Genomic Epidemiology of Invasive Methicillin-Resistant Staphylococcus aureus Infections Among Hospitalized Individuals in Ontario, Canada

Abstract

Background

Prevention and control of methicillin-resistant Staphylococcus aureus (MRSA) infections remain challenging. In-depth surveillance integrating patient and isolate data can provide evidence to better inform infection control and public health practice.

Methods

We analyzed MRSA cases diagnosed in 2010 (n = 212) and 2016 (n = 214) by hospitals in Ontario, Canada. Case-level clinical and demographic data were integrated with isolate characteristics, including antimicrobial resistance (AMR), classic genotyping, and whole-genome sequencing results.

Results

Community-associated MRSA (epidemiologically defined) increased significantly from 23.6% in 2010 to 43.0% in 2016 (P < .001). The MRSA population structure changed over time, with a 1.5× increase in clonal complex (CC)8 strains and a concomitant decrease in CC5. The clonal shift was reflected in AMR patterns, with a decrease in erythromycin (86.7% to 78.4%, P = .036) and clindamycin resistance (84.3% to 47.9%, P < .001) and a >2-fold increase in fusidic acid resistance (9.0% to 22.5%, P < .001). Isolates within both CC5 and CC8 were relatively genetically diverse. We identified 6 small genomic clusters—3 potentially related to transmission in healthcare settings.

Conclusions

Community-associated MRSA is increasing among hospitalized individuals in Ontario. Clonal shifting from CC5 to CC8 has impacted AMR. We identified a relatively high genetic diversity and limited genomic clustering within these dominant CCs.

antimicrobial resistance, genotyping, methicillin-resistant Staphyloccoccus aureus, population genomics, whole-genome sequencing

(See the Editorial Commentary by Long, on pages 1943–5.)

Infections caused by methicillin-resistant Staphylococcus aureus (MRSA) are a major cause of morbidity and mortality worldwide [1, 2]. The initial strong association of these infections with hospital and other healthcare settings gave rise to the epidemiological concept of healthcare-associated MRSA (HA-MRSA). As infections with this drug-resistant pathogen increased among healthy individuals without prior healthcare contact, the epidemiological-based term community-associated MRSA (CA-MRSA) was coined [3]. In Canada, national surveillance identified CA-MRSA in 23% of hospitalized individuals with MRSA [4], and one province with population-level estimates reported a doubling of the incidence of laboratory-confirmed MRSA infections associated with the introduction of CA-MRSA [5].

Healthcare-associated MRSA has traditionally been differentiated from CA-MRSA based on epidemiological information, and MRSA isolate characteristics such as antimicrobial resistance (AMR) patterns and genotyping [6]. Community-associated MRSA-specific and HA-MRSA-specific strain populations have been defined using genotyping methods such as staphylococcal protein A (spa) typing, pulsed-field gel electrophoresis (PFGE), and multilocus sequence typing (MLST), alone or in combination with staphylococcal cassette chromosome mec (SCCmec) typing and detection of virulence genes such as Panton-Valentine leukocidin (PVL) [7, 8]. The use of MLST initially demonstrated strong associations of (1) clonal complex (CC)5 strains with epidemiologically defined HA-MRSA and (2) CC8 strains with CA-MRSA. In North America, these 2 MRSA lineages are particularly common [9], whereas other CCs, such as CC59, that dominate CA-MRSA in East Asia, are seen less frequently [10].

As the epidemiology of MRSA continues to change, the strain characteristics (eg, genotypes) that initially distinguished HA-MRSA from CA-MRSA have blurred. Several studies have shown that genotyping methods alone are no longer able to differentiate HA- and CA-MRSA [4, 11, 12]. In this study, to understand MRSA epidemiological trends among hospitalized individuals in the Canadian province of Ontario, we examined MRSA isolates collected at 2 time points (2010 and 2016) by sentinel hospitals across the province. An integrative approach was applied, combining case-level clinical and demographic data with MRSA molecular typing and genomic results. We report (1) increased hospitalizations of individuals with CA-MRSA and (2) a clear shift in the strains causing invasive MRSA infections from CC5 to CC8 and other emerging CCs—ultimately impacting AMR patterns. We also show that the dominant CCs are genetically diverse, and that within hospitals we detected minimal clustering indicative of recent nosocomial transmission.

