Identification of genetic markers of resistance to macrolide class antibiotics in Mannheimia haemolytica isolates from a Saskatchewan feedlot

ABSTRACT

Mannheimia haemolytica is a major contributor to bovine respiratory disease (BRD), which causes substantial economic losses to the beef industry, and there is an urgent need for rapid and accurate diagnostic tests to provide evidence for treatment decisions and support antimicrobial stewardship. Diagnostic sequencing can provide information about antimicrobial resistance genes in M. haemolytica more rapidly than conventional diagnostics. Realizing the full potential of diagnostic sequencing requires a comprehensive understanding of the genetic markers of antimicrobial resistance. We identified genetic markers of resistance in M. haemolytica to macrolide class antibiotics commonly used for control of BRD. Genome sequences were determined for 99 M. haemolytica isolates with six different susceptibility phenotypes collected over 2 years from a feedlot in Saskatchewan, Canada. Known macrolide resistance genes estT, msr(E), and mph(E) were identified in most resistant isolates within predicted integrative and conjugative elements (ICEs). ICE sequences lacking antibiotic resistance genes were detected in 10 of 47 susceptible isolates. No resistance-associated polymorphisms were detected in ribosomal RNA genes, although previously unreported mutations in the L22 and L23 ribosomal proteins were identified in 12 and 27 resistant isolates, respectively. Pangenome analysis led to the identification of 79 genes associated with resistance to gamithromycin, of which 95% (75 of 79) had no functional annotation. Most of the observed phenotypic resistance was explained by previously identified antibiotic resistance genes, although resistance to the macrolides gamithromycin and tulathromycin was not explained in 39 of 47 isolates, demonstrating the need for continued surveillance for novel determinants of macrolide resistance.

IMPORTANCE

Bovine respiratory disease is the costliest disease of beef cattle in North America and the most common reason for injectable antibiotic use in beef cattle. Metagenomic sequencing offers the potential to make economically significant reductions in turnaround time for diagnostic information for evidence-based selection of antibiotics for use in the feedlot. The success of diagnostic sequencing depends on a comprehensive catalog of antimicrobial resistance genes and other genome features associated with reduced susceptibility. We analyzed the genome sequences of isolates of Mannheimia haemolytica, a major bovine respiratory disease pathogen, and identified both previously known and novel genes associated with reduced susceptibility to macrolide class antimicrobials. These findings reinforce the need for ongoing surveillance for markers of antimicrobial resistance to support improved diagnostics and antimicrobial stewardship.

INTRODUCTION

Bovine respiratory disease (BRD) is a major concern in the feedlot industry where it is estimated to cause more than US$3 billion in losses each year globally (1). BRD is commonly controlled by treating high-risk groups of animals with antibiotics on arrival at the feedlot in a practice known as metaphylaxis (2, 3), which has been associated with selection of antimicrobial-resistant bacteria [reviewed in reference (4)]. Concerns about antimicrobial resistance (AMR) resulted in the passage of legislation increasing oversight of antimicrobial use in animal agriculture in Canada and the European Union (5–8). To meet the challenge of reducing the use of antibiotics and using antibiotics more prudently, veterinarians and producers need rapid and accurate diagnostic tests and sampling strategies to provide the best evidence for treatment decisions.

Diagnostic sequencing is the application of sequencing technologies in medical and veterinary diagnostics (9). The biggest benefit of diagnostic sequencing over traditional methods is the speed at which it can provide an answer, taking hours to get a result where traditional methods often take days. In the specific case of detecting AMR, the reduced turnaround time is due to bypassing the requirement for time-consuming culture and antimicrobial susceptibility testing of bacteria in the laboratory. Identifying genetic markers of resistance allows veterinarians to make evidence-based decisions about appropriate antibiotics while avoiding prescribing ineffective antibiotics. Realizing the potential of diagnostic sequencing requires a comprehensive and informative database of all genetic markers of resistance (genes and mutations) in pathogens of interest.

Mannheimia haemolytica is the major bacterial agent associated with BRD in North America (10), and macrolide class antibiotics are often the choice for use in metaphylaxis or as first-line therapeutics, which has resulted in the evolution of increased resistance to these drugs (11–13). A number of macrolide resistance genes and mutations have been identified in M. haemolytica. However, the recent characterization of a macrolide esterase encoded on a plasmid carried by a Sphingobacterium faecium isolate recovered from a feedlot water trough, and widely distributed among animal pathogens including M. haemolytica, highlights the need for further studies to expand our knowledge of the genetic markers of macrolide resistance in cattle-associated bacteria (14–16). Antimicrobial resistance genes (ARGs) in M. haemolytica are frequently carried within mobile genetic elements known as integrative and conjugative elements (ICEs) that are capable of spreading ARGs to related species such as Pasteurella multocida (17, 18). In addition to ARGs, single-nucleotide polymorphisms (SNPs) in the genes encoding the cellular targets of macrolides (ribosomal RNAs and ribosomal proteins) have also been known to confer macrolide resistance in M. haemolytica and other species (19).

In the current study, we analyzed the whole-genome sequences of macrolide-resistant and susceptible M. haemolytica isolates collected over 2 years from a feedlot in Saskatchewan, Canada, to identify the genetic markers of macrolide resistance. Our objective was to identify known AMR genes and mutations and to identify potentially novel markers associated with macrolide resistance, with the goal of enhancing M. haemolytica databases used in diagnostic sequencing.

