ABSTRACT
Methicillin-resistant Staphylococcus aureus (MRSA) has become a common cause of skin infections and invasive infections in community dwellers in the United States since the late 1990s. Isolates characterized as USA300 by pulsed-field gel electrophoresis (PFGE) are the predominant strain type in these infections. USA100 and USA500 strains commonly cause health care-associated infections. We compared PFGE with a number of other methods of genotyping in a sample of 149 clinical MRSA isolates from the University of Chicago Medical Center. The 5 USA500 isolates yielded 3 spa types and 2 multilocus sequence types (MLSTs). Among the 24 USA100 isolates, 21 (88%) were of spa type t002, 19 (79%) were of ST5, 2 carried arcA and opp3, and 1 was Panton-Valentine leukocidin positive (PVL+). Among the 102 USA300 isolates, 96 (94%) were of ST8 and 94 (92%) were of spa type t008. The combination of traits that provided the best sensitivity (98%), specificity (97%), positive predictive value (PPV) (99%), and negative predictive value (NPV) (95%) for identifying USA300 isolates were the presence of the arcA gene and the presence of the PVL genes (area under the curve, 0.980; 95% confidence interval [CI], 0.955 to 1.0). PFGE did not delineate a homogeneous group of MRSA genetic backgrounds, as documented for other typing methods, particularly for USA500 and USA100 pulsotypes. Documenting the presence of arcA and PVL genes by PCR was an efficient and accurate means of identifying USA300 in a collection of MRSA isolates in which USA300 is common. None of the tested genotyping methods provided an accurate means of identifying the next most common PFGE-based backgrounds, USA100 and USA500.
INTRODUCTION
Numerous methods of strain typing to classify Staphylococcus aureus have been used in the past 3 decades. These modalities have traditionally been classified as phenotypic (e.g., antimicrobial resistance patterns, serotypes, and phage types) or genotypic. Genotypic methods have been divided into sequence-based (e.g., multilocus sequence typing [MLST]) and non-sequence-based (e.g., pulsed-field gel electrophoresis [PFGE]) techniques (1). Each method differs in the equipment, cost, and expertise required. Depending on the purpose of the typing, the optimal technique is likely to differ because the ability of the various techniques to discriminate among related isolates is not the same.
In the United States, many studies have demonstrated that, since 2003, there have been 2 predominant lineages of methicillin-resistant Staphylococcus aureus (MRSA) that commonly cause human infections, USA300 and USA100 (2). Both are genetic backgrounds that were defined in 2003 by a classification scheme based on PFGE patterns; the most common types were designated USA100 and USA200, etc., through USA1100 (3). USA300 is the predominant MRSA strain type causing infections outside the health care environment in the United States. It is becoming a common cause of MRSA infection in health care facilities as well (4, 5). USA300 is highly virulent, and epidemiologic studies have suggested that it is readily transmitted in jails (6), among athletes (7), and in other settings. It has been associated with skin and soft tissue infections (SSTIs) and invasive, sometimes fatal infections in previously healthy people (8). It also was the predominant genetic MRSA background causing SSTIs in U.S. emergency department patients in 2004 and in 2008 (9). USA300 MRSA usually carries the genes encoding Panton-Valentine leukocidin (PVL), a toxin that lyses human neutrophils; the staphylococcal cassette chromosome mec (SCCmec) type IV element (10); and the arginine catabolic mobile element (ACME) (11). USA100, in contrast, is predominant in the United States as a cause of health care-associated MRSA infections (2). It is generally PVL negative and carries the SCCmec type II element. Other genetic backgrounds, defined by pulsotype, have been reported in clinical studies, but they have generally been low in number relative to USA300 and USA100 during studies conducted in the past decade.
Which genotyping technique is best? Because PFGE is often accorded a gold standard status, we set out to compare PFGE with a number of well-established and widely used methods of genotyping with a collection of clinical MRSA isolates. We had 2 goals. The first was to determine how homogenous isolates classified by the PFGE nomenclature of USA type were with regard to classification by other typing modalities. Tested modalities included MLST, spa typing, SCCmec typing, determination of the presence or absence of the PVL toxin genes, and determination of the presence or absence of the arcA and opp3 genes (markers for ACME [11]). Second, because PFGE can be more difficult and expensive to perform than other tests, we were also interested in assessing the sensitivity and specificity of various molecular classification schemata, alone and in combination, to predict the PFGE USA300 profile, the most prevalent pulsotype in the collection.
