Genomic and Phenotypic Analysis of Heat and Sanitizer Resistance in Escherichia coli from Beef in Relation to the Locus of Heat Resistance

ABSTRACT

The locus of heat resistance (LHR) can confer heat resistance to Escherichia coli to various extents. This study investigated the phylogenetic relationships and the genomic and phenotypic characteristics of E. coli with or without LHR recovered from beef by direct plating or from enrichment broth at 42°C. LHR-positive E. coli isolates (n = 24) were subjected to whole-genome sequencing by short and long reads. LHR-negative isolates (n = 18) from equivalent sources as LHR-positive isolates were short-read sequenced. All isolates were assessed for decimal reduction time at 60°C (D_60°C) and susceptibility to the sanitizers E-SAN and Perox-E. Selected isolates were evaluated for growth at 42°C. The LHR-positive and -negative isolates were well separated on the core genome tree, with 22/24 positive isolates clustering into three clades. Isolates within clade 1 and 2, despite their different D_60°C values, were clonal, as determined by subtyping (multilocus sequence typing [MLST], core genome MLST, and serotyping). Isolates within each clade are of one serotype. The LHR-negative isolates were genetically diverse. The LHR-positive isolates had a larger (P < 0.001) median genome size by 0.3 Mbp (5.0 versus 4.7 Mbp) and overrepresentation of genes related to plasmid maintenance, stress response, and cryptic prophages but underrepresentation of genes involved in epithelial attachment and virulence. All LHR-positive isolates harbored a chromosomal copy of LHR, and all clade 2 isolates had an additional partial copy of LHR on conjugative plasmids. The growth rates at 42°C were 0.71 ± 0.02 and 0.65 ± 0.02 log(OD) h⁻¹ for LHR-positive and -negative isolates, respectively. No meaningful difference in sanitizer susceptibility was noted between LHR-positive and -negative isolates.

IMPORTANCE Resistant bacteria are serious food safety and public health concerns. Heat resistance conferred by the LHR varies largely among different strains of E. coli. The findings in this study show that genomic background and composition of LHR, in addition to the presence of LHR, play an important role in the degree of heat resistance in E. coli and that strains with certain genetic backgrounds are more likely to acquire and maintain the LHR. Also, caution should be exercised when recovering E. coli at elevated temperatures, as the presence of LHR may confer growth advantages to some strains. Interestingly, the LHR-harboring strains seem to have evolved further from their primary animal host to adapt to their secondary habitat, as reflected by fewer genes involved in virulence and epithelial attachment. The phylogenetic relationships among the isolates point toward multiple mechanisms for acquisition of LHR by E. coli, likely prior to its being deposited on meat.

INTRODUCTION

Cooking is an effective way of inactivating bacteria on/in meat, to ensure microbiological safety. In Canada, it is recommended that ground beef be cooked to an internal temperature of 71°C and nonintact beef be cooked to a minimum internal temperature of 63°C, with the meat being turned over at least twice to ensure even heating (1, 2), to mitigate risks from enteric pathogens such as Escherichia coli O157:H7. The bacterium E. coli is generally heat sensitive, with most strains of this species having a D_60°C value (the duration of heat treatment at 60°C required to reduce a bacterial population by 90%) of <2 min (3, 4). However, some recent work has shown that a strain of E. coli (AW1.7) originating from a beef carcass has greatly elevated heat resistance (5), which may lead to its postcooking survival in hamburgers, due to its harboring the locus of heat resistance (LHR), an ∼15-kb genomic island containing 16 open reading frames (ORFs) (6). The D_60°C values for AW1.7 ranged from 3 min to 71 min in different studies (5, 7–9), likely due to different experimental conditions, such as heating methods, growth media, and culture history, used in these studies. The LHR was first described as a gene cluster containing the gene for the novel ATPase ClpK in a nosocomial strain of Klebsiella pneumoniae, located on a conjugative plasmid which also contains genes for resistance to extended-spectrum beta-lactams (ESBL) (10). Subsequent studies have reported chromosomal LHR in E. coli (6, 9, 11). In addition, two transferable loci of heat resistance have been reported for an E. coli strain (FAM21805), one sharing 98 to 99% similarity with the LHR in E. coli AW1.7 (LHR1) and the other (LHR2), ∼19 kb, having a different ORF organization and sharing 88% identity with LHR1 (12).

Based on bioinformatics analysis of all genome assemblies and genome sequences of all E. coli in the NCBI database up to 2015, an overall prevalence of approximately 2% was predicted for the harboring of LHR in E. coli (6). When E. coli isolates from different sources were screened for the presence of LHR by PCR, various levels of prevalence were reported: 0.5% (n = 615) in human clinical isolates (8); 1.97% (n = 1,450) in isolates from cattle, beef, and beef production environment (13); 11.4% (n = 4,123) in isolates from various processing stages of meat of multiple animal species (9); 36.7% (n = 253) in isolates from thermized milk (14); and 59% (n = 70) in isolates from wastewater treatment plants (15). The different sample sizes in different studies may have contributed to the disparity in the prevalence of LHR; however, it is apparent that the different ecological niches may have exerted different selective pressure in the emergence and spread of LHR. In our laboratory, when 59 E. coli isolates collected from enrichment cultures (modified tryptic soy broth supplemented with 20 mg/liter novobiocin [mTSB]) were screened for the presence of LHR, 22 (37.3%) were found to be positive, much higher than that the prevalence of LHR (1.60%) in E. coli isolates from similar environments (meat processing plants) which were recovered by direct plating methods (13).

Most studies in which the prevalence of LHR was investigated used PCR targeting three different fragments in the genomic island (6, 16). However, the presence of LHR, as determined by PCR, does not always confer extreme heat resistance to an E. coli isolate (7, 13, 17), suggesting that the exact composition/number of LHR in a strain and/or other genomic components may also play a role in determining the level of heat resistance in E. coli. In addition, E. coli has a strong ability to adapt to different environments, owing much to its open genome, and this diversification is driven by both phylogenetic background and habitats (18, 19). Information on the evolution of heat-resistant E. coli in the meat processing environments would be of great value for its control and prevention. The objectives of this study were to (i) determine factors contributing to the disparity in heat resistance in LHR-positive E. coli, (ii) determine phylogenetic relationships of E. coli from meat with respect to the emergence of heat-resistant strains, and (iii) determine whether harboring LHR confers to E. coli a growth advantage in mTSB and resistance to sanitizers commonly used in meat processing plants.

