Factors associated with mosquito pool positivity and the characterization of the West Nile viruses found within Louisiana during 2007

The data used in the modelling study were from several parishes in Louisiana during 2007. There were 611 positives reported by the Louisiana Animal Disease Diagnostic Laboratory, 165 in our target parishes. Classification regions were constructed based on land cover data from the Louisiana GIS Digital Map, May 2007(Figure 1). The land cover of a parish was determined by the majority rule. Ouachita and Caddo parishes are, for example, 50-75% forest lands; East Baton Rouge is a majority developed area comprised mostly of the urbanized capital of Baton Rouge [17]. Iberville Parish is classified as wetlands, defined as low lying areas saturated with moisture.

Before building the WNV model, we tested several factors via multinomial regression to determine whether these were significant predictors of the mosquito species found, regardless of WNV positivity. Interestingly, no factors (land cover, month) were significant for the 2007 data. Therefore, it was determined that no noteworthy correlation existed between these variables and thus no colinearity issues would arise by inclusion of all variables in the model.

The model originally included the effects of month, mosquito genus and species; but through a stepwise selection procedure, the variable species was not significant at the alpha = 0.05 level and thus eliminated from the model. This is likely due to the overwhelming number of Culex quinquefasciatus, which comprised over 61% of the total pools (N_total = 3246) and 88% of the positive pools (Table 1). This is consistent with earlier studies where Cx. quinquefasciatus was found to be the most abundant and likely epizootic vector for the virus [18, 19]. In addition, all interactions were not significant and therefore removed from the model.

Locations of isolates were coded according to land cover. The model included land cover, genus, and month. Type of regional land cover was a significant effect (p < 0.0001) indicating that there is an ecological component driving WNV activity. Additionally, month was significant (p < 0.0001); and thus there is a temporal component that contributes to the probability of getting a positive mosquito pool. This trend can be seen in figure 2.

The pair wise differences in least squares means are given in table 2 with Tukey adjusted p-values showing where the significant differences lay. Forested lands are clearly more likely to have a positive mosquito pool as compared to developed areas and wetlands. Similarly, August appears to be the month where a positive mosquito pool is more likely. August and July are eight and six times more likely to see a positive mosquito pool than June, respectively. Interestingly, August and July are not significantly different from the months of September-November. All months are significantly different from June, which is the least likely month during the accepted transmission season to have a positive pool; the odds of a positive pool is over seven times more likely in July than June. The overall trends of land cover and month are shown together in Figure 2.

The phylogenetic tree resulting from the contiguous segment comprised of the envelope and NS3 sequences is shown in figure 3. All samples are, unsurprisingly, of the 2002 lineage. However, there appears to be some diversity within both forested and wetland areas. Particularly, there is a possible small sub-group with samples 4893, 8441, and 3077. The samples from the wetlands show the possibility of a distinctive group; in particular, sample numbers 3766 and 3767. There is a clear delineation between forest and wetland samples based on two nucleotide substitutions in the NS3 gene: an adenine to guanine at positions 5760, and a cytosine to uracil at position 6324., Samples were grouped as follows: Israel 1998 as the root; the wetlands as a group; a small grouping of forest samples; a NY99 group; and the remaining were grouped together as representative of the North American 2002 clade. The between and within distances were computed according to the Jukes-Cantor model (table 3). Genetic diversity within Louisiana was modest when compared to isolates from a wide geographic range. The genetic distances within the eco-regions (forest vs. wetlands) in Louisiana were greater than the distances comparing Louisiana and those strains from outside of Louisiana. Table 4 identifies sample origins, classifications, and accession numbers.

According to the mosquito surveillance data, there are two primary species of mosquitoes that serve as possible WNV vectors to people in Louisiana: Culex quinquefasciatus and Aedes albopictus. There were twenty six species from eight genera captured and submitted for testing. Of these, fifteen species were found to be positive for WNV (Table 1).

Data was provided by parish mosquito control departments. Each parish operates independently with its own trapping protocols and methods. Not all parishes actively sampled throughout the year due to considerations of the local mosquito activity levels. We therefore analyzed the months that all target parishes had in common (June-November), which captured the majority of the WNV transmission period [21].

Mosquitoes were sexed and the females were pooled according to genus and species; that is, one pool consisted of a single species. The pools were then homogenized and submitted to the Louisiana Animal Disease Diagnostic Laboratory (LADDL) at the Louisiana State University (LSU) School of Veterinary Medicine (SVM). LADDL is the state testing facility for mosquito pools for all parishes, so criteria for positive pools is the same across all parishes.

Pools were obtained from the LADDL at the LSU SVM. Viral RNA was extracted from 140 μl of the supernatant from the mosquito pool homogenate using the QIAmp Viral RNA Extraction Kit following manufacturer's instructions (Qiagen, Valencia, CA). One microliter of the extracted viral RNA suspension was used as template for the reverse-trascriptase polymerase chain reaction (RT-PCR) using Superscript™ III RT-PCR kit (Invitrogen, Carlsbad, CA) with the previously described protocol [22]. Upon confirmation of the presence of amplified viral DNA by gel electrophoresis, the remaining sample was cleaned using Qiagen PCR Cleanup kit following manufacturer's instructions. Cleaned DNA was then sent to the Gene Probes and Expression Systems Laboratories of the Division of Biotechnology and Molecular Medicine at the Louisiana State University School of Veterinary Medicine for sequencing. Sequencing was performed on a Beckman Coulter 8800 (Pasadena, CA) using the manufacturer's reagents and methods.

SAS version 9.1.3 was used to code the data as a binary response, where a mosquito pool that was positive for was coded as '1,' while the negative pools were coded as '0.' The probabilities reported are the probabilities of the event = 1 (WNV positive). A confidence level of 95% was used for all tests; a stepwise selection process was invoked to cull out non-significant effects from the model. PROC GLIMMIX with a binary distribution specified was used to construct the model and obtain the odds ratios, as well as to obtain differences in least squares means for the effects of parish and month. PROC GLIMMIX is a useful alternative to PROC LOGISTIC when wanting to obtain differences in least squares and/or modeling random effects, which is not easily done in PROC LOGISTIC. Predicted means and odds ratios are the same between the two procedures. Confidence intervals and tests for significance will vary when random effects are included the model in PROC GLIMMIX, but there were no random effects modelled here [23].

The complete envelope (E) and non-structural protein 3 (NS3) genes were aligned separately for all successfully recovered isolates as well as 24 reference strains, which represented the Israel 1998 strain and isolates spanning the contiguous United States and one Mexican isolate [15]. Alignments and the creation of the E-NS3 contigs were done using Vector NTI software and exported to GeneDoc for trimming. Alignments were imported into MEGA 4 and converted to MEGA format. Bootstrap analysis (n = 1000) using maximum parsimony was performed and a tree produced. As the topologies of the E and NS3 phylogenies were the same, our tree represents a contiguous sequence of these two genes. The topology tree was collapsed with a node confidence of 70%. Genetic distances and means were also obtained using Mega 4 [24]. Not all samples in the statistical model were available for the phylogenetic analysis.