Transient and persistent population dynamics of Haemophilus influenzae in healthy and om-prone pediatric cohorts

To examine the colonization and infection dynamics of H. influenzae within and among human subjects over time and to identify specific H. influenzae accessory genes associated with OM disease risk, we sequenced and analyzed the genomes of clinical isolates collected in Rochester, NY from the upper respiratory tracts of children during the first 3 years of life. Strains in this ongoing long-term large-scale biorepository are prospectively collected: multiple specimens from the nasal/nasopharyngeal tract (NP) are collected during 6 healthy pediatric visits between 6 and 36 months of age. In addition, during any acute OM episodes experienced by enrolled subjects, specimens are collected from both the NP and directly from middle-ear fluid (MEF) obtained by tympanocentesis.

We selected longitudinal series of isolates from the biorepository collected between 2008 and 2016 from 9 healthy and 12 OM-prone subjects. Selected series were chosen based on H. influenzae being isolated at ≥ 3 healthy visits for all subjects and at ≥ 4 sick visits for OM-prone subjects. We obtained 206 high-quality whole-genome shotgun assemblies from these isolates; all were sequenced using short-read Illumina technology, and a subset were also sequenced using long-read Pacific Biosciences technology. We next annotated their genes and clustered protein-coding genes based on homology to determine a matrix of gene presence/absence across isolates (Fig 1A; subject and genome information in S1 Table and S2 Table, respectively; the column dendrogram uses the phylogenetic tree shown in Fig 1B) [45,64]. Codon-aware nucleotide alignments of single-copy genes were used to produce a phylogenetic tree modeling the population structure of the isolate collection (Fig 1B, S1 File) [45,65,66]. Finally, to track distinct strains within and among subjects over time, highly related isolates were grouped into 63 distinct “clonal lineages” (CLs) using PopPunk [67] (Fig 1C). Other methods to define CLs gave nearly identical groupings (S1 Table). Notably, although many CLs were sampled more than once (48 of the 63 CLs), the overall genetic diversity captured in this isolate set spanned species-level diversity, as illustrated by a phylogeny with 39 complete genomes from this study together with 38 complete genomes retrieved from RefSeq collected from diverse sources (S1 Fig).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Colonization and infection over time by distinct clonal lineages (CLs) in 9 healthy and 12 OM-prone children.

A. Presence/absence of homologous protein-coding gene clusters (rows) in the genomes of H. influenzae isolates collected from healthy and OM-prone children (columns). Dark blue indicates gene presence; pale blue indicates absence. Rows were hierarchically clustered to group genes with similar distributions across isolates, and the column dendrogram uses the species-level phylogeny shown in Fig 1B. B. Phylogenetic tree inferred from 3013 single-copy gene alignments (1315 of them core). Closely related isolates grouped into clonal lineages (CLs) have the same tip color, except for those labeled in light grey, which indicates singlet lineages that appeared only a single time. C. Colonization and infection dynamics in two cohorts (NOP: not OM-prone; OP: OM-prone) with each row indicating a single subject and circles indicating doctor’s visits from which H. influenzae was isolated. Filled circles correspond to isolation at OM episodes, and open circles indicate a scheduled pediatric check-up; distinct colors correspond to distinct CLs (except light grey), as described in B; the size of each circle is proportional to the number of distinct isolates (from 1 to 6) collected from each sampling point.

https://doi.org/10.1371/journal.ppat.1014649.g001

The pattern of distinct CLs seen within and among subjects over time illustrates the complexity of H. influenzae population dynamics, showing evidence of transience, persistence`, and recurrent colonization and infection (Fig 1C, S1 Table). The top-3 most-abundant CLs were biased to OM visits (45 of 50 isolates in the top-3 CLs were from OM) whereas “transient” CLs observed at only one visit were biased towards healthy visits (only 3 of 12 singlet CLs were from OM). However, many CLs were found at both healthy and OM visits, and many isolates from the same CL were often sampled from the same subjects and time points, particularly during OM. These results indicate that OM-causing isolates are not restricted to specific branches of the H. influenzae phylogenetic tree. Although it is possible that some lineages may be more prone to causing OM than others, because many CLs were re-sampled from OM-prone children, this could also simply be a property of the study design.

Since more than one isolate was often sequenced from a single subject and time point (from 2 to 8 isolates at 51 of 120 visits), we evaluated the rate of polyclonal colonization by multiple strains, albeit with a high limit of detection. This found that in 11.8% of visits with multiple isolates, more than one lineage was identified (6 of 51 visits had > 1 CL collected from distinct specimens). Thus, single CLs tended to predominate within time point and subject, even though distinct CLs had variable persistence over time. These patterns corroborate previous analyses of longitudinal series collected from health and disease, as well as studies that show middle ear fluid (MEF) isolates frequently closely match isolates from the nasopharynx and nasal passages (NP) that were collected during the same OM episode (~ 85%) [45,46,62,68–70].

Co-colonization and co-infection with H. influenzae by other bacteria could be partially assessed, since other OM-associated species were routinely isolated and cultured during specimen collection (Fig 1, S2 Table), particularly the other two common OM pathogens S. pneumoniae and Moraxella catarrhalis. Isolation of H. influenzae without any other species was more frequent during OM than at healthy visits (55.7% of OM visits vs. 16.1% of healthy visits, Fisher test p-value < 0.01). Similarly, isolation of H. influenzae by itself was higher in MEF over NP specimens (84.5% of MEF samples vs. 16.0% of NP samples, p-value < 0.01). This effect was comparable when healthy visits were excluded (15.6% of NP samples collected during OM, p-value < 0.01). Collectively, these data show that although polymicrobial colonization of the NP is common, OM infections by H. influenzae are often predominated by single strains within both the ME and NP, consistent with other studies.

Association testing between OM disease and H. influenzae accessory genes

Because the scope of genomic variation among clinical isolates is extensive but the set of genomes sequenced here was relatively small, we focused only on whether the presence or absence of annotated genes/operons was associated with OM. Although this likely missed important associations with specific alleles and may have suffered from artifacts of gene annotation and homolog clustering, it drastically reduced the number of statistical tests, simultaneously enabling focus on the predicted molecular functions of the genes identified. In the set of 206 genome assemblies, 1315 of 3267 were core genes, with one or more homologs found across isolates (Fig 1A, see legend). The 1951 accessory genes that remained could be further collapsed into 724 ‘segregation patterns’ or ‘features’, since many accessory genes were identically distributed across all 206 genomes (typically due to co-inheritance as operons). Focusing on these 724 accessory gene features thus excluded the many thousands of single-nucleotide, indel, and more complex genetic variants distinguishing these isolates, in favor of association testing with a reduced set of binary protein-coding gene presence/absence encodings.