MATERIALS AND METHODS

Study Design and Clinical Data Collection

The Canadian Nosocomial Infection Surveillance Program (CNISP) conducts country-wide prospective surveillance of laboratory-confirmed MRSA infections in hospitalized individuals in sentinel hospitals [13]. In this study, we analyzed data and isolates from MRSA infections occurring in 2010 and 2016 in CNISP hospitals from the province of Ontario. The medical records of newly identified MRSA cases were reviewed for relevant clinical and epidemiological information. Infections were categorized by infection control using CNISP standard criteria [14] as healthcare-associated—when there was no indication of MRSA infection at the time of admission, or if there was evidence to support MRSA acquisition in another healthcare setting in the previous 90 days. Cases were classified as community-associated when there was no prior healthcare exposure and they did not meet the criteria for HA-MRSA. Hospitals were assigned numeric identifiers (H# for a single hospital or HG# for a hospital group where there were multiple hospitals under the same administration). For simplicity, we refer to both hospitals and hospital groups as “hospitals.” The primary infection source was categorized as skin or soft tissue, pulmonary, surgical site, medical device, urinary, other site (eg, joint), or bacteremia without focus when the source of infection could not be determined. A 30-day, follow-up period was used to record all-cause mortality. The CNISP MRSA surveillance is within the scope of routine hospital quality control activities for which Research Ethics Board approval is not required.

Isolates and Antimicrobial Susceptibility Testing

Standard laboratory methods were used to identify MRSA isolates at participating Ontario CNISP hospitals. For each included MRSA infection, 1 isolate was forwarded to a central laboratory for further molecular and antimicrobial susceptibility testing to oxacillin (OXA), erythromycin (ERY), clindamycin (CLD), ciprofloxacin (CIP), fusidic acid (FUS), tetracycline, trimethoprim-sulfamethoxazole, gentamicin, mupirocin (MUP), and rifampin. Antimicrobial susceptibilities were tested by broth microdilution (or, for inducible clindamycin, by D-test) in accordance with the Clinical Laboratory Standards Institute (CLSI) guidelines [15]. For MUP, we defined (1) low-level resistance by minimum inhibitory concentration (MIC) values of 8–256 µg/mL and (2) high-level resistance by MICs ≥512 µg/mL. An MIC value ≥2 µg/mL was used to define FUS resistance. We defined multidrug resistance as resistance to ≥3 antimicrobial classes in addition to beta-lactams.

Whole-Genome Sequencing and Genotying

To prepare deoxyribonucleic acid (DNA) for genome sequencing, bacterial culture was added to 800 μL TE buffer with micro-organism lysing 0.1 mm glass beads (Bertin Corp., Rockville, MD) and shaken in a Precellys24 tissue homogenizer (Bertin Corp.) for 10 seconds at 2500 rpm. Samples were then centrifuged at 13 000 rpm for 3 minutes, 200 μL supernatant was added to 200 μL AL buffer (QIAGEN, Hilden, Germany), and DNA was extracted from this pretreated sample using the QiaAMP Blood kit (QIAGEN) according to manufacturer’s directions. Quantification of extracted DNA was performed with the dsDNA Broad Range Assay on a Qubit 2.0 Fluorometer (Thermo Fisher Scientific, Carlsbad, CA). Fragment length was estimated using the Agilent High Sensitivity DNA kit (Agilent Technologies, Santa Clara, CA). Genomic libraries were prepared with Nextera XT Library Prep Kits (Illumina, San Diego, CA) and sequenced as multiplexed libraries on the Illumina HiSeq4000 platform as paired-end (2 × 125 base pair [bp]) reads by the McGill University and Génome Québec Innovation Centre (Montreal, QC, Canada). Strains were demultiplexed using onboard software. Sequence data for all genomes are available at the National Center for Biotechnology Information Sequence Read Archive (BioProject PRJNA595884).

Genotyping methods including spa typing, PFGE, and SCCmec typing were performed as previously described [16, 17]. spa types were determined using BioNumerics version 5.1 (Applied Maths, Inc., Sint-Martens-Latem, Belgium), which automatically synchronized repeat and spa type signatures with the SeqNet/Ridom Spa Server (http://spaserver.ridom.de), and the Canadian MRSA typing database was used to assign epidemic PFGE designations [16]. The sequence type (ST) of each isolate was determined directly from whole-genome sequencing (WGS) data using MOST [18] and the S. aureus MLST database (https://pubmlst.org/saureus). Clonal complex was assigned based on the MLST database CC designations and previous studies [19–21]. We reclassified the original MLST database CC assignments of ST6 (CC5) and ST72 (CC8) to CC6 [21, 22] and CC72 [21], respectively. Novel STs sharing ≥5 alleles to the MLST pattern of an established CC were considered to belong to that CC. Presence of the pvl gene was determined directly from the WGS data using a custom SRST2 [23] database containing a 433-bp pvl sequence described previously [24].