MATERIALS AND METHODS

Selection of M. haemolytica isolates

Study isolates were selected from an existing collection of M. haemolytica isolates (20). The collection was derived from deep nasopharyngeal swabs collected from 800 cattle at the Livestock and Forage Centre of Excellence in the fall of 2020 and in the fall of 2021 at three time points: on arrival, 13 days after arrival, and 36 days after arrival. Animals were recently weaned steers of various beef breeds sourced from a local auction market and originating from a combined total of 292 unique farming operations (20). Samples were cultured to select for M. haemolytica; identification was confirmed by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (Bruker Daltonics, Billerica, MA), and susceptibility of one M. haemolytica isolate per animal was determined using the Sensititre system (Bovine/Porcine BOPO7F Vet AST Plate) at Prairie Diagnostic Services (PDS, Saskatoon). Isolates were classified as either resistant, intermediate, or susceptible in accordance with the breakpoints established by the Clinical Laboratory Standards Institute [Table S1; see reference (21)]. Isolates classified as intermediate were categorized as resistant (non-wild type) in all subsequent analyses.

We selected 52 macrolide-resistant and 47 susceptible isolates for further study. Fifteen isolates were selected in the fall 2020 collection period, with 1 collected on arrival and the remaining 14 at 13 days post-arrival. Eighty-four isolates were selected in the fall 2021 collection period, with 3 collected on arrival, 49 at 13 days post-arrival, and 21 at 36 days post-arrival, 2 from sick animals, 1 from a dead animal, and 8 from a collection of isolates selectively cultured on 5% sheep blood agar plates supplemented with 16 µg/mL of tulathromycin. Six of the selectively cultured isolates were derived from samples collected 36 days post-arrival, one from a sick animal and one from a dead animal. Antimicrobial susceptibility testing of the selectively cultured isolates was performed as described above. To maximize diversity, each of the selected isolates was from an animal sourced from a different farm of origin, and whenever possible, we selected a susceptible isolate from the same time point and pen for each resistant isolate. The 52 selected resistant isolates had a variety of phenotypes, although all were resistant to at least one of the four tested macrolides: gamithromycin (GAM), tulathromycin (TULA), tildipirosin (TILD), and tilmicosin (TILM). Many of the isolates were also resistant to tetracycline. We included the results for tetracycline resistance in our analysis due to its common use in feedlots for the treatment and prevention of BRD and the strong association of tetracycline resistance genes with ICEs (11).

Genomic DNA extraction

Isolates were cultured aerobically on 5% sheep blood agar plates at 37°C for 24 h, and one colony was selected and grown aerobically in 7 mL of brain heart infusion broth at 37°C for 24 h with shaking at 200 rpm. Two 3-mL aliquots of each broth culture were pelleted, and genomic DNA was then extracted using a salting-out protocol as previously described (22). DNA concentration of extracts was determined using spectrophotometry (NanoDrop 2000) and electrophoresis in 0.8% agarose was used to assess fragmentation and to detect RNA contamination. Extracts that were found to have visible RNA contamination were then incubated with 5 µL of RNase A (100 mg/mL) at 37°C for 1 h, followed by the sodium acetate and ethanol precipitation steps of the salting-out protocol.

To confirm the identity of the genomic DNA as M. haemolytica, cpn60 PCR and amplicon sequencing were performed using either universal cpn60 PCR primers (23) or M. haemolytica specific cpn60 primers (forward: GCG ATT GTA AAC GAG GGC TT, reverse: TGA CCG ACT GTT GAA TCT GAG). Sequencing was performed by the Sanger method, and the forward and reverse primer reads were combined using the pregap4 and gap4 software to produce a consensus sequence that was aligned to the cpnDB database (24) using the FASTA-based sequence comparison tool. Sequence identity of at least 90% over the length of the amplicon sequence was the minimum threshold for identification as M. haemolytica.

As a final quality check, all extracts were assessed for DNA concentration immediately prior to Illumina library preparation using fluorometry (Qubit dsDNA Broad Range Assay Kit, Invitrogen). A minimum DNA concentration of 20 ng/µL was required for an extract to be included in Illumina library preparation.

Whole-genome sequencing

Illumina sequencing libraries were prepared using the Nextera XT Library Prep Kit (Illumina) according to the manufacturer’s instructions. Libraries were first diluted to a DNA concentration of 0.2 ng/µL. Following indexing and purification, fragment size distribution in prepared libraries was assessed (Bioanalyzer 2100, Agilent) and quantified using the Qubit ds Broad Range assay kit (Invitrogen). Libraries were manually normalized to a concentration of 4 nM and then pooled and diluted to 8 pM with a final concentration of 1% PhiX DNA and sequenced using the 500 cycle MiSeq Reagent Kit (v.2) (2 × 250 bp) on the Illumina MiSeq platform. Up to 24 genomes were sequenced per flow cell.

Isolates were also sequenced on a GridION platform using the ligation sequencing kit (SQK-LSK109) and the FLO-MIN106 flow cell (PDS). MinKnow (v.22.03.2) was the operating system and Guppy (v.6.06) was used for base calling. A Qscore cut-off threshold of 9 was utilized for all reads, with those below the cut-off being excluded from the sequencing output.

Whole-genome sequence assembly

Raw, demultiplexed Illumina reads were processed using Trimmomatic with a sliding window of 4, quality score threshold of 20, and minimum sequence length of 36 bp (25). Raw nanopore reads were processed using Filtlong (v.0.2.1) with a minimum length of 1,000 bp. Quality-filtered reads from both Illumina and Nanopore sequencing were utilized for hybrid de novo whole-genome assembly using Unicycler (v.0.5.0) with the default settings (26). Whole-genome sequences were then assessed for quality using seqkit (v.2.3.0) (27) and annotated using Prokka (v.1.14.5) (28).

