fitness cost may be high as such strains are rarely described. By contrast, ΔOmpK35OmpK36GD mutants are little disadvantaged in vivo or in optimal growth conditions in vitro (Figs 2 and 3 and S7 Table). Expression of OmpK36 is unaffected (Fig 1 and S6 Table) as is that of other porins such as OmpK35 (Fig 1  and S6 Table), presumably because OmpK36 'rescue' is not required.
The precise loop 3 variation in OmpK36 is best explained by a convergent evolutionary process, as a range of different variants occur within genetically distant K. pneumoniae populations, all with an extra negatively charged aspartate (D) residue that significantly constricts the inner channel (Fig 5). The most common solution is the extra glycine and aspartate (PEFGGD to PEFGGDGD in the critical region) which we recreated in isogenic mutants for our experiments. The next most frequent, an extra TD (rather than GD), is similarly likely to spontaneously arise (S8 Fig) but is much less common, including in STs in which both GD and TD are found (Figs 6 and 7), implying a less optimal conformation. A recent survey of nearly 500 ertapenem-resistant Klebsiellae lacking specialised carbapenemases [101] supports our own finding of the extra aspartate in that position, most commonly as a GD pair, with TD and SD much less often, and other variants being rare. We found no examples of similarly acidic (glutamate) residues occurring in this position, perhaps reflecting the fact that even simple sequence changes (here, GAY to GAR) add an additional step to a simple duplication event, or the fact that glutamate's extra carbon makes it slightly less compact than an aspartate in this position.
Other Enterobacteria face the same challenge of excluding bulky anionic carbapenem antibiotics in order to survive high concentrations, even in the presence of a specialist carbapenemase. High level antimicrobial resistance has been ascribed to similar variations in L3 of OmpK36 homologues in Enterobacter aerogenes, Escherichia coli (S9 Fig) and Neisseria gonorrhoeae [102,103,104,105,106]. In comparison with their E. coli homologues (OmpF and OmpC), OmpK35 and OmpK36 permit greater diffusion of β-lactams [107]. Specifically, OmpK35 appears to be highly permeable to third-generation cephalosporins such as cefotaxime due to its particular L3 domain, which is also seen in Omp35 in E. aerogenes but not in other species, and has been proposed as an explanation for the high proportion of K. pneumoniae clinical isolates that lack this porin [84,107]. Our findings of increased MICs in OmpK35 mutants are consistent with those of others [107] but we show here that the more permeable OmpK35 is not important in the mammalian host. Rather, the much less permeable OmpK36 (equivalent to E coli OmpC) [107] is the bottleneck for large anionic antibiotics.
The term 'high risk clone' [108,109] is given to host-adapted/pathogenic strains that dominate the epidemiology of (antibiotic resistant) infections, presumably because they are more transmissible, more pathogenic and/or more tolerant of host-associated stresses (including antibiotics). Here, we see a range of unrelated clonal groups already identifiable as high-risk clones that are dispensing with the OmpK35 porin (Fig 7). The minimal antibiotic resistance advantage in nutritious media is only evident with carbapenems and is unlikely to arise in the presence of an existing OmpK36 loss mutation because the fitness cost is substantial. The loss of OmpK35 through low-level carbapenem exposure in environmental conditions is possible [110] but has a marked fitness cost and the exposure to carbapenems in the environment is expected to be limited, as they are a still a minority class of prescribed antibiotics and are not yet as common in environmental waters as the sulfonamides, quinolones, macrolides, tetracyclines and other beta-lactams [111].
A recent review of antibiotic resistance in Klebsiella pointed out that "The exact role of porins in antimicrobial resistance is difficult to determine because other mechanisms. . .are commonly present . . ." [112]. We suggest that host-adaptation in K. pneumoniae is widespread and that many K. pneumoniae have dispensed with the OmpK35 matrix porin required for an environmental life cycle. Bacteria are expected to adapt effectively to major stress such as antibiotic pressure or high concentrations of bile salts in the intestinal lumen [113]. Our hypothesis of adaptive loss of OmpK35 is based on results presented in this study and on strong evidence from others: i) toxic agents as antibiotics and bile salts diffuse better through the larger OmpF channel (homolog of OmpK35) than the narrower OmpC (equivalent to OmpK36 in K. pneumoniae) [114]; ii) high osmolarity, high temperature, low pH and anaerobiosis (typical conditions in gut environment) induce the production of OmpK36 but inhibit the expression of ompK35 [115] [116] [117] and iii) E. coli mutants with reduced permeability (decreased ompF and increased ompC mRNA and protein levels compared with parental strain) can be easily recovered from intestinal gut of germ-free mice after few days of colonization [118]. In addition, we have shown that the highly specific variation in the inner channel of OmpK36 provides carbapenem resistance at no cost to colonising ability, competitiveness or pathogenicity and can be expected to be an increasingly common feature of host-adapted 'high-risk clones'.