Whole-Genome Sequencing-Based Phylogenetic Analysis

Newly generated Illumina short-reads and reads from a selection of publicly available genomes (Supplementary Table S1) were analyzed using a custom bioinformatics pipeline (SNPensemble) (Supplementary Materials). In brief, reads were mapped to appropriate CC5 or CC8 reference genomes (GenBank accession nos. CP000736.1 and NC_010079.1, respectively), and single-nucleotide polymorphisms (SNPs) were identified. The SNPs in the core genome were then used to generate maximum likelihood phylogenetic trees using FastTree 2.1.8 [25]. Isolates within the same genomic cluster, defined using a 10-SNP threshold [26], were given a unique cluster identifier. Detailed methods are available as Supplementary Materials (Appendices and Supplementary Table S2).

Statistical Analysis

All analyses were performed in R (v3.5.1). Differences across the study periods for categorical variables were evaluated by χ ² or Fisher’s exact test. Genotypic diversity was calculated using Simpson’s index of diversity (D) [27]. We used univariable logistic regression to identify clinical and epidemiological factors associated with CC (CC5 vs CC8) and calculated odd ratios (ORs). Multivariable logistic regression including sex and age, along with variables associated with CC in univariable analysis (included if P ≤ .2), was used to estimate the adjusted OR (aOR) and 95% confidence intervals (CIs). A complete-case analysis strategy (excluded records with missing data: n = 4 [1.1%]) was used, with stepwise backward selection of variables following Akaike Information Criterion minimization [28]. Pairwise SNP distances were compared using the Mann-Whitney U test.

RESULTS

Clinical Characteristics of Methicillin-Resistant Staphylococcus aureus Infections in Ontario

A total of 426 confirmed MRSA infections (212 identified in year 2010 and 214 in year 2016) with isolates submitted by CNISP hospitals for further testing were included in our analyses (Supplementary Figure S1), accounting for 32.6% of the total MRSA cases diagnosed in these hospitals during these 2 years. Each case was represented by a single isolate. The largest number of cases were observed in persons ≥60 years of age (55.3%) and in males (64.6%) (Table 1), and there were no significant differences in the distribution of age or sex in the 2 study years. Relative to 2010, infections in 2016 were more likely to be nonsurgical site skin and soft tissue infections and less likely to be medical device associated. The number of infections determined to be healthcare associated decreased from 139 (65.6%) in 2010 to 97 (45.3%) in 2016, whereas the proportion determined to be community associated increased from 23.6% to 43.0%.

Table 1.

Demographic and Clinical Characteristics of the Study Sample Overall and by Year, Ontario, Canada

Characteristics	Total n (%)	Year		P Valuec
		n (%)a
		2010b	2016b
Overall	426 (100)	212 (49.8)	214 (50.2)
Age Groupd, Years
0–4	34 (8.0)	15 (7.1)	19 (8.9)	.321
5–19	16 (3.8)	5 (2.4)	11 (5.1)
20–39	39 (9.2)	16 (7.6)	23 (10.7)
40–59	101 (23.8)	51 (24.2)	50 (23.4)
60+	235 (55.3)	124 (58.8)	111 (51.9)
Sex—male	275 (64.6)	145 (68.4)	130 (60.7)	.121
Isolate Type
Blood	243 (57.0)	101 (47.6)	142 (66.4)	<.001
Nonblood	183 (43.0)	111 (52.4)	72 (33.6)
Primary Infection Source
Skin or soft tissuee	133 (31.2)	60 (28.3)	73 (34.1)	<.001
Pulmonarye	66 (15.5)	41 (19.3)	25 (11.7)
Surgical site	58 (13.6)	35 (16.5)	23 (10.7)
Medical device	57 (13.4)	32 (15.1)	25 (11.7)
Urinary	28 (6.6)	21 (9.9)	7 (3.3)
Other sites	33 (7.7)	12 (5.7)	21 (9.8)
Bacteremia without focus	51 (12.0)	11 (5.2)	40 (18.7)
BSI 30-Day Outcomef
Survived	185 (76.1)	75 (74.3)	110 (77.5)	.110
Location of Acquisition
Hospital/healthcare	236 (55.4)	139 (65.6)	97 (45.3)	<.001
Community	142 (33.3)	50 (23.6)	92 (43.0)
Unknown	48 (11.3)	23 (10.8)	25 (11.7)

Abbreviations: BSI, bloodstream infection.

