Antibiotic Susceptibility Profiles and Frequency of Resistance Genes in Clinical Shiga Toxin-Producing Escherichia coli Isolates from Michigan over a 14-Year Period

ABSTRACT

Shiga toxin-producing Escherichia coli (STEC) is an important foodborne pathogen that contributes to over 250,000 infections in the United States each year. Because antibiotics are not recommended for STEC infections, resistance in STEC has not been widely researched despite an increased likelihood for the transfer of resistance genes from STEC to opportunistic pathogens residing within the same microbial community. From 2001 to 2014, 969 STEC isolates were collected from Michigan patients. Antibiotic susceptibility profiles to clinically relevant antibiotics were determined using disc diffusion, while epidemiological data were used to identify factors associated with resistance. Whole-genome sequencing was used for serotyping, examining genetic relatedness, and identifying genetic determinants and mechanisms of resistance in the non-O157 isolates. Increasing frequencies of resistance to at least one antibiotic were observed over the 14 years (P = 0.01). While the non-O157 serogroups were more commonly resistant than O157 (odds ratio, 2.4; 95% confidence interval,1.43 to 4.05), the frequency of ampicillin resistance among O157 isolates was significantly higher in Michigan than the national average (P = 0.03). Genomic analysis of 321 non-O157 isolates uncovered 32 distinct antibiotic resistance genes (ARGs). Although mutations in genes encoding resistance to ciprofloxacin and ampicillin were detected in four isolates, most of the horizontally acquired ARGs conferred resistance to aminoglycosides, β-lactams, sulfonamides, and/or tetracycline. This study provides insight into the mechanisms of resistance in a large collection of clinical non-O157 STEC isolates and demonstrates that antibiotic resistance among all STEC serogroups has increased over time, prompting the need for enhanced surveillance.

INTRODUCTION

Shiga toxin-producing Escherichia coli (STEC), a Gram-negative foodborne pathogen, has been linked to ∼265,000 illnesses every year in the United States (1). Hundreds of STEC serotypes have been identified, though O157:H7 infections have predominated in the United States until recently. Infections caused by non-O157 STEC, particularly the predominant “big six” serogroups (O26, O45, O103, O111, O121, and O145) have steadily increased over time (2, 3). This increase has been attributed in part to changes in laboratory surveillance and detection, as the non-O157 serogroups were not reported to the National Notifiable Disease Surveillance System in the United States until the year 2000 (2).

STEC infections are usually self-limiting, resulting in watery diarrhea, nausea, abdominal pain, and vomiting; however, they can progress to severe infections resulting in hemorrhagic colitis and hemolytic uremic syndrome (HUS) (4). While antibiotics are not recommended to treat patients with STEC due to the increased risk of HUS development (5), antibiotic resistance has been documented in clinical STEC isolates (6–8). Nonetheless, resistance frequencies are likely to be underestimated since susceptibility testing is not routine and not all laboratory-confirmed cases yield isolates for testing. Additional population-based studies are therefore needed to more accurately measure resistance frequencies in clinical STEC isolates recovered from different geographic locations.

The emergence of antibiotic resistance in STEC can be attributed to the extensive use of antibiotics in clinical and agricultural environments. Use of antibiotics in food animals such as cattle, which are the primary reservoir for STEC (9), is likely to be a critical factor in resistance emergence and maintenance. Indeed, resistant STEC have been recovered from animals and foods (7, 10) along with other resistant E. coli populations. This extensive use of antibiotics creates a strong selection pressure on bacterial populations, resulting in the transfer, selection, and propagation of antibiotic-resistant organisms and antibiotic resistance genes (ARGs) (11). The evolution of antibiotic resistance occurs due to mutations in existing genes or the acquisition of mobile genetic elements containing ARGs, which render the activity of antibiotics ineffective against specific types of bacteria. Rapid and reliable tools combined with enhanced surveillance activities are critical to identify resistant pathogens and control the spread and emergence of resistance. Moreover, the use of next-generation sequencing has gained prominence in clinical microbiology (12) and has proven to be useful for the detection of ARGs in bacteria (13, 14).

Here, we sought to determine whether antibiotic-resistant STEC infections have increased in frequency in Michigan over a 14-year period. Because the National Antimicrobial Resistance Monitoring System (NARMS) does not monitor resistance among STEC isolates from Michigan or in any of the non-O157 STEC serotypes in the United States, data are sparse. In addition to estimating resistance frequencies, we used whole-genome sequencing (WGS) to determine the distribution and frequency of circulating ARGs in clinical non-O157 STEC isolates and evaluated its use as an alternative to phenotypic antimicrobial susceptibility testing. Enhancing surveillance of resistant STEC in specific populations is important for the development of region-specific strategies aimed at controlling the dissemination and emergence of resistance in STEC and other bacterial populations.

RESULTS

Antibiotic susceptibility profiles and isolate characteristics.

Among the 969 STEC isolates recovered by the Michigan Department of Health and Human Services (MDHHS) from 2001 to 2014, 62 (6.4%) were resistant to at least one of the three antibiotics evaluated. The greatest number of isolates (n = 56; 5.8%) had resistance to ampicillin, followed by trimethoprim-sulfamethoxazole (n = 22; 2.3%); only one (0.1%) isolate was resistant to ciprofloxacin. Sixteen isolates were resistant to more than one drug class.

Of all 969 isolates, 65.1% (n = 631) were classified as O157, and 34.9% (n = 338) represented non-O157 serogroups; 3 (0.8%) non-O157 isolates were nontypeable (NT). The H-antigen type was determined for only a subset of O157 isolates and, hence, was not included in the analysis. Significantly more non-O157 isolates than O157 isolates had resistance to at least one antibiotic (P = 0.0007), despite their lower frequency (Table 1). This was true for isolates resistant to only ampicillin (P = 0.03) and only trimethoprim-sulfamethoxazole (Fisher’s exact test P = 0.02). Any resistance to ampicillin was observed in 8.6% (n = 29) of the non-O157 isolates and in 4.3% (n = 27) of the O157 isolates, whereas any trimethoprim-sulfamethoxazole resistance was observed in 13 (3.8%) non-O157 and 9 (1.4%) O157 isolates. In all, 16 (1.7%) isolates were resistant to more than one antibiotic; one O102:H6 isolate was multidrug resistant with resistance to ciprofloxacin, ampicillin, and trimethoprim-sulfamethoxazole.

