Seroprevalence and associated risk factors of mosquito-borne alphaviruses in horses in northern Queensland

Survey and study area

Ethics approval was obtained from James Cook University Human and Animal Ethics Committees (approval nos. H5536 and A1997, respectively) to carry out a cross-sectional study, commencing September 2013 through June 2014, of the domestic horse population of northern Queensland (regions north of Rockhampton and Longreach). Serum samples were collected from horses of consenting owners participating in regional equestrian events and from horses admitted to regional equine veterinary practices. The majority of samples were collected at participating equestrian events. The study excluded horses that were severely ill or had a history of chronic infections or underlying immunosuppressive disease. The exclusion criteria included horses that were presented for emergency evaluation or surgery, had clinical evidence of severe systemic metabolic or cardiovascular disturbance, an infectious disease process, lymphopenia and/or neutropenia or had been diagnosed with an immune-mediated condition. Horses that had not been in the study area for at least 6 months prior to the time of blood collection were also excluded from sampling.

The number of horses to be included in the survey was based on the sensitivity (99%) and specificity (84%) of the serum microneutralisation test,8 an assumed prevalence of 90% RRV-seropositive horses² and an allowable error of 5 % using a confidence level of 95%. Using these assumptions, a sample size of 189 horses was calculated for the prevalence survey.9, 10 The assumed prevalence of 90% for RRV was based on a pilot study of 10 horses tested at James Cook University prior to the survey. The pilot horses were used for teaching purposes and had been resident in northern Queensland for several years.

Sample collection and storage

A veterinarian collected a 10-mL red stopper (plain glass) and a 10-mL green stopper (EDTA) vacutainer blood sample from the jugular vein of each horse using a 19-gauge needle. The red stopper samples were stored at +4°C overnight and then centrifuged at +20°C for 15 min at 18,000 rpm. Serum from the red stopper tubes was aliquoted into sterile cryovials and stored at −20°C until analysis. The EDTA plasma samples were not used in this study and will not be discussed further.

Questionnaire survey

A structured questionnaire survey was completed by each horse owner at the time of blood collection or via a telephone interview thereafter. The questionnaire covered the previous 12-month period of the horse and comprised 32 questions designed to acquire information on the horse’s demographic profile (age, sex, breed), associated geographic and environmental factors (location of the property on which the horse was housed in reference to coastal areas, marshlands, waterways and wetland and the type of agricultural activities prevalent in the area), management factors (water sources, housing conditions, other animal species present in the horse paddock and in adjoining paddocks) and observance of clinical signs that might be associated with viral infection (ataxia, submandibular lymphadenopathy, distal limb and/or ventral abdominal oedema, muscle stiffness, pyrexia, abortion, still births). Information regarding the annual average rainfall and temperature corresponding to each property was also obtained from the Australian Bureau of Meteorology (Climate statistics for Australian sites, Commonwealth of Australia 2014; http://www.bom.gov.au/).

Sample analysis

Serum samples were tested at James Cook University, Townsville, for the presence of antibodies to RRV T48, BFV K5540 and WAV M78 using the microneutralisation assay method. The Reed and Muench tissue culture infectious dose 50 calculation method was used to establish the working dilution of viral antigen for each of the three viruses.11 Each serum sample was transferred in batches of 80 to an 8 × 12-well clear polypropylene plate with a non-treated surface, labelled accordingly and frozen at −20°C until required.

Sera samples were serially diluted in a 96-well flat-bottom microtitre plate (Sarstedt, Germany) using Dulbecco’s modified eagle medium (DMEM with 5% fetal bovine serum [FBS]) to give final dilutions between 1 : 10 and 1 : 1280.

DMEM with 5% FBS was added to column 12 to act as a cell control while column 11 was the viral control. Plates were covered and incubated for 1 h at 37°C in a 5% CO₂ atmosphere. Aliquots of 50 μL containing a suspension of 20,000 Vero cells (4 × 10⁵ cells/mL) in DMEM with 5% FBS was added to all wells of the plates, which were then incubated at 37°C in a 5% CO₂ atmosphere and monitored daily for 6 days when cytopathic effect was recorded and serum antibody titres were calculated. This process was repeated for each virus until all serum samples had been tested for all three viruses.