^aPercentages have been rounded and may not total 100%.

^bParticipating hospitals in 2010 (n = 15) and 2016 (n = 16).

^cχ ² test for statistically significant difference between the years 2010 and 2016.

^dData unavailable for 1 case in year 2010; therefore, the age group denominator is reduced accordingly.

^eReported in 2010 were necrotizing fasciitis (n = 1) and necrotizing pneumonia (n = 2).

^fOutcome for individuals with BSIs only (n = 243).

Methicillin-Resistant Staphylococcus aureus Genotype Replacement Across Study Periods

Although 15 different MLST CCs were identified, including some unique to each study period, the vast majority (n = 370, 86.9%) of isolates belonged to CC5 or CC8. Clonal complex 5 isolates decreased from 59.9% in 2010 to 45.3% in 2016 (P = .004), whereas CC8 isolates increased from 27.4% in 2010 to 41.1% in 2016 (P = .004) Table 2. There was a 100% increase in the number of infections caused by CC59, and a decrease in CC30 isolates from 2.8% (n = 6) in 2010 to 0.5% (n = 1) in 2016, albeit the numbers remain small compared with CC5 and CC8. We also observed an increase in the number of STs within CC8 in 2016—4 of which represented novel STs—suggesting ongoing diversification of CC8 (Figure 1 and Supplementary Table S3). The clonal shift among study periods was also apparent by other genotyping methods (Table 2). Staphylococcal cassette chromosome mec types changed over time, with type II—which dominated (58.6%, 123 of 210) in 2010—representing only 25.4% (54 of 213) of isolates in 2016, whereas type IV increased to represent 69.5% (n = 148) of all SCCmec types in 2016. This was mostly due to a shift in SCCmec types within CC5 (Supplementary Figure S2). Of 74 spa types identified, the dominant genotypes were t002 (n = 158 of 421, 37.5%, 157 of which were CC5) and t008 (n = 118 of 421, 28.0%, all were CC8) (Supplementary Table S4). Examining all markers together for the study population revealed a relatively high genetic diversity (D = 0.897) with 123 distinct genotypes. The 2 most common genotype profiles were CC5-ST5-t002-CMRSA2(USA100, USA800)-SCCmec-II-PVLneg (92 of 426; 81.5% were in healthcare-associated infections) and CC8-ST8-t008-CMRSA10(USA300)-SCCmec-IV-PVLpos (84 of 426; 56.0% were in community-associated infections) (Supplementary Table S5).

Table 2.

Molecular Characteristics of Methicillin-Resistant Staphylococcus aureus Clonal Complexes by Year for Isolates Collected in Sentinel Hospitals in Ontario, Canada (n = 426)

Characteristics	Year		P Valueb
	n (%)a
	2010	2016
Clonal Complex 5	127 (59.9)	97 (45.3)
PFGE Epidemic Typec
CMRSA2 (USA 100/800)	127 (100.0)	96 (99.0)	.433
Other types	0 (0.0)	1 (1.0)
SCCmec Typed
II	113 (90.4)	36 (37.1)	<.001
IV	12 (9.6)	59 (60.8)
V	0 (0.0)	1 (1.0)
Other types	0 (0.0)	1 (1.0)
PVL Gene
Present	1 (0.8)	3 (3.1)	.319
Clonal Complex 8	58 (27.4)	88 (41.1)
PFGE Epidemic Typec
CMRSA10 (USA300)	53 (91.4)	83 (94.3)	.519
Other types	5 (8.6)	5 (5.7)
SCCmec Typed
II	2 (3.4)	14 (16.1)	.025
IV	54 (93.1)	67 (77.0)
V	0 (0.0)	3 (3.4)
Other types	2 (3.4)	3 (3.4)
PVL Gene
Present	51 (87.9)	74 (84.1)	.685
Clonal Complex 59	5 (2.4)	10 (4.7)
PFGE Epidemic Typec
USA1000 (China/Taiwan)	5 (100.0)	10 (100.0)	–
SCCmec Type
II	0 (0.0)	3 (30.0)	.006
IV	1 (20.0)	7 (70.0)
V	4 (80.0)	0 (0.0)
Other types
PVL Gene
Present	0 (0.0)	2 (20.0)	.524
Other Clonal Complexes  e	22 (10.4)	19 (8.9)
PFGE Epidemic Typec
CMRSA2 (USA 100/800)	0 (0.0)	2 (10.5)	.209
Other types	22 (100.0)	17 (89.5)
SCCmec Type
II	8 (36.4)	1 (5.3)	.037
IV	13 (59.1)	15 (78.9)
V	1 (4.5)	2 (10.5)
Other types	0 (0.0)	1 (5.3)
PVL Gene
Present	4 (18.2)	9 (47.4)	.091

