Identification of MRSA carriers
Under routine hospital procedures, a double-headed swab (double-headed BBL Culture Swab Liquid Stuart; Becton-Dickinson, Sparks, Maryland) was inserted into the anterior nares of each patients’ nostril and rotated at least 5 times. The swabs were transported to VMC’s Clinical Microbiology Department. One swab was tested for MRSA using the polymerase chain reaction (PCR) with the BD GenOhm® to detect the SCCmec gene, which confers methicillin resistance. A positive result from the rapid screen identified patients as MRSA carriers.
Eligibility criteria and sample size
The following eligibility criteria were applied to cases and controls: age 18-65 years, resident of a top swine producing area in North Carolina, English or Spanish speaker, and screened for MRSA nasal carriage. Cases and controls were restricted to residents of swine producing North Carolina zip codes in which the number of swine permitted for production by the North Carolina Division of Water Quality was equal to or greater than the median for the state. There were 176 eligible zip codes. The age and geographic restrictions were used to increase the prevalence of occupational and livestock exposures.
Sample size goals were based on consideration of participant enrollment time, molecular typing expenses, and estimated numbers of participants needed to achieve 77% - 87% power, assuming 20% - 40% exposure prevalence and an odds ratio of 2.3.
Cases were defined as MRSA nasal carriers, based on a positive result from the rapid PCR screen that was administered at hospital admission. Eligible cases were identified by reviewing daily electronic medical record reports of all admitted patients. The reports listed MRSA screening results, age, gender, and other demographic information.
Controls were non-MRSA carrying patients based on a negative PCR screen. They were identified using the same daily electronic medical records that were used to identify cases. One control was matched to each case based on age (±5 years) and gender. When more than one patient was an eligible match for a case, a random number generator was used to select the potential control.
Interviews and medical record review
Structured interviews were administered to participants (cases and controls) in their hospital rooms. The questionnaire included information about current employment; job title; work address; number of household members; household member occupation; home address; direct (touching) and indirect (working near but not touching) contact with cows, pigs, chickens, turkeys, and horses at work or outside of work; demographics; ability to smell odor from animal farms when at home; living on a farm with animals; and handling meat at home or at work in the past 2 weeks.
Medical records were reviewed to identify the first-listed diagnosis for the current hospitalization and to determine whether participants were hospitalized for any reason within one year of the current admission.
One author (L.S.) administered all the interviews and abstracted all the data from medical records.
ArcMap10® (ESRI, Inc., Redlands, CA, USA) was used to geocode home and work addresses. Five participants reported a home address that could not be geocoded; however, the address listed in their medical records was different and could be assigned coordinates. For these 5 participants, coordinates were assigned according to the address in the medical record.
Human and swine population densities and rural area classifications
Topically integrated geographic encoding and referencing® shapefiles from the 2010 United States Census were used to identify the census block group of each home or work address, and to classify home addresses as being in rural areas; in urban clusters, which contain at least 2,500 people; or in urbanized areas, which contain 50,000 or more people. Urban areas and clusters were combined to form a single urban category.
Densities of total, farrowing, and non-farrowing permitted swine in each census block group (number of swine/square miles in the block group) were calculated using a publicly available database from the North Carolina Division of Water Quality, which lists the type and address of livestock facilities in North Carolina that hold non-discharge wastewater permits, as well the number of permitted animals at each. Farrowing swine include breeding sows and pigs from birth to weaning. Density calculations were categorized by developmental stage because of evidence that livestock associated MRSA is more prevalent in the youngest pigs[9, 10].
In addition, 2010 census data was used to assign human population densities to each block group (number of people/square miles in the block group). Block groups are subdivisions of census tracts; in North Carolina, they contain an average of 1,549 residents.
Satellite imagery in Google Earth™ was used to identify swine or poultry CAFOs located within 1 mile of each participant’s home and work addresses.
Duplicate MRSA nasal swabs collected from positive patients were stored at 4°C for up to 48 hours and then transported to VMC’s infection control laboratory. Nasal specimens were streaked onto a CHROMagar® MRSA plate (CHROM agar Microbiology, Paris, France) and incubated for 24 - 48 hours at 37°C. According to manufacturer recommendations, mauve colored colonies were identified as MRSA. One colony from each plate was selected and grown onto sheep’s blood agar (Remel, Lenexa, KS).
The diversilab® system
Molecular typing of isolates was performed using the Diversilab® system (bioMérieux, Inc., Durham, NC), and conducted according to manufacturer recommendations. DNA was extracted from a pure culture using the UltraClean™ Microbial DNA Isolation Kit (Mo Bio Laboratories, Solana Beach, CA). The NanoDrop® ND-1000 Spectrophotometer (Isogen, Ijssel stein, The Netherlands) was used to estimate the genomic DNA concentration. Sample DNA was diluted to a final concentration of 35 ng/μl.
Repetitive-element based PCR was performed using the Diversilab® Staphylococcus kit. Amplicons were separated using a Diversilab® DNA LabChip kit with microfluidic technology, as described previously. The analysis was performed using DiversiLab® software (version v.r.3.3.40). The data for each sample consisted of a dendogram, a virtual gel image (banding pattern), a graph of fluorescence corresponding to each banding pattern, and a similarity matrix. MRSA isolates were classified as CA or HA associated by comparing rep-PCR profiles with samples in the DiversiLab® MRSA library, which contains 70 samples of 14 representative USA pulsed field gel electrophoresis types. Strain relatedness was defined as >95% similarity with up to one band difference in the virtual gel image and determined by the similarity matrix and the pattern overlay function of the DL software. Isolates that did not match any samples in the library according to the above criteria were classified as non-matches.