To test for associations between accessory gene features and OM disease, we used the Scoary pipeline [71]. First, we applied Fisher’s Exact Test to identify accessory genes with skewed isolation from MEF compared to NP. Next, to account for phylogenetic dependencies that might spuriously associate markers with OM-associated clades, we tested for evidence of multiple independent ancestral gene-trait co-occurrences across the phylogeny [71–73]. Table 1 lists the seven accessory gene features most strongly associated with isolation from the middle ear. Although many isolates were clustered into clonal lineages (often from the same subject or sampling point), the top candidate associations did not show simple statistical associations with health or disease states, particular subjects, or clonal lineages. Instead, distinct evolutionary histories were observed for each candidate, in which recurring gene acquisition and/or loss along the phylogenetic tree correlated with gains and/or losses of health or disease states at the tips of the tree. These results suggest that the presence of these accessory genes could directly affect the risk of OM infection.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Associations between accessory gene features and isolation from middle-ear fluid (MEF) vs. nasopharynx/nasal passages (NP) by naïve Fisher’s exact test with a Benjamini-Hochberg adjusted p-value < 0.05 (FDR), as well as a “best” p-value for phylogenetic gene-trait co-occurrence of < 0.05 (Pairwise Comparisons – Best) from Scoary. All MEF isolates were collected from subjects during acute OM episodes, but NP isolates come from both health and OM disease. The numbers in parentheses indicate the subset of NP isolates collected during OM episodes instead of healthy visits. Further details are in S3 Table, including for two other clinical “traits”: whether isolated from an OM versus healthy visits or (b) whether from an OM-prone versus healthy subject. Only genes/operons that passed both the FDR-adjusted Fisher’s test and “best” comparisons of phylogenetic contrast at p < 0.05 were included. Genes strictly co-inherited across strains were treated as single tests and are separated by the double dashes (--).

https://doi.org/10.1371/journal.ppat.1014649.t001

The absence of H. influenzae polyamine metabolism genes is associated with isolation from OM disease states

Notably, of the seven associations, five were “anti-OM” associations, where the presence of the gene or operon correlated with healthy carriage rather than disease (i.e., the associations with negative log₂(odds ratios) in Table 1). One positive association with OM was to an accessory gene toxT predicted to encode the toxin in a recently characterized competence-regulated toxin-antitoxin system [74]. This putative “pro-OM” gene would be an interesting candidate for future follow up studies on its potential role in pathogenesis, but we noted that most candidates were “anti-OM” genes/operons, each with a plausible role in nitrogen/polyamine metabolism and pH homeostasis:

1. (1). The top candidate, potE-speF, is further investigated below (Table 1, row 1). Evidence for a genetic association of potE-speF absence and OM disease comes from the nearly 16-fold depletion of this operon in MEF compared to NP isolates (Fig 2A; Table 1, row 1), as well as a history of gains and losses that correlates with changes in health status at the tips of the phylogeny. A parsimonious model of potE-speF acquisition in seven independent lineages would suggest that five are associated with loss of virulence in several different subjects (Fig 2B). The presence-absence calls of these two genes were not an artifact of annotation or clustering method, since all copies had > 95% amino acid identity, tblastn found no partial copies on the genome assemblies, and the local gene neighborhood was consistent in strains that lacked potE-speF (S2 Fig). The potE-speF operon is predicted to encode machinery that exchanges and interconverts polyamines with the extracellular environment: potE encodes a predicted inner membrane putrescine-ornithine antiporter that exchanges neutral extracellular ornithine for alkaline cytosolic putrescine, and speF encodes a cytosolic ornithine decarboxylase converting ornithine into putrescine (Fig 2C). In E. coli, the orthologous potE-speF operon is known to confer ornithine-dependent acid tolerance, since the neutral polyamine ornithine is converted into weakly basic polyamine putrescine, thereby allowing E. coli to directly neutralize an acidic environment when in the presence of ornithine [75–78].

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. H. influenzae accessory operon potE-speF is negatively associated with otitis media.

A. Mosaic plot showing the distribution of potE-speF operons across isolates. AOM = acute otitis media; MEF = middle-ear fluid; NP = nasal or nasopharyngeal. B. Phylogenetic tree has the same topology as Fig 1B and S1 File: filled tips indicate presence of potE-speF and open tips indicate absence; tip colors indicate isolation source as in mosaic plot, red = isolation from a subject with AOM from MEF, black = isolation from a subject with AOM from NP, light grey = isolation from the NP of a subject without AOM; distinct outer ring colors indicate isolation from distinct subjects. C. Diagram of predicted functions of PotE and SpeF proteins. PotE acts to take up neutral extracellular ornithine and secrete alkaline intracellular putrescine, while SpeF decarboxylates ornithine into putrescine.

https://doi.org/10.1371/journal.ppat.1014649.g002

1. (2). A putative cyclase-encoding candidate gene, “cluster_1982”, was nearly always co-inherited with potE-speF though not as part of the same operon or in the same gene neighborhood. The cluster_1982 gene was absent from only a single isolate with potE-speF and was present in only one without it (Table 1, row 2). Thus, this association with OM is highly correlated with potE-speF. No further information about potential biochemical functions was found.
2. (3). The iciA-cadA candidate operon was identified in 10.1% of all isolates but only 1.4% of OM isolates and none of the MEF isolates (Table 1, row 4). The canonical iciA gene from E. coli encodes a transcriptional regulator, and the canonical cadA encodes lysine decarboxylase, which converts lysine into the polyamine cadaverine (weakly basic) while consuming a proton and releasing CO₂ [78–82]. Like potE-speF, these genes have been implicated in both pH homeostasis and pathogenesis, and their activity impinges on central metabolism through their interaction with the tricarboxylic acid pathway [83,84].
3. (4). The allC accessory gene is predicted to encode an allantoate amidohydrolase, which breaks down the tetraamine allantoate into the triamine S-ureidoglycolate, consuming protons and water, while releasing carbon dioxide and ammonium ion (weakly acidic). In E. coli, the allC gene product is involved in the breakdown of purines for anaerobic nitrogen utilization [85,86]. Although we found no specific reports demonstrating a role in pH homeostasis or pathogenesis, allC was found in 19.2% of all isolates but only 5.9% of OM isolates, with a relative depletion of nearly 13-fold in MEF compared to NP isolates (Table 1, row 5).
4. (5). Finally, a diverged paralogous copy of pyrG (pyrG_2) (Table 1, row 6) encodes a putative CTP synthetase, which in E. coli converts UTP to CTP, consuming an ATP and using glutamine or ammonia as a nitrogen source. Notably, whenever this candidate was present, an unannotated accessory gene cluster_1202 was absent (Table 1, row 7), and vice versa, so these two accessory genes are mutually exclusive and have reciprocal distributions across isolates. No predicted functions could be assigned to cluster_1202 so this may be an additional “pro-OM” gene, but the pyrG_2 gene, like other “anti-OM” candidates, could plausibly affect nitrogen cycling, pH homeostasis, or have other direct or indirect effects on central metabolism.

