Materials and Methods

Study population—We performed a longitudinal study over 3 seasons of EHDV transmission (ie, from 2000 through 2002) on the cattle population of Illinois and western Indiana as previously described.¹⁴ The study was conducted over a 3-year period because of anticipated variability in EHDV transmission from year to year. Herds were selected on the basis of geographic location and willingness of herd owners to participate in the study. In total, 60 beef and dairy herds were recruited, 52 in Illinois and 8 in Indiana (Figure 1). Selected herds were located in 34 Illinois and Indiana counties. During the study period, there were approximately 500,000 cattle on 7,059 farms in these counties.¹⁵ In previous studies within temperate zones, temperature has been found to have a strong influence on Culicoides spp distribution and activity,^16,17 as well as on EHDV amplification within the vector.¹⁸ For this reason, mean minimum winter temperature was used to categorize the study area into 4 zones.¹⁹ To estimate 1% zonal seroprevalence with 1% error and 95% confidence, the target sample size per zone was 375 cattle. This sample size calculation assumed perfect sensitivity and specificity of the diagnostic test. Zone delineation reflected natural gaps in the spatial distribution of herds and an attempt to reach a minimum sample size of 375 cattle/zone. We attempted to test more cattle in the 3 northernmost zones because seroprevalence was expected to be lower. The geographic location of each herd was determined with a global positioning system receiver.^a Coordinates for dairy herd locations were recorded at their milking parlors. Coordinates for beef herd locations were recorded at the primary corrals of each operation. The research in this study complied with all relevant animal use federal guidelines and institutional policies.

View Full Size

Distribution of herds in Illinois and Indiana tested for antibodies against EHDV following the transmission seasons 2000 to 2002. Zones are defined by mean minimum temperature during the month of January (zone 1 < −10°C; zone 2 = −10° to −8°C; zone 3 = −7° to −6°C; zone 4 = −5° to −3°C). Each dot (n = 60) represents a beef or dairy herd from which samples were obtained during ≥ 1 year of the study.

Citation: American Journal of Veterinary Research 69, 10; 10.2460/ajvr.69.10.1286

• Download Figure
• Download figure as PowerPoint slide

Sample collection and testing—Blood samples were collected from cattle during winter and early spring months, typically November through March, when no EHDV transmission was assumed to occur. Thus, seropositive cattle represented transmission events in the previous transmission seasons. Cattle were selected by convenience on each operation. Cattle that were included in the study were ≥ 6 months of age at the time of the previous vector season, June through October, so that remaining maternal antibodies would not prevent seroconversion to EHDV following natural exposure. Cattle were categorized by age into those that were 7 to 18 months of age at the time of sample collection and those that were > 18 months of age. This categorization was performed to separate cattle that had lived through only a single EHDV transmission season. Each animal was individually identified so that it could be retested the following year.

Blood samples were stored at 4°C until processing, which was done within 24 hours of collection. Collection tubes were centrifuged at 3,000 × g for 5 minutes at 4°C. Serum was removed with a sterile transfer pipette and stored at −20°C until testing. Each serum sample was tested for antibodies against the serogroup of EHDVs by use of a competitive ELISA with EHDV serotype 1 antigen and EHDV-specific monoclonal antibody (18B2).²⁰ The optimal dilution of EHDV antigen and monoclonal antibody 18B2 was determined to generate an optical density value (at a wavelength of 405 nm) of 1.0 in a check-board titration indirect ELISA. In the competitive ELISA, EHDV antigen at the proper dilution was coated in microtiter plates overnight. After blocking the plates, properly diluted 18B2 was mixed with equal volume of testing serum diluted 1:5 before addition to the EHDV antigen–coated plates. The presence of anti-EHDV antibodies in testing sera was determined by inhibition of 18B2 binding to the solid-phase EHDV antigen. Bovine serum that generated values of > 40% inhibition of 18B2 binding to EHDV antigen was considered positive for antibodies against EHDV on the basis of the results of testing of 1,200 cattle that were seronegative for EHDV.