Multi locus sequence typing
Multi locus sequence types (MLST) were assigned using a bash script applied to assembled Illumina data. Briefly, Illumina short read sequences were assembled into contigs using the SPADES assembler. Quality of the assembly was determined by the N50 parameter as well as by mapping that reads back to the assembly. Nucleotide-Nucleotide BLAST (Version 2.2.25+) was used to compare the housekeeping gene against each of the assembled genomes. Sequence similarity matches of genes were determined using thresholds of 100% nucleotide identity and 100% coverage of the query sequence length. The script then used the matched genes and MLST profile data to determine the final MLST type.
Phyloviz software was used to draw a minimum spanning tree using the in silico predicted MLST types of 48 isolates. The plot was drawn to scale.
MRSA CC398 was not identified among the MRSA isolates collected in this study. However, members of CC5 were identified. A high proportion of broiler chickens have been shown to carry S. aureus CC5. Therefore, Illumina whole-genome sequence data sets were aligned against the chromosome of a published poultry associated sequence type (ST) 5 S. aureus reference genome (strain ED98; GenBank accession no. NC_013450 ) using the short-read alignment component of the Burrows-Wheeler Aligner. Each alignment was analyzed for single-nucleotide polymorphisms (SNP) using GATK. To avoid false calls due to sequencing errors, SNP loci were excluded if they did not meet a minimum coverage of 10X and if the variant was present in <90% of the base calls for that position. SNP calls were combined for all of the sequenced genomes such that, for the locus to be included in the final SNP matrix, it had to be present in all of the genomes. SNPs falling in the duplicated regions on the reference genome were discarded.
Phylogenetic trees were generated using the maximum-parsimony method in PAUP v4.0b10 using only the High confidence SNPs. Published CC5 genomes from various sources were used to characterize the nature of the MRSA isolates. Details about the genomes are given in Additional file1. A published ST80 strain (strain 11819-97; GenBank accession no. NC_017351.1) was selected as an outgroup to root the whole genome sequence tree. Isolates in the clade nearest to this bifurcation point were used to root subsequent trees.
scn gene detection
NCBI Blast was used to detect the scn gene in the isolate assemblies.
For a number of reasons, not all eligible patients were available to be interviewed- they were discharged from the hospital, unconscious, sleeping, or receiving medical treatments at the time of interviewer contact, for example. Also, not all invited patients agreed to participate. Therefore, after data collection was complete, some participants did not have a matching case or control. To avoid double loss of information in analysis, case and control matched sets were pooled; the gender and age matching case or control who was admitted to the hospital within the shortest amount of time of unmatched participants was selected for the pooled set. Using the same control for more than one case has been described as a valid approach that should not bias measures of association[30, 31].
Conditional logistic regression models, which adjusted for the matching variables age and gender, were used to derive odds ratios (OR) and 95% confidence intervals (CI). A term representing education (<high school degree vs. high school degree or more) was entered into all models; this variable was selected a priori based on the belief that it could serve as a proxy measure for lifestyle factors that might confound relationships.
Associations between MRSA carriage and the following features and characteristics were explored: residence within 1 mile of a swine or poultry CAFO, swine densities (total, farrowing, and non-farrowing) in the block group of residence, ability to ever smell odor from an animal farm when at home, handling of uncooked meat at work and/or at home in the 2 weeks before hospital admission, indirect contact at work or direct contact at home with horses, indirect contact at work or direct contact at home with livestock (pigs, cows, chickens, turkeys), human population density in the census block group of residence, residence in a rural vs. urban area, living with others vs. alone, and participants’ employment status. Any participant who worked > 0 hours per week within 2 weeks of their hospital admission was considered employed.
Coding decisions were based on variable distributions and comparison of Akaike information criterion statistics. Except variables representing human and swine population density, all exposures were coded as binary terms. Human population density was coded as a linear term. Variables representing densities of total, farrowing, and non-farrowing swine were categorical (0 swine/square-mile, referent vs. > 0 to ≤ 149 swine/square mile vs. > 149 swine/square mile). Zero was the median and mode of the distribution of total swine density and 149 was the 25th percentile of the distribution of observations with non-zero total swine density values.
Because of small numbers within categories of non-farrowing and farrowing swine density, these variables were also categorized according to their own distributions (0 swine/square-mile vs. > 0 to ≤ 77 swine/square-mile vs. > 77 swine/square-mile for farrowing and 0 swine/square-mile vs. > 0 to ≤ 616 swine/square-mile vs. > 616 swine/square-mile for non-farrowing swine). The cut-points 77 and 616 were the median of the distribution of non-zero values for farrowing and non-farrowing densities, respectively.
Models were rerun to compare cases whose nasal swabs grew MRSA colonies with matched controls. Also, isolates were classified as CA or HA based on their USA pulsed field gel electrophoresis (PFGE) types, which were identified by the Diversilab® system. The classification of the USA types as CA or HA was based on the historic origins of these isolates in the United States. CA and HA carriers were compared with matched controls.
Statistical analysis software version 9.3 (Cary, NC) was used to conduct all analyses.
This study was approved by the non-biomedical institutional review board at the University of North Carolina at Chapel Hill (Study #11-0907), and by East Carolina University’s University and Medical Center Institutional Review Board office (Study #11-0257). All participants provided written informed consent and signed Health Information Portability and Accountability Act authorization forms.