TABLE 1.

Differences in resistance frequencies among non-O157 versus O157 Shiga toxin-producing E. coli (STEC) isolates from Michigan

Antibiotic(s)	No. of resistant isolates (%)		OR (95% CI)a	P value
	Non-O157 (n = 338)	O157 (n = 631)
At least one antibiotic (any resistance)	34 (10.1%)	28 (4.4%)	2.4 (1.43–4.05)	0.0007
Ampicillin only	21 (6.2%)	19 (3.0%)	2.1 (1.13–4.03)	0.03
Trimethoprim-sulfamethoxazole only	5 (1.5%)	1 (0.2%)		0.02b
Ampicillin and trimethoprim-sulfamethoxazole	7 (2.1%)	8 (1.3%)	1.6 (0.59–4.59)	0.49
Ampicillin, ciprofloxacin, and trimethoprim-sulfamethoxazole	1 (0.3%)	0 (0.0%)		0.34b

^aOR, odds ratio; CI, confidence interval.

^bFisher’s exact Chi-square test was used to examine associations when fewer than 5 isolates were included in one cell.

Stratifying the subset of 338 non-O157 isolates by serogroup demonstrated that the proportion of resistance in the “big six” serogroups (n = 27 of 304; 8.9%) was significantly lower than that in all other serogroups combined (n = 6 of 30; 20.0%) (P = 0.05). Among the 304 big six isolates, resistance was most common in O111 (n = 9; 2.9%), O103 (n = 6; 1.9%), and O45 (n = 6; 1.9%), followed by O26 (n = 3; 1.0%), O145 (n = 2; 0.6%), and O121 (n = 1; 0.3%). The non-big six resistant isolates were classified as O102:H6 (n = 1), O123:H11 (n = 1), O71:H11 (n = 1), O21:H2 (n = 1), O8:H7 (n = 1), and O8:H19 (n = 1). One additional resistant isolate was ONT:H45.

Variation in resistance frequencies was also observed when isolates were stratified by stx subtypes. Indeed, STEC strains with stx₁ were more likely to be resistant to at least one antibiotic (n = 28; 8.9%) than strains with stx₂ alone plus those with both stx₁ and stx₂ (n = 34; 5.2%) (odds ratio [OR], 1.8; 95% confidence interval [CI], 1.05 to 2.99). The same trend was observed when resistance to ampicillin and trimethoprim-sulfamethoxazole was examined separately. Strains with stx₁ alone had significantly higher frequencies of ampicillin (n = 23; 7.3%) and trimethoprim-sulfamethoxazole (n = 11; 3.5%) resistance than isolates with stx₂ alone or in combination with stx₁ (ampicillin: n = 33; 5.0%; trimethoprim-sulfamethoxazole: n = 11; 1.7%). This gene profile, however, is likely linked to serogroup, as most non-O157 isolates (n = 302; 95.9%) possessed only stx₁.

Screening for the presence of virulence gene sequences among the 324 non-O157 isolates with WGS data available detected eae (n = 312; 96.3%) and ehxA (n = 305; 94.1%) in most. Additionally, 30 (9.6%) of the eae-positive and 28 (9.2%) of the ehxA-positive isolates were resistant to at least one antibiotic. Among the resistant isolates with eae, subtypes eae_epsilon (n = 13; 43.3%) and eae_theta (n = 10; 33.3%) were most common, while the ehxA subtypes C (n = 17; 60.7%) and F (n = 10; 35.7%) predominated among all ehxA-positive resistant isolates (see Table S1 in the supplemental material).

Resistance trends over time.

Over the span of 14 years, the frequency of resistance to at least one antibiotic increased significantly among all 969 STEC isolates examined (Mantel-Haenszel P = 0.01) (Fig. 1). Significant increases were also observed for both trimethoprim-sulfamethoxazole and ampicillin resistance (Mantel-Haenszel P = 0.02 and P = 0.03, respectively), though yearly fluctuations were observed. Indeed, the frequency of resistance to at least one antibiotic was highest in 2010 (12.3%), followed by 2012 (11.6%). Stratifying by serogroup also indicated changes over time. Among the O157 isolates, for instance, an increasing trend in trimethoprim-sulfamethoxazole resistance was detected (Mantel-Haenszel P = 0.04) (Table S2). No increasing trend was observed among the non-O157 isolates (Table S3), even after limiting the analysis to the big six serogroups.

FIG 1.

Frequencies and trends in resistance in all 969 Shiga toxin-producing E. coli (STEC) isolates recovered from Michigan by year. The orange line shows resistance to at least one antibiotic over the 14-year period. Resistance frequencies to ampicillin (AMP; blue line), ciprofloxacin (CIP; gray line), and trimethoprim-sulfamethoxazole (SXT; black line) are also shown. Significant trends were identified over time using the Mantel-Haenszel χ² test.

A comparison of resistance frequencies between STEC O157 isolates from Michigan (n = 631) to those recovered via NARMS (n = 2,795) (15) also revealed differences. Overall, significantly higher frequencies of ampicillin resistance were observed in Michigan (n = 27; 4.3%) than nationally (n = 74; 2.6%) (P = 0.02) (Fig. 2A). While both ampicillin resistance (Fig. 2B) and trimethoprim-sulfamethoxazole resistance (Fig. 2C) increased in Michigan over the same time frame, only the latter was significant (Mantel-Haenszel P = 0.04). No increase was observed to either drug in the NARMS isolates. Low frequencies of ciprofloxacin resistance were observed in the NARMS isolates with no significant changes over time (Fig. 2D).

FIG 2.

Comparing resistance frequencies between Shiga toxin-producing E. coli (STEC) O157 clinical isolates from Michigan and those from the National Antimicrobial Resistance Monitoring System (NARMS), 2001 to 2014. (A to D) The frequency (%) of (A) any resistance by antibiotic, resistance to (B) ampicillin (AMP), (C) trimethoprim-sulfamethoxazole (SXT), and (D) ciprofloxacin (CIP) by year. *, P = 0.03.

