Ovine serum samples submitted to Animal and Plant Health Agency (APHA) (formerly the Animal Health and Veterinary Laboratories Agency) – Weybridge regional laboratories in England and Wales for diagnostic and monitoring purposes between 2005 and 2012 were investigated for possible spatial and temporal variations in seropositivity to Toxoplasma gondii infection. Of the 4354 samples tested by latex agglutination, 2361 (54.2 per cent) were seropositive. No correlation between seropositivity and climatic conditions was identified by mixed-effects modelling using meteorological data summaries. The proportion of seropositive samples collected during November was found to be significantly lower than those collected during other months and samples from the North West England and North Wales Regions had significantly lower odds of being positive. Spatial cluster analysis identified a significantly higher proportion of seropositive animals in East Anglia and the South, East and Midlands of England. Spatio-temporal cluster analysis detected a single significant cluster of seropositive animals dating from January 2006 to January 2011, which covered a large proportion of the farm locations. As well as confirming high overall levels of infection within the national flock, these findings also indicate possible temporal and regional variations in exposure of sheep to T. gondii.
- Public health
Statistics from Altmetric.com
Toxoplasmosis, a disease caused by the protozoan parasite Toxoplasma gondii, is considered to be the most common parasitic zoonosis worldwide (EFSA 2007). The parasite has a complex life cycle, and although virtually all warm blooded animals can be infected as intermediate hosts, sexual reproduction can only occur in intestinal epithelium cells of the definitive host species, namely domestic cats and other felids. T. gondii has three main stages, all of which are infective to both definitive and intermediate hosts. Infection can result from ingestion of sporozoites found in faecal oocysts of infected cats and bradyzoites that form tissue cysts in intermediate hosts, while congenital infection may occur through transplacental passage of circulating tachyzoites in an acutely infected pregnant animal (Tenter and others 2000, Dubey 2004, PHE 2011).
Human postnatal infection can result from ingestion of either viable tissue cysts in undercooked meat or sporulated oocysts in food or water contaminated with faeces from infected cats. Although serological tests capable of identifying sporozoite-specific antigen have recently been reported (Hill and others 2011), commonly used diagnostic tests are unable to differentiate between oocyst and tissue cyst-derived infection and the relative importance of each of these sources of infection remains unknown (Cook and others 2000, Dubey 2004, Kijlstra and Jongert 2008).
Infection is usually asymptomatic in immunocompetent people and only mild flu-like symptoms are experienced during the acute stages in about 10 per cent of cases. However, infection of a non-immune woman during pregnancy can result in congenital disease, the clinical severity of which is inversely proportional to gestational age. Severe disease, including chorioretinitis and encephalitis, can also occur as a result of primary infection or reactivation of chronic infection in immunosuppressed or immunocompromised individuals (Montoya and Liesenfeld 2004, PHE 2011, FSA 2012).
Seroprevalence of Toxoplasma infection in people varies considerably according to age and geographical location. Antibodies to Toxoplasma can be detected in 20–40 per cent of UK adults, whereas seroprevalences as high as 77.4 per cent have been reported in other areas of Europe (FSA 2012). Variation is also apparent at the national level, with blood donor sample surveys revealing higher seroprevalences in western areas of Great Britain compared to the east. These studies also indicate a positive, non-linear relationship between age and seroprevalence in the population of the UK, although in the absence of further studies it is not clear whether this effect is due to falling levels of human infection over time or whether susceptibility to infection increases with age (EFSA 2007, FSA 2012). Although seroprevalence data can be used to estimate levels of T. gondii infection in the population, significant problems have been encountered when trying to quantify the public health burden of toxoplasmosis, largely as a result of under-detection and under-reporting of clinical, and in particular, congenital toxoplasmosis (B. Bénard and L. R. Salmi, unpublished report, EFSA 2007, FSA 2012).