RESULTS

Genome assemblies and pangenome.

A total of 1,312,117 to 7,863,795 short sequencing reads were obtained for the isolates, with coverage estimated at >70× for each genome. The assembled draft genomes ranged from 4.6 to 5.3 Mbp, with the GC content being between 50.5% and 51.0% (see Table S1 in the supplemental material). Interestingly, the median size of the LHR-positive genomes (5.0 Mbp) was significantly (P < 0.001) larger than that of the LHR-negative genomes (4.7 Mbp) (Fig. 1). The range of and mean sizes for the genomes were 4.64 to 5.16 Mbp and 4.71 to 5.30 Mbp and 4.77 and 5.02 Mbp for the LHR-negative and -positive isolates, respectively. Nanopore sequencing generated 31,235 to 239,622 long reads for the LHR-positive isolates, resulting in an approximate coverage of >50× for each genome. The GC content was 50.5 to 51.1% (Table S2), which is consistent with the Illumina assemblies. All chromosomal genomes were circularized, with length ranging from 4.74 to 4.99 Mbp. Circularized plasmids, up to seven per genome, were found in 22 genomes but were absent in two of the genomes (CX08 and CX12) (Tables S2 and S3). The GC content for plasmids ranged from 35.7 to 57.9%.

FIG 1.

Box plots comparing the sizes of draft genomes of the LHR (locus of heat resistance)-positive (n = 24) and LHR-negative (n = 18) Escherichia coli isolates recovered from meat. Central horizontal lines and “×” symbols indicate medians and means, respectively; whiskers indicate the lowest data point within the 1.5 interquartile range (IQR) of the first quartile and the highest data point within 1.5 IQR of the third quartile.

For pangenome analysis, E. coli AW1.7 was also included. A total of 10,835 genes were found in the 43 E. coli isolates, of which 3,319 and 7,516 were core (shared by all isolates) and accessory (present in at least one but not all isolates) genes, respectively.

Phylogenetic relatedness based on core genome, accessory genome, and core genome multilocus sequence typing (cgMLST).

The LHR-positive and -negative E. coli isolates clustered separately in the phylogenetic tree based on core genomes (Fig. 2). Overall, 22/24 of the LHR-positive beef isolates included in this study were separated into three clades, with most isolates from combo bin beef trimmings (11/13) (CX04 to CX07, CX11, CX14 to CX16, and CX19 to CX21) in clade 1 and all isolates (6/6) from dressed carcasses before chilling (0H15, 0H18, and 0H23 to 0H26) in clade 2. The third clade, however, contained isolates from various origins: beef products at one facility (CT01) and hide-on carcasses after hide-on wash (AH02 and AH04) and combo bins (CX08 and CX12) at another facility. The heat resistance reference strain, AW1.7, clustered more closely with clade 1 and 2 isolates. The D_60°C values for isolates in clade 2 were largely ≥5 min and for those in clade 3 were largely ≤4 min, while clade 1 isolates had D_60°C values ranging from 3 to 7 min. The two LHR-positive isolates that were not included in the three clades were CT02 and AH21, the latter of which clustered with LHR-negative isolates.

FIG 2.

Core genome phylogenetic tree (A) and heat resistance (D_60°C; B) of the 43 E. coli isolates included in this study, rooted with S. Typhimurium LT2. The tree was constructed based on the concatenated alignment of nucleotide sequences of core genes (n = 1,602) shared by S. Typhimurium LT2 and E. coli isolates. The scale bar in panel A represents substitution(s) per site. A node in the tree is in green if the bootstrap support value is >70. The outgroup branches, 0.37 in length, are truncated to show a better resolution of the right side of the tree (A). E. coli isolates harboring LHR are labeled with one (one LHR) or two (two LHR) asterisks. Phylogroup A and B1 isolates are in purple and blue, respectively. The D_60°C values in panel B for non-combo bin (CX) isolates were produced in our previous study (13).

The phylogenetic relatedness among the isolates determined by core genome alignment is largely congruous with the phylogenetic tree of accessory genomes (Fig. S1), with all clade delineation determined in core genome alignment reflected in the accessory genome tree as well but better defined within clade branches. One notable difference between the relative positioning of isolates was for CT02, which clustered more closely with LHR-positive isolates in core genome tree but with the LHR-negative isolates in the accessory genome tree. The alignment of the LHR-positive genomes against the pangenome was visualized in GView (Fig. 3).

FIG 3.

Genome alignment for the locus of heat resistance (LHR)-bearing E. coli isolates included in this study. The pangenome was constructed using GView by appending unique regions to the seed genome (AW1.7). Gaps indicate regions where genes are missing but are present in other genomes.

The cgMLST analysis showed that the genetic distances among the LHR-positive isolates within clades 1 and 2 were 0 to 7 alleles and 0 or 1 allele (Table 1; Table S4), respectively, indicating that isolates within each clade were very closely related. Among clade 3 isolates, the allele difference ranged from 0 to 115. In this clade, CT01, AH02, and AH04 (0 or 1 allele), and CX08 and CX12 (0 alleles) (Table S4) were closely related, respectively, despite the fact that the former three were recovered from two different beef plants more than 300 km apart. The two lone LHR-positive isolates (AH21 and CT02) were not closely related to any of the other isolates, similar to the LHR-negative isolates except for the LHR-negative combo bin isolates, which differed by 0 to 10 alleles. The genetic distance between LHR-positive and -negative isolates ranged between 622 and 2,163 alleles (Table S4).

TABLE 1.

Subtyping of LHR-positive E. coli isolates

Isolate(s)	Allelic distancea	Serotype	Clermont phylogroup	MLST type (scheme 1)	MLST type (scheme 2)b
Clade 1 (CX04-CX07, CX11, CX14-CX16, CX19-CX21)	0–7	O10:H25	A	635	682
Clade 2 (0H15, 0H18, 0H23-0H26)	0–1	O154:H12	A	399	661
AW1.7	NA	O128:H12	A	6002	Unknown
CT02	NA	O120:H5	A	1072	378
Clade 3 (CT01, AH02, AH04, CX08, CX12)	0–115	O8:H21	A	3202	Unknown
AH21	NA	O−:H34	A	1415	419

^aNA, not available.