Next, we sought to compare resistance frequencies among the clinical STEC isolates recovered in Michigan to all E. coli isolates from food sources available through NARMS (16). Among the 18,300 food-associated E. coli isolates recovered by NARMS from 2002 to 2017, 24.5% (n = 4,491) were resistant to ampicillin and 3.7% (n = 672) were resistant to trimethoprim-sulfamethoxazole (Fig. S1). While these frequencies were higher than those observed in Michigan, the NARMS database included data from a lengthier time frame and for all types of E. coli, including STEC; hence, the numbers are not directly comparable.

Epidemiological associations with resistant STEC infections.

Factors associated with antibiotic resistance were identified by univariate and multivariate analyses with any resistance to at least one antibiotic as the dependent variable (Table 2). The univariate analysis showed that non-O157 isolates had greater odds of resistance to at least one antibiotic compared to O157 isolates (OR, 2.4; 95% CI, 1.43 to 4.05). Even though non-O157 isolates belonging to serogroups other than the big six were more commonly resistant in this study, the numbers were small (n = 6; 20%). Moreover, the odds of a resistant STEC isolate carrying stx₁ genes was also higher relative to strains carrying both stx₁ and stx₂ (OR, 2.4; 95% CI, 1.28 to 4.66). Multivariate logistic regression controlling for age, sex, and season identified non-O157 versus O157 serogroup to be the only significant predictor of resistant STEC in this analysis (OR, 2.5; 95% CI, 1.46 to 4.21) (Table 3).

TABLE 2.

Univariate associations identifying predictors of resistance in 969 patients with Shiga toxin-producing E. coli (STEC) infections in Michigan, 2001–2014

Characteristic	Total no. of strainsa	No. resistant (%)	OR (95% CI)b	P valuec
Pathogen factors
Serotype
O157	631	28 (4.4)	1.0
Non-O157	338	34 (9.8)	2.4 (1.43–4.05)	0.0007
stx profile
stx1	315	28 (8.9)	2.4 (1.28–4.66)	0.005
stx2	263	19 (7.2)	1.9 (0.97–3.91)	0.05
stx1stx2	391	15 (3.8)	1.0
Demographics and other factors
Residence
Urban	453	23 (5.1)	0.7 (0.39–1.15)	0.14
Rural	501	37 (7.4)	1.0
Age (yrs)
0–18	416	32 (7.7)	1.5 (0.88–2.65)	0.12
19–64	466	24 (5.1)	1.0
≥65	86	6 (6.9)	1.4 (0.55–3.49)	0.49
Sex
Male	430	23 (5.4%)	1.0
Female	520	37 (7.1%)	1.3 (0.79–2.32)	0.26
Antimicrobial drug prescription rates by county
High	272	17 (6.2%)	0.9 (0.55–1.77)	0.97
Low	682	43 (6.3%)	1.0
Cattle density per county
High (>8,400)	466	32 (6.9%)	1.0 (0.53–1.75)	0.90
Low (<8,400)	272	18 (6.6%)	1.0
Season
Winter and spring	260	21 (8.1%)	1.4 (0.83–2.47)	0.19
Summer and fall	709	41 (5.8%)	1.0

^aDepending on the variable examined, the number of isolates does not add up to the total (n = 969) because of missing data.

^bOR, odds ratio; CI, confidence interval.

^cThe P value was calculated with the Chi-square test or Fisher’s exact test for variables with <5 in a cell.

TABLE 3.

Multivariate logistic regression for the data in Table 2^a

Characteristic	OR	95% CIb	P value
Sex: Female	1.4	0.82–2.45	0.21
Age (yrs): 0–18 and ≥ 65	1.6	0.93–2.78	0.09
Season: winter and spring	1.4	0.77–2.39	0.29
Serotype: non-O157	2.4	1.43–4.15	0.001

^aLogistic regression was performed using forward selection while controlling for variables that yielded significant (P ≤ 0.05) and strong (P ≤ 0.20) associations with hospitalization in the univariate analysis. The Hosmer and Lemeshow goodness-of-fit test was insignificant (P = 0.994). All variables were tested for collinearity by analyzing the Eigen values and condition numbers.

^bWald 95% confidence intervals (CI).

Antibiotic resistance gene profiles.

The 321 non-O157 STEC with complete WGS assemblies were evaluated for the presence of ARGs and single nucleotide polymorphisms (SNPs) conferring resistance. Isolates belonging to the big six serogroups O26 (n = 61), O45 (n = 97), O103 (n = 95), O111 (n = 38), O121 (n = 4), and O145 (n = 5) were included, as well as 18 other serogroups (n = 21). In all, four types of chromosomal mutations were observed; two were nonsynonymous SNPs resulting in an amino acid substitution. These mutations were detected in genes conferring resistance to ciprofloxacin and ampicillin as well as colistin and spectinomycin. No mutations conferring resistance to tetracycline, aminoglycoside, macrolide, sulfonamide, or rifamycin were detected. Notably, three isolates had a Ser-83-Leu amino acid change in gyrA, though these isolates were phenotypically susceptible to ciprofloxacin. Eight additional isolates had a Val-161-Gly amino acid change in pmrB, a mutation known to confer resistance to colistin. A SNP in the ampC promoter that has been shown to modify the ampC promoter transcription initiation site, –42 (C→T), was also observed. Furthermore, 32 isolates had a mutation on site 1192 (G→A) in the 16S rRNA gene (16S rrsB), which has been linked to spectinomycin resistance (17).