MATERIALS AND METHODS
At the University of Chicago Medical Center (UCMC), consecutive MRSA isolates were collected prospectively from the Clinical Microbiology Laboratory in 2008 from patients in multiple clinical settings (emergency department and inpatient and outpatient clinics). Only the first MRSA isolate obtained from each patient was used. A 20% random sample of the MRSA isolates was selected (n = 168), including isolates obtained from infections and from cultures obtained to assess colonization status. The study was approved by the Institutional Review Board (IRB) of the Division of the Biological Sciences of the University of Chicago. Attempts were made to contact each patient to obtain informed consent for characterization of the isolate. If a potential subject could not be contacted despite five attempts, the patient isolate was used with a waiver of consent. Nineteen patients refused enrollment, and a total of 149 patient isolates were used.
Each MRSA isolate underwent characterization by PFGE, using SmaI as a restriction endonuclease, at the Minnesota Department of Health (3). Patterns were compared by using Dice coefficients, and those that were within 80% similarity were considered part of the same pulsotype and characterized according to USA group (3). Those isolates with a banding pattern that was a 100% match were considered indistinguishable. Other genotyping methods were performed at the University of Chicago MRSA Research Center. These methods included MLST (12) and detection of the PVL genes (13), arcA, and opp3 (11) by PCR, as described previously. In addition, the SCCmec type was determined by a series of PCR assays as described previously (14), and each isolate underwent spa typing as described previously (15).
Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) for each genotyping modality and for each combination of genotyping strategies were assessed for their ability to predict the PFGE USA300 profile, which was considered to be the gold standard (3). To assess the value of the various genotyping modalities as “tests” to identify USA300 isolates in the collection under study, the area under the receiver-operator curve (ROC) (plotting the sensitivity by 1 − specificity) was calculated for each genotyping test and combination of genotyping tests (Stata 11, College Station, TX). The greater the area under the curve (AUC), the better the predictive criteria.
RESULTS
Among the 149 MRSA isolates from unique patients, by PFGE, 68.5% were defined as USA300 (n = 102), 16.1% were defined as USA100 (n = 24), and 15.4% belonged to other USA types (n = 8) or could not be classified under a known USA pulsotype (n = 15).
Among the 149 MRSA isolates, 15 MLSTs were represented. The 2 most common MLSTs were ST8 (n = 101; 68%), the ST commonly associated with USA300, and ST5 (n = 29; 20%), the ST commonly associated with USA100. Together, they accounted for 88% of the isolates. The 2 most common spa types were t008 (n = 96, 64%), commonly associated with USA300, and t002 (n = 31, 21%), commonly associated with USA100. Together, they accounted for 85% of the 149 isolates.
The majority of isolates were PVL positive (PVL+) (n = 108; 73%), and an even larger majority carried SCCmec type IV (SCCmec IV) (n = 114; 77%). All 104 isolates that carried the opp3 gene (104/149; 70%) also carried the arcA gene and vice versa.
PFGE yielded 3 pulsotypes containing >4 isolates and a number of other genetic backgrounds represented by fewer isolates. We evaluated the large USA100, USA500, and USA300 pulsotype groups further.
There were 5 USA500 isolates; none were indistinguishable by PFGE. These isolates were heterogeneous when typed by other methods and included 3 spa types and 2 MLST types. One USA500 isolate carried SCCmec type II, and 4 carried SCCmec type IV. Three USA500 isolates were PVL+, and one carried the arcA and opp3 genes. Thus, the USA500 pulsotype included isolates with varied strain characteristics (Table 1).
Molecular characterization of 149 methicillin-resistant Staphylococcus aureus case isolates, University of Chicago Medical Center, 2008e
Of the 24 USA100 MRSA isolates, 7 were indistinguishable by PFGE. All but 3 (21/24; 87.5%) of the 24 USA100 isolates were of spa type t002; the others belonged to spa types t242, t067, and t010. All but 1 of the USA100 MRSA isolates (23/24; 96%) carried SCCmec type II. Among the USA100 isolates, 19/24 (79%) were of ST5, and the 5 (21%) that were not of ST5 belonged to 4 related sequence types (ST105, ST231, ST1730, and ST2513, all single-locus variants of ST5). Two USA100 isolates carried arcA and opp3, and one was PVL+.
