Figure 1

Figure 1. Geographic distribution and PRNT₉₀–validated West Nile virus seroprevalence among equids per sampling site in Bahia State, Brazil. Inset shows location of Bahia State in northeastern Brazil. Sample sizes...

We collected serum samples from 713 equids, including horses and mules, sampled as part of routine veterinary surveillance activities during 2013–2018 in the state of Bahia in northeastern Brazil. The animal ethics committee of the Federal University of Bahia approved the sampling and analyses (authorization no. 55/2017). Sampling covered a large area that connects the location of the human case from 2014 and the 2018 horse epizootic (9,10). The area is adjacent to the Atlantic, northeastern, and central avian migratory routes (Figure 1).

Figure 2

Figure 2. WNV seroprevalence among equids, Brazil. A) ELISA absorbance values displayed as sample to cutoff ratio, as previously described (2). We increased the ELISA cutoff by 10% above which...

For antibody screening, we used an experimental WNV IgG ELISA based on a fusion loop envelope antigen containing mutations. We chose this ELISA to decrease the chances of cross-reactivity with antibodies elicited by other flaviviruses (2). Among 713 serum samples, 47 (6.6%, 95% CI 4.9%–8.7%) yielded positive ELISA results (Figure 2, panel A). Beyond WNV, horses in Latin America frequently are infected with Saint Louis encephalitis virus (SLEV), Cacipacoré virus (CPCV), Rocio virus (ROCV), and Bussuquara virus (BSQV) (12); and WNV, CPCV, and SLEV all belong to the Japanese encephalitis serocomplex (Appendix Figure 1). Serologic analyses for WNV-specific antibodies in horses could be compromised by cross-reactive antibodies against other flaviviruses, eliciting potentially false-positive test results (11). Therefore, we confirmed ELISA-based WNV antibody detection by comparing the endpoint titers for the 90% plaque-reduction neutralization tests (PRNT₉₀), considered the standard for arbovirus serologic testing, for WNV, CPCV, SLEV, BSQV, and ROCV in all 47 ELISA-positive serum samples. Of the 47 samples, 20 (44.7%) neutralized WNV only in PRNT₉₀; another 22 (46.8%) showed heterotypic reactions for WNV, CPCV, or SLEV (Figure 2, panel B). Averaged endpoint titers were significantly higher for WNV than for the other flaviviruses (p<0.0001) and exceeded those for CPCV, SLEV, BSQV, or ROCV by >4-fold in 12/22 heterotypic samples (Figure 2, panel C), a titer difference commonly considered decisive in flavivirus serology. Thus, 68.1% (32/47) of ELISA-positive samples had WNV-specific antibody responses (Figure 2, panel C); 4 samples were seronegative for all 5 flaviviruses by PRNT₉₀, potentially because of differential sensitivity of ELISA and PRNT. No samples had higher SLEV-, BSQV-, or ROCV-specific PRNT₉₀ titers compared with WNV, but 2 ELISA-positive samples had >4-fold endpoint titers for CPCV compared with WNV and other flaviviruses (Appendix Table 1). These findings substantiated WNV and CPCV cocirculation among equids in northeastern Brazil, which is consistent with previous data on CPCV circulation in another region of Brazil (12), and high specificity of the ELISA-based screening algorithm.

PRNT₉₀ validated the overall WNV seroprevalence of 4.5% (32/713 samples; 95% CI 3.1%–6.3%), which we used for downstream analyses (Table 1). We noted seropositive animals in 11/28 municipalities distributed over ≈900 km², suggesting wide geographic spread of WNV (Table 1; Figure 1). We observed a concentration of positive samples in 2018 (Appendix Figure 2), and in 1 municipality (Figure 1), comprising 9 different seropositive farms with an average seroprevalence of 13.9% (95% CI 8.3%–21.4%). Antibody levels typically decline over time after flavivirus infection (13), which might bias positivity rates of serologic tests. However, WNV-specific PRNT₉₀ endpoint titers were significantly lower in 2018 than in the preceding years by Mann-Whitney test (p = 0.002), excluding a bias from hypothetically more recent WNV infection in the animals sampled in 2018 (Appendix Figure 3).

We performed generalized linear model analyses and principal component analysis to compare 12 environmental, ecologic, and demographic factors potentially affecting WNV seroprevalence (Table 2; Appendix Figure 4). Anthropogenic changes of pristine habitats can increase the abundance of zoonotic pathogens (14), potentially including arboviruses like WNV (15). Indeed, the model considering forest loss, but not the model considering tree cover alone, was significantly associated with higher WNV seroprevalence (odds ratio [OR] 5.106, 95% CI 1.318–31.796; p = 0.005) (Table 2). Model results were consistent with a higher proportion of WNV-seropositive sites in disturbed areas compared with pristine areas by χ² test (p = 0.009) (Figure 2, panel D). Higher WNV seroprevalence in disturbed areas was not biased by the number of animals living in those sites compared to sites from pristine areas. By Student t-test, neither the overall number of animals nor the number of animals per site differed significantly between disturbed (p = 0.9) and pristine areas (p = 0.2894) (Figure 2, panel D; Appendix Figure 5).

Because the geographic distribution of both the 2018 horse epizootic and the only known human case might be linked geographically to the northeastern and coastal avian migratory routes (Figure 1), we included distances to avian routes in model analyses of WNV seroprevalence. Proximity to the central avian migratory route was associated with higher WNV seroprevalence (Table 2; Appendix Figure 4). This finding was consistent with other seroprevalence studies, indicating the presence of WNV in horses in the central region in Brazil (7,8), but failed to connect the WNV detections in Brazil to geographically adjacent avian migratory routes. Our data were consistent with prior studies of WNV ecology, but the explicatory power of our models was low despite statistical significance (Table 2; Appendix Figure 4).

Our study was limited by absence of longitudinal samples from individual sampling sites, lack of information on animal trade and animal age, and relatively low numbers of seropositive animals from individual sites. Thus, we cannot exclude biases affecting the accuracy of our modeling approach. However, our large sample and the combination of thorough serologic analyses and mathematical modeling enabled robust estimates of WNV spread that can guide prospective studies.

Our findings of substantial WNV seroprevalence in equids in Brazil warrants WNV surveillance in cases of acute neurologic disease in humans and horses. In addition, blood products should be screened in areas of Latin America with high risk for WNV.