A Multi-level Analysis of the Interactions betwen Vectors, Hosts, and Habitats of EEEV by Laura Kristin Estep A disertation submited to the Graduate Faculty of Auburn University in partial fulfilment of the requirements for the Degree of Doctor of Philosophy Auburn, Alabama May 9, 2011 Key words: dispersal kernel, forage ratio, arbovirus, mosquito, avian reservoir host, GIS Copyright 2011 by Laura Kristin Estep Approved by Geoffrey E. Hil, Chair, Profesor of Biological Sciences Thomas R. Unnasch, Profesor of Global Health, University of South Florida Bertram Zinner, Asociate Profesor of Mathematics and Statistics F. Stephen Dobson, Profesor of Biological Sciences Arthur G. Appel, Profesor of Entomology and Plant Pathology ii Abstract Eastern equine encephalitis virus (EEV) is a rare and dangerous mosquito-borne pathogen with an elusive patern of ocurrence across its range in North America. The primary basis of our understanding of its transmision cycle betwen the vector Cs. melanura and avian reservoir hosts stems from studies in the northeastern United States. In this disertation, I draw upon a range of innovative statistical approaches to study the transmision of EEV in the southeastern United States in relation the ecology of EEV vectors and avian reservoir hosts. I estimate the dispersal distance of Culex. eraticus, a mosquito species potentialy involved in transmision of the virus, using a novel approach rooted in Bayesian statistics and borrowed from the sed dispersal literature. I also consider the distribution of avian reservoir hosts of the virus and their habitat asociations, with the goal of estimating rates of utilization of avian host species by EEV mosquito vectors. Such estimates of host utilization have great potential utility in revealing relative contribution of various hosts species to transmision of the virus. Indeed, I provide evidence that the most highly preferred host of Cs. melanura, the common yelowthroat (Geothylpis trichas), has the strongest support of influencing transmision as a dilution host among species considered. Finaly, I develop model a to predict rates of contact of avian hosts with Cx. eraticus based on host characteristics using data on host traits available from the ornithological literature. Together these results of my studies provide a strong basis for the future development of predictive models for occurrence of the virus and provide a framework for iii future research of the transmision of this virus in the Southeast, and potentialy throughout its range in North America. iv Acknowledgments This disertation, from its inception to the final wording of its sentences, has been a collaborative efort. I thus have many people to thank for their guidance and support through this proces. My graduate advisor and commite chair, Dr. Geoffrey Hil, constantly went above and beyond my expectations of a graduate advisor in terms of his encouragement, academic savoir-faire, and, in the critical periods of approaching deadlines, skilful editing. Above al else, I am grateful to him for teaching me to strive for and achieve clarity in my scientific thinking and writing. Dr. Thomas Unnasch provided me with a tremendous opportunity to focus my graduate studies by alowing me to collaborate on his research of eastern equine encephalitis virus. Without this opportunity I would not have been able to conduct the research I present in this disertation. For this and his service on my commite, I am very grateful. I also am extremely grateful to Dr. F. Steve Dobson and Dr. Bertam Zinner for their wilingnes to collaborate on my graduate research as commite members, and Dr. Arthur Appel, for serving as an outside reader, as their collective service has ensured that this disertation mets a high standard of quality for ecological and quantitative research. I also thank Dr. Hasan Hasan, Dr. Nathan Burket-Cadena, Dr. Robert Unnasch, and PhD candidate Chris McClure, al of whom were involved in the data collection, analysis, and editing for one or more of these chapters. I would like to expres my gratitude in particular to my student colleagues, Chris and Nathan, for demonstrating to me the esence of collaboration in scientific research during the preparation of this disertation. I also thank Brian Rolek, Lisa McWiliams, v Austin Mercadante, Katherine Gray, Chris Porterfield Nathan Click, and Dr. Christy Voakes for their roles in the field research and collection of data used for analyses in this research, and the labs of Dr. Hil and Dr. Wendy Hood for extensive editing of drafts of the chapters of this disertation. Finaly, I thank my family and loved ones, Diane, Kely, and David Estep, and Brian Butler and Justin Hayes, for their steadfast support during my comprehensive examinations and the writing of this disertation. vi Table of Contents Abstract.........................................................................................................................................ii Acknowledgments........................................................................................................................iv List of Tables.............................................................................................................................vii List of Figures...............................................................................................................................x I. Estimation of dispersal distances of Culex eraticus in a focus of eastern equine encephalitis virus in the southeastern United States.............................................................................1 Abstract ...........................................................................................................................1 Introduction ......................................................................................................................1 Materials and Methods .....................................................................................................4 Results ............................................................................................................................11 