Identification of antibiotic resistance genes

Whole-genome assemblies were screened for known antibiotic resistance genes and mutations using the Resistance Gene Identifier (RGI) algorithm provided by the Comprehensive Antibiotic Resistance Database (CARD) using the default settings and database (v.3.2.5) (29, 30).

The macrolide esterase estT was not present in the utilized version of the database, and an alternative approach was therefore taken to examine the whole-genome assemblies for the presence of this gene. In Geneious Prime (v.2023.2.1), a BLAST database was constructed consisting of the whole-genome assemblies of the study isolates. The reference sequence for estT was then downloaded from the reference sequence for Sphingobacterium faecium strain WB1 plasmid pWB1 (CP094932.1) and compared against this database using the Geneious built-in BLAST software to identify homologous regions of the genome assemblies.

Identification of 23S rRNA, L4, and L22 mutations

Barrnap v 0.9 (31) was used to extract all copies of 23S rRNA gene sequences from M. haemolytica genome assemblies and then 23S rRNA gene sequences were aligned using snippy v 4.6.0 (32) to the reference sequence from M. haemolytica USDA-ARS-USMARC-183 (Genbank Accession NR_103087) with nucleotide numbering according to this sequence.

For the L4 and L22 ribosomal proteins, two custom shell scripts were utilized that extracted the L4 and 22 ribosomal protein amino acid sequences from the Prokka annotations (Supplemental File 1). The ribosomal protein amino acid sequences were then aligned to the reference sequences from M. haemolytica strain USMARC_2286 (L4: AGR7646 and L22: AGR73803) using BLASTp, and any amino acid differences were recorded manually.

Pangenome analysis

All whole-genome sequences were submitted to Roary analysis software (v 1.7.7) alongside a reference strain for the two M. haemolytica genotypes 1 and 2 (genotype 1 = CP017518, genotype 2 = CP017519) (33). Gene clusters were sorted into one of four categories: core, soft-core, cloud, and shell, based on how many of the isolates in the examined data set the gene cluster appeared in. With a total input of 101 genomes (2 references and 99 study isolates), core genes were those found in 99–101 isolates, soft core in 95–98 isolates, shell in 15–94 isolates, and cloud in less than 15 isolates. A multiFASTA alignment of core genes was produced using MAFFT using the -e and -n flags. The roary_plots.py python script was then utilized to generate three outputs: a pangenome matrix, a frequency plot, and a pie chart. A phylogenetic tree was additionally generated using the core gene multiFASTA alignment as the input for FastTree (34). The genotype of each isolate was inferred based on proximity to the genotype 1 and 2 reference sequences in the phylogenetic tree.

Whole-genome SNP analysis

To reduce the impact of assembly errors in the identification of SNPs, all whole-genome sequences were polished using Pilon (v 1.24) (35). Each genome underwent a minimum of five rounds of pilon polishing prior to whole-genome SNP analysis following the method described in reference (36). In brief, genomes were indexed using Hisat2 (v.2.2.1) (37) to produce a SAM file, which was converted to a BAM file, and then sorted and indexed using Samtools (v.1.17) (38) with the resulting BAM file being used as input to Pilon.

For whole-genome SNP analysis, we utilized the snippy-multifunction of snippy (v.4.6.0) to generate a .VCF file for each analyzed genome containing all the identified SNPs (32). To reduce the number of SNPs that were related to genotype, we chose to analyze the two genotypes of M. haemolytica separately and used a susceptible isolate of either genotype as the reference genome (37787 S-06 for genotype 1 and 34455 S-12 for genotype 2). Finally, pyseer (v.1.3.10) was used to identify SNPs that were associated with phenotypic resistance to macrolide class antibiotics for each of the individual genotypes using a P value cut-off of 0.05, a maximum allele frequency equivalent to the proportion of phenotypically resistant isolates included in the analysis (0.5 for genotype 1 and for 0.54 for genotype 2), and the elastic net whole-genome model (39, 40). The odds ratio of each SNP was then determined using Microsoft Excel for Microsoft 365 MSO (v.2308, build 16.016731.20052) 64-bit. SNPs that possessed an odds ratio lower than 0.8 or greater than 1.2 were considered as potentially explanatory for the observed phenotype.

Association of phenotypic macrolide resistance with known AMR genes and candidate genes

Generalized linear models (GLMs) with binomial errors were used to confirm the associations between known AMR genes and phenotypic resistance to each of the five antimicrobial drugs (GAM, TULA, TILD, TILM, and TET) using the glm() function in R (41). Resistance to each of the five drugs was coded as a binomial variable as susceptible (0) or resistant (1). The known AMR genes tet(H), estT, APH(3′)-1a, APH(3″)-1b, and APH 6)-1d were present in the same 44 isolates. Similarly, the known AMR genes msr(E) and mph(E) were present in the same eight isolates. As statistical analyses cannot include identical explanatory factors, only tet(H) and msr(E) were included in the GLMs [i.e., the other five genes were not included in the analysis because they were redundant with tet(H) and msr(E)]. Genotype was also included as an explanatory factor as it had previously been shown to be associated with phenotypic resistance (42). For these GLMs, a P value of ≤0.05 was significant.