Thirty-two distinct horizontally acquired ARGs were also detected. These ARGs were categorized based on one of the three main resistance mechanisms—antibiotic inactivation, efflux pumps and cellular protection, or drug target replacement (Fig. 3A). Most ARGs encoded antibiotic-inactivating gene products (n = 17; 53.12%), followed by products that resulted in protection and/or replacement of cellular targets (n = 9; 28.12%) and efflux pumps (n = 6; 18.75%). When stratified by serogroup, the frequency of ARGs classified by each resistance mechanisms varied. Of the 45 non-O157 isolates with at least one antibiotic inactivation gene, for example, most belonged to serogroups O45, O103, and O111 (Fig. 3B). Serogroup O103 and O111 isolates also had more ARGs important for efflux (Fig. 3C) and cellular protection (Fig. 3D). High frequencies of isolates with more than one ARG were also observed. Roughly 12.0% of the 321 isolates carried ≥1 aminoglycoside ARG, and 8.1% carried β-lactam ARGs, while 10.0% and 11.2% carried ARGs conferring resistance to sulfonamides and tetracycline, respectively (Fig. S2). Overall, a high degree of variation was observed in the distribution of horizontally acquired ARGs among the sequenced non-O157 isolates (Fig. 4). Among the 243 ARGs detected, those encoding resistance to the aminoglycosides predominated, followed by the tetracyclines and sulfonamides. With the exception of fosfomycin and the macrolides, two or more ARGs were detected. For example, eight distinct ARGs were detected for aminoglycoside resistance, with antibiotic inactivation genes encoding aminoglycoside O-phosphotransferases predominating (Table 4). A high frequency of genes encoding β-lactam TEM penicillinases was also detected, as well as genes for dihydropteroate synthases and tetracycline efflux that have been linked to sulfonamide and tetracycline resistance, respectively.

FIG 3.

Distribution of horizontally acquired antibiotic resistance genes (ARGs) (n = 32) by their mechanism of action in 321 STEC genomes by serotype. (A) The proportion of genes classified by resistance mechanism. (B to D) The proportion of isolates that carried at least one (B) antibiotic inactivation gene (n = 45), (C) antibiotic efflux gene (n = 36), and (D) cellular protection/target replacement gene (n = 34).

FIG 4.

Diversity and frequency of horizontally acquired antibiotic resistance genes (ARGs) detected in the non-O157 Shiga toxin-producing E. coli (STEC) genomes by antibiotic class. ARGs are indicated for each antibiotic class; the numbers represent the percentage of isolates containing each gene within each class.

TABLE 4.

Horizontally acquired antibiotic resistance genes (ARGs) detected in 321 non-O157 Shiga toxin-producing E. coli (STEC) genomes by antibiotic class^a

Antibiotic class	Description of mechanistic action	Protein family	ARG identified	ARG mechanism of action	No. (%)b
Aminoglycoside	Aminoglycoside N- acetyltransferases	AAC(3) group	aac(3)-IId	Antibiotic inactivation	1 (1.0)
			aac(3)-via	Antibiotic inactivation	2 (2.1)
	Aminoglycoside O-nucleotydyltransferases	ANT(3″)-Ia group	ant(3′')-Ia/aadA1	Antibiotic inactivation	6 (6.2)
			aadA2	Antibiotic inactivation	10 (10.3)
			aadA5	Antibiotic inactivation	2 (2.1)
	Aminoglycoside O-phosphotransferases	APH(3′)-Ia group	aph(3′)-Ia	Antibiotic inactivation	22 (22.7)
		APH(3″)-Ib group	aph(3′')-Ib/strA	Antibiotic inactivation	27 (27.8)
		APH(6)-Id group	aph(6)-Id/strB	Antibiotic inactivation	27 (27.8)
β-lactam	TEM-1 penicillinases	BlaTEM-1	bla TEM-1a	Antibiotic inactivation	2 (7.4)
			bla TEM-1b	Antibiotic inactivation	14 (51.8)
			bla TEM-34	Antibiotic inactivation	1 (3.7)
	β-Lactam resistance AmpC	BlaCMY	bla CMY	Antibiotic inactivation	8 (29.6)
		BlaDHA	bla DHA	Antibiotic inactivation	1 (3.7)
	β-Lactamase class A CARB-1	BlaCARB	bla CARB	Antibiotic inactivation	1 (3.7)
Fosfomycin	Glutathione transferase	FosA fosfomycin resistance protein	fosA	Antibiotic inactivation	2 (100)
Quinoxaline	Multidrug efflux pump	OqxAB	oqxA	Antibiotic efflux	1 (50)
			oqxB	Antibiotic efflux	1 (50)
Macrolide	Macrolide 2′-phosphotransferase I	Mph(A)	mph(A)	Antibiotic inactivation	5 (100)
Phenicol	Chloramphenicol acetyltransferase	CatA2	catA2	Antibiotic inactivation	1 (7.7)
	Chloramphenicol exporter	FloR	floR	Antibiotic efflux	12 (92.3)
Quinolones	Quinolone resistance determinant	QnrA	qnrA	Cellular protection/target replacement	1 (50)
		QnrS	qnrS	Cellular protection/target replacement	1 (50)
Sulphonamide	Dihydropteroate synthase	Sul	sul1	Cellular protection/target replacement	14 (37.5)
			sul2	Cellular protection/target replacement	26 (62.5)
Tetracycline	Tetracycline efflux	Tet(A)	tet(A)	Antibiotic efflux	26 (61.9)
		Tet(B)	tet(B)	Antibiotic efflux	10 (23.8)
		Tet(D)	tet(D)	Antibiotic efflux	1 (2.4)
	Tetracycline resistance protein	Tet(M)	tet(M)	Cellular protection/target replacement	5 (11.9)
Trimethoprim	Trimethoprim-resistant dihydrofolate reductase	DfrA	dfrA1	Cellular protection/target replacement	2 (15.4)
			dfrA12	Cellular protection/target replacement	8 (61.5)
			dfrA17	Cellular protection/target replacement	2 (15.4)
			dfrA23	Cellular protection/target replacement	1 (7.7)

^aThe mechanism of action and specific ARGs are shown by antibiotic class.

^bThe number and percentage of isolates containing each gene within each class.

Correlation between resistance phenotypes and ARG profiles uncovered by WGS.