Herbivores acquire T. gondii infection by ingestion of sporulated oocysts in contaminated pasture, feed or water, or congenitally by transplacental passage of tachyzoites from dam to foetus. It is estimated that infection in sheep can result from ingestion of as few as 200 sporulated oocysts (McColgan and others 1988). In a similar manner to human infection, exposure of immunologically naive sheep during early or mid-pregnancy can cause fetal death, stillbirth or the birth of weak lambs, while clinically normal but infected lambs can occur as a result of congenital infection during the later stages of gestation (Buxton and others 2007, Innes and others 2009). Toxoplasmosis is therefore a major concern to the UK sheep industry, with estimated losses of £12–24 million resulting from early embryonic death and abortion of approximately 0.5 million lambs annually (Defra n.d., Innes and others 2009). Veterinary Investigation Diagnosis Analysis data for 2012 indicated that T. gondii infection was responsible for 18.5 per cent of all diagnosed cases of sheep and goat fetopathy in Great Britain, with only Chlamydophila abortus and Schmallenberg virus accounting for a greater percentage of losses (Defra 2013).
A recent serological survey of breeding ewes in Great Britain identified seroprevalences of 74 per cent and 100 per cent at individual animal and flock level, respectively, while a similar study of Scottish sheep indicated the same seroprevalence at flock level and an overall seroprevalence of 56.6 per cent among individuals. The fact that both studies also showed an increasing seroprevalence with age provides further evidence to suggest that most infections occur in the postnatal period (Hutchinson and others 2011, Katzer and others 2011). Significant regional variation in within-flock seroprevalence to T. gondii infection in sheep has also been identified in some studies. In Scotland, for example, median within-flock seroprevalence ranged from 42.3 per cent in the south of the country to 69.2 per cent in the north (Katzer and others 2011). Similar gradients of seroprevalence have also been reported in sheep in France, in sheep and cervids in Finland and in roe deer and moose in Norway (Vikøren and others 2004, Halos and others 2010, Jokelainen and others 2010).
Oocysts in freshly voided cat faeces are non-infective until sporulation occurs, usually one to five days after shedding, depending on environmental conditions. Viability of sporulated oocysts is in turn influenced by temperature, moisture and aeration, and they can remain infective for up to 18 months at optimal climatic conditions (Lindsay and others 2002, Dubey 2004, Buxton and others 2007, EFSA 2007, Innes and others 2009, Meerburg and Kijlstra 2009). It has therefore been speculated that geographical differences in seroprevalence to T. gondii infection in herbivores could be related to climatic variation and its influence over oocyst sporulation, viability and dispersal, and the ecology of arthropod and annelid transport hosts (Dubey 2004, Buxton and others 2007, Innes and others 2009, Meerburg and Kijlstra 2009).
This paper describes the results of a study that used serological data derived from sheep blood samples submitted to the Animal and Plant Health Agency (APHA) (formerly the Animal Health and Veterinary Laboratories Agency) – Weybridge (APHA – Weybridge) laboratories in England and Wales during an eight-year period. The aim of the study was to identify overall levels of seropositivity in sheep serum samples received for Toxoplasma serology by APHA – Weybridge between May 2005 and December 2012 and to investigate possible temporal and spatial variation in seropositivity to T. gondii infection. Statistical analysis using meteorological data was also used to explore possible associations between climate and seropositivity in sheep.
Materials and methods
Study population and sample selection
Sheep serum samples submitted by veterinary practitioners and farmers to the 14 APHA – Weybridge regional laboratories in England and Wales between May 2005 and December 2012 were used for the study. All samples submitted for abortion investigation or disease monitoring purposes where T. gondii serology was subsequently performed were included in the study.
The date of submission was recorded for each submission. Reason for submission, County Parish Holding (CPH) number, farm address and postcode were also to be captured on the submission paperwork. CPH or farm address details were available to identify an individual farm for 824 of the 1043 submissions received during the study period. This information allowed for the identification of multiple submissions received from 73 holdings. Assuming that there were no multiple submissions from holdings that submitted samples without CPH details, the dataset included 951 farms (Table 1). Data on age, breed and vaccination status of the sampled animals were not recorded.
Each submission was linked to 1–47 samples (mean=4.2, median=3), giving a total of 4354 samples tested for Toxoplasma.