The same GLM approach was used to test the association between phenotypic resistance and the candidate genes that were identified in the pangenome analysis. The sign of the slope relating AMR to the presence/absence of a gene indicates whether the gene confers resistance (positive slope) or susceptibility (negative slope) to a given drug. For these GLMs, a Bonferroni-corrected P value was used, where 0.05 was divided by the number of candidate genes. Both unifactorial and multifactorial GLMs were performed. The unifactorial GLMs included only the candidate gene of interest, whereas the multifactorial GLMs also included genotype, tet(H), and msr(E).

Protein structure predictions and alignments

Amino acid sequences were aligned with MUSCLE in Geneious Prime using default settings. Secondary structures were predicted with PSIPRED and DISOPRED3, and protein structure prediction from primary amino acid sequences was performed with DMPFold within the PSIPRED Protein Analysis Workbench (43). Structure visualization and alignments were generated with Pymol (Schrodinger, Inc.).

Mobile genetic element analysis

All 99 genome sequences of the study isolates were submitted to the online web tool implementation of ICEfinder for identification of putative ICEs (44). The sequences of identified putative ICEs and integrative and mobilizable elements (IMEs) were downloaded, and information such as the insertion site and direct repeat sequence was recorded in an Excel spreadsheet. Putative ICE sequences that were found to possess ≤10 type 4 secretion system (T4SS) components and/or lacked an identified insertion site were removed from further analysis. Manual curation and comparisons of the putative ICE sequences were done in Geneious Prime. To determine if the identified putative ICE sequences were novel or homologs of the previously identified ICEMh1 (45), the representative consensus sequence for each ICE was aligned to ICEMh1 from M. haemolytica 42548 using BLASTn (ICEberg database ID ICEO_0000728) (46).

RESULTS AND DISCUSSION

Description of isolates

Study isolates were initially characterized by their antimicrobial resistance phenotype resulting in six groups (Table 1; Table S2).

TABLE 1.

Antibiotic resistance phenotypes of study isolates

Phenotypea	Number of isolates
Susceptible	47
GAM/TULA/TET/TILM/TILD	27
TULA/TET/TILM/TILD	11
GAM/TULA	8
TET/TILM/TILD	5
GAM/TULA/TILM	1
Total	99

^a GAM, TULA, TILM, and TILD are all macrolide class antibiotics. TET is a tetracycline class antibiotic. GAM, gamithromycin; TET, tetracycline; TILD, tildipirosin; TILM, tilmicosin; TULA, tulathromycin.

Genome sequencing and assembly

The average length of the assembled genomes (n = 99) was 2.9 Mb (range 2.5–5.2 Mb) with an average N50 score of 2.6 Mb (range 0.6–2.7 Mb). The median number of contigs per assembly was 2 (range 1–26). The average estimated coverage for assemblies was 158× (median 131×, range 23×–578×). Isolate 38241 S-35 was not included in the whole-genome SNP analysis due to insufficient coverage with high-quality Illumina reads. Based on phylogenetic analysis of core genomes identified in the pangenome analysis described below, most isolates (82 of 99) were identified as genotype 2, and the remainder (17 of 99) as genotype 1 (Fig. S1). A detailed summary of the characteristics of the genome assemblies is provided in Table S3.

Antibiotic resistance gene identification

Antibiotic resistance genes were identified in each whole-genome sequence utilizing the RGI software provided by CARD, where “perfect” hits were considered to represent the presence of a given gene, and “strict” hits were considered to represent the presence of a gene providing the resulting hit had a percent identity of at least 90% and a percent length between 99% and 101%, as calculated by dividing the length of the query protein by the length of the reference protein (Table 2). This analysis revealed two distinct genetic profiles that together accounted for 94% (49 of 52) of the phenotypically macrolide-resistant isolates. One genetic profile consisted of the known macrolide resistance genes msr(E) and mph(E), which encode a ribosomal protection protein and a macrolide inactivating phosphotransferase, respectively (47, 48), and the sulfonamide resistance gene sul2. The other genetic profile included the tetracycline resistance gene tet(H) and its regulator, tetR, the aminoglycoside resistance genes APH(3″)-1b, APH(3′)-1a, and APH(6)-1d, and sul2.

TABLE 2.

Antibiotic resistance genes detected in study isolates

Phenotypea	No. of isolates	msr(E)	mph(E)	erm42	sul2	tet(H)	APH(3″)-1b	APH(3′)-1a	APH(6)-1d	estT
GAM/TULA	8	8/8	8/8	0/8	7/8	0/8	0/8	0/8	0/8	0/8
GAM/TULA/TET/TILM/TILD	27	0/27	0/27	0/27	25/27	27/27	27/27	27/27	27/27	27/27
TULA/TET/TILM/TILD	11	0/11	0/11	0/11	11/11	11/11	11/11	11/11	11/11	11/11
TET/TILM/TILD	5	0/5	0/5	0/5	5/5	5/5	5/5	5/5	5/5	5/5
GAM/TULA/TILM	1	0/1	0/1	0/1	1/1	1/1	1/1	1/1	1/1	1/1
Susceptible	47	0/47	0/47	0/47	0/47	0/47	0/47	0/47	0/47	0/47

^a GAM, gamithromycin; TET, tetracycline; TILD, tildipirosin; TILM, tilmicosin; TULA, tulathromycin.

