Epidemiology of methicillin resistant Staphylococcus species carriage in companion animals in the Greater Brisbane Area, Australia

Background: Methicillin resistant Staphylococcus species such as S. aureus (MRSA) and S. pseudintermedius (MRSP) can be involved in life-threatening multidrug resistant infections in companion animals. Knowledge of methicillin resistant Staphylococcus (MRS) carriage and factors inuencing carriage in companion animals from South East Queensland is limited. Nasal and rectal swab samples were collected from dogs, cats and horses upon admission or within 24 hours of hospitalisation to several primary accession and referral veterinary practices between November 2015 and December 2017. MRSA and MRSP were identied using standard microbiological (Brilliance selective medium) and molecular (mecA gene PCR) methods. Risk factors associated with methicillin resistant Staphylococcus (MRS) carriage were quantied using Bernoulli logistic regression models. A Bayesian geostatistical model was developed to predict the probability of MRS carriage in Brisbane and surrounding areas. Results: Our results indicated that while the prevalence of MRSP carriage in dogs was 8.7% (35/402) no MRSP was isolated from cats (0/69) and horses (0/60); no MRSA was isolated in any species. MRSP carriage in dogs was signicantly associated with previous hospitalisation, previous bacterial infection, consultation type, average precipitation, and human population density. Our predictive map of MRSP carriage indicated that the probability of carriage was highest along the coastal areas of Greater Brisbane, particularly Brisbane city, Sunshine Coast and Gympie areas. Conclusions: This study determined that MRSP carriage in dog populations from South East Queensland is geographically clustered and associated with both clinical and environmental factors.


Background
Methicillin resistant Staphylococcus aureus (MRSA) and methicillin resistant S. pseudintermedius (MRSP) are important opportunistic pathogens in human and veterinary medicine and their broad range of antimicrobial resistance can hinder treatment in people and animals (1). Methicillin resistant S. aureus, a predominant human pathogen, has gained attention in the veterinary community due to its detection in healthy and diseased dogs, cats and horses (2,3). In contrast, MRSP is commonly isolated from dogs and cats but has been rarely identi ed in horses and human infections (4). Both MRSA and MRSP can exist commensally on the skin, nares and intestinal tract of animals and humans but can also become opportunistic pathogens (5). Case reports have shown risks for zoonotic transmission for both MRSA and MRSP (6,7). The rst case of a non-invasive infection due to zoonotic transmission was identi ed in a human with otitis externa who had the same strain of S. pseudintermedius (previously S. intermedius) as the carriage isolates from her pet dog (8).
Methicillin resistant S. aureus carriage in humans and MRSP carriage in dogs can be risk factors for infections (9)(10)(11). Studies have shown that decolonisation of methicillin resistant and susceptible S. aureus in human surgical patients reduce their risk for developing health care-associated infections (12)(13)(14). MRSP carriage was also shown to be a risk factor for surgical site infections in dogs (9). However, carriage of Staphylococcus may persist despite systemic antibiotic treatments and after successfully treating a MRSP infection (15). Finally, to show the impact of carriage not only in animals but also in humans, case reports can be found in the literature of animals with no clinical signs acting as reservoirs of bacteria for recurring infections in their owner. A mupirocin resistant MRSA infection in people was linked back to the family dog and further reoccurrence of the infection was prevented by decolonisation of MRSA from the dog's nares (16). Investigating the epidemiology of MRSA and MRSP carriage in companion animals is therefore a critical step to understand the possible implications for human and animal health.
Identifying populations at higher risk of carriage or infection and their determinants is critical information to target infection control strategies. For example, Chusri, Chongsuvivatwong (17) utilised spatiotemporal analyses to identify clusters of Acinetobacter baumannii infections within and between ward outbreaks in a human hospital indicating the need to improve infection control practices. Spatial studies regarding companion animals have focused more on disease distribution in communities rather than hospitals (18,19). Worthing, Abraham (20) described the geographical distribution of clinical MRSP isolates based on multilocus sequence typing (MLST) types across Australia; however, their study did not explore risk factors that could explain the spatial distribution and geographical clustering. The literature is lacking information overall regarding the spatial patterns of MRS carriage in animals.
In this study we, aimed to quantify the epidemiology of MRSA and MRSP carriage in dogs, cats, and horses in South East Queensland by identifying potential risk factors and the geographical distribution of at-risk areas.