Discussion ......................................................................................................................13 References ......................................................................................................................17 II. A multi-year study of mosquito feding paterns on avian hosts in a southeastern focus of eastern equine encephalitis virus....................................................................................28 Abstract .........................................................................................................................28 Introduction ....................................................................................................................28 Materials and Methods ...................................................................................................30 Results ............................................................................................................................36 Discussion ......................................................................................................................38 References ......................................................................................................................44 vii III. Developing models of avian host forage ratio models for Culiseta melanura and Culex eraticus using host characteristics.................................................................................56 Abstract .........................................................................................................................56 Introduction ....................................................................................................................56 Materials and Methods ...................................................................................................59 Results ............................................................................................................................62 Discussion ......................................................................................................................64 References ......................................................................................................................69 IV. Using atributes of avian communities to predict local enzootic transmision of eastern equine encephalitis virus.................................................................................................84 Abstract .........................................................................................................................84 Introduction ....................................................................................................................84 Materials and Methods ...................................................................................................88 Results ............................................................................................................................91 Discussion ......................................................................................................................93 References ......................................................................................................................95 vii List of Tables I. Table 1. Mean (95% credible intervals) for parameters and deviance of the models developed for Cx. eraticus abundance at 2006 adult sampling sites in Tuskegee National Forest, Alabama, with DIC of the models also presented. .......................................................21 II. Table 1. Predicted relative abundances of avian species observed during point-count surveys in TNF and used in forage ratio calculations. ............................................................................50 Table 2. Total number of blood meals derived from avian species for mosquitoes collected in TNF betwen March and October from 2001 through 2009. ....................................................51 Table 3. Forage ratios (95% CI) of the avian species from which blood meals were derived for Cx. eraticus, Cx. restuans, and Cs. melanura betwen March and October from 2001 through 2009. .............................................................................................................................52 Table 4. Forage ratios of avian species using al bloodmeals collected betwen March and October, or alternatively, strictly betwen May and August 15 th . .............................................53 III. Table 1. Predictor variables used in for Cs. melanura forage ratio and Cx. eraticus forage ratio candidate models. ..............................................................................................................75 Table 2. Bias-corrected AIC (AICc) table for Cx. eraticus forage ratio models with moderate or strong support.........................................................................................................76 Table 3. Results of model averaging for Cx. eraticus forage ratio models and importance weights of predictor variables (weight)......................................................................................77 Table 4. Bias-corrected AIC (AICc) table for Cs. melanura forage ratio models with moderate or strong support.........................................................................................................78 Table 5. Results of model averaging for Cs. melanura forage ratio models and importance weights of predictor variables (weight)......................................................................................79 ix IV. Table 1. Descriptive statistics of the relative abundances of avian species at sentinel sites in Walton County, Florida in 2009.............................................................................................99 Table 2. Summary of atributes of candidate models considered in developing a logistic regresion model of the log-odds of a sentinel site having a high- versus low-level of virus activity (>1 chicken seroconverted to EEV antibodies) in Walton County