Data analysis—Seroprevalence was defined as the proportion of cattle tested with positive ELISA results. Unadjusted seroprevalence and 95% CIs were calculated for each zone and for the entire study region. Multivariable mixed-effects logistic regression was used to assess differences in the odds of seropositivity among zones, study years, herd types, and age. The binary (positive, negative) test result for each individual animal was the outcome variable. Predictor variables were zone (4 levels), year (3 levels), herd type (dairy, beef), and age (adult, young [< 18 months of age at time of sample collection]), and all were modeled as fixed effects. A random effect for herd was used to account for the interherd variability in herd-level odds of seropositivity. Initially, all main fixed effects were forced into the logistic regression model. The significance (P < 0.05) of each variable, as well as the likelihood ratio test, was used to determine which covariates should remain in the final model. Effect modification was assessed by including all first-order interaction terms for the main effects retained in the model. Finally, the random effect for herd was added to the model, and fixed effect covariates and first-order interaction terms with values of P > 0.05 were excluded from the final model. For all analyses, variable estimates with a value of P b 0.05 were considered significant. Logistic regression analyses were performed with standard statistical software.^b,c

The spatial distribution of seroprevalence of antibodies against EHDV at the herd level was tested for global and local autocorrelation by use of the Moran I statistic and the LISA statistic, respectively.^21,22,d The null hypothesis of the global autocorrelation test is that herds are randomly distributed with respect to seroprevalence of antibodies against EHDV. A positive Moran I statistic indicates clustering, whereas a negative value indicates dispersion. However, the global Moran I statistic does not identify the locations of clusters. The LISA statistic identifies clusters of herds with similarly high or low seroprevalence (positive autocorrelation). It also identifies spatial outliers (negative autocorrelation); this includes herds with high or low seroprevalence that are surrounded by herds with low or high seroprevalence, respectively. To aid the understanding of seroprevalence at the herd level, continuous isopleth maps of estimated seroprevalence of antibodies against EHDV in cattle for each year were created by use of ordinary kriging.^23,24,e

The geostatistical method of kriging assumes a constant variance in the data. This assumption is violated by differences in precision of seroprevalence estimates caused by varying herd sample sizes. To counter heterogeneity in the variance, spatial empirical Bayesian smoothing methods were applied to seroprevalence at the herd level prior to kriging.^22,d This method calculates a weighted mean of the raw seroprevalence estimate and a local mean with a weighting scheme on the basis of k = 6 nearest neighbors. Seroprevalence estimates from small populations with large SEs are weighted more toward the neighborhood mean than estimates from large populations. Spatial empirical Bayesian smoothing imposes autocorrelation on the raw seroprevalence estimates, leading to bias in global and local autocorrelation tests. To avoid this bias, the Moran I and LISA statistics were calculated by use of the empirical Bayesian standardization method, again on the basis of k = 6 nearest neighbors. This method is similar conceptually to spatial empirical Bayesian smoothing but differs computationally.^25,26

Another assumption of ordinary kriging is data stationarity (lack of directional trends). Trend surface regression was used to assess the presence of first- and second-order trends in data.^22,d Kriging was performed on the regression residuals, and the trend was added back to the final results. Kriging model variables were derived by fitting empiric semivariograms to seroprevalence at the herd level over twelve 35-km lag intervals by iteratively manipulating nugget, range, and sill variables and by fitting spherical, exponential, and Gaussian models.^f Final models were chosen by minimizing a goodness-of-fit statistic and by evaluating prediction errors calculated through cross-validation.^e

Results

During the 3 years of the study, 9,414 serum samples were obtained from 60 herds and tested for antibodies against EHDV (Table 1). In 2000, 55 herds participated in the study. In 2001, 16 herds out of the original 55 were lost to follow-up and 4 new herds were added, providing a total of 43 herds that year. In 2002, 13 herds tested in the previous year were lost to follow-up. No new herds were added in 2002, but 11 of the herds lost after 2000 rejoined the study. Thus, 41 herds were tested in 2002. Twenty-seven of the herds participated in the study during all 3 years, 26 herds participated for 2 years, and 6 participated for 1 year only. The mean number of cattle tested per herd was 68 (range, 4 to 223 cattle/herd). The target sample size per zone per year (375 cattle) was met in all zones with the exception of zone 4 (southernmost zone), in which the target was not met during the second year (232 cattle) and third year (313 cattle).