Target population and study design
The target population of the study included companion animals presenting to veterinary clinics in South East Queensland. Samples from 402 dogs, 69 cats and 60 horses were collected with owner's consent at six collaborating clinics and/or hospital between November 2015 and December 2017 in Brisbane, Queensland, Australia. These included two private referral hospitals, two veterinary teaching hospitals, and two general practice clinics. The inclusion criteria included animals that had not been hospitalised for more than 24 hours to reduce the likelihood of isolating hospital associated MRS. Nasal and rectal swabs were collected from animals at the time of consultation or within 24 hours of admission. Studies have reported better detection of staphylococci using two samples; and that rectal swabs were better for MRSP and nasal for MRSA detection (15,21). A mixture of healthy and diseased animals were sampled (Supplementary Table 1).

Sample processing
Once samples were collected, the transport media swabs were stored at 4 °C. They were transported to the laboratory for processing within ve days from collection. All swab (larger M40 and smaller E-swabs) samples were initially incubated aerobically in Mueller Hinton broth containing 6.5% NaCl (w/v) overnight (Thermo Fisher Scienti c, Thebarton, South Australia, Australia). Staphylococcus isolation and identi cation was achieved using routine microbiological tests including selective MRSA 2 Brilliance™ agar (Thermo Fisher Scienti c), Gram staining, catalase and coagulase testing. S. aureus colonies culture as a dark denim blue colour, and S. pseudintermedius colonies a lighter denim blue colour on MRSA 2 Brilliance agar. Both organisms are gram positive, catalase and coagulase positive. A multiplex PCR was performed to distinguish between MRSA and MRSP (22). Methicillin resistance was determined phenotypically by performing disk diffusion antimicrobial sensitivity tests using 1 µg oxacillin or 30 µg cefoxitin following CLSI guidelines (23). Resistance was also determined genotypically by the presence of the mecA gene using polymerase chain reaction (PCR) (24).

Risk factor data
Animal demographic and clinical history data up to 12 months prior to sampling was extracted from each animal's medical record and stored in a MS Excel database. Recorded data included signalment, household geographical location, previous medication use, previous bacterial infections, previous hospital visits, hospitalisation, and surgeries (Table 1). Socio-Economic indexes for areas (SEIFA) data were retrieved from the Australian Bureau of Statistics (26). SEIFA data were provided for greater Brisbane postcodes as deciles ordered from lowest (1) to highest (10) scores. SEIFA deciles were extracted for each animal street address by performing a spatial join with the postcode of their place of residence. Due to the distribution of the SEIFA scores in the sample population every two decile categories were merged. Further to that, relevant environmental maps were downloaded from: average temperature and precipitation maps (with a resolution of ~ 1 km 2 ) were retrieved from WorldClim version 2 (27) for the elevation map (with a resolution of ~ 0.8 km 2 ) were retrieved from DIVA-GIS (28), human population density from 2017 (with a resolution of 3 arc approximately 100 m at the equator) was obtained from the WorldPop project site (29)  Risk factors for MRS carriage were evaluated using univariable and multivariable Bernoulli logistic regression models for each species with MRS carriage as the outcome of interest. Location of sampling was included as a random effect in both univariable and multivariable analyses to account for multiple observations at the same clinics and to standardise error. All other factors were included as xed effects. Firstly, univariable models were used to identify any association between the risk factors and MRS carriage based on a conservative p-value of 0.20. Pearson's correlation coe cient was used to test for any correlation between variables. Risk factors that had a p-value of 0.20 or less in the rst phase were then included in a manual backward stepwise variable selection process in a multivariable logistic regression analysis. Confounders were identi ed by assessing the impact of their removal on the coe cients of the remaining variables. If the coe cient of one variable changed by more than 25% when a variable was eliminated, then it was considered a confounder and put back into the multivariable model. Risk factors included in the nal multivariable model with a p-value of 0.05 or less were termed signi cant in this study.
Two multivariable models were created, one model including demographic and clinical factors only and one including demographic, clinical and environmental factors (average land temperature, average precipitation, elevation, human population density, socio-economic status and urban vs rural location data). Regression statistical analyses were performed using the statistical software Stata version 13.1 (Stata Corporation, College Station, TX, USA).