Thus, the majority of USA100 isolates shared many genetic characteristics. However, the group of USA100 isolates was also not homogeneous with respect to the other assayed genetic characteristics. Instead, this PFGE designation included isolates carrying a variety of potential virulence factors and heterogeneous strain types, as defined by the other genetic classification systems, particularly MLST (Table 1).
Among the 102 isolates identified as USA300 isolates, 76% (77/102) were indistinguishable by PFGE. All of the USA300 isolates (102/102) carried the PVL genes (PVL+), 98% (100/102) carried the SCCmec type IV element, 98% (100/102) were arcA and opp3 positive, 94% (96/102) were of ST8, and 92% (94/102) were of spa type t008. Results of genotyping of these isolates are shown in Table 1.
We wondered which of the 5 genotyping methods would have the most accurate characterization, i.e., the sensitivity, specificity, PPV, and NPV, to predict the USA300 pulsotype. We examined, individually and in combination, each of the 5 genotyping methods. The characteristics used to predict USA300 were spa type t008, PVL positivity, MLST ST8, arcA and opp3 positivity, and carriage of SCCmec type IV.
No single typing method or combination of methods had perfect sensitivity and specificity for USA300. While PVL+ status had 100% sensitivity, its specificity was only 85%, because isolates from other pulsotypes were sometimes PVL+. Similarly, SCCmec type IV carriage as a predictive USA300 characteristic had a high sensitivity (98%) but a low specificity (67%). spa type t008 and the arcA PCR assay both had high specificity (95% and 92%, respectively) and sensitivity (93% and 98%, respectively).
The sensitivity, specificity, NPV, and PPV of all combinations of the genotyping modalities are shown in Table 2. When any 3 or more tested genotyping modalities were combined to predict the USA300 pulsotype, many true USA300 isolates were not identified; the negative predictive value for each such combination was <92%. Therefore, the combination of >2 tested typing methods is not recommended for identifying USA300 isolates.
Test characteristics in identifying the USA300 pulsotype among 149 MRSA case isolates, University of Chicago Medical Center, 2008
Similarly, certain combinations of 2 typing methods had poor sensitivity for identifying USA300 isolates. For example, the t008 spa type combined with the ST8 MLST had excellent specificity (98%) and PPV (99%), meaning that this combination of genetic traits rarely identified a non-USA300 isolate; however, the sensitivity was not as high (87%), because the use of these 2 criteria excluded some USA300 isolates. Furthermore, the NPV for the combination of spa type t008 plus ST8 was relatively poor (78%), meaning that it did not accurately identify the entire group of USA300 isolates in our collection.
The combinations of traits that provided the best sensitivity, specificity, PPV, and NPV for identifying USA300 isolates were arcA positivity plus carriage of SCCmec IV and arcA positivity plus PVL positivity, which had sensitivity, specificity, PPV, and NPV that were all >90% (Table 2). ROC analysis showed that the best single combination of molecular predictors for USA300 (AUC, 0.980; 95% confidence interval [CI], 0.955 to 1.0) was the combination of the presence of arcA and the PVL genetic determinants. Other combinations yielded a lower AUC (Table 2).
No single method or combination of genotyping methods was sensitive or specific for identifying USA100 or USA500 isolates (results not shown).
DISCUSSION
Our study has 2 important findings. First, PFGE was unable to differentiate isolates that were genetically distinct based on other tests, particularly for the USA100 and USA500 pulsotypes. While the reasons for this are not known with certainty, it is possible that USA300 has not been in existence for as long or that it has not faced the same selective pressures as the other backgrounds examined and therefore that USA300 may not have evolved the heterogeneity that we observed, for example, in the USA100 background. The finding that PFGE did not distinguish heterogeneous strains among USA500 and among USA100 isolates should be considered when molecular epidemiologic studies are done.
Second, we demonstrated that among MRSA isolates at the UCMC, PCR assessment for the presence of arcA as well as lukS-PV and lukF-PV, the genetic determinants of the PVL toxin, provided the best means of identifying USA300 isolates defined by PFGE with near-perfect sensitivity and specificity. The procedures to test for the presence of these genes are much less time-intensive and less prone to difficulties of interpretation and comparison among laboratories than PFGE, and in our strain population, these 2 PCR assays can be used as a nearly accurate replacement for PFGE identification of the USA300 pulsotype.