Abbreviations: PFGE, pulse-field gel electrophoresis; PVL, Panton-Valentine leukocidin; SCCmec, staphylococcal cassette chromosome mec.

^aPercentages have been rounded and may not total 100%.

^bχ ² test (or Fisher’s exact test where appropriate) for statistically significant difference between the years 2010 and 2016.

^cAssigned based on spa type and/or PFGE pattern.

^dSCCmec typing was not available for CC5 isolates in 2010 (n = 2) and CC8 isolates in 2016 (n = 1); therefore, the SCCmec denominator is reduced accordingly.

^eOther clonal complexes: CC1, CC6, CC7, CC15, CC22, CC30, CC45, CC72, CC80, CC97, CC152, and CC361.

Figure 1.

Minimum spanning trees based on multilocus sequence typing and colored according to healthcare-acquired methicillin-resistant Staphylococcus aureus (HA-MRSA), community-acquired MRSA (CA-MRSA), and those where the location of MRSA acquisition was unknown, with the presence or absence of Panton-Valentine leukocidin gene (PVL), for the years 2010 and 2016. Partitions (gray area) were built from sequence types that differ by ≤2 alleles and are indicated by the clonal complex (CC). Pies are scaled to the number of isolates. Trees were generated using BioNumerics version 6.6 (Applied Maths, Sint-Martens-Latem, Belgium).

Overall, our genotype data showed that historical associations between infections identified using surveillance case definitions as HA-MRSA and CC5 and CA-MRSA and CC8 persist. However, both CCs were associated with HA- and CA-MRSA, and there was a notable increase in CC5 strains related to CA-MRSA infections in 2016 (27 of 97 [27.8%]) compared with 13 of 127 (10.2%) in 2010, P = .001. Moreover, although 86.8% of PVL-positive isolates belonged to CC8, PVL was identified across CCs and was associated with both HA- and CA-MRSA (Figure 1). Analysis of characteristics associated with CC5 and CC8 revealed that individuals <60 years of age had 5.0 (95% CI, 3.0–8.3) times the odds of infection with a CC8 isolate, and those with infections acquired in the community were more likely to have a CC8 strain (aOR, 4.2; 95% CI, 2.4–7.3) compared with infections acquired in a healthcare setting (Supplementary Table S6). We also noted an increased odds of CC8 isolation when the location of infection was unknown (aOR, 6.9; 95% CI, 3.1–15.2), indicating that these may represent CA-MRSA infections.

Impact of Genotype Shifting on Antimicrobial Resistance Patterns

Antimicrobial susceptibility results were available for 423 of the 426 study isolates (Table 3). Half of all isolates were multidrug resistant (n = 255, 60.3%). The most frequently observed resistance profile was OXA-ERY-CLD-CIP (188 of 423 isolates) (Supplementary Table S7), although the number of cases with this combination of drug resistance declined in 2016 (P = .017). Almost all of the isolates with the OXA-ERY-CLD-CIP resistance profile were CC5 (n = 178), whereas CC8 isolates were more diverse in resistance profiles with OXA-ERY-CIP (n = 59), OXA-ERY (n = 20), OXA-ERY-CIP-FUS (n = 17), and OXA-CIP (n = 12) as the most frequently observed. We observed a significant decrease in erythromycin (P = .036) and clindamycin resistance (P < .001) and an increase in resistance to FUS (P < .001) between 2010 and 2016. The increase in FUS resistance was largely associated with CC8 strains (from 6 of 58 [10.3%] in 2010 to 27 of 87 [31.0%] in 2016, P = .007) (Supplementary Table S8). The decrease in constitutive clindamycin resistance was associated with an overall decline in the proportion of isolates that were CC5-SCCmec-II (from 53.8% [113 of 210] in 2010 to 16.9% [36 of 213] in 2016, P < .001), as well as with a decline in the proportion of these isolates with constitutive resistance (95.6% [108 of 113] versus 44.4% [16 of 36], P < .001); see Supplementary Table S9 for an analysis of clindamycin by CC and SCCmec type. This was partially counterbalanced by an increase in the relative frequency of inducible clindamycin resistance in CC5 isolates, independent of SCCmec type, which increased from 2 of 119 (1.7%) isolates in 2010 to 43 of 97 (44.3%) in 2016 (P < .001).