Analysis of residual geographical clustering of methicillin resistant Staphylococcus species
Semivariograms of the residuals from the nal multivariable models (including and excluding environmental factors) were used to quantify the extent of geographical clustering of MRSA or MRSP carriage in dogs after accounting for all risk factors in the nal Bernoulli logistic regression model (31). The semivariogram is a graphical representation of the variation in observations between all pairs of dogs located at a series of de ned separating distances. The semivariogram is de ned by three parameters including the nugget, the partial sill and the range. The nugget is the spatially unstructured variation which represents the natural random variation, very small-scale spatial variability or measurement error. The partial sill is the spatially structured variance which can represent the tendency for geographical clustering. The range is an indication of the size of clusters as it represents the separating distance at which spatial autocorrelation ceases to occur. The proportion of the variance that is spatially structured was estimated by dividing the partial sill by the sum of the partial sill and nugget. This measure indicates the role of location in explaining the variation in MRSA and MRSP carriage in dogs in the greater Brisbane area (also known as the propensity for geographical clustering). Residual semivariograms were produced using the geoR package in R version 3.4.1 (The R foundation for statistical computing, Vienna, Austria).

Spatial risk prediction of methicillin resistant Staphylococcus species and model validation
A model-based geostatistical model of MRS carriage in client owned dogs was built using OpenBUGS version 3.2.3 rev 1012. A total of 391 individual observations were included in the analysis. Initial covariates included sex and age categories (< 1 y, 1-4 y, > 4-7 y, > 7-10 y and > 10 y) as xed effects, and the average values of precipitation, temperature, elevation and human population density extracted at the home address of dog owners as a spatial random effects. Environmental variables were standardised by subtracting the mean and dividing by the standard deviation. A predictive model was run with the age sex combination that had a higher in uence on carriage (being a female and age categories > 7-10 y and > 10 y). The outputs of the Bayesian models are distributions termed 'posterior distributions', which represent the uncertainty associated with the parameter estimates. Posterior means predictions and standard deviations of the mean prediction were categorised and mapped using ArcMap version 10.6.1 (ESRI 2018). The model's ability to correctly predict carriage was analysed using the receiver operating characteristic (ROC) analysis in Stata. Only dog data were analysed as there was no MRSP carriage identi ed in cats and horses sampled in this study. The univariable screening (p < 0.2) determined that sex, neuter status, prior clinic visit, prior bacterial infection, prior hospitalisation, prior surgery, chemotherapy, consultation type, urban or rural location, and all environmental and demographic factors should be retained in the full regression model ( Table 2). The overall p-value for categorical variables was not reported due to unbalanced data where at least one category had less than ve positive cases. In these cases, if the p-value of at least one category was smaller than 0.20 then it was included in the full regression model. The nal multivariable model including environmental factors; i.e. urban and rural data, elevation, temperature, precipitation, and human population density resulted in a better t to the data [i.e. lower Akaike information criterion (AIC) (AIC = 163) compared to the model with demographic and clinical factors only (AIC = 173)]. Dogs were used as the unit of analysis and the location of sampling as a random effect. P < 0.2 is considered signi cant. † Antimicrobial combinations categories: A = dogs received antimicrobials on day of sampling and 1 month prior, some dogs received antimicrobials up to 12 months prior to sampling, B = dogs received antimicrobials on sampling day but not 1 month prior and some dogs had antimicrobials between 1 and 12 months prior to sampling, C = dogs received no antimicrobials on sampling day but did a month prior and some dogs received antimicrobials 2 to 12 months prior to sampling, D = dogs received no antimicrobials on day of sampling and 1 month prior and some did 2 to 12 months prior, E = dogs received no antimicrobials up to 6-12 months before sampling. ‡ Corticoid combinations categories: A = dogs received corticoids on day of sampling and 1 month prior, some dogs received corticoids up to 12 months prior to sampling, B = dogs received corticoids on sampling day but not 1 month prior; and some dogs had corticoids between 1 and 12 months prior to sampling, C = dogs received no corticoids on sampling day but did a month prior and some dogs received corticoids 2 to 12 months prior to sampling, D = dogs received no corticoids on day of sampling and 1 month prior and some did 2 to 12 months prior, E = dogs received no corticoids up to 6-12 months before sampling.  Dogs were used as the unit of analysis and the location of sampling as a random effect. P < 0.05 is considered signi cant. † Antimicrobial combinations categories: A = dogs received antimicrobials on day of sampling and 1 month prior, some dogs received antimicrobials up to 12 months prior to sampling, B = dogs received antimicrobials on sampling day but not 1 month prior and some dogs had antimicrobials between 1 and 12 months prior to sampling, C = dogs received no antimicrobials on sampling day but did a month prior and some dogs received antimicrobials 2 to 12 months prior to sampling, D = dogs received no antimicrobials on day of sampling and 1 month prior and some did 2 to 12 months prior, E = dogs received no antimicrobials up to 6-12 months before sampling. ‡ Corticoid combinations categories: and some dogs had corticoids between 1 and 12 months prior to sampling, C = dogs received no corticoids on sampling day but did a month prior and some dogs received corticoids 2 to 12 months prior to sampling, D = dogs received no corticoids on day of sampling and 1 month prior and some did 2 to 12 months prior, E = dogs received no corticoids up to 6-12 months before sampling.