USA300 has been identified as the predominant community-acquired MRSA (CA-MRSA) genetic background in many studies from North America (2, 4) and has been detected as a cause of clinical infection in many other countries (16, 17). While there has been some reported variability in antimicrobial susceptibilities (18, 19), typically, USA300 has been susceptible to clindamycin, tetracyclines, gentamicin, rifampin, and trimethoprim-sulfamethoxazole and is resistant to erythromycin (20). Some investigators have used antimicrobial susceptibility patterns as an inexpensive substitute for classifying isolates as USA300 isolates. This can be problematic, because antimicrobial susceptibilities can vary independently of other genetic characteristics of an isolate (21). In addition, other phenotypic methods such as biochemical tests or colony appearance are also relatively undiscriminating and may not identify strains that share specific genetic characteristics.
Our findings are likely now applicable in the United States, but over time, there may be changes in the predominant USA300 background. In Latin America, a variant of pulsotype USA300-0114 (the standard USA300 pulsotype strain used by the CDC and in the present study) that lacked arcA was identified in 2006 to 2008 (22), demonstrating the possibility of the emergence of novel isolates sharing the USA300 pulsotype. Continued surveillance is thus needed to assess USA300 MRSA isolates in the United States for such changes. Another PCR-based assay for the detection of isolates with the USA300-0114 pulsotype was proposed previously (23), although it has not been explicitly tested for sensitivity or specificity to detect isolates with the USA300-0114 pulsotype in a group of unselected clinical MRSA isolates.
Genotypic classification systems include examination of genomic DNA fragments after cleavage by specific endonucleases (e.g., PFGE [24]), and PCR-based DNA sequencing methods include MLST, multilocus variable-number tandem-repeat analysis (MLVA), and coagulase gene (coa), agr, and spa typing. PFGE is thought to be useful in outbreak situations and in hospital epidemiology because it can often differentiate isolates that have undergone minor genetic changes, but it is labor-intensive and requires specialized equipment. Criteria have been established to compare PFGE band patterns (24). A limitation of sequence-based methods and of PFGE is that they may suggest that strains are closely related when, as we show in our study, they may in fact differ in ways that are critical to pathogenesis.
Our data and the observations of others demonstrate that no two genotyping methods examining distinct genetic loci will consistently provide identical results in classifying MRSA isolates. This is because these methods assess genetic differences that can evolve independently. Classification systems often employed for epidemiologic research have created competing nomenclatures that are useful for assessing relatedness of strains but are unfortunately often not directly commensurable. For example, our data illustrate that isolates belonging to a single MLST may include isolates with >1 spa type, and a specific spa type may include >1 MLST.
In the future, the optimal means of distinguishing MRSA isolates may be whole-genome sequencing. Recently, five studies have demonstrated the power of whole-genome sequencing to compare strains of S. aureus by identifying single-nucleotide polymorphism profiles and other properties of their genomes (25–29). At present, this technology remains relatively expensive and requires expertise for interpretation. The potential of whole-genome sequencing as an epidemiologic and nosologic tool has not yet been realized, in part because computational methods to interpret efficiently the massive amount of data produced, rapidly annotate genomes, accurately account for errors in these data, and identify the critical differences among isolate genomes are still formidable challenges. Until such tools are readily available, other methods of genotyping will remain useful.
Our study has certain limitations. We examined isolates from a single center in the United States. The PPV and NPV of any diagnostic test varies with the prevalence of true positives in a population. Therefore, the very high NPV and PPV for USA300 isolates when PVL and arcA genes were found may not hold in regions where USA300 is less common. However, in the United States, USA300 is the predominant pulsotype, so it is likely that these 2 PCR assays in combination would have a similar predictive power in much of the country. Evaluation of our criteria in different populations would be useful to assess its utility in future molecular epidemiology studies.
ACKNOWLEDGMENTS
This work was supported by grants from the National Institutes of Health (grants R01 A140481-01A1 and 2R56AI040481-08A1 to R.S.D. and S.B.-V., 1 U01 GM087729-01/B270JA to R.S.D., and 1K23 AI095361-01 to M.Z.D.), the Grant Healthcare Foundation to S.B.-V., and the Lawrence Livermore Laboratory to S.B.-V.
We thank Adriana Cadilla for assistance with genotyping of isolates.
FOOTNOTES
- Received 11 September 2012.
- Returned for modification 8 October 2012.
- Accepted 3 December 2012.
- Accepted manuscript posted online 26 December 2012.
- Copyright © 2013, American Society for Microbiology. All Rights Reserved.