Table 3.

Frequency of Phenotypic Resistance to Antimicrobial Agents in Ontario, Canada, Overall and by Year

Susceptibility Results	Total Resistant n (%)	Year		P Valuea
		n (%)
		2010	2016
Number testedb	423	210	213
Resistant toc
Oxacillin	423 (100.0)	210 (100.0)	213 (100.0)	.622
Erythromycin	349 (82.5)	182 (86.7)	167 (78.4)	.036
Clindamycin (constitutive)	184 (43.5)	132 (62.9)	52 (24.4)	<.001
Clindamycin (induced)	52 (14.0)d	2 (1.3)d	50 (23.5)	<.001
Clindamycin (total)	236 (63.4)d	134 (84.3)d	102 (47.9)	<.001
Ciprofloxacin	343 (81.1)	174 (82.9)	169 (79.3)	.419
Fusidic acid	67 (15.8)	19 (9.0)	48 (22.5)	<.001
Tetracycline	23 (5.4)	12 (5.7)	11 (5.2)	.812
Trimethoprim-sulfamethoxazole	16 (3.8)	5 (2.4)	11 (5.2)	.131
Gentamicin	6 (2.9)e	6 (2.9)	ND	–
Mupirocin
Low-level	11 (5.2)e	11 (5.2)	ND	–
High-level	5 (2.4)e	5 (2.4)	ND	–
Rifampin	4 (0.9)	2 (1.0)	2 (0.9)	1.000

Abbreviations: ND, not done.

^aχ ² test (or Fisher’s exact test where appropriate) for statistically significant difference between the years 2010 and 2016.

^bAntimicrobial susceptibilities unavailable for 3 study isolates.

^cNo resistance observed in either year for daptomycin, linezolid, and vancomycin.

^dThe D-test was not done for 51 erythromycin-nonsusceptible isolates in 2010 and reduced the number tested for clindamycin (induced) and clindamycin (total) to n = 159 in 2010 and the overall total tested to n = 372.

^eThe total number of isolates tested for gentamicin and mupirocin was n = 210.

Increasing Genetic Diversity Within Dominant Methicillin-Resistant Staphylococcus aureus Clonal Complexes

We identified 72 315 SNPs among all MRSA isolates relative to the core genome sequence of a CC5 reference strain. The majority of isolates grouped in 2 major lineages comprising CC5 and CC8 strains (Supplementary Figure S3). We next mapped CC5 and CC8 WGS reads to reference genomes matching their respective CCs. We identified 7981 SNPs relative to the CC5 reference among CC5 isolates and 4531 SNPs relative to the CC8 reference among CC8 isolates. Within the CC5 population, isolates were relatively diverse, 58.5% all isolates were >40 SNPs from any other CC5 isolate collected in the same year, whereas CC8 isolates were genetically more closely related to each other, with 44.1% of isolates >40 SNPs of another CC8. There were also differences in the median pairwise SNP distances between study years. For CC5 isolates, the minimum pairwise distance increased from 37 SNPs (interquartile range [IQR], 23–70) in 2010 to 65 (IQR, 41–96) in 2016 (P < .001). A similar increase was observed for CC8 isolates for which the minimum median pairwise SNP distance was 30 (IQR, 20–45) in 2010 and 43 (IQR, 22–72) in 2016 (P = .004).

Initial inspection of whole genome-based phylogenies for CC5 and CC8 isolates identified that the trees shared a common topology—both having a number of distant outgroup isolates (Figure 2). We hypothesized that these outgroup isolates may represent infections by CC5 or CC8 strains normally circulating elsewhere [29], whereas the remaining isolates were likely to represent “local” MRSA strains. Supporting the hypothesis, these phylogenetically distant isolates had more variable AMR profiles compared with other CC5 and CC8 isolates (Figure 3). Furthermore, when we included several publicly available CC5 and CC8 MRSA genome sequences from other countries, we found that for both CC5 and CC8, sequences from MRSA recovered in the United States nested within the “local” Ontario population, whereas those from isolates recovered outside of North America nested within the outgroup isolates (Supplementary Figure S4).