Serum samples were tested by latex agglutination test (LAT); the standard ISO 17025 (UKAS) accredited test used by APHA – Weybridge to detect T. gondii-specific antibody (both IgM and IgG) in animal serum samples. A series of doubling dilutions (1:16 to 1:2048) was made for each serum sample. Samples exhibiting significant agglutination at a serum dilution factor (antibody titre) of 1:64 were defined as positive (Barker and Holliman 1992, MAST diagnostics 2004).
Statistical analysis: identification of risk factors
A dataset for statistical analysis was produced with the following explanatory variables for each record: month, season and year of submission and the Met Office geographical region (Region) of the sampled farm. Additionally, sinusoidal components (sine and cosine terms) for 3-, 6- and 12-month period temporal trends were added to each record to allow for the individual modelling of quarterly, half-yearly and yearly cycles (Chatfield, 2003).
Meteorological data of monthly regional summaries, including actual and ‘anomaly’ (difference from long-term averages) records, were gathered from the Met Office website (http:www.metoffice.gov.ukclimateukindex.html) for minimum, maximum and mean temperature (°C), hours of sunshine, rainfall (mm), days of rain and days of frost. These 14 explanatory variables were linked to the Toxoplasma dataset by the Region of the farm and the month of sample collection. Records that did not contain sufficient farm location information (CPH, postcode) to be linked to a Region were omitted from risk factor analysis.
Univariable logistic modelling was completed on the dataset in Stata V.10 (StataCorp), using the xtmelogit command, to analyse the spatial, meteorological and temporal variables to detect statistically significant associations with sample-level Toxoplasma presence (positive or negative). To account for multiple samples being collected from the same holding, mixed-effects modelling was used, with the submission reference included as the random effect to account for this non-compliance with the independence assumption. Any variables with a P value under 0.20 were included in a multivariable model that could determine which variables remained significantly (P <0.05) associated with Toxoplasma serological-positive results while accounting for the variance accounted for by the other variables included in the model. A backwards stepwise procedure was used to remove variables from the model utilising a likelihood ratio test to test whether the removal of a variable would not significantly (P >0.05) reduce the fit of the model. At each step, the variable that had the least effect on the model fit was removed until only those variables remained for which removal would cause a significant difference to the model.
Statistical analysis: identification of clusters (spatial and spatio-temporal)
The CPH of each record was linked to data recorded in APHA – Weybridge databases to determine the X and Y geographical coordinates for each sampled farm location to allow spatial analysis. Where the CPH was missing from a record, the farm address and postcode information was used to determine the X and Y coordinates. Records that did not contain either farm location or postcode information could not be linked to an X and Y coordinate and were omitted from the scan statistic analysis.
To examine whether statistically significant spatial and spatio-temporal clusters were present in the results, analyses utilising the scan statistic were completed using SaTScan (Kulldorff and Nagarwalla, 1995, Kulldorff, 1997). The analysis identifies the position and size of the most likely clusters of high proportions of Toxoplasma seropositive samples by comparing the relative risk of being Toxoplasma positive within a circular spatial window in comparison with the relative risk outside of the area. The most likely main cluster is identified when the relative risk is above that expected and the cluster has the maximum likelihood of representing the study population. Secondary clusters are also detected in this manner but only if they do not overlap the main cluster. The standard maximum cluster size was used (50 per cent of the population at risk) and Monte Carlo simulation using 999 replicates was completed to assess the significance (P value) of the identified clusters (Kulldorff, 1997).
A discrete Poisson-based model, assuming that the number of seropositive samples was proportional to the population of samples at each location, was used to model each sample Toxoplasma result at a farm location rather than classifying the farms into a binary positive or negative status (Kulldorff, 1997). This model structure was used for both spatial and spatio-temporal analyses. For the spatio-temporal cluster detection, time was aggregated to the level of months. The spatio-temporal analysis does not require population data to be specified at each time point and rather estimates the population using a linear interpolation based on the population at the time points immediately preceding and immediately following to calculate the population at risk at each time point.