The overall mean prevalence of antibodies against EHDV in all herds in transmission seasons 2000, 2001, and 2002 was determined (Table 1). Annual seroprevalence of antibodies against EHDV in cattle by zone generally increased from north to south. We used 2 final multivariable logistic regression models to describe the risk of seropositivity for EHDV. One model (model 1) included zone, year, age, and the interaction between zone and year (Table 2). This model was based on 7,033 test results, with the remaining 2,381 results excluded because of missing data on age. To assess the effect of excluding those test results, a second model (model 2) was constructed excluding the age variable and including all 9,414 test results. This model included the variables zone, year, and interaction between zone and year. Herd, considered a random effect, explained 31.7% and 32.0% of the total variance in risk of seropositivity for EHDV in models 1 and 2, respectively.

In the age-adjusted model (model 1), odds of seropositivity for EHDV varied from year to year (Table 3). In 2000, no significant differences were found between zone 2 and zone 1 and between zone 3 and zone 1. The odds of seropositivity in 2000 in zone 4 were 37 times as great as in zone 1. Odds of seropositivity in 2001 in zone 2 and zone 3 were both significantly higher than in zone 1 and the odds of seropositivity in zone 4 was 274 times as great as the odds of seropositivity in zone 1. However in 2002, the odds of seropositivity in zone 3 were no longer significantly different than zone 1, whereas the odds of seropositivity in zone 2, compared with zone 1, remained similar to the previous year. The odds of seropositivity in zone 4, compared with zone 1, were lower than the previous 2 years, but remained high.

The interaction between zone and year was strongest in zone 4 (Figure 2). The probability of seropositivity in zone 4 was higher in 2001 than 2000 but lower in 2002 than 2000. No significant differences were found across years within zones 1, 2, and 3 except for zone 3, where the probability of seropositivity was slightly lower in 2002 than 2000.

View Full Size

Probability of epizootic hemorrhagic disease seropositivity in cattle estimated from logistic regression model 1.

Citation: American Journal of Veterinary Research 69, 10; 10.2460/ajvr.69.10.1286

• Download Figure
• Download figure as PowerPoint slide

Characteristics of the 2,381 cattle that were excluded from the age-adjusted model were assessed. In total, 1,009 cattle from zone 1; 755 from zone 2; 449 from zone 3; and 168 from zone 4 were excluded because of missing information on age. These numbers represented 37%, 21%, 21%, and 18% of the total number of cattle tested in zones 1, 2, 3, and 4, respectively. The numbers of serum samples with positive test results that were excluded because of missing information on age were 17, 83, 100, and 76 for zones 1, 2, 3, and 4, respectively. These numbers represented 28%, 24%, 35%, and 18% of the total number of serum samples with positive test results from zones 1, 2, 3, and 4, respectively. When these cattle were included in the mixed-effects logistic regression model, with the variable for age dropped from the model, odds of seropositivity in zones 2, 3, and 4 relative to zone 1 varied from year to year in a similar pattern to that already described (Table 3). The only difference was that the odds of seropositivity in zone 3 were significantly higher than in zone 1 in 2000 (odds ratio, 4.50; 95% CI, 1.61 to 12.56), possibly because 87 of the 100 serum samples with positive test results that were eliminated from zone 3 were from 2000.

Global Moran I statistics were 0.278, 0.552, and 0.255 for 2000, 2001, and 2002, respectively, suggesting moderate to strong autocorrelation in the data. All 3 Moran I statistics had pseudo values of P < 0.01. The LISA statistic identified spatial clusters of low seroprevalence in the northern part of the study area and high seroprevalence in the southern part of the study area in all 3 years, supporting the results of the logistic regression model (Figure 3). However, the LISA statistic also identified spatial outliers of high seroprevalence in northern Illinois in 2000 and 2002 and in central Illinois in 2001. Outliers of low seroprevalence were identified in southern Illinois and Indiana in 2000 and 2002. These outliers suggest local variation in seroprevalence at the herd level that is not evident in the logistic regression model.