in 2009. ..........100 Table 3. Variable weights and model-averaged estimates of intercept and variable coeficients for EEV models in Walton County, Florida. .....................................................101 x List of Figures I. Figure 1. Map of the EEV study area in the TNF study area during 2006. ..........................22 Figure 2. Map of the EEV study area the TNF study area during 2007 and 2008. ..............23 Figure 3. Histograms of the total number of female Cx. eraticus collected at adult sampling sites in the TNF study area in either 2006 or 2007 and 2008 cumulatively. .............................24 Figure 4. Density and trace plots of MCMC samples from the posterior densities of parameters in the best-fiting model for Cx. eraticus abundance at 2006 adult sampling sites, Model area. .....................................................................................................................25 Figure 5. Scaterplot of the total number of Cx. eraticus females collected at 41 adult sampling sites betwen 2007 and 2008 in the TNF study area versus the total height of dispersal kernels as parameterized in the area model centered on each larval pond, weighted by pond area, and overlapping at a site...................................................................................................................26 Figure 6. Scaterplot of Spearman?s Rank Correlation Coeficient (r S ) for asociations betwen the relative density of Cx. eraticus at adult sampling sites in the TNF study area in 2007 and 2008 and the total number of overlapping larval buffers at the site. ........................27 II. Figure 1. Forage ratios for avian species present in at least two of the total bloodmeal samples collected Cs. melanura, Cx. eraticus, and Cx. restuans in TNF betwen March and October from 2001 through 2009. ...........................................................................................54 Figure 2. Relative frequencies of blood-engorged Cs. melanura (N=70), Cx. eraticus (N=1457), and Cx. restuans (N=28) collected in TNF each month betwen 2001 and 2009. ...55 III. Figure 1. Cs. melanura(a) and Cx. eraticus (b) forage ratios of avian host species nesting either at ground/low-levels or in the mid-story/canopy..............................................................80 xi Figure 2. Scaterplot of Cs. melanura(a) and Cx. eraticus (b) forage ratios for avian host species versus body mas, with best-fit line from simple linear regresions overlain????.81 Figure 3. Scaterplot of Cs. melanura(a) and Cx. eraticus (b) forage ratios for avian host species versus nestling stage length, with best-fit line from simple linear regresions overlain.......................................................................................................................................82 IV. Figure 1. Sentinel site locations in Walton County, Florida used in model development. EEEV-positive sites are those where at least one chicken seroconverted to EEV antibodies in 2009. ....................................................................................................................................102 Figure 2. Image of 2009 avian point-count-locations centered on individual sentinel sites in Walton County, Florida. ..........................................................................................................103 Figure 3. Inferred rates of EEV transmision among sentinel sites of variable avian community sizes in Walton County, Florida, with estimated function of the probability of a site having a high level of virus activity overlain. ..........................................................104 Figure 4. Relative abundances of common yelowthroat at EEV-positive and EEEV-negative sentinel sites in 2009 in Walton County, Florida. ........................................105 Figure 5. Asociation betwen weights asociated with variables of relative abundances of avian species vs. estimated forage ratios of those species (r S (5) = 1.000, p= 0.0167). .........106 1! I. ESTIMATION OF DISPERSAL DISTANCES OF CULEX ERRATICUS IN A FOCUS OF EASTERN EQUINE ENCEPHALITIS VIRUS IN THE SOUTHEASTERN UNITED STATES Abstract Paterns of mosquito dispersal are important for predicting the risk of transmision of mosquito- borne pathogens to vertebrate hosts. We studied dispersal behavior of Culex eraticus (Dyar & Knab), a potentialy significant vector of eastern equine encephalitis virus (EEEV) that is often asociated with foci of this pathogen in the southeastern United States. Using data on the relative density of resting adult female Cx. eraticus around known emergence sites in Tuskegee National Forest, Alabama, we developed a model for the exponential decay of the relative density of adult mosquitoes with distance from larval habitats through parameterization of dispersal kernels. The mean and 99 th percentile of dispersal distance for Cx. eraticus estimated from this model were 0.97 km and 3.21 km per gonotrophic cycle, respectively. Parameterized dispersal kernels and estimates of the upper percentiles of dispersal distance of this species can potentialy be used to predict EEV infection risk in areas surrounding the TNF focus in the event of an EEV outbreak. The model that we develop for estimating the dispersal distance of Cx. eraticus from collections of adult mosquitoes could be applicable to other mosquito species that emerge from discrete larval sites. Introduction Urbanization and the acompanying modification of natural landscapes are increasing human exposure to mosquito-borne pathogens (Norris 2004, Pimentel et al. 2007, Patz et al. 2! 