Of the 4354 samples tested, 2361 (54.2 per cent) were seropositive and came from 807 (77.4 per cent) of the 1043 submissions (1–18 per submission). Reason of submission was available for 3048 samples. Of these, 2558 (84.0 per cent) were classified as diagnostic and 489 (16.0 per cent) as monitoring. There was no appreciable difference in the proportion of seropositive samples within each group (46.1 per cent for diagnostic samples compared with 45.1 per cent for monitoring samples).
Identification of risk factors
A population of 899 submissions could be linked to a Region with meteorological summaries (2067 positives from 3705 samples). This included 18 samples (0.49 per cent) submitted to English and Welsh APHA – Weybridge laboratories that originated from farms located in Scotland. Univariable mixed-effects modelling was completed and three of the explanatory variables in the dataset had a P value <0.20. These were Region, month of submission and the monthly ‘anomaly’ summary of the number days of rain (≥1 mm of rain) within that Region. After backwards stepwise selection, only a single variable (Region) was retained in the model.
A significantly lower odds of a sample being positive was detected when the sample originated from the North West England and North Wales Met Office Region (OR 0.50: P<0.05, Table 2). This was supported but the multivariable model, which showed that samples from these regions had a significantly lower odds of being seropositive compared with the baseline (OR 0.45: P<0.05, Table 3). The univariable and multivariable models also detected significant differences between months of submission (November: OR 0.49: P<0.05, Table 2, and OR 0.41: P <0.03, Table 3, respectively). Although the sample results indicated a possible seasonal variation in Toxoplasma seropositivity (Fig 1), a significant association was not detected for year, season or any of the three sinusoidal components (yearly, half-yearly and quarterly temporal cycles). However, the results by year did show that there was a non-significant reduction in the percentage of seropositive results in 2012 (48.7 per cent), whereas the results from 2005 to 2011 ranged from 51.4 to 56.8 per cent. None of the monthly meteorological summaries were found to be significantly associated with the Toxoplasma serology results, although the ‘anomaly’ summary of the number of days of rain was positively associated with a greater odds of being seropositive, although not below the 0.05 significance level (Tables 2 and 3).
Spatial and spatio-temporal cluster analysis
The Scan statistic was completed on the 884 submissions (2040 positives from 3649 samples) with X and Y coordinates. The spatial cluster analysis identified a significant cluster (P<0.05) with a significantly higher proportion of seropositive animals being located in East Anglia and the South, East and Midlands of England (Fig 2). This cluster included 180 submissions (148 farms, 583 positive samples) with a relative risk of 1.27, indicating that samples within the cluster were 1.27 times more likely to be seropositive than those submissions collected outside of it. Reducing the size of the maximum cluster size to 40 per cent did not change the size of the detected cluster.
The spatio-temporal cluster analysis detected a single significant cluster dating from January 2006 to January 2011, including 510 submissions (457 farms, 1099 positive samples) with a relative risk 10.50. The cluster covered a large proportion (58 per cent) of the farm locations in the studied population, apart from those in South-Western and Northern England, Western Wales and the small number of locations in Scotland (Fig 2). A reduction of the maximum cluster size to 40 per cent reduced the spatio-temporal cluster to a single cluster dating from January 2009 to January 2010, including 374 submissions (330 farms, 404 positive samples) with a relative risk 30.24. The cluster remained disperse, including the same spread of regions as the original spatio-temporal cluster.
It is acknowledged that some degree of selection, submission and testing bias is inevitable when using data obtained from clinical samples rather than a random cross-sectional population. For example, since the majority of samples were taken for diagnostic purposes, it is likely that flocks with a recent history of abortion will be over-represented. Furthermore, previous surveys have shown that seroprevalence in sheep tends to increase with age (Halos and others 2010, Hutchinson and others 2011, Katzer and others 2011), and although it is probable that most of the samples used in this study will have been taken from adult breeding female sheep, possible age-related effects could not be accounted for due to lack of data.