^bUnknown, the MLST type was not assigned.

Subtyping of the isolates.

All LHR-positive isolates within a clade were of the same serotype and of the same MLST type by both MLST schemes (Table 1; Table S5), with clades 1, 2, and 3 belonging to serotypes O10:H25, O154:H12, and O8:H21, respectively, and belonging to sequence type 635 (ST635), ST399, and ST3202, respectively, in MLST scheme 1 and to ST682, ST661, and an unknown type, respectively, in MLST scheme 2. CT02, AH21, and AW1.7 were identified as O120:H5, O−:H34, and O128:H12, respectively. They belong to ST1072, ST1415, and ST6002 (MLST scheme 1) and ST378, ST419, and an unknown type (MLST scheme 2).

LHR-negative isolates were mostly of different serotypes, including O−:H34, O−:H7, O−:H9, O141:H4, O149:H2, O174:H45, O3:H21, O8:H19, O8:H₂O, O9:H4, and O93:H16 (Table S5). As with serotypes, LHR-negative isolates had much more diverse ST types, including ST10, ST46, ST58, ST101, ST388, ST398, ST708, ST1415, ST2521, and ST2973 in scheme 1 or ST2, ST87, ST339, ST419, and ST819 in scheme 2. In addition, all 22 LHR-positive isolates belonged to phylogenetic group A, while most LHR-negative isolates (13/18) were determined to be of phylogenetic group B1 (Fig. 2; Table S5).

Characterization of LHR.

The ORFs in the LHR region(s) in each genome were compared against the LHR1 in E. coli AW1.7 and LHR2 in E. coli FAM21805. They were subsequently assigned as LHR1 or LHR2 based on both nucleotide identity and ORF composition (Fig. 4; Table S6). All LHR-positive isolates in this study had a full-length LHR1 containing orf1 to -16 located on the chromosome, with nucleotide identity of 96.9 to 100% with the LHR1 ORFs in AW1.7. The full-length LHR1 all had 62.2% GC content and were all flanked by mobile elements.

FIG 4.

Organization of the open reading frames (ORFs) in LHR1 and LHR2 in representative E. coli isolates. The LHR1 in E. coli AW1.7 and LHR2 in E. coli FAM21805 are included as references. All LHR-positive isolates in this study have a similar composition of a full-length LHR1 (CX16). An additional partial copy of LHR1 was found in CT02. An additional copy of LHR2 (0H15) with a similar composition was found on a plasmid in all clade 2 isolates. The ORF colors are as follows: black, mobile element proteins; gray, ORFs in both LHR1 and LHR2; blue, ORFs unique to LHR2; purple, complete orf11 that was found only in LHR1; red, ORFs of the mcsSIAB cluster in the LHR2 of FAM21805. Partially disrupted ORFs (the remnants of orf11 and orf11r) are in yellow.

In addition, all clade 2 isolates (0H15, 0H18, and 0H23 to 0H26) also had a partial LHR2 containing orf1 to orf3, orf7 to orf11, and orfA to orfC (Fig. 4); the nucleotide identity of these ORFs with the corresponding ORFs in FAM21805 LHR2 was in the range of 99.2 to 100%. A partial LHR (containing orf1 to orf4, orf14, and orf15) was found in one isolate, CT02, which was designated LHR1 as it did not contain ORFs unique to LHR2. However, orf15 of this partial LHR1 shared a much higher identity to LHR2 (97.9%) than to LHR1 (75.8%) (Table S6). All of the second-copy LHR was found on plasmids (Table S3). All clade 2 isolates have largely similar plasmid composition. The partial-LHR2-bearing plasmids (IncHI/pESA2) all have 48.4% GC content and an average identity of >99.9% among them. The GC content of the partial LHR2 was 60.8%. In contrast to the isolates in this study which had a complete copy of LHR on chromosome, the LHR in AW1.7 was located on a plasmid (IncFI), which shared <50% identify with the plasmid bearing the partial LHR1 in CT02 (IncFI/IncY) (Table S3).

All plasmids bearing a partial LHR2 were conjugative, and the partial LHR1 bearing plasmid in CT02 was mobilizable but not conjugative. Interestingly, the LHR-bearing plasmid in E. coli AW1.7 was neither conjugative nor mobilizable.

In the phylogenetic tree constructed using single nucleotide polymorphism (SNP) sites of LHR1, AW1.7 and AH21 were separated from the other isolates (Fig. 5). The relative position of AW1.7 in relation to the clade isolates is different from the core genome tree and accessory genome tree. Alignment of the amino acid sequences of each ORF revealed two distinct differences in these ORFs. A region in orf3 (encoding ClpK, positions 918 to 928; EPEQPDAAKAT) present in clade 1 and 2 isolates as well as in CT02, was missing in the other isolates, including AW1.7 (Fig. S2). FtsH, an ATP-dependent metallopeptidase, has its N- and C-terminal fragments encoded in two ORFs in AW1.7 (orf5, 228 amino acids [aa]; orf6, 47 aa). In all isolates in this study, they were present in one ORF encoding 575 aa except for CT02 (163 aa) and AH21 (429 aa), in which they were truncated at the C terminus. It is noteworthy that a chromosomal copy of the FtsH (644 aa) gene was present in all LHR-positive isolates, including AW1.7, which had 100% identity (amino acid) among them. A number of SNPs were found in AW1.7 LHR. LHR expression in E. coli was dependent on a chromosomal copy of evgA (6). evgA was found in all the isolates harboring LHR, and the amino acid sequences were 100% identical among these isolates. A number of silent mutations of evgA were found in AW1.7, AH21, and CT02.

FIG 5.

Maximum-likelihood tree of LHR1. The tree was constructed based on single nucleotide polymorphisms and rooted with LHR2 in E. coli FAM21805 (KY416992.1). The scale bar represents substitutions per site. The outgroup branches, 1.5 in length, are truncated to improve the resolution of the right side of the tree. A node in the tree is in green if the bootstrap support value is >70.