Resistance genes msr(E) and mph(E) were identified in only 8 of 52 isolates, all of which had the GAM/TULA resistance phenotype (Table 2). Monomethyltransferase encoding gene erm42, previously associated with macrolide resistance (16), was not detected in any isolate. In all the remaining macrolide-resistant isolates, the tet(H) gene was detected alongside the three aminoglycoside resistance genes APH(3”)−1b, APH(3’)−1 a, and APH (6)−1d (Table 2). The presence of tet(H) has been strongly associated with the presence of ICEs, and the identified aminoglycoside resistance genes have frequently been identified on ICEMh1 (17, 18, 49, 50). The sulfonamide resistance gene sul2 was identified in all but three phenotypically macrolide-resistant isolates, which may reflect sul2 having been incidentally selected for alongside macrolide resistance as part of the same mobile element. Regardless of the origin of the sul2 gene, Clinical and Laboratory Standards Institute presently lacks a breakpoint for sulfadimethoxine for M. haemolytica, so no determination on the resistance profile of the isolates was made. No known antibiotic resistance genes were identified in the phenotypically macrolide-susceptible isolates (Table 2).

Macrolide esterase gene estT was detected in all resistant isolates except for those with the GAM/TULA resistance phenotype. estT was consistently annotated by Prokka as rdmC, a methylesterase involved in the biosynthesis of the anthracycline antibiotic aclacinomycin (51). The identification of estT offers a partial explanation for the observed phenotypic resistance; however, it has been demonstrated biochemically that the enzyme encoded by estT has no activity on GAM or TULA (14). Thus, its presence fails to explain the observed resistance to these drugs in the GAM/TULA/TET/TILM/TILD, TULA/TET/TILM/TILD, and GAM/TULA/TILM isolates. We used GLMs with binomial errors to test whether genotypes tet(H) (identical presence/absence pattern to estT) and msr(E) [identical presence/absence pattern to mph(E)] were associated with phenotypic resistance to each of the five individual antimicrobial drugs (Table 3). The unifactorial GLM found that genotype was significantly associated with resistance to TET, TILD, and TILM, and that tet(H) and msr(E) were significantly associated with phenotypic resistance to all five tested antimicrobials. The results changed when all three explanatory factors were included in the multifactorial GLM; tet(H) was significantly associated with phenotypic resistance to all five tested antimicrobials, and msr(E) was significantly associated with phenotypic resistance to GAM and TULA, consistent with the observations in Table 2. The association of tet(H) with macrolide resistance is almost certainly due to the strong linkage of tet(H) with ICEs and not due to any role of the tet(H) gene product in reduced susceptibility to macrolides. In the multifactorial GLM, genotype was not associated with phenotypic resistance to any antimicrobial drugs. The association between genotype and resistance to TET, TILM, and TILD in the unifactorial GLM was probably driven by differences in the prevalence of tet(H) in genotypes 1 and 2 [0.0% (0 of 17) and 52% (44 of 84), respectively]. The lack of a relationship between genotype and resistance to any of the antibiotics tested somewhat contradicts previous reports that genotype 2 isolates are more likely to carry known antibiotic resistance genes (42); however, the limited number of genotype 1 isolates included in this study means this lack of association should be interpreted with caution.

TABLE 3.

Results of unifactorial and multifactorial GLMs for each antibiotic^a

Analysis	Drug	Genotype	tet(H)	msr(E)
Unifactorial	GAM	0.33	4.9e-07	3.2e-05
Unifactorial	TULA	0.97	1.67e-14	0.0004
Unifactorial	TET	2.6e-06	<2.2e-16	0.002
Unifactorial	TILM	1.8e-06	<2.2e-16	0.002
Unifactorial	TILD	2.6e-06	<2.2e-16	0.002
Multifactorial	GAM	0.52	1.4e-10	1.2e-06
Multifactorial	TULA	0.52	< 2.2e-16	1.2e-06
Multifactorial	TET	0.52	< 2e-16	>0.99
Multifactorial	TILM	0.52	< 2e-16	>0.99
Multifactorial	TILD	0.52	< 2e-16	>0.99

^a Shown are the P values for the three explanatory factors: genotype, tet(H), and msr(E).

23S rRNA, L4, and L22 polymorphism identification

Mutations in the cellular targets of macrolides, 23S rRNA and ribosomal proteins L4 and L22, have previously been associated with reduced susceptibility (47, 52). Six copies of the 23S rRNA gene were identified in each of the study isolates, and no resistance-associated polymorphisms were identified relative to a reference sequence (NR_103087) (Table S3). No resistance-associated differences were identified in the amino acid sequences of the L4 ribosomal protein, however, in the L22 ribosomal protein, a D94A amino acid difference occurred in 23% (12/52) of the macrolide-resistant isolates and none of the susceptible isolates (Table S3). Of the twelve isolates with the D94A L22 variant, all were genotype 2, and seven isolates had the TULA/TET/TILM/TILD resistance phenotype, three isolates had the TET/TILM/TILD phenotype, and two isolates had the GAM/TULA/TET/TILM/TILD phenotype.

Among the 12 isolates with the L22 ribosomal protein D94A sequence variant, the only commonality was phenotypic resistance to the 16-membered lactone ring macrolides TILM and TILD, which suggests this variant may alter the binding affinity of 16-membered lactone ring macrolides but not the binding affinity of GAM (a 15-membered lactone ring macrolide) or TULA [a triamilide macrolide comprising a 90:10 mixture of 15-membered lactone ring and 13-membered lactone ring regioisomers (53)].