Candidate “Anti-OM” Operon potE-speF confers ornithine-dependent alkaline sensitivity

Taken together, all five strong “anti-OM” associations had predicted molecular functions in similar or related pathways (Table 1). This suggests that natural bacterial genetic variation in polyamine metabolism and pH homeostasis affects the risk of middle-ear disease. Several candidates have known or plausible roles in the ability of bacteria to survive acid stress, whereas reports in the literature have indicated that middle-ear fluids from OM can sometimes have an alkaline pH, instead of the acidic pH often associated with other purulent secretions [54–58]. In particular, the predicted molecular functions of genes in the top candidate accessory operon potE-speF (Fig 2C) led us to hypothesize that the absence of potE-speF increases bacterial survival in an alkaline environment. To test this, we evaluated growth and survival of a wild-type (WT) strain (the low-virulence laboratory strain Rd KW20, which naturally possesses potE-speF) and a mutant derivative carrying a ∆potE ∆speF double knock-out mutation. During strain construction, a ∆potE single mutant was also produced, but the ∆speF single knock-out was unable to be generated, suggesting that this gene may be conditionally essential when potE is present. To identify genetic differences between WT and ∆potE ∆speF outside the knockout constructs that could have phenotypic effects, we used high-coverage genome sequencing. This found many sites with high within-strain genetic heterogeneity (i.e., 222 sites had 5–25% variant allele frequencies within both cultures), reflecting the many phase-variable simple sequence repeats (SSRs) in the H. influenzae genome (and possibly some persistent sequencing artifacts). Only five positions had asymmetric allele frequency distributions between WT and ∆potE ∆speF samples, but in all five cases, none was a fixed difference that distinguished the strains (S4 Table). Thus, phenotypic effects of these putative genetically unstable sites would affect both strains. Experiments reported here focused attention on comparing the WT and ∆potE ∆speF double mutant strain, since both genes were always both present or absent across the set of genomes analyzed.

Since we hypothesized that the presence of potE-speF would lead to ornithine-dependent effects on growth and survival through effects on pH, we first measured differences between WT and mutant growth curves (changes in OD₆₀₀ over time) for cells in rich media adjusted to three different initial pHs (6.0, 7.5, or 9.0), with ornithine supplementation at four concentrations (0, 12.5, 25, or 50 mM) across a physiologically relevant range [83]. Representative shaking culture growth curves are shown in Fig 3A and areas-under-the-curves are summarized in Fig 3B. All culture conditions supported growth over time, but WT cells grown with high concentrations of ornithine showed an initial extended exponential phase, followed by decreased OD₆₀₀ after the cultures entered stationary phase, potentially indicative of cell death and lysis. Furthermore, this effect was more pronounced at higher pHs (Fig 3A, pink lines; Fig 3B, blue lines). By contrast, the ∆potE ∆speF double mutant showed no ornithine-dependent differences in growth at all three pHs (Fig 3A and Fig 3B, lower panels), consistent with a requirement for these genes in converting neutral ornithine in the extracellular environment into putrescine.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Effect of ∆potE ∆speF on planktonic growth in shaking culture; representative experiment shown.

A. Growth over time for subcultures measured by OD₆₀₀ every 15 min over 16 h in WT and ∆potE ∆speF strains (rows), at 3 initial pHs (columns), and 4 ornithine concentrations (colors). B. Areas-under-the curve (y-axis) with varying ornithine (x-axis) at three initial pHs (colors) and two genotypes (panels). C. Changes in culture pH after overnight growth.

https://doi.org/10.1371/journal.ppat.1014649.g003

Measurements of culture pH after overnight growth found a strong effect of potE-speF on final pH when in the presence of ornithine (Fig 3C). As expected, cultures of both genotypes without supplementation became more acidic overnight, due to the accumulation of acidic byproducts of primary metabolism in the medium, regardless of initial pH. With ornithine supplementation, WT cultures showed concentration-dependent increases in the final pH overnight, as predicted. By contrast, deletion of potE-speF showed insensitivity to supplemental ornithine, and so all cultures at all starting pHs showed comparable pH decreases towards acidity during overnight growth.

In parallel, we examined how supplementation with six other polyamines that had previously been detected in Haemophilus [87] affected the growth of WT and of ∆potE ∆speF mutant strains. In addition to ornithine, we also tested cadaverine, diaminopropanol, putrescine, spermidine, spermine, and tyramine. Supplementation with all tested polyamines had distinct pH-specific effects on growth but none showed a strong difference between WT and ∆potE ∆speF, except ornithine and (weakly) putrescine (S3, S4 and S5 Figs). These results show that polyamine content and the pH of the environment have strong effects in H. influenzae growth and show a specific relationship between potE-speF activity and growth in the presence of ornithine or putrescine.

Since the growth curves in Fig 3A used OD₆₀₀ biomass measurements and so did not directly measure viability, we next measured CFU/ml titers after 18 h overnight growth (Fig 4). In the absence of supplemented ornithine, both wild type and ∆potE ∆speF showed significantly and substantially higher titers (>30-fold) after overnight growth when inoculated into medium with a higher initial pH (mean 7.5x10⁷ CFU/ml at starting pH 6; 6.8x10⁸ at pH 7.5; and 2.8x10⁹ at pH 9; two-way ANOVA: p-value < 10^-5 for pH effect; p-value = 0.51 for genotype effect). The increased survival of cells in overnight cultures at higher pHs is likely because acidic waste products of metabolism were more effectively neutralized. By contrast, supplementation with ornithine (50 mM) had strong genotype-dependent effects on survival at the neutral and alkaline pHs. No detectable CFU were seen for wild type overnight grown with added ornithine (limit-of-detection <10² CFU/ml), but the ∆potE ∆speF mutant remained viable with supplemented ornithine (mean 2.4x10⁸ CFU/ml for starting pH 7.5; 1.5x10⁹ for pH 9). At both pH 7.5 and pH 9, two-way ANOVA found significant effects of ornithine supplementation (p-value < 10^-4). Although no effect of genotype alone was seen (p-value > 0.7), a strong genotype-by-treatment interaction was observed (p-value < 10^-3). At a starting pH 6, no effects were seen by two-way ANOVA, however overnight viability of cells in media at starting pH 6 was substantially lower than for higher pHs with some titers below the limit-of-detection (Fig 4, downward triangles).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Ornithine-dependent cell death in alkaline conditions is bypassed by deletion of potE-speF.