View Full Size

Results of the LISA test showing significant clusters (pseudo P value < 0.05) of low and high seroprevalence and spatial outliers across Illinois and Indiana.

Citation: American Journal of Veterinary Research 69, 10; 10.2460/ajvr.69.10.1286

• Download Figure
• Download figure as PowerPoint slide

Isopleth maps of EHDV were developed (Figure 4). The spatial regression results suggested a first order trend in the data for all 3 years. The nugget, range, sill, and type of model fitted to the empirical semivariograms varied by year. In 2000, the semivariogram was fitted to an exponential model with a range of 415 km, nugget of 0.014, and sill of 0.022. In 2001, the semivariogram was fitted to a Gaussian model with a range of 305 km, nugget of 0.005, and sill of 0.027. In 2002, the semivariogram was fitted to a spherical model with a range of 140 km, nugget of 0.0010, and sill of 0.0016. Cross-validation revealed that all 3 models underestimated the variability in the data. The root mean square standardized errors for 2000, 2001, and 2002 were 1.2, 1.5, and 1.2, respectively. The large nugget-to-sill ratios in 2000 and 2002 suggested that there were large amounts of variability at close distances that were not accounted for in the kriging models.

View Full Size

Isopleth risk maps of seroprevalence of antibodies against EHDV at the herd level, 2000 to 2002, across Illinois and Indiana. Each dot represents a beef or dairy herd from which samples were obtained during the year depicted.

Citation: American Journal of Veterinary Research 69, 10; 10.2460/ajvr.69.10.1286

• Download Figure
• Download figure as PowerPoint slide

The interpolated maps revealed a pattern of increasing seroprevalence from north to south, but with considerable annual variation in the southern two thirds of the study region. The maps reflected the results of the logistic regression model by demonstrating that the seroprevalence in zones 1 and 2 did not vary much from year to year, whereas seroprevalence in zones 3 and 4 were significantly lower in 2002 than in 2000. The maps also revealed a more complex picture of seroprevalence that is not evident in the logistic regression model. Areas of higher seroprevalence were found extending northward along the eastern and western edges of the study area that surround an area of lower seroprevalence in central Illinois. In 2000, areas of higher seroprevalence extended along both the east and west edges. In 2001, the area of higher seroprevalence extended only along the western side, whereas in 2002 it extended only along the eastern side. In 2001, the overall direction of increasing seroprevalence appeared to shift to a southwest direction from a general north to south direction the other 2 years.

Discussion

Historically, EHDV has been considered a more important pathogen of wild deer than of cattle in the United States,² and few estimates of seroprevalence of antibodies against EHDV in cattle have been published.^12,13 The present study provides an estimate of seroprevalence of antibodies against EHDV in cattle and describes the spatial and temporal distribution of EHDV in Illinois and western Indiana in greater detail than has been previously reported in cattle. Study results support the hypothesis that seroprevalence of antibodies against EHDV in cattle is higher in the southern part of the study area. However, the temporal and spatial pattern of seroprevalence is complex. This is reflected by the interaction between year and zone in the logistic regression model; odds of seropositivity in zones 2 and 3 changed relative to zone 1 from year to year. The logistic regression model also described annual variation within zones. Seroprevalence within each zone did not change significantly between 2001 and 2000. The same was true for zones 1 and 2 in 2002, whereas seroprevalence within zones 3 and 4 were significantly lower in 2002 than in 2000.

Temporal variation in seroprevalence of antibodies against EHDV has been observed in other studies. In Georgia, seroprevalence in deer increased in the southern regions of the state and decreased in northern Georgia from 1989 to 1991. The authors attributed the decrease in the northern regions to waning immunity following a 1988 epidemic in that region.²⁷ The last known EHDV epidemic in Illinois prior to this study was in 1998.^g It is possible that the substantial decrease in seroprevalence in the 2 southernmost zones in 2002 is related to waning immunity. The decrease might also reflect a temporary absence of competent vector species within this region. For example, local rainfall fluctuations (a period of drought) can have an important impact on vector populations, such as Culicoides spp, and the pathogens that they transmit.²⁸