LHR and virulence factors.

To explore the potential co-occurrence of LHR and virulence factors in the E. coli genomes, the genomes were scanned for virulence genes often found in E. coli. Genes associated with specific E. coli pathotypes were not found in any of the genomes. All isolates had the siderophore enterobactin-encoding genes (entA to entF and entS) and genes for its associated ABC transporter protein (fepA to -D, fepG, and fes) and OmpA, an outer membrane protein involved in general stress response (20) (Table S7). There were some notable differences between LHR-positive and LHR-negative isolates (Table S7). The gene cluster (ecpABCDE and ecpR, also known as yagVWXYZ and ykgK, respectively) encoding the E. coli common pilus (21) was absent in all LHR-positive isolates except AW1.7 and AH21. On the other hand, some components (gspLM) of the gsp operon of the type II secretion system (22) and the gene for the adhesin FdeC (23) were missing in all or most of LHR-positive isolates but were present in most or all LHR-negative isolates (Table S7).

Genome-wide contrasts between heat-resistant and -sensitive isolates other than the LHR genes.

In all, 92 genes with predicted or known functions showed differential presence in the three groups, i.e., those with a D_60°C of ≥4.7 min, those with a D_60°C of <4.7 min and ≥2 min, and those with a D_60°C of <2 min (Fig. 6). Overall, clade 1 and 2 isolates, clade 3 isolates, and isolates with a D_60°C of <2 min had largely distinct patterns, with the first and last group having more contrast than the middle group. AW1.7, AH21, and CT02 had variable patterns, with the former two being closer to the LHR-negative profile than the rest of the LHR-positive isolates. Genes that had greater representation in LHR-positive isolates included those for maintaining plasmids (chpB, higA, higB, repB, ccdB, and pemL), utilization of glycerol under starvation (gldA) and plant-sourced substrates such as 3-(3-hydroxyphenyl)propionic acid (mph cluster) and S-methylmethionine (mmuM and mmuP), for the production of the global regulator ppGpp under stress (diguanylate cyclase [dosC]), and genes located on several cryptic prophages, such as e14, CP4-57, and CP4-6 (Fig. 6A). In addition to the ecp operon described in “LHR and virulence factors,” operons encoding a putative aldehyde dehydrogenase (paoABC), an alternative ribosomal protein expressed under zinc starvation (ykgMO), the toxin-antitoxin system YafQ-DinJ (potentially involved in the cell death process in biofilm formation), p-aminobenzoyl-glutamate hydrolase (abgABR), and homolog proteins (spaQRS) involved in a secretion pathway responsible for the surface presentation of determinants needed for the entry into mammalian cells were underrepresented in the LHR-positive isolates, compared to the LHR-negative isolates (Fig. 6B).

FIG 6.

Differential presence of genes in E. coli isolates with different heat resistance. (A and B) Genes that are overrepresented in heat-resistant (D_60°C ≥ 2 min) and heat-sensitive (D_60°C < 2 min) isolates, respectively. The numbers 0, 1, and 2 are gene copy numbers.

Genomic analysis for potential novobiocin resistance.

None of the DNA gyrase sequences had mutations at sites (D73, G77, I78, R136, and T165) that would lead to novobiocin resistance. Among parE, clsA, cysB, cysE, alaS, argS, ileS, rpoN, and soxR, mutations were identified in alaS. Most clade 1 isolates had G702-to-S and A705-to-T mutations (Fig. S3). However, the A705-to-T mutation was also noted in LHR-negative isolates. No differences were noted between the LHR-positive and -negative combo bin isolates, between LHR-positive and -negative isolates from other sources, or between combo bin isolates and isolates from other sources.

All clade 2 strains and CT02 had two copies of mdtA, while all other strains had one copy (Table S8). The clade 2 strains and an LHR-negative strain (0H11) had three instead of two copies of clsA, as in other strains.

Growth of selected LHR-positive and -negative isolates in mTSB at 42°C.

The growth rates of E. coli in mTSB were found to be different among the strains (P < 0.05) (Table 2), varying from 0.58 (CX10 and AH09) to 0.82 (CT02) log optical density (OD) h⁻¹. The mean growth rate was slightly higher numerically for the LHR-positive group, 0.71 ± 0.02 and 0.65 ± 0.02 log OD h⁻¹. Although strain variation in the maximum OD (MaxOD) values (P < 0.05) was also observed, no correlation with LHR was found (P > 0.05). No lag phase was noted for any of the 14 strains.

TABLE 2.

Growth characteristics of selected E. coli isolates in modified tryptic soy broth supplemented with novobiocin at 42°C^a

Strain	LHR	Min OD	Max OD	Growth rate [log(OD) h−1]
CX10	−	0.02	2.21 b	0.58 ± 0.00 a
AH09	−	0.03	2.48 bc	0.58 ± 0.01 a
CX16	+	0.02	2.69 c	0.62 ± 0.00 ab
0H01	−	0.02	2.45 bc	0.63 ± 0.01 ab
CX01	−	0.01	2.2 b	0.66 ± 0.06 ab
0H17	−	0.02	2.30 bc	0.69 ± 0.01 abc
CX08	+	0.02	2.35 bc	0.70 ± 0.00 abc
0H18	+	0.01	2.22 b	0.70 ± 0.03 abc
AH04	+	0.02	2.30 bc	0.70 ± 0.01 abc
AH14	−	0.02	1.76 a	0.71 ± 0.01 abc
0H24	+	0.01	2.14 ab	0.71 ± 0.03 abc
0H04	−	0.02	2.39 bc	0.71 ± 0.0 abc
AH21	+	0.02	2.37 bc	0.72 ± 0.03 bc
CT02	+	0.02	2.22 b	0.82 ± 0.10 c

^aBoldface indicates LHR-positive isolates. Means that do not share a letter are statistically significantly different (P < 0.05). For comparison among strains using ANOVA, F-test values were <0.0002 and <0.0005 for MaxOD and growth rate, respectively. For comparison of LHR-positive and -negative group using unpaired t test, unpaired-t-test values were 0.546 and 0.081 for MaxOD and growth rate, respectively.

Sanitizer resistance of LHR-positive and -negative isolates.