Cultures were grown in sBHI medium with 0 (blue) or 50 mM (red) supplemented L-ornithine at three different initial pHs for 18 h before dilution and plating to sBHI agar plates for colony counts and CFU/ml calculations. Independent replicates were carried out on separate days with independent cultures; circles indicate CFU/ml titers and downward triangles indicate cultures with no detected viable CFU at a limit-of-detection of ~10² CFU/ml. Boxes indicate the mean, and error bars indicate the standard deviation.

https://doi.org/10.1371/journal.ppat.1014649.g004

These results show a strong ornithine-dependent viability defect in WT cells when grown in media starting at neutral or alkaline pH. This viability defect depends on the presence of potE-speF, since deletion of the operon bypasses ornithine-dependent cell death at high pH. No significant difference in acid tolerance was observed between genotypes as previously reported for E. coli [75–78], possibly reflecting the particulars of these assays and/or organism-specific differences. As a notable aside, these results also show substantially higher survival of H. influenzae cells after overnight culture when inoculated into standard rich media adjusted to a high pH, which has practical implications for working with H. influenzae strains, which can often have low viable CFU/ml after overnight culture.

The molecular functions of potE-speF confirmed by mass spectrometry

We next sought to confirm that the phenotypic effects observed for in vitro cultures were associated with the predicted function of the potE-speF operon to take up extracellular ornithine, convert it into putrescine, and secrete putrescine. We used liquid chromatography mass spectrometry (LCMS) to determine the relative quantities of different polyamines in cell pellets and lyophilized media supernatants of WT and mutant cultures grown at three pHs, with or without supplemented ornithine. Results after subculture for 5 h at different starting pHs and supplements are shown in Fig 5 (time points at 2, 4, 5, and 24 h are shown in S6 Fig). In the absence of supplemented ornithine, ornithine levels in both pellets and supernatants were low after 5 h for both genotypes at all three pHs (Fig 5A, top panels). With supplementation, ornithine levels remained low, except for cell pellets of the ∆potE ∆speF mutant, which accumulated ornithine to high levels (>10-fold; ANOVA p < 10^-5 for an ornithine-by-genotype interaction; p > 0.5 for effect of pH). These results are consistent with the known function of speF, and they further indicate that ornithine uptake does not necessarily require potE, likely instead supported by the partially redundant functions of the potABCD transporter or similar uptake systems.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Mass spectrometric analysis of intra- and extracellular ornithine (A) and putrescine (B) levels in H. influenzae WT and ∆potE ∆speF mutants grown at different pHs, with or without supplemented ornithine (50 mM), after 5 h of growth.

Cell pellets and lyophilized culture supernatants were derivatized and extracted before LCMS to quantify relative polyamine levels. The y-axis indicates the micromolar (µM) concentration of each polyamine in culture-equivalent volumes for the extracellular environment (supernatant) and cells (pellet). Results shown are for three independent replicates conducted on different days. Boxes indicate the mean, and error bars indicate the standard deviation.

https://doi.org/10.1371/journal.ppat.1014649.g005

Consistent with the predicted function of speF, putrescine levels for ∆potE ∆speF were low in all conditions tested, indicating that putrescine production depends upon potE-speF. In further support of the predicted functions of the operon, in WT cultures, ornithine supplementation led to a > 2-fold increase in intra- and extracellular putrescine levels (Fig 5B, bottom vs. top panels, p = 1.1x10^-5 by ANOVA; all time points shown in S7 Fig). In summary, ornithine can still be taken up in the absence of potE, as expected, but putrescine levels depend on speF. Notably, the starting pH had little overall effect on the transport and interconversion of ornithine and putrescine by cells. In conclusion, potE-speF causes ornithine-dependent production and secretion of putrescine, leading to increased media pH and survival defects of H. influenzae cells associated with alkalinity.

The effects of potE-speF on other polyamines

In parallel, LCMS was also used to examine the relative levels of other polyamines previously detected in Haemophilus as part of a survey for polyamine markers that distinguish members of the Pasteurellaceae family [87] (S8 to S12 Figs). The relative abundances of spermidine, spermine, diaminopropanol and tyramine were unaffected by the ∆potE ∆speF mutation; however, cadaverine showed markedly decreased levels in the absence of potE-speF, indicating that cadaverine levels are directly or indirectly affected by potE-speF activity (S8 Fig). These results show that the potE-speF gene products affect the level of metabolite(s) other than ornithine and putrescine.

Inactivation of potE-speF increases disease burden in a mouse model of otitis media but not lung infection

Our genetic association study found that H. influenzae strains lacking potE-speF are enriched in OM disease, and in vitro growth assays showed that potE-speF confers ornithine-dependent sensitivity to alkalinity likely due to the production of alkaline putrescine. However, these data alone do not show a direct role for these genes in modulating OM pathogenesis. To test for differences in pathogen burden by WT and ∆potE ∆speF mutant strains during infections, we employed two established mouse models of H. influenzae in vivo fitness, one for acute middle-ear infection after influenza viral infection, and another for lung infection [88,89].

To test whether the potE-speF genes affect OM disease burden, wild-type and mutant strains were tested in a mouse model of otitis media, in which mice are infected with influenza virus a week prior to intranasal inoculation of bacteria. Three days post-infection, burden was evaluated from nasal lavage, ME washes, and ME bullae. The wild-type Rd KW20 strain had a low bacterial burden in all three specimen types (~10⁴ to 10⁵ CFU), whereas the ∆potE ∆speF mutant showed substantial and significant increases in bacterial burden (Fig 6A; ~ 32-fold for nasal lavage, ~ 63-fold for ME washes, and 6.2-fold for homogenized ME bullae, p < 0.05 for all sites or each site by one-way ANOVA on log-transformed CFU). These results indicate that the presence of the potE-speF operon limits the ability of H. influenzae to cause middle-ear disease after influenza viral infection.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Mouse disease burden in middle ear and lung infection models.