2008). Predictions of such increases follow directly from Pavlovsky?s theory of the natural nidality of transmisible diseases, which states that transmision of a vector-borne pathogen to humans occurs via asociation with the natural focus of the pathogen, with the focus defined as the specific conditions of habitat and geography that alow for maintenance of the natural transmision cycle of the pathogen (Pavlovsky 1966). Development adjacent to natural habitats such as wetlands increases the frequency of asociation betwen humans and isolated pathogen foci, either through the encroachment of human populations on foci or through areal expansion of foci themselves via habitat alteration that creates novel breeding sites for mosquito vectors (Norris 2004). Delineation of the geographic boundaries of mosquito-borne disease foci is thus necesary for acurate quantification of the degree of spatial asociation betwen humans and disease-causing pathogens. Such delineation is particularly relevant in the case of eastern equine encephalitis virus (EEV), a mosquito-borne pathogen that exhibits relatively stable foci and that is the most severe of the arboviral encephalitides in the United States. The human mortality rate of persons infected with EEEV is 30 - 40% (Whitley and Gnann 2002). Survivors of infection suffer mild to severe neurological damage and commonly require expensive, long-term medical care (Vilari et al. 1995). Horses and gamebirds are also commonly infected with the virus and experience mortality rates over 80% (Scott and Weaver 1989). Although it is an extremely pathogenic disease of humans and horses, EEV is one of the rarer viral encephalitides causing clinical infections in the United States. The rarity of EE cases may be due, in part, to the geographic isolation of the virus from areas of human habitation. EEV is endemic to freshwater swamp habitats where its primary enzootic vector, Culiseta melanura (Coquilet), and avian reservoir hosts are sympatric (Scott and Weaver 1989). As 3! such, delineation of the boundary zones surrounding these foci is critical to identifying high-risk areas as population growth and expansion in the Southeast potentialy lead to human encroachment on EEV foci (Wear and Greis 2001, Alig et al. 2004). In the southeastern United States, the mosquito species Culex eraticus (Dyar & Knab) has recently been identified as a bridge vector that may play a key role in the transmision of EEV to humans and horses (Cupp et al. 2003, Cupp et al. 2004, Cohen et al. 2009). This mosquito is a moderately competent vector of EEV and feds on both birds and mamals (Chamberlain et al. 1954, Hasan et al. 2003). Its typical larval habitat in the southeastern United States is permanent bodies of fresh water overgrown with surface plants (Horsfal 1955). These larval habitats can overlap areas where EEEV is endemic, and in such areas, relative densities of adult Cx. eraticus are high compared to other potential bridge vector species of the virus (Cupp et al. 2003, Cupp et al. 2004). A previous study of the dispersal behavior of Cx. eraticus indicated that it is a long- distance flier with a maximum flight range of 1.4 - 2.2 km and a mean dispersal distance of 0.73 (+-0.61) km (Morris et al. 1991). One approach to delineating the boundaries of areas where humans would be at risk of EEEV infection is to buffer al larval sites in an EEEV focus with a distance equivalent to the upper limit of the maximum flight range of Cx. eraticus. Flight distance estimates for Cx. eraticus reported by Morris et al. (1991) were based upon a mark- release-recapture approach, but these results must be viewed with caution for several reasons. First, the mosquitoes used in the study were not dispersing from their natural emergence or oviposition site, but from an arbitrary location in the middle of a wastewater treatment facility. Because the flight range of mosquitoes is known to vary with habitat (Silver 2008), mosquitoes released in this environment may display diferent paterns of dispersal relative to mosquitoes 4! dispersing from a natural emergence site. Second, mosquitoes were released in the morning in an open, sunny area. Culex eraticus seks resting sites during the morning hours to avoid desication (Gray et al. 2010) and would likely undertake an initial movement in search of such a suitable resting site. Finaly, marking, trapping, and handling mosquitoes may alter mosquito dispersal behavior (Silver 2008). These potential complications with the previous mark-release- recapture study highlight the need for the development alternative approaches to estimating dispersal distances of Cx. eraticus. In the current study we developed new estimates of the dispersal distance of Cx. eraticus that do not rely on mark-release-recapture methods. Because Cx. eraticus typicaly oviposits in rather large, easily located bodies of water, we were able to identify the most likely sites of emergence in a study area in central Alabama. We then parameterized a dispersal kernel for this species using the straight-line distances betwen the sampling sites where adult females were collected and their putative sources of emergence. Dispersal kernel parameterization is an approach to the study of dispersal in other organisms, most notably angiosperms. We develop a novel application of such ?sed dispersal? models to the study of dispersal of Cx. eraticus. Our goal was to both estimate the dispersal distances of Cx. eraticus and to ases the general utility of ?sed dispersal? models for studying mosquito dispersal and predicting relative mosquito densities on a local scale. Materials and Methods Field Methods We estimated the dispersal distances of female Cx. eraticus using data on the relative density of adult and larval mosquitoes collected betwen 2006 and 2009 in Tuskegee National 5! Forest (TNF) in Macon County, Alabama. TNF has served as the site of an ongoing study of the vector and vertebrate host dynamics of EEEV since 2001 and is described more fuly in Cupp et al. (2003). Briefly, the study site encompased a 28-km 2 circular area centered on a core wetland complex located approximately 3.0 km from the town square of the city of Tuskegee (32?38?40?N, 85?25?59?W). In 2006 - 2008, we sampled the adult mosquito population within the study site by aspirating resting mosquitoes from artificial shelters. In 2006, we used a variety of shelter types including fiber pots, resting boxes, and 50-galon plastic cans, but in 2007 and 2008 we exclusively used 50-galon black plastic cans, the most atractive