Spatial dependence and geographical risk prediction of MRSP carriage
The residual semivariogram for the model excluding environmental factors showed clustering of MRSP carriage with an average cluster size of 5.52 km and 47% propensity for clustering, after adjusting for the environmental factors average clusters increased to 6.23 km and the propensity for clustering increased to 69% (Fig. 1 and Table 4). During the learning phase of our geographical model of MRSP carriage risk we found that female dogs over the age of 7 years old (combined 7-10 years and > 10 years age categories) had a higher probability of MRSP carriage compared to other sex/age groups thereby providing a justi cation for our predictive analysis to be performed for this age/sex combination (Supplementary table 1). Our predictive map revealed that the probability of MRSP carriage rates was highest along the coast, ranging from 50% to over 70% (Fig. 2). Areas predicted to have a higher probability of MRSP carriage (> 50% to > 70%) include suburbs belonging to the Brisbane City Council, Sunshine Coast Regional Council, Gympie Regional Council and Gold Coast City areas. Lower carriage rates were estimated further inland, ranging from under 10-20%. Standard deviations associated with the predicted means were between 0.24 and 0.28 in the high-risk areas in Brisbane City and Sunshine Coast areas. The standard deviations were higher (0.28-0.32) in areas where predicted carriage was high in the Northern and Southern ends of the map (Noosa Shire, Gympie region and Gold Coast City). The ROC analysis revealed a ROC area of 0.99940 with 95% con dence intervals 0.99841 to 1.0 indicating a very good discriminatory ability of the model.

