Detection of Specific Antibodies against Toscana Virus among Blood Donors in Northeastern Italy and Correlation with Sand Fly Abundance in 2014

Toscana virus (TOSV) is a Phlebovirus transmitted by phlebotomine sand flies and is an important etiological agent of summer meningitis in the Mediterranean basin. Since TOSV infection is often asymptomatic, we evaluated the seroprevalence in blood donors (BDs) in the Bologna and Ferrara provinces (Northeastern Italy)—the areas with the highest and lowest numbers of TOSV neuroinvasive cases in the region, respectively. A total of 1208 serum samples from BDs were collected in April–June 2014 and evaluated for the presence of specific TOSV-IgG by ELISA. The IgG-reactive samples were confirmed by indirect immunofluorescence assay (IIF) and by microneutralization test (MN). Serum samples were defined as positive for anti-TOSV IgG when reactive by ELISA and by at least one second-level test; TOSV seroprevalence was 6.8% in the Bologna province, while no circulation of TOSV was detected in the Ferrara province. Sand fly abundance in 2014 was also estimated by a geographic information system using a generalized linear model applied to a series of explanatory variables. TOSV seroprevalence rate was strongly associated with the sand fly abundance index in each municipality, pointing out the strong association between sand fly abundance and human exposure to TOSV.

All images were resampled at 250 m resolution and cropped inside the study area using QGIS 2.18. Abundance model were carried out in R version 3.3.2 (R Development Core Team 2005) using the lme4, lattice, Tweedie, boot, raster, sp and rgdal packages. A GLM with log link and poisson error was used for modelling sand fly abundance from the beginning of July to the end of August 2014 (specimens sum) in 62 CO2 traps correlated to the 12 explanatory variables. A 6-fold Cross Validation was used to validate the model and the deviance explained to evaluate accuracy. This means that data set was split into 6 equal parts, the model was applied to all data except one part, data of excluded part were used as control; the procedure was applied for each part of the data.

Correlation of human TOSV-seroprevalence and sand fly abundance
A binomial logistic regression model was used to evaluate the correlation between the TOSV seroprevalence at municipality level and the sand fly abundance (SA), estimated as average of sand flies per trap per municipality according to the obtained model. To reduce the high variability typical of the vector population, the abundance was log10 transformed before binominal logistic regression model analysis. For each municipality, results from the prevalence observed in blood donors were analyzed separately by means of binomial logistic regression with log10 (SA) as covariate. Intercooled Stata 7.0 software (Stata Corporation, College Station, TX, USA) was used for statistical data analysis. Significance was established at p <0.05. The binomial logistic regression model was highly significant (p<0.001).

Medialization of the sand fly abundance
The first four variables ranked by the model are reported in Table S1. A 92% value was obtained by the 6-fold Cross Validation, and the cumulative deviance explained by the first 4 variables to evaluate accuracy is reported in Table S2. The output of the model (estimated number of sand flies per trap) is graphically represented in Figure S1.  Figure S1. Graphic representation of estimated number of sand flies per trap with reference of location of the traps (black dots).

Correlation of human TOSV-seroprevalence and sand fly abundance
A binomial logistic regression model was used to evaluate the correlation between the TOSV seroprevalence at municipality level and the sand fly abundance (SA) (1).

Logistic (p+) = Ln[p+/(1-p+)] = a + b log10(SA) (1)
To avoid 0 prevalence result and to consider the uncertainty introduced by sampling on the true prevalence estimation the mean of a beta function was used. The beta transformation was applied to the prevalence observed in each municipality. Therefore, if 0 positive donors were observed out of 27 donors analyzed (Poggio Renatico municipality) the real prevalence (p+) estimated was 0.04 and the logistic transformation of (p+) was -3.3. Using Bayes' theorem where no prior knowledge of the prevalence (P) in blood donor population is available, the fraction of positive donor could be assumed to follow a Beta distribution (Vose 2006). If we assume a uniform [0.1] prior distribution for P+ (the probability of being positive) and find that K of M positive donors, the posterior distribution of donors was modeled as reported in (2).
Finally, the logistic regression included only municipalities where at least 20 donors were examined for TOSV IgG; data are summarized in Table S3.  The 75% of the variability observed in the prevalence of positive BD per municipality is explained by the abundance of sand flies, expressed as log (10) following the equation (3).
The coefficient 0.29 has a 95 coefficient interval (CI) of 0.208-0.395 and the intercept has a 95 CI of-2.791-2.410. The Figure S2 represents the logistic transformation of the prevalence (p+) of BD in each municipality (y axis) against the log(10) of sand fly mean abundance (x axis). The equation indicates that the prevalence of TOSV in BD (logistic transformation) is positively associated to the abundance of the vectors in the different municipalities. Figure S2. Binomial logistic regression of TOSV prevalence (P) in BD and sand fly abundance index (log10SA).