All non-O157 isolates with phenotypic susceptibility to ciprofloxacin had no known ARGs or chromosomal mutations conferring resistance to ciprofloxacin. While one isolate has a point mutation in gyrA (Ser-83-Leu) and carries the plasmid-associated gene qnr, this isolate was not resistant; the presence of qnr was previously shown to confer reduced susceptibility to quinolones but not at the clinical level (18). The resistance mechanism was not determined for the only ciprofloxacin-resistant non-O157 isolate, which was PCR-positive for stx genes upon isolation but was excluded from the genomic analysis due to our inability to detect stx in silico. Thus, for ciprofloxacin, the major error of genotypic resistance profiling was calculated to be 0 (Table 5). In the case of trimethoprim-sulfamethoxazole, one isolate with phenotypic resistance to trimethoprim-sulfamethoxazole carried a dfr gene but not sul. Therefore, the very major error of genotypic resistance profiling for trimethoprim-sulfamethoxazole was calculated as 0.31%. Notably, three isolates with phenotypic resistance to ampicillin lacked any of the known ARGs that promote resistance to the β-lactams. One of these isolates, however, had a –42 (C→T) mutation in the chromosomal ampC promoter. In contrast, three isolates were phenotypically susceptible to ampicillin yet were found to carry horizontally acquired ARGs such as an AmpC β-lactamase (bla-_CMY) and bla-_TEM genes (bla-_TEM-1b and bla-_TEM-34).

TABLE 5.

Correlation between phenotypic and genotypic resistance among 321 non-O157 Shiga toxin-producing E. coli (STEC) isolates

Antibiotic	No. with resistant phenotype	No. with resistant genotype and resistant phenotype	Very major errora	No. with susceptible phenotype	No. with susceptible genotype and susceptible phenotype	Major errorb
Ampicillin	26	23	3 (0.93%)	295	292	3 (0.93%)
Ciprofloxacinc	0	0	NAe	321	321	0
Trimethoprim-sulfamethoxazoled	12	11	1 (0.31%)	309	309	0

^aA very major error occurs if a phenotypically resistant isolate is genotyped as susceptible. In other words, this error is the failure to detect phenotypic resistance using genotypic methods.

^bA major error occurs if a phenotypically susceptible isolate is genotyped as resistant. In other words, the genotypic tests predict there is resistance when there is none.

^cGenotypic resistance to ciprofloxacin is defined by the presence of at least one SNP in gyrA and a second in gyrA, gyrB, parC, or parE. The presence of plasmid-mediated determinants such as qnr or qepA plus the first SNP in gyrA does not confer clinical resistance, but reduced susceptibility, to quinolones.

^dGenotypic resistance to trimethoprim-sulfamethoxazole is defined as the presence of both dfr and sul.

^eNA, not applicable; no isolates were phenotypically resistant to ciprofloxacin.

Associations between non-O157 sequence types (STs) and antibiotic resistance.

Among the 324 non-O157 isolates with WGS data available, the 7 multilocus sequence typing (MLST) genes were extracted from 323 isolates, grouping them into 28 different STs. The most prevalent lineage was ST-119 (n = 185; 57.3%), followed by ST-106 (n = 102; 31.6%); 19 STs were represented by only one isolate each, highlighting the diversity of the strain population (Fig. 5). It is notable that isolates belonging to the most common lineages, ST-106 (n = 13; 41.9%) and ST-119 (n = 12; 38.7%) were most commonly resistant to at least one antibiotic. Three resistant isolates, however, were among the 15 non-O157 isolates that could not be evaluated by MLST due to poor sequencing quality. Overall, the distribution of resistance by lineage was not significant (Mantel-Haenszel P = 0.58). STs 80, 86, 104, 182, 274, and 343 had only one resistant isolate each.

FIG 5.

Evolutionary relationships among 323 non-O157 isolates representing 28 multilocus sequence types (STs) and correlation with antibiotic resistance phenotypes. The evolutionary relationship was inferred by the neighbor-joining method based on 3,732 nucleotides per sequence. Evolutionary distances were calculated using the p-distance method and represent the number of base differences per site. Branches with >70% bootstrap support using 1,000 replicates are indicated at some nodes. Branches containing strains with ampicillin-resistant isolates are indicated by the red boxes, while blue boxes represent STs with strains containing trimethoprim-sulfamethoxazole resistance. The serotypes represented among each ST, which are noted at the end of each branch, are indicated along with the number (n) of isolates per serotype.

DISCUSSION

The emergence of antibiotic resistance in clinical STEC has been documented in several studies outside the United States (6, 7, 19) and was reported in the novel E. coli O104:H4 outbreak strain linked to unusually high morbidity and mortality rates (20). Resistant STEC have also been recovered from animals and foods (7, 10) along with other resistant E. coli populations. Indeed, the high use of antibiotics in agricultural and veterinary settings is likely to be a major contributing factor to the emergence of antibiotic resistance in STEC populations. Although antibiotics are not recommended for the treatment of STEC (5), empirical use of antibiotics to treat STEC O157 infections has been documented in clinical settings in the United States (21). Nelson et al., for example, reported that 62% of patients with STEC O157 infections were given fluoroquinolones, trimethoprim-sulfamethoxazole, or β-lactam antibiotics, with 29% receiving the drugs after culture confirmation (21). Not only does antibiotic use select for resistance in STEC, but it also enhances the transfer of resistance genes to other bacterial populations residing in the same community (22), which can happen during an infection or colonization of reservoir hosts (e.g., cattle). These findings therefore raise clinically important concerns about the spread of resistant STEC as well as the transfer of resistance genes to commensal populations and opportunistic pathogens.

In this study, we observed an increasing trend in antibiotic resistance among STEC recovered from 2001 to 2014, with higher ampicillin resistance frequencies in O157 isolates from Michigan patients than NARMS (15). Our study is unique, as it is the first to comprehensively examine antibiotic resistance among STEC recovered from this area over a 14-year period. Despite being restricted to Michigan, our population-based data are representative of resistance frequencies and trends across STEC serogroups. These data would not have been captured by national data presented via NARMS, which examines only a subset of O157 isolates from Michigan for susceptibility testing; <50 O157 isolates were tested via NARMS from 2001 to 2004. In addition, NARMS does not screen for resistance in non-O157 STEC isolates.