Considering the meat-processing-plant origin of the isolates and previous reports on association of LHR and oxidative stress resistance in E. coli (15, 24), sanitizer resistance was assessed. The MICs for LHR-positive and -negative isolates were 3.7 ± 0.7 and 3.1 ± 0.0 ppm against the quaternary ammonium compound (QAC)-based sanitizer and 6.5 ± 0.5 and 7.2 ± 1.5 ppm against the peroxyacetic acid (PAA)-based sanitizer, respectively. The values for the LHR-positive and -negative groups were significantly different (P < 0.05), with variations primarily among individual combo bin isolates. When the combo bin strains were compared with non-combo bin strains, the mean MICs were 3.6 ± 0.8 and 3.3 ± 0.3 ppm for QAC and 7.4 ± 1.4 and 6.3 ± 0.0 ppm for PAA, respectively, with the latter pair being statistically significantly different (P < 0.05). None of the slightly elevated MICs was correlated with the difference in efflux-related genes (Table S8). Despite the statistical difference, the differences between the highest and lowest MICs for individual strains against either sanitizer and between the means of the paired groups were well within 2-fold (the sensitivity of the MIC test).

DISCUSSION

Analysis of phylogeny, genetic diversification, and habitat-association of commensal E. coli has demonstrated widespread associations between phylogenetic groups and isolation sources (18). Also, the evolution or diversification of E. coli is shaped by its phylogenetic background and habitat to similar extents through frequent acquisitions and deletions of fragments of DNA. The species has a strong phylogenetic structure, with phylogroups (A, B1, B2, and D) representing the majority of strains and A and B1 strains being predominantly found in humans and domesticated animals, respectively (25). Different phylogroups may have different survival rates in their secondary environments (26). Despite these differences, strains of both groups are often isolated from secondary habitats. In the present study, we analyzed the phylogeny of LHR-positive and LHR-negative E. coli from meat-packing-plant environments, to better understand the microevolution of LHR-bearing E. coli in relation to meat-packing-plant environments and food safety. The LHR-positive E. coli and LHR-negative strains from the same source showed different phylogroup patterns, with LHR-positive isolates belonging exclusively to phylogroup A and most LHR-negative isolates belonging to phylogroup B1. In a few studies where LHR-harboring E. coli strains were phylotyped using the Clermont method, they mostly belonged to phylogroup A and the strains included those recovered from meat (6), municipal wastewater (24), and dairy products (14, 27). The 24 LHR-positive E. coli strains investigated in this study all had a complete chromosomal copy of LHR, and some had an additional partial LHR on a conjugative/mobilizable plasmid. On the other hand, the LHR in E. coli AW1.7 was located not on the chromosome but on a nonmobilizable and nonconjugative plasmid, which was different from its chromosomal location previously reported (6) but consistent with a recent report on the genome of E. coli AW1.7 closed by Illumina short reads and PacBio long reads (28). The combination of long reads and short reads may have afforded a better ability to resolve the structure of the LHR-bearing element in AW1.7 than the previous work. For the isolates in this study, there were two variants of partial LHR which were carried by different types of plasmids (IncHI/pSEA2; IncFI/IncY) and were both different from the E. coli AW1.7 LHR-bearing plasmid (IncFI), despite the immediate origin of these isolates being meat. This suggests that these strains may have acquired their LHR from different origins. Whether the different location of LHR, i.e., chromosomal or plasmid, affects the rate of spread of LHR into new strains warrants further study. Nevertheless, all LHR were flanked by genetic mobile elements, suggesting their transmissible nature.

Interestingly, a much more limited genetic diversity was observed among the LHR-positive E. coli isolates than the LHR-negative isolates, as evidenced by serotyping and by both schemes of MLST and cgMLST. Strains within clade 2 or clade 3 were very closely related, i.e., <10-allele differences (29). Despite the fact that the isolates in clade 3 were from different sources, the genetic distance within the clade, <115 alleles, was much smaller than the differences between LHR-positive and -negative isolates, >622 alleles. In addition, all clade 3 isolates were of the same serotype, despite their different origins and degrees of heat resistance as indicated by D_60°C. These findings together with the clear separation of LHR-positive and LHR-negative isolates by core genome alignment suggest that E. coli isolates of a certain genetic background are more efficient at acquiring and maintaining LHR. This is in line with the observation that epidemiological clones of E. coli tend to emerge from specific phylogenetic groups in the presence of pervasive horizontal gene transfer across the species of E. coli (30, 31). These findings are corroborated with a larger median genome size and overrepresentation of genes carried by cryptic prophages as well as genes involved in plasmid maintenance in LHR-positive isolates than LHR-negative isolates from the same source. The different clades and some lone LHR-positive isolates together with the previous report of genetically different LHR-positive E. coli strains recovered from the same beef processing facility also suggest multiple mechanisms of acquiring LHR by E. coli.

It is undisputable that as a genomic island, LHR confers a high level of heat resistance to its host bacterium; however, the exact functions of the gene products in various forms of LHR are far from clear. Functional analysis suggests that none of the three fragments, the homeostasis module (orf1 to -7), envelope stress module (orf8 to -10), and oxidative stress module (orf11 to -16) of E. coli AW1.7 LHR, when present alone, confers the extreme heat resistance phenotype (6). On the other hand, the stand-alone ClpG in E. coli ST10 and Pseudomonas aeruginosa clone C isolates seems to be deterministic for their extreme heat tolerance (32, 33). In the present study, clonal strains with identical LHR composition based on cgMLST and assigned to the same clade based on core genome alignment differed in D_60°C, suggesting that genes in the accessory genomes in addition to LHR genes may also contribute to the high-level heat resistance in E. coli. This is in agreement with the study by Boll et al. (12) in which the heat resistance of E. coli K-12 containing the two heat resistance loci from FAM21805 was significantly lower than that of the wild-type strain FAM21805, but the two heat resistance loci from C604-10 made K-12 as heat resistant as the wild type C604-10. These findings suggest that background genes may enhance heat resistance in LHR-positive strains in a strain-dependent manner, which could explain the variable heat resistance of LHR-positive strains determined by PCR (6, 7, 13). Then, the relatively low heat resistance of isolate CT02 (D_60°C, 2.2 min), which has a complete copy of an LHR closely related to the LHR in clade 1 and 2 isolates and a partial LHR1, could be attributed to its different accessory genome. This is supported by the different patterns of overrepresented genes in the three heat resistance groups (Fig. 6).