Similarly, since no details of vaccination status were available, it is possible that a number of false positive results may have occurred through the inclusion of vaccinated animals in the survey. The LAT method is unable to differentiate between vaccine-induced and naturally occurring antibody and studies have shown that antibody titres in previously naive sheep peak at 1/64 within two weeks of vaccination and fall to 1/32 or below within 12 weeks (Maley and others 1997). However, it was considered likely that the attending veterinary surgeon will have considered the probability that the animal was vaccinated before submitting the sample and therefore, although the proportion of vaccinated animals was not known, it is likely to be lower than the 6.2 per cent of vaccinated animals identified in a recent similar serological survey of British breeding ewes (Hutchinson and others 2011). The use of LAT on sheep sera was further complicated by the lack of sensitivity and specificity data. Although a test sensitivity of 99 per cent and specificity of 81 per cent has been described by Barker and Holliman (1992), these data related to the use of human sera and are not applicable to sheep. It was therefore not possible to estimate the possible effect of false positive or false negative results on the overall levels of seropositivity. The possibility of false positive results arising due to cross-reaction with other coccidian parasites should also be considered since reactions with unspecified IgM have been reported when using LAT on human sera (Barker and Holliman 1992, EFSA 2007). However, notwithstanding these limitations, the overall proportion of 54.2 per cent positive samples identified by this study provides further evidence of high levels of exposure to T. gondii within the national flock as described previously (Hutchinson and others 2011, Katzer and others 2011).
The univariable and multivariable models did not differ for which responses of the variables were significantly associated with Toxoplasma seropositivity, although the multivariable model presented adjusted outcomes once the effect of the other variables in the model had been accounted for. The sample results indicated that a possible seasonal variation in Toxoplasma seropositivity was present in the results, with samples collected in spring (March–May) shown to have a low proportion of seropositives and those collected in winter (December–February) having the highest. However, this was not found to be significant in the univariable model when multiple samples from within the same submission were taken into consideration. It is interesting to note that although the multivariable model indicated that samples taken during November were significantly less likely to test positive than the baseline, the percentage of positive samples was not consistently the lowest in November of every year. It should also be noted that samples collected in February had a low odds of being positive, which was approaching significance (P 0.096). Seasonal variation in the prevalence of acute toxoplasmosis has previously been reported in people, but there is little or no evidence to suggest that similar fluctuations occur in sheep, and since the LAT test is not capable of distinguishing between IgM and IgG antibody, it was not possible to distinguish between recently acquired and historic infection in this study (Meenken and others 1991, Logar and others 2005, Bobić and others 2010).
The ‘anomaly’ summary of rainfall was the only meteorological summary variable included in the multivariable model and was positively associated with seropositive results, although the results were only approaching significance (P=0.062). This may, however, have been influenced by a lack of resolution as the weather data came from monthly regional summaries rather than specific weather conditions at that farm prior to submission. The multivariable model also highlights that North West England and northern Wales had the lowest odds of being seropositive and that this was significant when compared against the baseline of East Anglia. No other Regions had a significant difference detected, although the low odds in the Midlands was approaching significance (P=0.089).
As not all submissions had information provided that could link them to an individual farm, it was not possible to use a farm identifier in the model to account for multiple submissions from the same farm. Although the use of the submission identifier improved the fit, it is possible that the addition of a farm identifier as a random effect may have further improved the model fit.
The significant spatial cluster of seropositive samples in Southern and Eastern England appears to reflect geographical variation in seroprevalence as previously identified in Scotland, France, Finland and Ghana (Halos and others 2010, Jokelainen and others 2010, Katzer and others 2011, Van der Puije 2000). Although differences in experimental approach, local husbandry practices and cat population density preclude direct comparison of previous surveys with the results of this study, it is nevertheless possible that these findings may in part be related to the influence of environmental temperature and precipitation on oocyst sporulation, viability and dispersal (Dubey 2004, EFSA 2007, Meerburg and Kijlstra 2009). Although the relative risk for the cluster was small (1.27), it was related to a large population of samples and that may explain why this cluster was statistically significant.
The spatio-temporal cluster detected covered over half of the farm locations in the dataset and ranged across a five-year period. The relative risk of a sample from the cluster being seropositive was large compared with those submitted outside of it. The cluster shape and size provides a useful description (especially after graphical presentation) that presents such issues as the reduction in seropositivity in the final year of study and in samples from the geographical locations not covered by the cluster. In view of the fact that the ‘anomaly’ summary of rainfall results were approaching significance, it is tempting to speculate this cluster of seropositive samples may have resulted from increased exposure due to favourable climatic conditions for oocyst survival and dispersal in the five-year period (Dubey 2004).