Discussion
This study is the rst to analyse the prevalence, associated risk factors and geographical distribution of MRS carriage in South East Queensland, Australia. We identi ed a carriage rate of 8.7% for MRSP and 0% for MRSA in dogs. No MRS was isolated in cats or horses. These ndings are in agreement with a recent study in two Sydney veterinary hospitals that isolated MRSP from 8% of veterinary personnel owned dogs and 7% of general canine hospital admissions (32). Other studies in non-clinical Australian settings have yielded much lower MRSP carriage estimates; only one MRSP was isolated from 117 healthy dogs in Victoria (33) and no MRSP on dogs from remote Aboriginal communities in Western Australia and New South Wales (34,35).While the Victorian and Sydney study did not detect MRSA in cats or dogs, the two studies on dogs from remote Aboriginal communities in Western Australia and New South Wales identi ed MRSA in 2.6% of the dogs (34,35). MRSA has been detected in Australian horses (3.7%) (2) so perhaps the lack of MRSA isolation in horses could be explained by the small sample size and single sampling location in this study, indicating the need to enrol more horses from multiple locations in future carriage investigations.
Our results also agree with existing literature from others countries where carriage rates of MRSP in dogs ranged between 3-34% for studies reported from Spain, Finland, Iran and Canada (15,(36)(37)(38). MRSP is isolated less often in cats with carriage prevalence between 4-19% as reported in the United States, Brazil and Iran (38,39,40). There is currently no MRSP carriage being reported in horses; however, it has been identi ed from clinical infections (41)(42)(43). The low isolation rate of MRSA in dogs and cats was also expected. The reported prevalence of MRSA carriage in dogs ranges between 0.5-9% in Canada, Portugal and the United Kingdom (6,(44)(45)(46). A lower prevalence of MRSA ranging between 0-4% has been reported in cats from Brazil, Portugal and the United Kingdom (6,46,47). MRSA reported in horses is 2-5% in Canada and the United Kingdom (46,48).
Our ndings suggest that MRSP carriage in dogs is associated with previous history of health-care contact including prior hospitalisation, prior bacterial infection and consultation type. The causal link between MRSP carriage and previous hospitalisation and prior bacterial infection is likely to be confounded by the frequency and length of stay. Indeed, previous studies on risk factors for MRS infections had demonstrated an increased risk linked to more frequent and longer admissions to veterinary clinics (37,49,50). This can partly be explained by the increase the likelihood of being exposed and colonised by nosocomial bacteria such as MRS as a function of longer periods spent in healthcare settings or the in uence of treatments. Speci c consultation types, internal medicine and dermatology, seemed to in uence carriage. MRS positive dogs that visited dermatologists had odds twice as high as MRS positive dogs attending internal medicine consultations. Resistant staphylococci are predominately a skin pathogen and are frequently treated by dermatologists (51). Dogs visiting dermatologists are more likely to have a history of persistent MRS infections and increased exposure to antibiotics and corticoids, which could increase the likelihood of MRS isolation from these animals (52). Another point to consider is that higher risks of carriage associated with hospitalisation and dermatology/internal medicine visits do not necessarily mean that transmission is occurring at these locations as dogs visiting these facilities usually experience underlying diseases that expose them to antimicrobials which may facilitate resistance in methicillin susceptible S. pseudintermedius already carried on their skin. This is supported by our nding that dogs with previous bacterial infections had higher odds of carriage. A previous study also indicated that MRSP carriage was higher in dogs that had pyoderma either previously or during sampling (52).
Our analysis is the rst to reveal important ecological risk factors associated with MRSP carriage. population could function as a proxy of a higher pet dog population density. Crowds have been known to be at higher risk of bacterial infections, and so the higher density in human and associated dog populations could result in higher skin contact rates between individuals thus facilitating the dissemination of MRSP through the population (54).
Our ndings also demonstrate that MRSP is highly clustered in southeast Queensland. After adjusting for individual, clinical and ecological risk factors our results still indicated the presence of residual spatial autocorrelation in MRSP carriage with average cluster sizes of 6 km. The propensity for clustering increased from 47-69% when environmental variables were included in the analysis. This result indicates that MRSP transmission is likely to be spatially structured and could be driven by other environmental or behavioural factors not included in our models. MRSP can be transmitted between and persist in household pets and can survive in the environment for prolonged periods of time (55). These factors could result in transmission in the community outside of the hospital environment such as dog parks or beaches. Indeed, our predictive probability of MRSP carriage map (Fig. 2) indicates that there is higher risk of carriage along the eastern coast of Queensland from the Gold Coast to the Sunshine Coast. Risk also increases north along the coast, which is where average precipitation and temperature are higher too. These high-risk areas included the Brisbane City council, Sunshine Coast council, Noosa Shire, and Gympie regional council. Surveillance strategies associated with MRS in companion animals in South East Queensland should target these areas rst. The uncertainty in the spatial prediction (as measured by the standard deviation (SD)) associated with the Brisbane and Sunshine Coast high-risk areas are low (SD: <0.20 to 0.26), which indicates a higher level of certainty associated with these results. The areas with high uncertainty (SD: >034) surrounding the predictions were linked to the northern areas, including Noosa Shire and Gympie, where predicted carriage was between 50 and over 70%. Higher uncertainty (SD: >0.28-0.3) was also associated with the Gold Coast area where there was a high predicted risk of carriage. These areas should be targeted by future sampling studies to ll the gap in knowledge.
The results of our study should be interpreted in light of some limitations. First, there could be bias associated with the clinical sample of individuals we had available for our study. The inclusion of dogs presenting to referral and teaching hospitals might not be entirely representative of a typical dog population. Further, medical records are characterised by incomplete information, which could have contributed to the lack of data for areas found to be associated with high predictive uncertainty for MRSP carriage in South East Queensland such as the Gold Coast and Northern Sunshine Coast. Future studies should aim to investigate MRS carriage in companion animals in this part of Australia. Nevertheless, the ndings from this study are useful to uncover the landscape epidemiology of MRS carriage in Australian dogs. Second, the ecological nature of our model could have contributed to a lack of statistical support for the socioeconomic variable which as previously been postulated to be an important factor in antimicrobial resistant bacteria epidemiology, including MRSA (56,57). While SEIFA scores were signi cant at the univariable analysis the fact that the SEIFA scores were available at the postcode level could have contribute to the loss of statistical support due to presence of regression dilution bias (58).
Further studies should consider obtaining detailed information of owners SES to account for the variation in MRSP risk identi ed in this study. In addition, our results indicate that our model failed to account for all the residual geographical clustering of MRSP and future studies should consider adjustment for additional ner-scale factors that could in uence MRSP carriage status in dogs such as in-house pet contacts, presence of an MRSP infected dog, use of disinfectants at home, as well as human carriage.

Conclusions
In conclusion, MRSP carriage in dog populations of Brisbane and surrounding areas contrasts with other Australian ndings but is within the range of previous published studies worldwide. The associated risk factors include prior bacterial infection, prior hospitalisation, consultation types, average precipitation and human population density. Clustering of 6 km is evident in Brisbane and surrounding areas and higher risk of carriage was found along the coast, particularly Brisbane city, Sunshine Coast and Gympie Availability of data and materials The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.