A potential bias was introduced through repeated sample collection from individual cattle during the 3 years of the study. The 9,414 samples tested in this study represent 6,872 individual cattle: 5,038 cattle were tested once during the 3 years, 1,514 cattle were tested in 2 of the study years, and 320 cattle were tested in all 3 study years. Three hundred and eighty-eight samples did not have individual animal identifications. Among cattle that were tested multiple times, 172 were seropositive on the first test, and 54 of these remained seropositive on subsequent tests. There were 130 cattle tested multiple times that seroconverted during the study. Because a small proportion (9.7%) of seropositive cattle had undergone repeated sample collection, data were assumed to be an independent cross-sectional sample each year with nearly all cattle at risk of exposure and seroconversion to EHDV.

The isopleth maps highlight variation in seroprevalence, especially in the areas approximating zones 2 and 3. The maps also show temporal changes in seroprevalence patterns that might be caused by climatic influences on vector activity, such as changes in wind patterns, temperature, or precipitation. This variation in seroprevalence of antibodies against EHDV in cattle contrasts with the pattern of antibodies against BTV observed in the same cattle population during the same study period.¹⁴ Seroprevalence of antibodies against BTV followed a more straightforward pattern than against EHDV, with increasing seroprevalence along a gradient of decreasing latitude. No year-by-zone interactions were found in seroprevalence of antibodies against BTV, and there were clearly identifiable areas of zero seroprevalence in zone 1, < 1% seroprevalence in zones 2 and 3, and 6% to 14% seroprevalence in zone 4. The Georgia serosurvey of deer also reported a difference in seroprevalence of antibodies against EHDV and BTV, with EHDV-2 having much higher prevalence than EHDV-1 and the 5 BTVs that were detected.²⁷

The geostatistical methods used to create the isopleth maps are based on spatial autocorrelation (spatially close objects are more alike than spatially distant objects). Although the global Moran I test suggested a high degree of autocorrelation of seroprevalence at the herd level, considerable variability was found among herds at a local scale. For example, in 2001 the LISA statistic identified an outlier of high seroprevalence among a group of 9 low-seroprevalence herds, all within a 10-km radius. This smallscale variability may have contributed to the poor fit of the data to the kriging models.

It should be noted that the interpolated results were applied to a rectangular plane defined by the spatial extent of sample collection points. Therefore, extrapolated results in western Illinois and eastern Indiana should be interpreted with caution. Also, results can be less accurate in areas where there are large gaps in sample collection points, such as the area between the northernmost herds and the herds in central Illinois. In addition, differences among years in the number and location of herds tested may artificially influence the results. For example, the shift to a southwestern direction of increasing seroprevalence observed in 2001 may have been influenced by the absence of 2 high-seroprevalence herds in southern Indiana that were tested in 2000 and 2002, but not 2001. Finally, it is also important to note that although seroprevalence is treated as a continuous surface in these maps, predicted risk is dependent on the presence of cattle herds. These factors limit the usefulness of EHDV risk maps, which are based solely on geostatistical methods, for predictive purposes.

Despite the predictive limitations of kriging in this study, the isopleth maps provide an effective way to demonstrate the general pattern of seroprevalence and remove visual bias created by artificial boundaries such as administrative borders or temperature zones. The isopleth maps are also useful for hypothesis generation. The maps emphasize spatial patterns that are not readily apparent in traditional statistical methods and can help guide the investigation of spatial risk factors associated with seroprevalence of antibodies against EHDV in cattle, including temperature, precipitation, and proximity to habitats that are associated with Culicoides vectors or wild deer. The year-to-year variability in central Illinois seen in these maps suggests that investigation of environmental risk factors in this region may be appropriate. Ideally, an effort to model the EHDV transmission cycle would include data on Culicoides spp. However, collection and analysis of data on Culicoides spp is expensive and labor-intensive. An analysis of the serologic data reported here in relation to spatial environmental risk factors would be helpful for focusing future studies of Culicoides spp with the purpose of optimizing limited resources. In addition, generation of wild deer population spatial distributions and EHDV antibody distribution in deer may help explain much of the spatial variation of seroprevalence of antibodies against EHDV in cattle observed in this study.