Data analysis

Apparent prevalence (AP) and its corresponding 95% confidence interval (CI) of each virus were calculated using proportions of properties testing positive to the specific virus. Any property with at least one horse reacting positively to the virus at 1 : 10 dilution was assumed to be seropositive. Similarly, the AP of each virus on horse properties located within the Far North Coast–Tableland region, the Townsville–Burdekin region and the Mackay–Whitsunday region, together with their 95% CIs, were also determined and compared. The CIs for AP were calculated using formulae adopted from Cameron.12

Survey data were entered into Microsoft Excel (2007) and the computer software program NCSS (NCSS Inc., Kaysville, UT, USA) for analysis. The associations between seropositive horses/properties and various risk factors were analysed using Chi-square tests (i.e. univariate analysis). Risk factors with a P-value < 0.20 on univariate analysis were selected to be included in the multivariable logistic regression models.13 A hierarchical stepwise forward elimination process with switching of variables based on log likelihood values was used to determine the best fitting logistic regression model. NCSS has a built-in algorithm that does the selection process and several iterations were run before the final model was selected, which contained the best fitting log likelihood value and only the variables with a P-value < 0.05 on the Wald test.14, 15

Ross River virus

Distribution of the microneutralisation assay titres for RRV ranged from 10 to 1280, with the most frequently recorded titre being 80 (29% of serum samples, Figure 2). Of the 287 horses, 263 showed exposure to RRV at serial two-fold dilutions (1 : 10 to 1 : 1280) of serum, correlating to a seropositivity of 91% (95% CI 88–95%). The samples originated from 147 properties across northern Queensland and 134 (91%) of these were found to have horses that had been exposed to the virus (Figure 3). Adjusting for clustering among properties,10 the true prevalence of RRV antibodies in horses within the areas surveyed was found to be between 87% and 96%.

From the Far North Coast–Tableland region, 38 horses residing on 28 properties were sampled, of which 31 from 23 properties were found to be seropositive to RRV. In the Townsville–Burdekin region, 201 horses residing on 89 properties were sampled and 190 from 83 properties were found to be seropositive, while in the Mackay–Whitsunday region 49 horses residing on 30 properties were sampled and 42 from 28 properties were seropositive. The proportion of seropositive properties in the Far North Coast–Tableland, the Townsville–Burdekin region and the Mackay–Whitsunday regions was 82% (95% CI 74–90%), 93% (95% CI 90–96%) and 90% (95% CI 88–98%), respectively. No significant difference could be shown between each of the regions sampled (χ² = 3.5, df = 2, P = 0.16).

The seroprevalence of RRV by demographic, geographic and management factors is presented in Table 1. Seroprevalence of RRV did not differ significantly between the sexes (χ² = 0.10; df = 2; P = 0.95). The age of the sampled horses ranged from 1 to 33 years, with a median age of 11 years. No significant association between seropositivity and age group was detected (χ² = 2.94, df = 5, P = 0.71). Horses were also categorised by breed encompassing the highest represented purebred breeds; crossbred animals and the remaining horses not represented with enough numbers to be considered as a category were grouped as ‘other’. No significant association between seropositivity and breed was observed (χ² = 4.36, df = 5, P = 0.50). However, with a P-value of < 0.2 for Thoroughbred horses and crossbred horses (P = 0.17 and P = 0.11 respectively), both were included in the multivariable logistic regression models.

On the horse properties included in the survey the annual average temperature ranged between 19.1°C and 25°C; the median temperature was 24.4°C. The 147 horse properties were divided into one of two categories according to their annual average temperature (< 23°C and ≥ 23°C). The horse properties were also divided into one of two categories according to their annual average rainfall (< 1000 mm and ≥ 1000 mm). On the horse properties included in the survey the annual average rainfall received ranged between 393 and 3573 mm. There was no significant association between seropositivity and either annual average temperature (Yates’ χ² = 3.04, df = 1, P = 0.11) or rainfall (Yates’ χ² = 1.01, df = 1, P = 0.17). P-values for both temperature and rainfall were < 0.2 in the univariate analysis and were therefore included in the multivariable logistic regression models. In order to account for possible collinearity, temperature and rainfall were analysed separately within the models.

Association with seropositivity and horse properties with large numbers of mosquitoes in the previous 12 months was found to be significant (χ² = 4.23; df = 1; P = 0.03) in the univariate analysis and this variable was analysed further within the multivariable logistic regression models. It was found to be the only remaining risk factor associated with RRV seropositivity (P = 0.06, Table 2) if a significance level of 10% was regarded as acceptable. The presence of other animal species, including avian (domestic and wild), bovine, porcine (domestic and feral), canine (including dingoes) and marsupial, on the property and/or on those within the neighbourhood, and their influence on the status of the property being seropositive to the virus, were also analysed. Avian species had a P-value < 0.2 in the univariate analysis, but the variable was removed during multivariable logistic regression analysis because of a combination of the small category size and all outcomes being positive, which in turn disrupted the model. Permanent or temporary flying fox roosts on or near the properties were not identified or analysed.