A. Infection of WT Rd and ∆potE ∆speF mutants in a C57BL/6 mouse OM model. Three specimen types were recovered 3-days after intranasal inoculation of 10⁵ CFU into adult animals that were pre-infected with influenza virus 7-days earlier. Infectious burden was measured as total CFU recovered from each specimen type (n = 4 animals per genotype), plotted on a log-scale. ME washes were recovered from both ears separately resulting in 2 washes per mouse; ME bullae were pooled before homogenization. B. Infection of WT and ∆potE ∆speF mutant strains in a CD1 mouse model of lung infection, in which animals (n ≥ 5 per genotype) were intranasally inoculated with 10⁸ CFU. Two specimen types were collected for dilution and plating to measure bacterial burden 24 or 30 h later, Bronchoalveolar lavage fluid (BALF) and homogenized lungs. Infectious burden was measured as total CFU recovered from each specimen type. Boxes indicate the mean, and error bars indicate the standard deviation. Statistically significant differences were evaluated using one-way or multi-factor ANOVA on log-transformed CFU. Significant p-values < 0.05 are shown.

https://doi.org/10.1371/journal.ppat.1014649.g006

Results from the OM model contrast with results from the lung infection model (Fig 6B). CD1 mice were intranasally inoculated with ~10⁸ CFU of WT or mutant bacteria. Lung disease burden after 24 or 30 h was evaluated by recovering bacteria from bronchoalveolar lavage fluid samples (BALF) or from homogenized lungs, diluting, and plating to sBHI agar solid media to determine total CFU counts per BALF and per lung. At 24- and 30-h post-infection, CFU titers were >100-fold lower than the original inoculum, but no significant differences were seen between clearance of the WT and ∆potE ∆speF strains (multi-factor ANOVA p-value>> 0.05), although the BALF titer for ∆potE ∆speF was slightly lower at the 30-h timepoint. These data suggest little to no role for the potE-speF operon in lung infection, sharply contrasting with the OM model that, instead, showed much higher disease burden in the double knockout mutant.

Middle ear fluids have increased alkalinity when infected with H. influenzae

While testing the WT and ∆potE ∆speF strains for pathogen burden in the mouse OM model, we also measured the pH of the middle-ear washes, finding that they had significantly increased pH after infection by the ∆potE ∆speF mutant compared to WT (Fig 7A). Although this result initially seemed inconsistent with the pH-increasing effects of potE-speF in ornithine-supplemented in vitro cultures, importantly, the in vivo results also showed an orders-of-magnitude difference in disease burden (Fig 6A). Thus, we reason that the increased pH of middle ear fluid could be part of a host response to infection, and the increased pathogen load in the infections by ∆potE ∆speF could explain this finding.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. The pH of middle ear fluids during H. influenzae OM infection.

A. The pH of middle ear washes from the mouse OM experiment shown in Fig 6. B. The pH of human middle ear fluids collected into PBS buffer that were infected by three different OM pathogens. Boxes indicate the mean, and error bars indicate the standard deviation. Statistical significance was assessed using Kruskal-Wallis test, followed by Wilcoxon Rank Sum tests to distinguish differences among groups, as indicated. Only p-values < 0.05 are shown.

https://doi.org/10.1371/journal.ppat.1014649.g007

Previous reports have noted that middle-ear fluids from human otitis media infections can be alkaline, raising the question of whether the pH of human MEF specimens differs depending on the infecting pathogen [54–58]. Furthermore, the association study here identified a potential role in OM disease for H. influenzae genes involved in pH homeostasis. To test whether the pH of infected middle ear fluid varied among the three main OM pathogens (H. influenzae, S. pneumoniae and M. catarrhalis), we selected 10 stored MEF specimens from the biorepository for which there was clear evidence of infection by only one of the three species. The pH of these PBS-buffered MEF specimens had marginally higher pH in the H. influenzae specimens compared to those of M. catarrhalis and S. pneumoniae (Fig 7B). Importantly, at collection during tympanocentesis, these specimens had been directly collected into phosphate-buffered saline (PBS) at pH 7.5 (in a ratio of ~1:10 to ~1:1, depending on the approximate volume of MEF recovered, ranging from ~50–500 µl), so the differences seen here likely underestimate the true variation in pH among specimens. Together with the high alkalinity of mouse ME washes in the high-burden infections caused by the ∆potE ∆speF mutant, this result supports the existence of a pathogen-specific host response to H. influenzae infection that leads to alkaline middle-ear fluids [56], and they indicate a role for H. influenzae pH homeostasis genes.

Ethics statement

The long-term sampling study was approved by the Institutional Review Board of the Rochester General Hospital Research Institute (approval number 1141B), and written informed consent was obtained from parents prior to child enrollment.

Bacterial cultures

H. influenzae were cultured using standard methods [111,112]. Liquid cultures were grown at 37ºC and 5% CO₂ on a roller drum or shaking at 200 RPM in sBHI, brain-heart infusion broth (Difco) supplemented with hemin (10 μg / mL; ICN Biochemicals) and β-nicotinamide adenine dinucleotide (NAD) (2 μg / mL; Sigma). Solid media cultures were grown in a 37°C incubator with 5% CO₂ on chocolate (Biomérieux) or sBHI (Condalab) agar. Erythromycin (11 µg / mL; Erm₁₁) was used to supplement media for H. influenzae as required. Escherichia coli was grown in Luria Bertani (LB) broth or on LB agar at 28°C or 37°C, supplemented with ampicillin (100 µg / mL; Amp₁₀₀) or erythromycin (150 µg / mL; Erm₁₅₀) as required.