The cgMLST, core genome alignment, accessory genome analysis, and SNP analysis of LHR1 for clade 1 and clade 2 strains suggested recent divergence of these two clades. A marked difference between the two clades was an additional copy of LHR2 in clade 2 carried by the same type of plasmids. The elevated heat resistance of clade 2 isolates could be attributable to orf7 to -10 containing partial LHR2. The additional LHR fragment can contribute to heat resistance, as can an additional full-length LHR (9, 12). One of the major differences between the LHR1 and LHR2 is that the FtsH protease gene (orfB; 609 aa), functional in LHR2, was found to be split and much truncated (orf5 and orf6; 228 and 47 aa) in LHR1 in previous reports (12, 34). The FtsH gene in the LHR of most of isolates in this study is more similar in length (575 aa) and composition to the LHR2 FtsH gene. In addition, orf5 and orf6 in E. coli AW1.7 do not seem to encode any protein product (35). This, together with the universal presence of an identical genomic copy of FtsH in all isolates, suggests that this gene is not essential for heat resistance, and it would be unlikely that the low level of heat resistance of CT02 resulted from the truncation of FtsH (163 aa).

In addition to heat resistance, the LHR in E. coli AW 1.7 also contributes to its resistance to oxidative stress from chlorine and peroxides (24). In the present work, the LHR in various strains did not seem to provide any advantage against the PAA- or the QAC-based sanitizer, as determined by MIC. The mode of action of QAC is through outer membrane lipid bilayer perturbation (36). The findings of similar MICs for various groups E. coli in the present study were consistent with the lack of difference in the efflux pump-related genes in the respective groups. A number of genes on cryptic prophages in E. coli have been reported to be involved in stress response, including resistance to antibiotics such as novobiocin (37–39). None of these genes were overrepresented in LHR-positive isolates, despite the overrepresentation of cryptic prophage genes in general in this group. Sublethal heat shock at 42°C unfolds or misfolds cellular proteins, resulting in protein aggregates if these proteins are not refolded or degraded (40, 41). The sHsp20, ClpK, and sHSP encoded by LHR do not confer the heat resistance phenotype in E. coli AW1.7 but are sufficient to confer the phenotype related to disaggregation (34, 35). These findings suggest that LHR, in particular the protein quality control aspect of it, rather than the resistance to novobiocin may have contributed to the slight growth advantage of LHR-positive strains over the LHR-negative strains in mTSB at 42°C observed in this study. Interestingly, a study by Guragain et al. (9) noted a much higher prevalence of LHR in E. coli from various stages of processing of diverse meat animals, some of which were recovered via an enrichment step at 42°C, than in E. coli from similar sources recovered by direct plating at 37°C (13) and E. coli in general (6). This is also in agreement with the much higher LHR prevalence (36.7%) in E. coli in cheese where thermization of milk is applied (14).

Unlike the municipal wastewater treatment plants, in which LHR-positive E. coli strains emerged through coselection with chlorine resistance (15), the meat plant environments from which the strains used in the present study were isolated may not be the environmental driver for the emergence of these LHR-positive strains. The phylogenetic relationship among the core genomes and LHR in the strains suggests that these strains may have acquired the LHR differently. The underrepresentation of genes for the E. coli common pilus (ECP) and the adhesion factor FdeC, which are in involved in adhesion to epithelial cells (21, 23), in most LHR-positive isolates may suggest that the LHR-bearing strains are further evolved to adapt to survival in secondary environments.

MATERIALS AND METHODS

Bacterial isolates.

A total of 42 generic E. coli isolates were included in this study (Table S1). They were recovered from beef at various stages of processing in beef packing plants in Canada, including E. coli from beef carcasses after a hide-on wash (AH02, AH04, AH06, AH09, AH14, and AH21) and combo bin beef trimmings (CX01 to CX22) in one facility and dressed carcasses before chilling (0H01, 0H04, 0H11, 0H15, 0H17, 0H18, and 0H21 to 0H26) and beef cuts and trimmings (CT01 and CT02) in another facility over a period of 3 years (42–45). Combo bins are corrugated cardboard boxes (96 by 115 by 115 cm) with a plastic lining bag, each containing up to 906 kg of beef trimmings. These isolates included those which were found to be negative (n = 18) and positive (n = 24) for LHR from equivalent sources, as determined by a PCR method (13). All isolates were recovered by direct plating on lactose monensin glucuronate (LMG) and buffered-methylumbelliferyl-β-glucuronide (MUG) agar except for those from combo bins, which were recovered through enrichment in mTSB, due to the low level of contamination of the combo bin beef trimmings with E. coli (45). The non-combo bin LHR-positive isolates (n = 11) were all among the 700 isolates from beef processing plants screened for LHR in a previous study (13). An initial PCR screening of combo bin isolates (n = 59) for LHR was carried out using primers that target all three fragments of LHR, as was done in a previous study (13) which found 22 isolates positive for LHR. The combo bin LHR-positive isolates (n = 13) were selected based on the following criteria: one isolate was selected from each enrichment culture if all five isolates were of the same genotype as determined by multiple locus variable-number tandem repeat analysis (MLVA); one isolate was selected from each MLVA type if more than one MLVA type was found among the five isolates from each enrichment culture (45).

In addition to PCR screening, heat resistance of all isolates selected for this study was also assessed by their D_60°C values. The D_60°C values for non-combo bin isolates have been reported in our previous work (13), and the combo bin isolates were similarly assessed for their heat resistance at 60°C, using a water bath method (46).

Whole-genome sequencing.

All E. coli isolates were subjected to shotgun whole-genome sequencing, and all LHR-positive isolates were also subjected to long-read sequencing to achieve more complete genomes to better understand the structure of LHR (Table S2). The benchmark heat-resistant E. coli strain AW1.7 was included for sequencing, as a closed genome for this strain was not available when this work was carried out.