Whole-genome SNP analysis

To identify potentially novel sequence variants associated with reduced macrolide susceptibility, we conducted a genome-wide SNP analysis of genotype 1 and genotype 2 isolates. A total of 325 sequence variants (SNPs, insertions, or deletions) were initially identified as having a significant association (P ≤ 0.05) with observed phenotypic macrolide resistance (any of the phenotypes in Table 1) in the genotype 1 isolates; however, all odds ratios were between 0.8 and 1.2. For the examined genotype 2 isolates, a total of 1,031 sequence variants were identified, with all odds ratios falling between 0.8 and 1.2. These results suggest that SNPs are not an explanation for the observed phenotypic macrolide resistance (Tables S4 and S5).

Pangenome analysis

A total of 101 isolates were included in the pangenome analysis, 99 study isolates and 2 reference strains representing genotype 1 and genotype 2. A total of 8,325 gene clusters were identified in the pangenome, with each isolate possessing an average of 2,765 (median = 2,766) gene clusters (range = 2,486–5,181) (Table S6). Of the 8,325 gene clusters identified, 1,651 were identified as core genes (i.e., were present in 99–101 of the 101 examined isolates), which were used in a phylogenetic analysis to identify isolates as genotype 1 (17 isolates) or genotype 2 (82 isolates) based on their distance to and from the two reference genomes (Fig. S1).

To discover novel AMR genes, we used the same GLM approach for each of the five macrolide antimicrobials, as described previously. The unifactorial GLM only included the candidate gene of interest as the single explanatory factor. The multifactorial GLM included not only the candidate gene of interest but also genotype, tet(H), and msr(E) as explanatory factors. Of the 8,135 genes identified in the pangenome, 1,413 were part of the core genome (i.e., they were present in all isolates), and these genes were not analyzed for their association with AMR. Of the remaining 6,912 genes, 24 genes were excluded because they had the same presence/absence profile as tet(H), msr(E), or genotype, so that 6,888 candidate genes remained for GLM analysis (Table S7). A Bonferroni-corrected P value of 0.05/6,888 = 7.26 × 10⁻⁶ was used to determine whether a given candidate gene was significantly associated with phenotypic resistance to each of the five macrolide antimicrobials.

The unifactorial GLMs found 1,223 candidate genes that were significantly associated with phenotypic resistance to one or more of the tested macrolides (Table S8). In comparison, the multifactorial GLMs found only 150 candidate genes that were significantly associated with resistance to GAM and no candidate genes that were significantly associated with resistance to any of the other four examined macrolide antimicrobials (Table S9). Thus, most of the genes in the unifactorial GLM were significant only because they were associated with either tet(H) and/or msr(E) in the genome of M. haemolytica. Below, we further investigate the 150 genes identified in the multifactorial GLMs.

Of the 150 GAM-resistance associated genes, 79 were classified as putative resistance genes (GLM slope was positive) and 71 as putative susceptibility genes (GLM slope was negative). Of the 71 putative susceptibility genes, 5 had functional annotations: clpP_2 (protease subunit), mutS (DNA mismatch repair protein), rplW (ribosomal protein L23), ssb_4 (single-stranded DNA binding protein), and tldD (metalloprotease) (54–58), and the remainder were annotated as hypothetical/putative proteins.

Of the 79 putative resistance genes, 4 had functional annotations: glpX_1 (fructose-1,6-bisphosphatase), mnmE (tRNA modification GTPase), rusA (endonuclease that corrects defects during genetic recombination and some DNA repair), and ybcO (putative nuclease), and the remainder were annotated as hypothetical/putative proteins (59–63). One possible connection to macrolide resistance is that the mnmE-encoded GTPase may introduce a tRNA modification that prevents inhibition of protein translation by macrolide class antibiotics; however, as the motifs that macrolide class antibiotics recognize are in mRNA, this seems unlikely (64). Similarly, it is possible that the DNA repair activity of the rusA-encoded endonuclease may result in sequence changes that alter or eliminate macrolide activating motifs (64).

Further examination of the candidate genes identified in the GLM analysis revealed that the genes mnmE, mutS, tldD, rplW, and glpX_1 had a variant with the inverse association (i.e., associated with susceptibility). These variants were identified as distinct genes rather than variants due to the nature of the pangenome tool used (Roary). In identifying genes shared across multiple genomes, an all-vs-all comparison of all coding sequences is conducted. Based on a defined threshold for “same,” coding sequences are grouped into genes and their distribution among genomes is reported in the gene presence-absence table output from Roary. If multiple variants of a gene are sufficiently different from each other so that they exceed the threshold for clustering, they will be assigned to different gene groups. In this case, group_753 is a variant of mnmE, group_508 for mutS, group_2294 for tldD, group_4230 for rplW, and group_452 for glpX_1. Of particular interest is the rplW gene, which encodes the L23 ribosomal protein, a protein that is in close proximity to the nascent polypeptide exit tunnel of the ribosomal complex (57). As mutations in other ribosomal proteins that surround the exit tunnel (L4 and L22) are known to confer macrolide resistance, we further investigated the amino acid sequences of both rplW variants and determined that the group_4230 variant has a 14-amino acid insertion relative to the rplW sequence identified in susceptible isolates (Fig. 1A). To determine the location of the insertion within the three-dimensional structure of the protein, we aligned the M. haemolytica rplW and group_4230 sequences with the L23 sequence of Thermus thermophilus, for which a structure has been determined (65). The protein consists of a well-ordered part containing antiparallel beta-strands and a large loop that forms part of the peptide exit tunnel wall (65). The predicted structure of the susceptibility-associated rplW-encoded L23 aligns well with T. thermophilus L23, but the insertion in the group_4230 variant nearly doubles the size of the loop (Fig. 1B). This group_4230 variant of L23 was only identified in the 27 isolates with the GAM/TULA/TET/TILM/TILD resistance phenotype, suggesting that alterations in the structure of the peptide exit tunnel may contribute to the observed resistance to gamithromycin and/or tulathromycin in these isolates.