Study population and clinical isolates of H. influenzae

Clinical isolates of H. influenzae were retrospectively selected from the biorepository at the Rochester General Hospital Research Institute. These had originally been collected from pediatric subjects between ~6–36 months of age between mid-2006 and late 2016, as previously described [61,62,113]. Enrolled subjects were prospectively sampled at six regularly scheduled pediatric checkups (“healthy” visits), as well as at any acute OM episodes (“AOM” visits) they experienced. At any visit, nasal wash specimens were collected, along with nasopharyngeal and throat swabs. At AOM visits, middle ear fluid specimens were also collected into phosphate-buffered saline (1xPBS) from one or both ears by tympanocentesis, as indicated, and standard-of-care treatment was prescribed. Approximately a week after AOM episodes, follow-up visits were also conducted with sampling as above. Follow-ups were typically labeled as “healthy” visits unless also associated with a persistent or rapidly recurring infection. Nasal wash and middle ear fluid (MEF) samples were subsequently partitioned into three aliquots: one directly plated for bacterial isolates, a second inoculated into overnight broth cultures that were stored and then plated for bacterial isolates, and a third into Trizol for later bulk analysis of the original specimen. For conciseness, we collectively refer to isolates originally collected from the throat, nasopharynx, nasal washes, or nasal wash broths (i.e., not from the middle ear) as “NP” isolates, while those collected from middle ear fluids, either directly or after overnight culture, as “MEF” isolates. Individual isolates were cultivated, identified, and stored. Typically, only a single isolate was picked for a given species from a given specimen, but several isolates of the same species were often collected from distinct specimens taken from the same subject and time point. In addition to H. influenzae, S. pneumoniae, M. catarrhalis, and alpha-hemolytic Staphylococcus were also commonly isolated.

At the time clinical isolates were retrospectively sampled (late 2016), the collection comprised bacterial isolates from 1178 subjects spanning 5545 doctor visits and 1410 OM episodes, with 2271 H. influenzae strains isolated from 1083 doctor visits, which included 579 OM episodes. Longitudinal patient series were selected from the total collection based on the following criteria: (a) approximately half from OM-prone children (OP, or stringently-defined recurrent acute OM, defined as experiencing ≥3 episodes within 6 months or ≥4 within 12 months along with microbiological confirmation from MEF [13,63,113]) and half from healthy children (NOP, at most 1 OM episode with no H. influenzae isolation); and (b) high prevalence of H. influenzae across the series (≥3 healthy visits with an H. influenzae isolate and, for OP subjects, ≥ 2 healthy visits and ≥3 OM episodes with H. influenzae isolation). This resulted in a set of 109 longitudinal series, from which 21 were randomly chosen (9 from NOP and 12 from OP subjects). This set had originally comprised 267 specimens collected from the 21 subjects at 123 distinct visits. Details about subjects and individual isolates that were successfully sequenced are in Supplementary S1 and S2 Tables.

Whole-genome sequencing and clonal lineage assignments

Of the original 267 specimens, 223 single-colony isolates were successfully revived from cryo-storage, cultured, and had high-quality genomic DNA extracted. Stocks were streaked on chocolate agar petri dishes and grown overnight in a 37°C incubator with 5% CO₂. For most isolates (n = 204), a single colony was picked and grown overnight in 5 mL sBHI in a 37°C shaker (250 rpm). Two 1 ml aliquots were used for 15% glycerol stocks at -80°C, and the remaining cells were pelleted at 3000 RPM for 5 min, decanted, and stored at -80°C for DNA extractions. For isolates with poor growth (n = 19), cells were scraped off lawns grown on chocolate agar, resuspended in sterile water, pelleted at 3000 RPM for 5 min, and stored at -80°C for DNA extraction. Genomic DNA was extracted from pellets using DNeasy Blood & Tissue Kits (QIAGEN). DNA integrity was verified by agarose gel electrophoresis, purity was evaluated spectrophotometrically (NanoDrop 1000, Thermo Scientific), and concentration was estimated fluorometrically (Synergy HT plate reader, BioTek; Quant-It dsDNA Assay Kit, Molecular Probes).

DNA extractions were processed into Illumina sequencing libraries using Nextera XT tagmentation according to manufacturer guidelines (1 ng genomic DNA input), followed by pooling libraries to 96-plex, and sequencing pools on a NextSeq 500 using mid-output runs at 2x150 nt, targeting an average coverage of ~200-fold per genome. The genome assembly pipeline was as previously described [45]; briefly, technical sequence removal used Trimmomatic (v 0.33) [114], overlapping paired-ends were merged using COPE (v 1.1.2) [115], error-correction used a module from Allpaths-LG [116], and sequence assembly used SPADES (v 3.7.0) [117]. Contigs were removed if their length was ≤ 256 bp, > 98% of bp were homopolymers, or coverage was < 10-fold per million reads. Further quality control removed assemblies with >100 contigs, assembly lengths of <1.7 Mb or >2.3 Mb, and assemblies that were not identified to the species level with high confidence as Haemophilus influenzae by Taxator-tk (v1.2) [118]. In addition, polyclonal and contaminated samples were identified using ConStrains (v. 0.1.0) [119]. After excluding poor and contaminated assemblies, we used a set of 206 genomes from 120 visits for further analysis.

After clonal lineage (CL) assignments using the Illumina assemblies (see below), a subset of 39 genomic DNAs representing each CL with >1 isolate in the collection were also sequenced using long-read Pacific Biosciences (PacBio) technology, either RSII or Sequel. SMRTbell libraries were prepared according to PacBio recommendations and pooled to target >50-fold genomic coverage. RSII data were assembled using HGAP2, and Sequel reads were assembled using Falcon. Polishing used Circlator (v1.0.2) [120] (with permutation to the dnaA gene) with error correction at the permuted junction using the PacBio resequencing pipeline. For PacBio assemblies that did not circularize, assemblies were still permuted to dnaA by splitting single un-circularized contigs into two at dnaA and arranging these to the ends of the assemblies. Assemblies used the similar quality control criteria as for Illumina-based assemblies, except required <10 contigs after removing those with low quality scores (<35 quality for RSII; < 40 quality and <0.15 coverage for Sequel). Final error correction used Pilon with Illumina reads against the PacBio assemblies.

Clonal lineages (CLs, sets of closely related strains) were determined using three related methods. PopPunk was used to define clusters of strains based on a combination of nucleotide divergence and genomic content differences (minimum k-mer = 15 under the DBSCAN clustering model without refinement) [67]. Mashtree was also applied, also including an additional 38 complete H. influenzae genomes from RefSeq [121]. Multi-locus sequence type (MLST) was determined from the assemblies using Torsten Seeman’s tool https://github.com/tseemann/mlst (v. 2.11) [122] to access the pubMLST.org database [123]. All three lineage assignment methods largely agreed, although MLST assignments failed to classify some strains to an existing MLST due to novel alleles of some MLST genes not in the database.