Earlier studies utilizing older STEC isolates also observed increasing frequencies in resistance among STEC isolates. A CDC study of O157:H7 isolates from 1983 to 1985 observed no resistance to 12 antimicrobial agents (23), while low rates of resistance (2.9%) were reported among O157:H7 isolates recovered from the United States and Canada in a 1988 study (24). Although the increasing frequency of resistance in Michigan from 2001 to 2014 is worrisome, it is important to note that these isolates were recovered through both sentinel and active surveillance. A sentinel surveillance system, which is dependent on data and clinical specimens submitted from sentinel sites voluntarily and not through an active approach, was used prior to 2006 (25). Hence, sentinel surveillance could have contributed to an underestimate of both disease and resistance frequencies in the earlier years, thereby contributing to the trends observed. Similarly, the use of enhanced detection methods for non-O157 STEC (3) could have contributed to an apparent increase in frequency. Regardless of these limitations, however, the frequency of resistance to at least one antibiotic among all STEC isolates recovered only during the active surveillance period increased from an average of 37.3% in 2007 to 2009 to 62.8% in 2010 to 2014.

It is also notable that fluctuating resistance frequencies were observed over time, with increases in certain years, such as those from 2006 to 2007 and 2012 to 2013. This year-to-year variation could be due to the occurrence of outbreaks or case clusters that resulted in an increase in the same resistant strain in certain years. For instance, most of the resistant isolates recovered in 2007 and 2012 belonged to serogroup O157. Because outbreak data were not available for all cases, we cannot rule out the possibility that isolates lacking these data originated from the same source or were part of a cluster. For cases with data available, however, only one case per outbreak was included in the analysis to prevent overestimating the traits of a single isolate. Another explanation for these fluctuations could be seasonal variation in antibiotic usage by year (26) and/or turnover of antibiotic resistant lineages or genes in the environment (27).

It was estimated that 23 million kg of antibiotics are used every year in the United States (28). This high level of use in both clinical and agricultural settings undoubtedly plays an important role in the emergence and maintenance of resistance in bacterial populations. The use of antibiotics in food animals promotes selection of resistant foodborne pathogens that can have negative effects on human health. Such negative effects could be due to the consumption of food contaminated with resistant bacteria, direct contact with animals harboring resistant bacteria, or the mobile transfer of resistant genes to other clinically important pathogens. As a result, it is critical to enhance surveillance of antibiotic resistance in food animals worldwide. Since high frequencies of ampicillin and trimethoprim-sulfamethoxazole resistance were observed in all E. coli isolates from food and animal sources evaluated via NARMS, future studies should focus on further classifying these E. coli isolates, since only a subset represent STEC. Livestock such as cattle, sheep, and pigs are important reservoirs for STEC, and hence, the emergence of antibiotic resistance among STEC residing in animals and the subsequent transmission to humans is very likely. In Michigan, trimethoprim-sulfamethoxazole (Bactrim) and penicillin (Amoxil, Pen V) were the top drugs prescribed to adults and children in 2009 (29), while β-lactams and sulfas have been widely used in both food- and non-food-producing animals (30). Accordingly, the efficacy of these two classes of important antimicrobial drugs is limited due to increasing resistance frequencies. Since many ARGs conferring resistance to these classes of drugs are horizontally transferred, pathogens such as STEC can bring these genes into gut microbial communities where they get transferred to commensals or other clinically important pathogens. Although ciprofloxacin is widely prescribed in adults (29), low frequencies of resistance in STEC were observed. This result may be partly explained by the finding that multiple mutations are required in the E. coli genome to acquire clinically significant levels of resistance (31).

We also observed differences in antibiotic resistance frequencies by serotype. Using multivariate logistic regression, the non-O157 isolates were independently associated with resistance relative to O157. This finding differs from data generated in a Spanish study that observed similar resistance frequencies to 26 agents in both O157 (41%) and non-O157 (41%) isolates (7). Since we previously reported an increasing incidence of non-O157 infections in Michigan from 2001to 2012 (32) and recovered a greater number of non-O157 isolates among 1,096 cattle sampled across 11 Michigan herds in 2011 to 2012 (33), the more recent increases in resistance are likely driven by overall increases in the frequency of non-O157 STEC. Although the reason for variation in resistance frequencies across serogroups is not known, antibiotic use in agricultural environments could more readily select for resistance in common resident bacterial populations such as STEC. Other possibilities include differences in fitness, which was shown for nontyphoidal Salmonella sp. serovars (34), as well as strain variation.

Among the 323 non-O157 STEC genomes examined by MLST, a greater proportion of resistant isolates belonged to STs 106 and 119, representing two of the most dominant non-O157 lineages in this and other strain populations (35, 36). While the higher resistance frequencies in these lineages could be coincidental given their increased prevalence, it could also mean that some lineages more readily acquire resistance than others. Most (95.8%) of the resistant isolates representing STs 106 and 119 (n = 24) also possessed ehxA (ehxA-F or ehxA-C) and eae (eae_theta, eae_epsilon, or eae_beta), which are typically acquired via horizontal gene transfer along with many of the ARGs detected. Regardless, a larger, more diverse sampling of STs and more comprehensive genomics analyses are required before definite conclusions can be drawn. As noted here, the widespread distribution of ampicillin and trimethoprim-sulfamethoxazole resistance across serogroups and on different branches of the phylogeny further illustrate the importance of lateral transfer. Additional studies are needed, however, to determine how factors such as antibiotic prescription rates and antibiotic use, as well as specific evolutionary events and various distributions of STEC lineages and ARGs, affect resistance frequencies in geographically distinct regions.