Figure 2.

Phylogenetic trees of methicillin-resistant Staphylococcus aureus isolates from Ontario, Canada representing clonal complex (CC)5 (n = 224) and CC8 (n = 146). The midpoint-rooted maximum-likelihood phylogenetic trees were built using FastTree v2.1.8 (1000 bootstrap replications) based on 7981 (CC5) and 4531 (CC8) core-genome single-nucleotide polymorphisms and visualized in R (v3.5.1) using the ggtree package [30]. Year of isolation (tip points) and hospital identification (outer ring) are indicated.

Figure 3.

Cladogram of all methicillin-resistant Staphylococcus aureus study isolates (n = 426) built using FastTree v2.1.8 (1000 bootstrap replications) based on 72 315 core-genome single-nucleotide polymorphisms and aligned to a clonal complex (CC)5 reference (GenBank accession no. CP000736.1). Visualized in R (v3.5.1) using the ggtree package. Clonal complex (CC) is indicated by tip color. The year of isolation is represented in the innermost ring. Subsequent rings indicate phenotypic susceptibility results (HLR, high-level resistance; I, indeterminant; LLR, low-level resistance; R, resistant; S, sensitive) for rifampin (RIF), mupirocin (MUP), gentamicin (GEN), trimethoprim-sulfamethoxazole (SXT), tetracycline (TET), fusidic acid (FUS), ciprofloxacin (CIP), clindamycin (CLD), erythromycin (ERY), and oxacillin (OXA). No color for susceptibilities indicates the antibiotic testing was not done. Astericks (*) highlight phylogenetically distant outgroup isolates, please see the text for further details.

Although there was relatively high strain variation overall within CC5 and CC8, there were groups of phylogenetically related isolates particular to certain hospitals, notably HG06, H10, and HG21, in both CC5 and CC8 (Figure 2). Overall, when we examined the pairwise SNP difference of study isolates within and between hospitals, we found that isolates within the same hospital were more closely related than those from different hospitals. The median pairwise SNP distance for CC5 isolates was lower within hospitals (137; IQR, 93–200) compared to a median of 154 (IQR, 129–233) between hospitals, P < .001 (Figure 4). Likewise, for CC8 isolates, a slightly lower median pairwise SNP difference was observed within hospitals (83; IQR, 67–98) compared to between hospitals (85; IQR, 69–114), P < .001.

Figure 4.

Boxplots representing the pairwise single-nucleotide polymorphisms (SNP) distances of methicillin-resistant Staphylococcus aureus isolates collected within the same hospital and between different hospitals for clonal complex (CC)5 and CC8 strains. Pairwise distances were calculated using snp-dist v0.6.3 (https://github.com/tseemann/snp-dists).

Minimal Genomic Clustering

Among 224 CC5 isolates, we identified 3 genomic clusters consisting of pairs of closely related isolates (6 isolates, 2.7%). Among CC8, we also identified 3 genomic clusters (2–3 isolates/cluster), representing 4.8% (7 of 146) of isolates. No genomic clusters were found within the other CCs. Examining clusters within the context of clinical, laboratory, and epidemiological data indicated that they may potentially represent recent transmission—isolates were genomically related, collected in the same year, and within the same hospital (Supplementary Table S10 and Supplementary Figure S5). All 3 CC5 clusters involved persons ≥60 years of age, and all had been determined to have acquired MRSA within healthcare—typical of traditional CC5 infections. Likewise, the CC8 clusters largely followed traditional characteristics—community-associated for 5 of 7, with a lower age demographic (median, 46 years; IQR, 44–54) compared to CC5 clusters (median, 72 years; IQR, 69–74). All but 1 of the genomic clusters occurred in 2010.