There are three direct and immediate implications. Firstly, efforts to control the spread of such strains will be facilitated to some extent by the loss of environmental hardiness resulting from OmpK35 deletion, and should shift slightly more toward managing interpersonal transmission. Secondly, K. pneumoniae can be expected to become more antibiotic resistant overall, and organisms expressing currently circulating plasmid-borne carbapenemases will more commonly be untreatable with carbapenem antibiotics (e.g. ST258 strains with bla KPC ); the second (higher MIC) peak in the bimodal distribution of carbapenem MICs in K. pneumoniae populations will become more prominent. Finally, the mobile carbapenemase gene pool can be expected to flourish in the protected niche provided by host-adapted K. pneumoniae populations under strong carbapenem selection pressure in human hosts, thereby increasing the general availability of highly transmissible carbapenem resistance plasmids among hostadapted pathogens in the Enterobacteriaceae.   [82,83]. Red boxes, residues involved in the pore eyelet based on [84]. Black box, L3 variants. (TIF) S7 Fig. Distinct channel restrictions of OmpK36 two amino-acids mutants (-GD, -TD,  and -SD). Comparison of the reference OmpK36 structure under PDB accession 5nup1A (WT, wild type) against predicted structural models of mutants harbouring a two amino-acid insertion in loop 3 after G113, namely GGDGD, GGDTD and GGDSD. For each predicted structure, the 2 most protruding amino-acids resulting from the insertion were marked and coloured according to their backbone structure (carbons in yellow, oxygens in red and nitrogens in blue).  Table); ST: sequence type; number of predicted resistance genes encoded; carbapenamase gene encoded; ESBL: extended-spectrum beta-lactamase gene encoded. (TIF)

S11 Fig. Extended mosaic plot of the observed proportions of isolates with porins
OmpK35 and OmpK36 variations, across STs harbouring ompK36 GD/TD variants. The mosaic plot shows the relationships between 3 variables; ST (in purple) and presence/absence of ompK35 (in black) on the x-axis; and presence/absence and mutations of ompK36 (in grey) on the y-axis. The size of each plot tile is proportional to counts. Plot tiles are colored according to their standardized Pearson residuals, as determined by a log-linear model. Deeper shades of red and blue corresponding to a standardized residual less than -4 or greater than +4, respectively, can be interpreted as combinations observed significantly less or more than expected (under the assumptions that proportions have equal levels). (TIF) S12 Fig. Extended mosaic plot of the observed proportions of isolates with porins OmpK35 and OmpK36 variations versus countries. The mosaic plots show the relationships between 2 variables; A) country of isolation on the x-axis and presence/absence of ompK35 on the y-axis; B) country of isolation on the x-axis, and presence/absence and mutations of ompK36 on the y-axis. The size of each plot tile is proportional to counts. Plot tiles are colored according to their standardized Pearson residuals, as determined by a log-linear model. Deeper shades of red and blue corresponding to a standardized residual less than -4 or greater than +4, respectively, can be interpreted as combinations observed significantly less or more than expected (under the assumptions that proportions have equal levels). (TIF) S13 Fig. Extended mosaic plot of the observed proportions of isolates with porins OmpK35 and OmpK36 variations versus years. The mosaic plots show the relationships between 2 variables; A) year of isolation on the x-axis and presence/absence of ompK35 on the y-axis; B) year of isolation on the x-axis, and presence/absence and mutations of ompK36 on the y-axis. The size of each plot tile is proportional to counts. Plot tiles are colored according to their standardized Pearson residuals, as determined by a log-linear model. Deeper shades of red and blue corresponding to a standardized residual less than -4 or greater than +4, respectively, can be interpreted as combinations observed significantly less or more than expected (under the assumptions that proportions have equal levels). (TIF)

S14 Fig. Distribution of resistance genes identified in ST harbouring OmpK36GD or
OmpK36TD mutants. Boxplots were used to display the distribution of resistance genes identified with Abricate within each ST with the following OmpK36 variants, namely isolates with-GD in bright red,-TD in brown, or no insertion (-) in grey. Mean comparison p-values are also shown for each ST (Wilcoxon test, with '-' used as a reference group; ns: p > 0.05; � : p < = 0.05; �� : p < = 0.01; ��� : p < = 0.001; ���� : p < = 0.0001). In addition, the corresponding underlying isolate population is also visualised as individual points, coloured according to OmpK35 type, (1) intact in turquoise or (0) disrupted in coral. (TIF) S1