The presence of horizontally transferred ARGs in STEC is of great significance due to the possibility of transfer to other STEC isolates or opportunistic pathogens in reservoir hosts or the human gut during infection. Considering their ease of transfer and the increasing frequencies of multidrug resistance, coselection of multiple resistance genes is also a major concern. Indeed, the mammalian gut is highly conducive to horizontal gene transfer due to the high density and diversity of microbiota (22), thereby providing an ideal environment for the emergence and persistence of resistance. The detection of ARGs encoding beta lactamases in several non-O157 genomes is concerning. These ARGs, which are often carried on plasmids that result in extended-spectrum beta lactamase (ESBL) production, are of particular concern if they are transferred to other members of Enterobacteriaceae. Indeed, the CDC has classified ESBL-producing Enterobacteriaceae as a “serious antibiotic resistant threat” (37). Nonetheless, additional bioinformatic tools are needed to confirm whether specific ARGs are found on plasmids or other mobile elements among the sequenced non-O157 isolates. More in-depth studies are also needed to determine if resistance enhances STEC virulence, which was shown for other pathogens, such as Pseudomonas aeruginosa and Klebsiella pneumoniae (reviewed in reference 38). Although we did not evaluate whether resistance affects toxin production or virulence in our isolates, prior studies in E. coli have observed no effect due to β-lactamase-mediated resistance with specific ARGs (39, 40).

Surveillance of antibiotic-resistant pathogens has an important impact on designing policies and strategies to control the spread of resistance. The affordability and rapidity of WGS has made it an attractive tool for detecting resistance (41, 42); hence, many studies have characterized resistance by sequencing. One study of 640 nontyphoidal Salmonella isolates, for example, used WGS to classify resistance and found a strong correlation between phenotypic and genotypic testing in 99% of the cases (43). Although WGS has been used to predict resistance in E. coli (44) and other foodborne pathogens, incorporating sequencing data into clinical decision making is hampered by the lack of published studies showing correlations between phenotypic and genotypic resistance. Indeed, the European Committee on Antimicrobial Susceptibility (EUCAST) suggests that the lack of published evidence in different bacteria is a major barrier to using WGS for antibiotic resistance detection in clinical settings (45). The EUCAST also calls for international standardization and quality control metrics for interpreting WGS-based resistance detection (45). Our findings, which indicate a high level of phenotypic and genotypic concordance (Table S4), may also be applicable to other clinically important pathogens. Use of WGS enables the rapid identification of pathogens as well as resistance profiles and mechanisms, which could be important for guiding infection control measures and the surveillance and detection of resistance as it emerges in different pathogens.

It is important to note that some discrepancies were identified between the susceptibility profiles and the presence of ARGs using WGS. Three isolates with phenotypic resistance to ampicillin lacked sequences for known ARGs linked to ampicillin resistance, though one stx₁-positive ST-86 serotype O8:H19 isolate possessed a –42 (C→T) mutation in the chromosomal ampC promoter that could impact susceptibility. Despite the recent observation that mutations important for microbial metabolism can also contribute to antibiotic resistance (46), these novel mutations were not included in the resistance database used for our analysis. Such observations highlight the dependency on a well-curated database and represent an important limitation of WGS, as it can only identify known ARGs and will not detect novel or highly diverse genes. Well-curated databases are critical to obtain high concordance between phenotypic and genotypic results and can improve the sensitivity of genotypic classifications.

In contrast, we also observed three isolates that had phenotypical susceptibility to ampicillin but carried bla genes. A possible explanation for this finding could be that these genes result in low levels of β-lactamase production contributing to MICs that are not classified as clinically resistant. Examination of two ampicillin-susceptible isolates using ampicillin Etest strips, for instance, detected MICs of <3 µg/ml, confirming susceptibility to ampicillin and discordance with the ARG data. We also detected plasmid-mediated resistance genes for quinolones in one isolate that was phenotypically susceptible to ciprofloxacin. The qnrA and qnrS genes, for instance, encode pentapeptides that protect DNA gyrase and topoisomerase IV from the effect of quinolones (47). The presence of these genes could result in small increases in the MIC of quinolones, which are not likely to confer clinical levels of resistance to ciprofloxacin. Because a complete analysis of plasmid sequences was not performed, however, we cannot assume that the plasmids were uniformly sequenced across isolates. Single point mutations in gyrA, which typically confer resistance to fluoroquinolones and quinolones (48), were also detected; however, these isolates were phenotypically susceptible to ciprofloxacin. In Gram-negative bacteria, resistance mutations resulting in amino acid substitutions first occur in the quinolone resistance-determining region (QRDR) of gyrA. Although the initial mutation can result in reduced susceptibility to quinolones, it was demonstrated that additional mutations in parC or gyrA are required to achieve clinical levels of resistance (31). While genes conferring resistance to both colistin and spectinomycin were also detected, phenotypic resistance profiles were not determined for these drugs.

Collectively, these findings highlight the importance of continuous monitoring of STEC infections, resistance frequencies, and trends in different locations. Identifying factors associated with resistant infections is also imperative, as these likely vary across geographic locations and could be impacted by the prevalence of specific lineages or strain types as well as antibiotic usage. The concomitant use of genotypic and phenotypic antibiotic susceptibility testing, especially in clinical settings, remains important to prevent the reporting of false positives or negatives. Although antibiotics are not recommended for STEC infections (5), understanding the correlation between specific ARGs extracted from WGS data and susceptibility profiles can inform treatment decisions for infections caused by other types of E. coli and enteric pathogens. Knowledge of common ARGs in circulation among STEC isolates in different locations can also provide clues about resistance mechanisms in other human pathogens, particularly given that many of these ARGs are transferred horizontally within bacterial populations.

MATERIALS AND METHODS

Study population and STEC isolates.

A total of 977 laboratory-confirmed STEC isolates from 2001 to 2014, were collected through active and sentinel surveillance systems developed by the MDHHS. Isolates were serogrouped and confirmed to be positive for stx by the MDHHS using PCR targeting Shiga toxin gene variants (stx₁, stx₂) as described previously (49). Of the 977 isolates confirmed to be STEC by the MDHHS using PCR and serotyping, 8 were mixed after reisolation at Michigan State University (MSU) and were excluded, leaving 969 STEC isolates available for characterization. Only one representative isolate per outbreak was included to avoid overestimating the frequency of antibiotic resistance and genetic traits in certain years. Outbreak data, however, were not available for all cases. Study protocols were approved by the institutional review board at MSU (IRB 10-736SM) and the MDHHS (842-PHALAB).