For short-read sequencing, each isolate was grown in Luria-Bertani broth (LB; BD Difco, Fisher Scientific, Canada) at 35°C with shaking at 80 rpm for 18 ± 2 h. The LB overnight cultures were used for DNA extraction, using a Qiagen DNeasy blood and tissue kit (Qiagen, Toronto, ON, Canada) following the manufacturer's instructions. The DNA of all isolates except for AW1.7 was subjected to shotgun library preparation and subsequent sequencing (150-bp paired-end) using an Illumina HiSeqX platform by Genome Quebec (Montreal, QC, Canada). E. coli AW1.7 was sequenced using an Illumina MiSeq (250-bp paired-end) platform at the Alberta Agriculture and Forestry food microbiology laboratory (Edmonton, AB, Canada).

Oxford Nanopore Technology (ONT) was used for long-read sequencing. Overnight bacterial cultures were subjected to DNA extraction using a MasterPure complete DNA and RNA purification kit (Lucigen, Middleton, WI, USA). DNA concentrations were measured with a Qubit fluorometer (Thermo Fisher Scientific, Ottawa, ON, Canada) using a Qubit double-strand DNA (dsDNA) HS (high-sensitivity) assay kit (Thermo Fisher Scientific). Preparation of genomic DNA library was performed using the ligation sequencing kit SQK-LSK109 (ONT, Cambridge, MA, USA) following the native barcoding genomic DNA protocol. Input of 1 μg of genomic DNA was used for DNA repair and end preparation. A total of 24 samples were multiplexed by native barcoding (EXP-NBD104 and EXP-NBD114). The DNA library was loaded onto an R9.4.1 flow cell following the manufacturer’s instructions and run for ∼45 h on a MinION Mk1B device. Data acquisition was carried out using MinKNOW v20.10.3 software. An additional run was performed (1 flow cell; 3 samples) to include AW1.7 and two other samples which were included in the previous run but did not yield enough coverage to allow the generation of closed genomes. The data from both runs were combined for further data analysis.

Genome assembly and annotation.

For short-read sequencing data, the quality control of raw sequencing reads was performed using FastQC v0.11.8 and MultiQC v1.8 (47, 48). Phix genes in the reads were examined and removed using Bowtie v2.3.4.3 (49). Trimmomatic v0.39 was used to remove adapter sequences and sequences with average quality scores of <20 or lengths of <100 bases (50). Each genome was assembled using SPAdes v3.13.0 using default parameters (51), and the quality of assembled genome was assessed using Quast v5.0.2 (52). Contigs with length of <500 bp or k-mer coverage of <10 were removed using a Python script (53). The remaining contigs were ordered using Mauve v2015-02-13 (54) by referencing the complete genome of E. coli strain K-12 substrain MG1655 (55).

The Nanopore long-read data (raw FAST5 files) were subjected to base-calling using Guppy v4.2.2, and the quality control of sequencing data was performed using PycoQC v1.0.α (https://github.com/a-slide/pycoQC). Adapter trimming was performed using Porechop v0.2.4 (https://github.com/rrwick/Porechop), and reads that were <500 bp and/or had a minimum average read quality score of <10 were removed using Nanofilt v2.3.0 (56). Trimmed reads were assembled using Flye v2.8.2 (57). The resulting contigs were polished with Nanopore-trimmed reads using Medaka v1.2.1 (https://github.com/nanoporetech/medaka) and with Illumina-trimmed reads using Pilon v1.22 (58).

The 16S rRNA gene of each isolate was extracted using Barrnap v0.9 (https://github.com/tseemann/barrnap) and submitted to the Ribosomal Database Project (http://rdp.cme.msu.edu/) (59) to confirm the species identity. Both draft and closed genomes from the two sequencing platforms were annotated using Prokka v1.14.6 (https://github.com/tseemann/prokka).

Conventional subtyping.

The serotype of each isolate was determined using ECTyper v1.0 (https://github.com/phac-nml/ecoli_serotyping) with default settings. The phylogroup (Clermont) of each isolate was determined via in silico PCR using a perl script (https://github.com/egonozer/in_silico_pcr) with the primers designed by Clermont et al. (60). Multilocus sequence typing (MLST) was performed using a command line tool, mlst v2.19.0 (https://github.com/tseemann/mlst), by both the Achtman seven-gene scheme and the Pasteur Institute eight-gene scheme.

Phylogenetic analysis using core genomes and cgMLST.

The draft genomes of the heat resistance reference strain E. coli AW1.7 (6) and the 42 E. coli isolates were included for both analyses. Salmonella enterica subsp. enterica serovar Typhimurium strain LT2 (61) was used as an outgroup for core genome alignment. The pangenome was parsed using Roary v3.13.0 with identity threshold set as 90%. Core genes shared by the 43 E. coli isolates and S. Typhimurium LT2 were concatenated. A maximum-likelihood tree was constructed using RAxML v8.2.12 (62) with the general time-reversible gamma nucleotide model (GTRGAMMA) and bootstrap analysis for 1,000 iterations. The tree was annotated using the R package ggtree (63). To perform cgMLST analysis, ChewBBACA v2.5.6 was used with default settings and the Escherichia schemes (2,513 genes; http://enterobase.warwick.ac.uk/schemes/). The pairwise allelic distance among E. coli isolates was calculated using cgmlst-dists v0.4.0 (https://github.com/tseemann/cgmlst-dists).

Characterization of LHR.

The E. coli AW1.7 LHR (LHR1) and FAM21805 LHR (LHR2) ORFs were scanned against the nucleotide sequences of each isolate obtained from long-read sequencing using ABRicate v1.0.1 (https://github.com/tseemann/abricate), with coverage threshold set at 80% and a relatively low identity threshold set at 70%, to minimize false-negative results. To explore the phylogenetic relatedness of LHR1 found in the isolates in this study and in AW1.7, the nucleotide sequences of LHR1 (from the start codon of orf1 to the stop codon of orf16) were aligned using MAFFT v7.475 (64). The single nucleotide polymorphism (SNP) sites in the core regions of LHR1 were extracted using Gubbins v2.4.1 (65), based on which a maximum-likelihood tree was constructed using RaxML as described before. The LHR2 of FAM21805 (12) was used as an outgroup to root the tree.