Fig 1.

(A) Alignment of the amino acid sequences of L23 ribosomal proteins from T. thermophilus (PDB accession 1N88) and M. haemolytica (rplW and group_4230). A 14-amino acid insertion in group_4230 is indicated by yellow highlight and the flexible loop of L23 by orange text. Secondary structure features (determined by nuclear magnetic resonance spectroscopy for T. thermophilus or predicted by PSIPRED and DISOPRED3 for M. haemolytica sequences) are indicated above and below the alignment according to the legend. Sequences were aligned using MUSCLE pairwise alignment in Geneious Prime with default settings. (B) Ribbon (left) and space-filling models (right) of the alignment of the NMR structure of T. thermophilus L23 with predicted structures of M. haemolytica rplW and group_4230. The loop structure that is extended by the insertion in group_4230 is indicated on the ribbon diagram.

Mobile genetic element analysis

ICEfinder initially identified 169 putative ICEs among the examined genome sequences (Table S10). After filtering the ICEfinder results as described in Materials and Methods, a total of 68 putative ICE sequences were identified across 66 of the 99 (66%) isolates. At least one putative ICE sequence was identified in 19 of 47 (40%) susceptible isolates and 47 of 52 (90%) phenotypically resistant isolates. Manual examination of these sequences allowed us to consolidate them into three putative ICEs (ICEMh1-like01, ICEMhGAMTULA, and ICEMhSusceptible) based on shared direct repeat sequences, and number and types of cargo genes. For 37 of the putative ICEs, ICEfinder found an insertion site in a ribose-phosphate pyrophosphokinase gene. However, manual inspection of the flanking regions of this gene showed that these ICEs were in proximity to a tRNA-Leu gene, which is the typical insertion site for ICEs in M. haemolytica (18, 66, 67). On this basis, the ICE sequence was extended for all 37 of these putative ICEs to include the section between the boundaries initially identified by ICEfinder and the adjacent tRNA-Leu.

The sequences of all putative ICEs categorized as ICEMh1-like01 were extracted, and a multiple sequence alignment of these sequences was performed to generate a consensus sequence. This process was repeated for ICEMhGAMTULA and ICEMhSusceptible (Fig. 2), and the distribution of the ICEs among the study isolates was assessed (Table 4). The failure to identify any putative ICEs in seven of the resistant isolates may be attributable to lower assembly quality for these isolates as all the genome sequences for these isolates consisted of six or more contigs.

Fig 2.

Putative ICE sequences identified in M. haemolytica isolates in the present study: ICEMh1-like01, ICEMhGAMTULA and ICEMhSusceptible. ICEMh1 (top) is included for reference. Regions containing known ARGs are expanded to show detailed gene organization. ARGs, ICE-associated genes (encoding relaxases, transposases, integrases, and other functions), and genes encoding hypothetical proteins are indicated by color according to the legend. Scale bar indicates 20-kb pairs.

TABLE 4.

Distribution of identified putative ICEs in M. haemolytica isolates in the present study

Phenotype	No. of isolates	ICEMh1-like01	ICEMhGAMTULA	ICEMhSusceptible
Susceptible	47	0	0	19
GAM/TULA/TET/TILM/TILD	27	24	0	0
TULA/TET/TILM/TILD	11	10	0	0
TET/TILM/TILD	5	5	0	0
GAM/TULA	8	0	7	0
GAM/TULA/TILM	1	1	0	0
Direct repeat sequence		cgtgtcggttcgagtccgacc	aaataataatgaaaa	cgtgtcggttcgagtccgacc
Total	99	40	7	19

ICEMh1-like01 was identified in 40 of 52 of the macrolide-resistant isolates (Table 4). This ICE carried the known ARGs including APH(3′)-1 a, APH(3″)-1b, APH(6)-1, estT, sul2, tet(H), and a gene encoding the tet(H) regulator TetR, as well as the copper resistance gene mco. ICEMh1-like01 was nearly identical in gene arrangement and sequence of the ARGs to the previously identified M. haemolytica ICEMh1 in the region containing the ARGs. The portion of ICEMh1-like01 that failed to match to ICEMh1 lacked any of the known ARGs. However, this region did include genes encoding a MobH relaxase, an integrase, and two transposases, which explains why this region was identified by ICEfinder as being part of a putative ICE (Fig. 2; Fig. S2).

ICEMhGAMTULA was found in seven of eight of the isolates resistant to gamithromycin and tulathromycin and was found to carry the known macrolide ARGs msr(E) and mph(E) alongside the sulfonamide resistance gene sul2. The msr(E) and mph(E) genes have previously been identified in ICEMh1-like ICEs; however, this has always been alongside other ARGs such as tet(H) (17, 18, 67). Further, ICEMhGAMTULA was found to share relatively little sequence similarity or synteny with ICEMh1, except for a region containing genes encoding a MobH relaxase, an integrase, and several ICE-associated genes (Fig. 2). The similarity of ICEMhGAMTULA to ICEMh1 in this region suggests that ICEMhGAMTULA is related to ICEMh1 but has since lost other genes carried by ICEMh1. It is also possible that ICEMhGAMTULA is not an ICE but rather an integrated bacteriophage or an IME (44, 68), a suggestion supported by the limited number of T4SS components it carries compared to ICEMh1 (9 vs 14), which potentially inhibits the ability of ICEMhGAMTULA to undergo autonomous conjugative transfer.