Epidemiological data associated with each case were extracted from the MDHHS Michigan Disease Surveillance System (MDSS). Season was classified based on when the case was reported. December, January, and February reporting dates were considered winter. March, April, and May were spring; June, July, and August were summer; and September, October, and November were fall. Some cases were missing a report date; hence, either the date of pathogen isolation or the referral date were used. Using data published by the National Center for Health Statistics (NCHS) (50), only 10 Michigan counties were classified as urban, while cattle densities per county were extracted from a 2019 U.S. Department of Agriculture report (51). Similarly, antibiotic usage data in Michigan were used to classify counties as high or low antibiotic users; a “high” rate had a 30% higher rate than the state average (29). Actual antibiotic usage data were not available for STEC cases in this study, as these data were not collected by the MDHHS during case interviews.

Antibiotic susceptibility testing.

The 969 STEC strains were evaluated for susceptibility to ampicillin (10 µg), ciprofloxacin (5 µg), and trimethoprim-sulfamethoxazole (25 µg), which are commonly used antibiotics for the treatment of E. coli infections (21, 52). The Kirby-Bauer disc diffusion test was performed as described previously (8). Isolates were classified as resistant or susceptible according to Clinical and Laboratory Standards Institute (CLSI) guidelines (53) using E. coli ATCC 25922 as the quality control organism. An isolate was defined as multidrug resistant if it was resistant to all three antibiotics representing the three different antibiotic classes.

WGS of non-O157 STEC.

All non-O157 STEC (n = 338) recovered from 2001 to 2014 were cultured aerobically overnight in Luria-Bertani broth (BD Diagnostics, Franklin Lakes, NJ) and centrifuged for DNA extraction using the Wizard genomic DNA purification kit (Promega, Madison, WI). Libraries were prepped with the Nextera XT kit (Illumina, San Diego, CA) and sequenced using the Illumina MiSeq platform (2 × 250 reads) as described previously (54). After trimming with Trimmomatic (55) and quality checking with FastQC v0.11.7 (56), Spades v3.15.2 (57) was used for de novo genome assembly with parameters described in our prior study (54). Of the 338 non-O157 isolates, 14 were negative for stx by WGS and were excluded from subsequent analyses.

To confirm the non-O157 serogroup determined by the MDHHS, assembled contigs were screened for the wzy and wzx (O-antigen lipopolysaccharide) and fliC (flagellar H-antigen) genes. These genes were extracted from each genome for comparison to databases hosted by the Center for Genomic Epidemiology as previously described (35, 54); isolates with incomplete sequences were classified as nontypeable (NT). Assembled contigs were also screened for the presence of virulence genes encoding Shiga toxin subtypes (stx₁ and stx₂) as well as 14 distinct intimin (eae) subtypes and 6 enterohemolysin (ehxA) subtypes using sequences available in the National Center for Biotechnology Information (NCBI) as described previously (35, 54). Sequences targeting the internal fragments of seven housekeeping genes were extracted to classify the ST or genotype of each isolate using the Whittam MLST scheme (58). The MLST loci include aspC (aspartate aminotransferase), clpX (ATP-dependent Clp protease), fadD (acyl-CoA synthetase), icdA (isocitrate dehydrogenase), lysP (lysine-specific permease), mdh (malate dehydrogenase), and uidA (β-d-glucuronidase). Gene alleles and STs were assigned using the EcMLST database (www.shigatox.net). For all genes, sequences were extracted using in-house bioinformatic scripts to parse results using a command line Basic Local Alignment Search Tool (BLAST) (59). An E value of 0.0001 was set to confirm the specificity of the sequences extracted.

Among the non-O157 genomes, 321 isolates had complete assemblies that could be investigated for ARGs and point mutations conferring resistance using ABRicate v0.8 (https://github.com/tseemann/abricate) and starAMR v0.7.2 (https://github.com/phac-nml/staramr). A comprehensive list of acquired ARG sequences was downloaded from ResFinder v4.1 (cge.cbs.dtu.dk/services/ResFinder/) on 13 May 2021. Thirteen chromosomal genes were examined for point mutations previously shown to be important for antibiotic resistance in E. coli using PointFinder (60) (Table S5).

Data analysis.

Statistical analyses were performed in SAS v9.4 (SAS Institute, Cary, NC, USA) and Epi Info v7. The Mantel-Haenszel χ² test was used for analyzing trends, while the likelihood χ² test was used to identify significant associations between the different variables and resistance. Fisher’s exact test was used for any variable with a sample size of less than 5 per cell. Those variables showing strong associations with antibiotic resistance in the univariate analysis (P ≤ 0.20) were included in the multivariate analysis along with potential confounders such as age and sex. Forward logistic regression was performed in the multivariate analysis to identify factors that were independently associated with antibiotic resistance; a P value of <0.05 was considered significant.

The NARMS Now data set (15) was used to compare resistance frequencies between clinical isolates recovered nationally to those observed in Michigan patients over the same time period. The NARMS Now integrated data set (16), however, was used to compare national resistance frequencies among all E. coli strains isolated from foods and animal sources over the 14-year period to our Michigan STEC isolates. For the latter analysis, the type of E. coli (e.g., STEC) was not indicated, despite recovery from the following food sources: retail chickens, retail ground turkey, retail ground beef, and retail pork chops. Because of this limitation, direct comparisons could not be made between STEC recovered from NARMS and Michigan.

To determine how well the ARGs predicted the phenotypes, the phenotypic antibiotic resistance profiles and the presence/absence of ARGs were converted into a binary (1/0) format. For phenotypic antibiotic susceptibility testing, 1 represented resistance to the respective antibiotic and 0 represented susceptibility to the respective antibiotic. For genotypic testing of antibiotic resistance, 1 and 0 represented the presence or absence, respectively, of each ARG or point mutation in genes resulting in resistance. The accuracy of susceptibility testing methods was determined by comparing genotypic and phenotypic resistance profiles to calculate the “very major error” and the “major error” rates (61).

Data availability.

All non-O157 STEC genome sequences were published previously (35, 54) and have been deposited in GenBank under BioProject numbers PRJNA596289, PRJNA514245, PRJNA218110, and PRJNA368991.