In addition, the 42 E. coli isolates and AW1.7 were arbitrarily assigned into three groups according to their degree of heat resistance: D_60°C ≥ 4.7 min, 4.7 min > D_60°C ≥ 2 min, and D_60°C < 2 min. The former two groups were all LHR positive, whereas the latter group were all LHR negative. Genes specific to each group were analyzed using the following criterion: presence in >90% of isolates in one group and <10% of isolates in either one of the other two groups.

Screening for virulence factors and classification of plasmids.

To explore the potential co-occurrence of LHR and virulence factors in E. coli, the nucleotide sequences of each genome were scanned for virulence genes using ABRicate with the virulence factor database (VFDB; http://www.mgc.ac.cn/VFs/main.htm) as references. Plasmid sequences were identified using PlasmidFinder (66), and LHR-bearing plasmids were classified into three categories (conjugative, mobilizable, and nonmobilizable) according to the protein machinery associated with DNA transfer, using Plascad v1.17 (67).

Determination of genes potentially related to novobiocin resistance.

Two major differences set mTSB, the commonly used enrichment broth for E. coli, and other nonenrichment media apart: the addition of novobiocin and elevated incubation temperature of 42°C. Novobiocin is a large lipophilic molecule and as such likely crosses the outer membrane by diffusion. Members of the resistance-nodulation-cell division (RND) family are the high-efficiency efflux systems in Gram-negative bacteria for extrusion of a broad range of compounds (68). It has been demonstrated that the MdtABC-Tol and the AcrAB-Tol efflux systems in the RND family are involved in novobiocin extrusion (69, 70). Genes encoding the two RND efflux systems were compiled (Table S8). The presence of these genes in E. coli isolates was investigated by searching the pangenome files parsed using Roary.

Novobiocin can interact competitively with ATP in B subunit of DNA gyrase (GyrB), inhibiting the ATPase and supercoiling of DNA (71). Mutation at R136 (72), D73, G77, I78, and T165 of GyrB (73) may lead to novobiocin resistance in E. coli. In addition to DNA gyrase, mutations in genes encoding topoisomerase IV and some tRNA synthetases (parE, clsA, cysB, cysE, alaS, argS, ileS, rpoN, and soxR) may also lead to increased novobiocin resistance in E. coli (74–78). Those sequences were aligned against the respective amino acid sequences in E. coli K-12 to identify meaningful mutations using MAFFT v7.475 (79). For comparison, the draft genomes of E. coli O111 strains which have been reported to be resistant (00-4748, TB226A, and 95-0586) or susceptible (98-8338) to novobiocin were also included in this analysis (80, 81).

Growth of selected E. coli isolates in mTSB.

To determine whether harboring LHR favors the growth of E. coli in mTSB (TSB plus 20 mg/liter novobiocin) at the recommended incubation temperature of 42°C, the growth of a collection of 14 E. coli strains (7 LHR positive and 7 LHR negative) representative of their positioning on the phylogenetic tree was assessed. Each strain was grown in 10 ml TSB at 35°C overnight with shaking at 80 rpm. The overnight culture (100 μl) was inoculated into 10 ml of mTSB. Aliquots (200 μl) of each culture were dispensed into a 96-well Nunc microplate. The optical density of the cultures at 600 nm (OD₆₀₀) was monitored at 20-min intervals for 24 h using a POLARstar Omega plate reader (BMG Labtech, Germany) held at 42°C. To minimize condensation, the microplate cover was treated with 0.05% Triton X-100 in 20% ethanol by adding 3 ml of the mixture to the lid and tilting it to ensure that the solution covered the lid completely (82). After 30 s, the solution was poured off and the lid was left to air dry. The plate was shaken for 120 s at 100 rpm before each measurement. After the run, the OD readings were subsequently blank-corrected. Two independent replicates, each including three technical replicates for each isolate (n = 6), were conducted.

Growth rates were calculated by fitting the growth curves with the dynamic Baranyi model (83) using DMFit v3.5 (https://www.combase.cc/index.php/en/8-category-en-gb/21-tools). The minimum and maximum OD values (MinOD and MaxOD), representing the initial and the maximum population density, were calculated as averages of the three lowest and highest OD values, respectively, of the growth curve from each replicate. An analysis of variance (ANOVA) was performed using R v4.0.4 to compare differences in growth rates and MaxOD. If the F-test result was significant (P < 0.05), a Tukey post hoc test was performed to identify pairwise differences between strains, and an unpaired t test was applied to compare the LHR-positive and LHR-negative E. coli populations.

MICs of sanitizers.

All 42 E. coli isolates were evaluated for their susceptibility to two commonly used sanitizers, E-SAN and Perox-E, which are a quaternary ammonium compound (QAC; Epsilon Chemicals, Edmonton, Alberta, CA) and a peroxyacetic acid (PAA)-based sanitizer (Epsilon Chemicals), respectively. They are referred to as QAC and PAA sanitizers in this work. The susceptibility was assessed by determining the MIC for each sanitizer against each E. coli isolate using a microplate assay, as described previously (84). Briefly, sanitizers were prepared as serial 2-fold dilutions in sterile Millipore water. Overnight cultures of E. coli diluted in LB (10⁵ to 10⁶ CFU/ml) were mixed in equal volumes (100 μl) with dilutions of the sanitizers in a 96-well Nunc microplate. Plates were incubated at 35°C for 24 h. The OD₆₀₀ before and after incubation was determined using the POLARstar Omega microplate reader. The MIC was defined as the lowest concentration of a sanitizer preparation at which there was no visible growth after the overnight incubation (85), for which an arbitrary cutoff of OD₆₀₀ of 0.1 was used. For each isolate and each sanitizer, the assay was conducted in two independent replicates each consisting of two technical replicates (n = 4), from which mean MICs were calculated. Isolates were grouped according to origin (combo bin and non-combo bin isolates), carriage of LHR, and method of isolation (enrichment and direct plating). Differences in mean MICs were assessed using one-way ANOVA, with a significance level of 0.05.

Data availability.

All sequencing reads and genome assemblies for the 42 E. coli isolates and E. coli AW1.7 obtained in this study were deposited to the NCBI database under BioProject no. PRJNA716667.