ICEMhSusceptible does not carry any known ARGs but does contain genes encoding a MobH relaxase and an integrase, and some ICE-associated genes that are similar to the corresponding genes in ICEMh1 in both arrangement and sequence identity (Fig. 2). ICEMhSusceptible also included the same direct repeat sequence as ICEMh1-like01 and the insertion site of tRNA-Leucine. Due to the similarity in the “core” ICE genes to ICEMh1 and the complete absence of ARGs, it is possible that ICEMhSusceptible represents a degraded version of ICEMh1, a bacteriophage or an IME. This suggestion is supported by the observation that ICEMhGAMTULA contained only 9 T4SS components compared to the 14 T4SS components of ICEMh1, which would potentially limit its ability to undergo autonomous conjugative transfer.

Having established that the known macrolide ARGs msr(E), mph(E), and estT, and the known tetracycline resistance gene tet(H) were all present within an ICE, we next investigated whether any of the resistance-associated candidate genes identified in the multifactorial GLM analysis of the pangenome were also within the boundaries of predicted ICEs. A representative of each ICE was selected (ICEMh1-like01 from isolate MH077, ICEMhGAMTULA from isolate MH017, and ICEMhSusceptible from isolate 36267_S_17) and then inspected to determine whether any of the candidate genes associated with phenotypic resistance (as shown by the multifactorial GLM analysis) were located within the ICE sequence. None of the 150 candidate genes associated with resistance/susceptibility to GAM were identified within the boundaries of any of the putative ICEs. The lack of any putative susceptibility/resistance genes within ICEs may be due to the inclusion of tet(H) as an explanatory factor in the multifactorial GLMs, as tet(H) is known to be strongly associated with the presence of ICE in M. haemolytica. Thus, including tet(H) in our multifactorial GLM would mean that any other genes also associated with the presence of an ICE would fail to be identified. Since macrolide class antimicrobials function by inhibiting translation (64), it is unlikely that the presence of any given gene would confer susceptibility to them. We therefore suggest that future investigations focus on the 79 putative GAM resistance genes identified using our multifactorial GLM analysis.

The identification of an ICE in susceptible isolates suggests that ICEs are widespread in M. haemolytica and that ICEs may confer some beneficial trait to their bacterial hosts other than AMR as, in most cases, ICEs are known to confer a fitness cost to their bacterial host. ICEs are known to contain genes that encode other beneficial phenotypes such as nitrogen fixation, aromatic compound degradation, and phage defense systems; however, it is not clear whether ICEMhSusceptible confers any traits to its host (69–72).

Identification of ICEs containing the macrolide resistance gene estT provides an explanation for the observed resistance to the 16-membered macrolides TILM and TILD. However, it fails to explain the observed resistance to GAM and/or TULA in isolates containing ICEMh1-like01 since the estT-encoded esterase enzyme had no activity on GAM or TULA in a previous report (14). While the L23 ribosomal protein variant we identified partially explains the observed resistance to GAM in the isolates with the GAM/TULA/TET/TILM/TILD phenotype, we still lack an explanation for the observed GAM and TULA resistance in isolates with the TULA/TET/TILM/TILD and GAM/TULA/TILM phenotypes. Further investigation of the 79 candidate resistance genes identified in the multifactorial GLM analysis may offer such an explanation.

Conclusions

Our results reinforce the strong association of ICEs with AMR in M. haemolytica, with 90% of the resistant isolates examined in this study containing a putative ICE. The presence of estT explains the observed phenotypic resistance to the 16-membered lactone ring macrolides TILM and TILD. However, based on previous reports of biochemical activity, it fails to provide an explanation for the observed resistance to GAM and/or TULA, in isolates with resistance to all four macrolides or those that exhibited resistance to TULA, TILM, and TILD. In the isolates that are only resistant to GAM and TULA, the presence of msr(E) and mph(E) explained this observed phenotypic resistance. By quantifying the relationships between the presence/absence profiles of genes in the M. haemolytica pangenome and antimicrobial resistance phenotypes, we identified candidate genes that may contribute to the observed resistance and that should be studied further. We were also able to identify a GAM resistance-associated insertion in the L23 ribosomal protein gene and a macrolide resistance-associated variant in the L22 ribosomal protein; however, neither of these associations has been functionally verified. Finally, we found that ICEs containing no identifiable ARGs are common among susceptible isolates, which suggests that these elements may confer other fitness benefits to M. haemolytica. Taken together, these results highlight that the success of diagnostic sequencing for detecting macrolide resistance will rely not only on the detection of well-established ARGs but also on ongoing monitoring for novel mutations in the cellular targets of these antibiotics.

DATA AVAILABILITY

Assembled genome sequences have been deposited to NCBI under the BioProject accession PRJNA1088094.

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/aem.00502-24.

File S1. aem.00502-24-s0001.txt.

Shell script for identifying L4 ribosomal protein sequences.

File S2. aem.00502-24-s0002.txt.

Shell script for identifying L22 ribosomal protein sequences.

Figure S1. aem.00502-24-s0003.pdf.

Phylogenetic tree of study isolates based on core genome.

Figure S2. aem.00502-24-s0004.pdf.

Synteny analysis of ICE sequences identified in the study.

Supplemental legends. aem.00502-24-s0005.docx.

Legends for supplemental material.

Supplemental tables. aem.00502-24-s0006.xlsx.

Tables S1 to S10.

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.