DISCUSSION

We report changes to the epidemiology of MRSA infections in Ontario, Canada, encompassing shifts from HA- to CA-MRSA and changes in the MRSA strain population structure, with a 1.5× increase in the relative frequency of CC8, a concomitant decrease in CC5, and rise of CC59. The reasons behind MRSA population changes are not fully understood, with hypotheses ranging from improvements in hospital infection control procedures and antibiotic stewardship programs to MRSA biological factors [31, 32]. Although our findings do not necessarily support any 1 hypothesis, the decrease we observed in infections commonly associated with healthcare settings (eg, medical devices) and HA-MRSA-associated strains are likely an indicator of improvements in hospital infection control. Although the reasons behind the increase in CA-MRSA in Ontario are unclear, overall, our results are consistent with previous North American studies describing a rise in CA-MRSA infections and the emergence of CC8 (CMRSA10/USA300) as a dominant clone [5, 33–36]. Shifts were also observed in minority CCs, including a decrease in CC30 and increase in CC59 strains. Given the small number of cases with these CCs, follow-up studies are needed to determine whether this trend persists.

The shift we observed in strain genotypes was reflected in AMR patterns, with significant decreases in erythromycin and clindamycin resistance related to CC5 strains—similar to reports from a US national surveillance study [33] and a longitudinal hospital-based study in Boston [37]. In addition, we identified an increase in inducible clindamycin resistance, which was also reported in the surveillance study. This rise in resistance should be monitored closely because empiric use of clindamycin for MRSA treatment is common. Furthermore, our finding of increased isolation of MRSA resistant to FUS in 2016 related to CC8 strains is of concern, and it may potentially represent expansion of community-associated strains in which resistance has resulted from historic usage of FUS in the community [38–41].

Our genotyping and WGS results revealed that MRSA genetic diversity was relatively high overall and within the dominant CCs. Within both CC5 and CC8, some phylogenetically more distantly related and/or genetically “unique” isolates were observed, which had AMR patterns different from the more closely related strains and may potentially represent travel-related infections or other MRSA introductions from abroad [29, 42]. This highlights the importance of considering epidemiological information such as travel history, in making decisions about empiric therapy. Recognizing that infections may be acquired overseas through genome sequencing of a large isolate collection demonstrates the value of population-based approaches to provide broader context when investigating potential sources of MRSA infection.

One important limitation was that detailed investigation data to support the identification of person-to-person transmission were unavailable. However, we speculate that MRSA transmission within this study population was likely minimal—only 6 small genomic clusters were identified, 3 of which were pairs of potential healthcare-associated transmissions. We were further limited by the availability of isolates—approximately one third of MRSA diagnoses had an isolate submitted—reducing our ability to identify source cases and draw conclusions regarding within hospital transmission. Previous studies have demonstrated limited within hospital spread due to stringent infection control and screening practices [43, 44], and this may explain the shift in CCs traditionally associated with CA-MRSA. Similar to studies elsewhere [44], we identified that in several instances, groups of phylogenetically related isolates had been collected in the same hospital—often spanning both study years—indicating that some strains may circulate in particular settings or geographic regions.

CONCLUSIONS

Combining genome analysis with case-level clinical and demographic patient data can identify changing MRSA epidemiological disease trends, define the underlying strain population structure, detect outbreaks, and characterize drug-resistant clones. With the limitation that our study population included only hospitalized individuals for which an isolate was available for WGS, and therefore the results cannot be readily generalized to the wider population, we show that CC5 MRSA strains are being replaced by CC8 strains in Ontario hospitals, thereby shifting AMR patterns. Whole-genome sequencing enhanced the resolution of classic genotyping techniques, revealed a relatively high MRSA strain diversity, identified the presence of strains potentially acquired abroad, and indicated that nosocomial transmission was likely minimal. Adoption of this type of integrative approaches can benefit MRSA surveillance programs and impact both clinical care and public health practice.

Supplementary Data

Supplementary materials are available at The Journal of Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes

Acknowledgments. We thank Vanessa Porter (Sunnybrook Health Sciences Centre) for assistance with the strain collection and molecular testing. We gratefully acknowledge all members of the Canadian Nosocomial Infection Surveillance Program (CNISP) for providing study isolates and data curation.

Financial support. This work was funded by Public Health Ontario through Internal Grant 2017-18-PIF-001 (to N. F. and S. N. P.).

Potential conflicts of interest. All authors: No reported conflicts of interest. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest.

The Ontario CNISP Hospital Investigators include the following: Johan Delport (London Health Sciences Centre), Gerald Evans (Kingston General Hospital), Susy Hota (University Health Network), Kevin Katz (North York General Hospital), Camille Lemieux (University Health Network), Dominik Mertz (Hamilton Health Sciences Centre), Michelle Science, (Hospital for Sick Children), and Nisha Thampi (Children’s Hospital of Eastern Ontario).