Genome annotation, gene clustering, and phylogenetic inference

Genome annotation, pan-genome clustering, and phylogeny inference followed previous methods [45]. In brief, gene annotations used Prokka (v1.11) [124], and clusters of homologous protein-coding genes were identified using Roary (v3.5.1) [64] with a blastp threshold of 75%. Clusters with at most one gene per strain were then re-aligned using the codon-aware aligner Prank [65], and alignments were concatenated for phylogeny inference with RAxML using the GTRCAT model (v8.2.4) [66,125]. RAxML treats gaps (i.e., indels and single-copy genes missing from one or more strains) as missing data. The tree was midpoint rooted, since this agreed with the root of outgroup-rooted trees that included sister species H. haemolyticus and H. parainfluenzae but included only ~10% of the number of genes.

Accessory gene association tests

To identify associations between gene possession (presence or absence of a protein-coding gene) and clinical isolation from health or disease, clinical isolates were classified by three clinical provenance “traits”: (1) isolation site, MEF vs. NP; (2) visit type, acute OM vs. healthy); and (3) subject class (OM-prone, vs. not OM-prone). Treatment outcome (success vs. failure) included only a few isolates, so these were not subjected to further analysis. For each of the three clinical provenance traits, we used Scoary [71] with our phylogenetic tree inference and the Roary gene presence/absence matrix, allowing for identically distributed accessory genes to be collapsed using the --collapse option. Results were deemed significant if the Benjamini-Hochberg adjusted p-value was < 0.05 and for which gene-trait co-occurrences on the phylogenetic tree gave a “best”-case binomial test p-value < 0.05.

H. influenzae gene inactivation

Strains, plasmids, and primers used are in Supplementary S5 Table. We generated ∆potE and ∆potE-∆speF knockouts in the laboratory strain Rd KW20. For disruption of potE (locus ID HI0590), a DNA fragment containing potE and its flanks (2,380 bp) was PCR amplified with Phusion polymerase (ThermoFisher Scientific) using Rd KW20 genomic DNA as a template and primers potE-F1 and potE-R1. This fragment was cloned into pJET1.2/blunt, generating pJET1.2-potE, followed by disruption of the cloned product by inverse PCR using primers potE-F2 and potE-R2. An internal 43-bp fragment was replaced by a blunt-ended Erm^R cassette excised by SmaI digestion from pBSLerm [126], generating pJET1.2-potE::ermC. This plasmid was used as a template to amplify the potE::ermC disruption cassette with primers potE-F1 and potE-R1. For disruption of the potE-speF locus (HI0591 and HI0590), a DNA fragment containing both genes and flanking sequences (3,575 bp) was PCR amplified using Rd KW20 gDNA as template and primers speFpotE-F1 and speFpotE-R1. The fragment was cloned into pJET1.2/blunt, generating pJET1.2-speF-potE, and the cloned product was disrupted by inverse PCR using primers speFpotE-F2 and speFpotE-R2. An internal 351-bp fragment was replaced by a blunt-ended Erm resistance cassette excised by SmaI digestion from pBSLerm, generating pJET1.2-speFpotE::ermC. This plasmid was used as a template to amplify the speFpotE::ermC disruption cassette with primers speFpotE-F1 and speFpotE-R1. Both potE and speFpotE disruption cassettes were used to transform the RdKW20 strain using the MIV method to induce natural competence. Transformants were selected on sBHI agar + Erm₁₁ to select for Rd KW20 ΔpotE and Rd KW20 ΔpotE ΔspeF mutants [111]. In all cases, mutants were confirmed by PCR. We also attempted to knock out the speF gene alone in Rd KW20 by a comparable approach but no transformants were recovered that had correctly targeted resistance cassettes, suggesting that knockout of ΔspeF in a strain expressing potE are inviable.

Genome sequencing analysis of the Rd KW20 WT and ΔpotE ΔspeF MUT strains

Single-colony passages of the WT and MUT strains were grown in 5-ml overnight cultures in sBHI, and genomic DNA was subjected to PacBio Sequel IIe sequencing after SMRTbell library prep v2.0 with a target coverage of ~350-fold genomic coverage by HiFi reads. The original Rd KW20 genome assembly (GenBank accession L42023.1) was modified to convert non-ACGTN characters into Ns and permute the genome to the dnaA gene. HiFi reads were then aligned to this reference using minimap2 [127] under recommended settings for HiFi reads. Reads spanning the breakpoint junctions at potE-speF were used to confirm the fixed insertional replacement in the mutant speFpotE::ermC by manual inspection of alignments with IGV. Additional differences in the distribution of variant alleles between strains were identified using FreeBayes with PacBio HiFi settings and in ‘pooled-continuous’ mode [128]. Variants showing strongly skewed allele frequencies between strains were further investigated, and the impact of each variant was determined by manual inspection against the Rd reference annotation.

Growth curve assays in varying pH and polyamine supplementation

To monitor the effects of pH and polyamine supplementation on bacterial growth, cultures were inoculated 1:50 into sBHI with cells revived from frozen log-phase stocks (frozen in 15% glycerol stocks at OD₆₀₀ = 0.3, defrosted, and resuspended in fresh sBHI) and grown in multi-well plates (Corning, Inc, Durham, NC) in a plate reader (Synergy HT, BioTek) at 37°C with medium shaking. Growth over time was determined by readings of optical density at 600 nm (OD₆₀₀) taken by the plate reader every 15 min for 18 h. Prior to inoculation, sBHI culture media was supplemented with varying concentrations of polyamines (ornithine, putrescine, L-ornithine and putrescine at 1:1, cadaverine, spermine, spermidine, and tyramine at 0, 12.5, 25, or 50 mM), followed by pH adjustment to 6, 7.5, or 9 using 3N hydrogen chloride or sodium hydroxide. Determination of pH before and after growth used pH paper (pHydrion papers; Micro Essential Laboratory Inc, Brooklyn, NY).

Growth and survival in varying pHs with ornithine supplementation

Cultures were inoculated with log-phase stocks at 1:25 into 5 mL pH-adjusted sBHI cultures ± 50 mM L-ornithine and incubated in a 37°C shaker (250 RPM). At 4, 5, or 18 h, aliquots were removed for: (a) measurement of pH, (b) bacterial titers of colony-forming units per ml (CFU/mL), and (c) extraction of polyamines for liquid chromatography-mass spectrometry (LC-MS). CFU/ml titers were determined by plating 50 μL of 4 cell dilutions (10^-5, 10^-6, 10^-7, 10^-8) on sBHI plates and counting colonies after overnight growth in a 37°C incubator with 5% CO₂. Individual replicates (n = 4) were performed on separate days. Significant differences were determined by multi-factor ANOVA of log-transformed CFU/ml given genotype (WT Rd KW20 ΔpotE, and ΔpotE ΔspeF), supplementation (± 50 mM ornithine), and starting pH (6, 7.5, or 9); differences were considered significant for p < 0.05.