Horse properties were further analysed to determine if the locality of the property in close proximity to waterways, marshlands, coastal regions, wetlands, creeks, rivers and/or dams had an influence on the status of the property being seropositive to RRV. No significant association between seropositivity and proximity to such water sources could be shown (Yates’ χ² = 0.18; df = 1; P = 0.34).

Within the questionnaire, horse owners were asked if their horses had exhibited signs related to RRV at any stage over the 12 months prior to the animals being tested. Only horses that demonstrated antibodies at a titre ≥ 1 : 640 were involved at this stage of the study. In total, 14 horses were identified with titres ≥ 1 : 640. Analysis of the data on ataxia, stiffness and soreness (both muscular and skeletal), inflammation of joints, swelling of the limbs, high fever, respiratory conditions, episodes of colic, lassitude and periods of abnormal irritability did not show any definitive association of these signs with seropositivity to RRV. Of the 14 horses assessed, four showed at least one of these symptoms. One horse was noted to have experienced inflammation of joints as well as stiffness and muscle tenderness, and another experienced joint stiffness and muscle tenderness only. The third displayed a prolonged period of lassitude and signs of very little energy, and the fourth showed signs of a respiratory condition.

Barmah Forest virus

A total of 22 horses of the 288 tested, originating from 22 of the 147 properties across northern Queensland, showed evidence of exposure to BFV (seropositivity 7.6%; 95% CI 6.00–9.00%). The total proportion of properties found to be seropositive to the virus was 15% (95% CI 12.00–18.00%).

From the Far North Coast–Tableland region, 38 horses residing on 28 properties were sampled, of which six horses from six different properties showed evidence of exposure to BFV. In the Townsville–Burdekin region 201 horses residing on 89 properties were sampled and 13 horses from 12 properties were found to be seropositive, while in the Mackay–Whitsunday region 49 horses residing on 30 properties were sampled and four horses from four properties were seropositive. The proportion of seropositive properties in the Far North Coast–Tableland, the Townsville–Burdekin region and the Mackay–Whitsunday regions was 21.4% (95% CI 13–30%), 13.5% (95% CI 10–17%), and 13.3% (95% CI 7–20%), respectively. No significant difference could be shown between each of the regions sampled (χ² = 1.14, df = 2, P = 0.57). The seroprevalence of BFV by demographic, geographic and management factors is presented in Table 3. Seroprevalence of BFV did not differ significantly between the sexes (χ² = 2.87; df = 2; P = 0.24). P-values for both female horses and geldings were < 0.2 and therefore included in the multivariable logistic regression models. No significant association between seropositivity and age groups was detected (χ² = 6.21, df = 5, P = 0.20). Horses aged between 21 and 25 years and those ≤ 5 years old recorded a P-value < 0.2 and were included in the multivariable logistic regression models. Analysis of the seroprevalence of BFV and breed of the horses showed a significant association between seropositivity and breed (χ² = 11.56, df = 5, P = 0.04). Quarter Horses, Thoroughbreds, crossbred horses, ponies and ‘other’ breeds were all noted to have P-values < 0.2 and were therefore included in the multivariable logistic regression models.

No association could be shown between the seroprevalence of BFV on the horse properties and annual average temperature (Yates’ χ² = 0.01, df = 1, P = 0.50), amount of rainfall (Yates’ χ² = 0.0, df = 1, P = 0.50) or the location of the horse properties in close proximity to waterways, marshlands, coastal regions, wetlands, creeks, rivers and/or dams (Yates’ χ² = 0.03; df = 1; P = 0.43). Unlike RRV, large numbers of mosquitoes were not found to be significantly associated with seropositive horse properties (χ² = 0.45; df = 1; P = 0.25) in the univariate analysis. The presence of other animal species, including avian, bovine, porcine, canine and marsupial, on the property and/or on those within the neighbourhood, and their association to BFV were also analysed. The presence of avian, bovine, canine and porcine species (domestic and wild) were all found to have P-values < 0.2 in the univariate analysis and were analysed further within the multivariable logistic regression models. Avian and porcine species were found to be the only remaining significant associations with BFV seropositivity (Table 4).

Whataroa virus

Previous exposure to Whataroa-like virus was detected in one horse in the Far North Coast–Tableland region and one donkey in the Mackay–Whitsunday region (3.6%; 95% CI −3.3–10.45% and 3.3%; 95% CI −3.3–9.76%, respectively). This was the only donkey bled in the study and it was included at the request of the owner. The proportion of properties found to be seropositive to WHAV was 1.36% (95% CI −0.51–3.23%). Because of the low prevalence and the high probability of false-positive results, analysis of risk factors associated with prevalence of WHAV was not conducted.