These spatial and spatio-temporal results highlight the benefit of using the scan statistic, which can detect significant ‘hot spots’ of positive results in time and space without being limited by artificial regional boundaries such as those used in the modelling analysis. These analyses could be used in the future to detect clusters that require further investigation, especially where a cluster is large and may not have been detected at a local level.
It is possible that the interpolation of the population in the spatio-temporal model overestimated the relative risk of the cluster by under-representing the population at risk around the submission periods within the study period. However, the analysis was still useful at identifying the most likely cluster. Although reducing the maximum cluster size from 50 to 40 per cent did not affect the spatial aspect of the spatio-temporal cluster, it did reduce the scale of the number of months in the cluster. This may indicate that the geographical size of an outbreak was more important than the temporal duration. Both the spatial and spatio-temporal analyses may have been limited by the use of only a circular scanning window as this would have reduced power in detecting long and narrow clusters, and a further improvement to the cluster detection method would have been the use of other cluster shapes.
The results of the analyses may have been biased by the selection of the population at risk. As only a minority of samples came from healthy animals tested for monitoring purposes, the risk may have been overestimated by using samples that were more likely to have Toxoplasma. However, the proportion of seropositives detected in the diagnostic samples from ewes with abortions was similar to the proportion from the monitoring samples. The studied population also did not cover many sheep farms and the Toxoplasma status on these farms may have influenced the size and shape of the detected clusters. The lack of submissions from these farms may indicate a lack of abortions and subsequently a reduced risk of Toxoplasma, and so if these ‘missing’ farms were present within the identified cluster, then it may have reduced the relative risk and significance. The dataset may also have contravened scan statistic Poisson model assumptions that the number of seropositives was proportional to the size of the population and that multiple submissions from the location would be independent. It would be likely that on a farm if one animal is seropositive then it is more likely that other animals were also infected. The management and conditions on a farm might also mean that the farm would be more likely to have multiple occasions of abortion problems and so would provide multiple submissions to the study population. These factors would have over-represented these submissions or farms in the scan statistic analysis.
Unfortunately, there was insufficient data to examine other potentially important variables that might influence levels of T. gondii exposure at a regional level. For example, husbandry practices such as presence of cats on the farm, use of surface drinking water sources, housing during lambing and size of farm have previously been identified as risk factors associated with seropositivity to T. gondii (Skjerve and others 1998, Van der Puije and others 2000, Vesco and others 2007, Dubey 2009, Katzer and others 2011). Despite considerable regional variation in preference for particular sheep breeds throughout England and Wales, there is no evidence to suggest that certain breeds are more susceptible to T. gondii infection than others and this was therefore considered unlikely to influence distribution of seropositive samples (Katzer and others 2011).
These findings are particularly relevant in view of the European Food Safety Authority recommendation that all EU Member States should monitor T. gondii infection in animals entering the food chain (EFSA 2007). While it is not possible to directly correlate seropositivity in sheep with the presence of viable tissue cysts in meat, the high levels of exposure to T. gondii identified in this study serve to reinforce public health advice aimed at reducing the risk of infection, particularly in pregnant women and other vulnerable groups. Likewise, the scale of the threat posed to pregnant sheep should not be underestimated, and although control of toxoplasmosis at the level of primary production is difficult, farmers and veterinary surgeons should continue to promote appropriate husbandry and vaccination strategies. In common with previous studies involving sheep and other species, this survey also indicates possible regional variations in levels of infection and provides a basis for future investigations, which may further inform our understanding of the epidemiology of T. gondii.
The authors wish to thank colleagues within APHA – Weybridge who were involved with this project, particularly Jonathan Drake for his assistance with the database and data extraction.
Provenance: not commissioned; externally peer reviewed
Funding This survey was funded by the UK Department for Environment, Food and Rural Affairs via project FZ2100.