Mass spectrometric analysis of polyamine levels in bacterial cell pellets and spent media supernatants

For extractions of polyamines for LC-MS analysis [129–131], 1 mL of culture was pelleted, supernatants were set aside, pellets were washed once in PBS, and both pellets and supernatants were stored at −20°C. After samples were thawed, supernatants were lyophilized, and both pellets and lyophilized supernatants were resuspended in 200 μL MOPS buffer, followed by 3 cycles of freeze-thaw using liquid nitrogen alternating with a 75°C water bath. Trichloroacetic acid (60 µl) was added, and tubes were incubated on ice for 5 min before centrifugation at 13,000 RPM for 5 min to pellet debris. Final supernatants were collected and stored at -20°C until preparation for LCMS.

For LCMS, polyamine standards and extracts of pellet and supernatant were pre-column-derivatized by addition of 5% v/v benzoyl chloride in acetonitrile and 100 mM sodium carbonate. Standards included ornithine, putrescine, cadaverine, spermine, spermidine, diaminopropanol, and tyramine, as previously detected in Pasteurellaceae species [87]. Pre-column-derivatized polyamine standards and pellet/supernatant extracts were analyzed on a Waters Acquity I-Class UPLC system coupled to an Acquity TUV detector and Synapt G2Si HDMS mass spectrometer in positive ion mode with a heated electrospray ionization source in a Z-spray configuration. LC separation was performed on a Waters Acquity UPLC BEH 1.7-µm, 2.1-mm, x 50-mm column using an 0.6 mL / min gradient of 95/5–15/85 A/B in 4 min followed by washing and reconditioning the column. Eluent A was 0.1% formic acid in water and B was 0.1% formic acid in acetonitrile. Conditions on the mass spectrometer were as follows: capillary voltage 0.5 kV, sampling cone 40 V, source offset 80 V, source 120°C, desolvation 250°C, cone gas 0 L / h, desolvation gas 1000 L / h, and nebulizer 6.5 bar. The analyzer was operated in resolution mode and low-energy data were collected between 100 and 2000 Da at 0.2-sec scan time. MS^e data were collected using a ramp trap collision energy of 20–40 V, and masses were extracted from the time-of-flight MS total ion currents using an absolute width of 0.005 Da.

Peak areas corresponding to each polyamine on the panel served as relative measures of polyamine concentration in the cells (pellets) and extracellular environment (lyophilized media supernatants). Significant differences were determined for each polyamine at each time point by multi-factor ANOVA of each polyamine’s peak area given the genotype (WT or ΔpotE ΔspeF), fraction (pellet or supernatant), supplementation (± 50 mM ornithine), and starting pH (6, 7.5, or 9); differences were considered significant for p < 0.05.

Mouse model of acute otitis media infection by H. influenzae

To assess the disease burden of H. influenzae in an animal model of acute OM, we employed an established influenza virus co-infection model [88]. C75Bl/6 mice were housed under pathogen-free conditions at the animal facilities at RGHRI and used at 6–8 weeks of age. Animal handling and procedures were in accordance with the Institutional Animal Care & Use Committee at RGHRI. Bacterial strains were grown in sBHI in a 37°C shaker (250 RPM) to an OD₆₀₀ of 0.6 and frozen at -80°C with 15% glycerol until use. Bacterial CFU/ml titers were determined when stocks were thawed. On Day 0 after acclimatization, mice (n = 4) were inoculated intranasally with influenza virus PR8 (TCID₅₀ = 10 µL). On Day 7, H. influenzae was intranasally inoculated into anesthetized mice at 5 x 10⁷ CFU / dose. On Day 10, mice were sacrificed and H. influenzae burden was measured in nasal lavages, ME washes, and combined ME bullae homogenates. Infections were performed in groups of 4 mice per genotype. Significant differences between genotypes collectively or for each specimen type alone were evaluated by one-way ANOVA on log-transformed CFU measurements, followed by Tukey’s HSD (“honest significant differences”) as indicated, and differences were considered significant for p < 0.05. The pH of ME washes was measured using pH test strips (7.5 to 9.5 range and 0–14 range; VWR Chemicals BDH), and significant differences (p < 0.05) were evaluated using Kruskal-Wallis tests and pairwise Wilcoxon rank sum tests.

Mouse model of lung infection by H. influenzae

For mouse lung infections, animal handling and procedures were in accordance with European (Directive 2010/63/EU) and National (RD118/2021) legislation, with authorization of the CSIC Animal Experimentation Committees and local Government (Protocol PI007/19). To determine differences in infectious burden during lung infection, a mouse model of H. influenzae lung infection was used, as previously described [132–136]. CD1 female mice were purchased from Charles River Laboratories (France), housed under pathogen-free conditions at the Institute of Agrobiotechnology facilities (registration number ES/31-2016-000002-CR-SU-US), and used at 22–25 g (6–7 weeks). Bacterial strains were revived and grown on chocolate agar before scraping from the plate and making a bacterial suspension in sBHI. For intranasal infection, 20 µL of a suspension containing 5 x 10⁹ CFU/mL (1 x 10⁸ CFU / mouse) was placed at the entrance of the nostrils until complete inhalation by the mouse, while the mouse was anesthetized (ketamine-xylazine, 3:1). At 24 or 30 h post-infection (hpi), mice were euthanized using cervical dislocation. BALF samples were obtained by perfusion and collection of 0.7 mL of PBS, with the assistance of a sterile 20G (1.1-mm-diameter) Vialon intravenous catheter (Becton-Dickinson) inserted into the trachea. Recovered BALF was serially 10-fold diluted in PBS and plated in triplicate on sBHI agar to determine the number of viable bacteria. By following published procedures [137], we determined a limit-of-detection of < 3–4 CFU/mL BALF. The left lung was individually weighed in sterile bags (Stomacher 80, Seward Medical) and homogenized 1:10 (w/v) in PBS. Each homogenate was serially 10-fold diluted in PBS and plated in triplicate on sBHI agar to determine the number of CFU per lung. Infections were performed in groups of ≥ 5 mice per genotype and time point. Significant differences among genotype, specimen type, and time point (BALF and lung homogenates) were evaluated by multi-factor ANOVA on log-transformed CFU/ml measurements, and differences were considered significant for p < 0.05.