type of shelter for resting mosquitoes in TNF (Burket-Cadena et al. 2008). Because the shelters used in 2006 varied in atractivenes to female mosquitoes, the data we used for the 2006 analysis were derived only from fiber pots and resting boxes, which showed no diference in atractivenes (Burket-Cadena et al. 2008). Restricting mosquito samples to the same type of collecting container within any year asured that atractivenes of shelters ultimately used in our analyses varied betwen, but not within, years. The number and locations of the adult sampling sites also varied betwen years. In 2006, six sampling sites were spaced approximately 0.43 km along each of five transects radiating out 3.0 km from a point at the center of the study site (Figure 1). In 2007 and 2008, seven sampling sites were located every 0.19 km along six transects radiating 1.5 km (Figure 2). We excluded six sampling sites, which were either moved betwen 2007 and 2008 or were adjacent to private lands on which we were unable to sample for larval mosquitoes. As such, the data used in our analyses originated from 26 of the 31 sampling sites where fiber pots and resting boxes were used for collection in 2006, and 41 of the 43 sampling sites from 2007 and 2008. 6! Adult mosquitoes were collected betwen March and October, with regular sampling occurring betwen June and September. In 2006, regular sampling consisted of seven sampling intervals, each of which was two weks in length and during which one mosquito collection was made at al 26 sampling sites. In 2007 and 2008, we collected mosquitoes once at al 41 sampling sites during each of 15 sampling intervals, with each sampling interval lasting one wek. Results of any sampling outside of these regular sampling intervals were excluded from our analyses, so that the number of mosquito samples collected is constant across sites sampled in the same year. Following collection in the field, mosquitoes were transported to the laboratory, anesthetized with CO 2 , and sorted on a chil table by species and gender. We also sampled al permanent ponds within the study site for mosquito larvae to identify Cx. eraticus larval habitats, and hence the sites of emergence of the population under study. We censused the ponds from mid-July to mid-September in 2009 by repeatedly dip-sampling along the perimeter of each pond. Thirty dips were taken at 100-m intervals along the perimeter of each pond using a larval dipper. Al larvae collected from each sampling point on a pond perimeter were transported back to the lab, heat-kiled, and sorted by species. Dispersal Distance Estimation Our approach to estimating the dispersal distances of Cx. eraticus was based on fiting a model of exponential decay with distance to Cx. eraticus relative density data from adult sampling sites. The model we used was a special case of the general set of models originaly developed to estimate the number of seds arriving at sed traps from multiple source trees distributed throughout a landscape (Clark 1998, Clark et al. 1998, Clark et al. 1999). Such sed dispersal models specify the number of seds at trap i as originating from a Poison distribution 7! with the mean and variance parameter ! i equal to the product of trap area A i and the sum, over al source trees j=1,,,n i that are located within the maximum sed dispersal distance of the trap, of the product of b ij , the estimated fecundity of each source tree j, a parameter !"!and g(d ij ), a probability density function (pdf) for sed arrival from each source, to an infinitely smal area centered on the location of the trap. g(d ij ) is understood to be a dispersal kernel (Nathan and Muller-Landau 2000), with a functional form that varies with the species for which dispersal is being modeled. Formaly, Y i !Pois ( i ) i =A i "b ij j1 n i # gd ij There is a clear correspondence betwen the problem of estimating the number of seds collected at a trap after dispersal from their parent trees and that of estimating the relative density of adult mosquitoes collected at a sampling site after dispersal from their larval habitats. As such, we adapted the above-formulated model to achieve the later goal. In the model that we developed for Cx. eraticus dispersal, Y i is a random variable of the total number of female Cx. eraticus collected over the course of the 2006 sampling season at an adult sampling site. While males were occasionaly collected at the adult sampling sites, we restricted our analysis to dispersal of female adults. Ponds throughout the TNF that had at least one larva collected during the first sixty dips of sampling, the number of dip samples taken at the smalest pond, were designated as suitable larval habitats. These ponds were clasified as the sources of dispersing female adults, and hereafter they wil be referred to as larval ponds. We estimated the distance betwen al adult sampling sites and al larval ponds by delineating the perimeter of each pond using the GPS coordinates of larval sampling points to create a polygon shapefile for each pond 8! in ArcGIS v.9.2. We then calculated the Euclidean distances betwen al adult sampling sites and larval pond centroids using UTM coordinates. A rigorous approach for estimating larval pond ?fecundity?, or rate of productivity of adult Cx. eraticus, was not available. Therefore, we developed a series of models with productivity parameterized in various manners: asumed constant across ponds (constant), proportional to pond area (area), proportional to pond perimeter (perim), proportional the number of larvae collected in the first sixty dips (larv), proportional to pond area*the number of larvae in the first sixty dips (area*larv), or proportional to pond perimeter*the number of larvae in the first sixty dips (perim*larv). The functional form of the dispersal kernel we used in the model was that of the exponential described in Clark et al. (1999) : g(d ij )= 1 2!" exp(#( d ij ). This dispersal kernel models the rate of decay of the!density with distance from the source as exponential. Parameterization of this part of the model efectively alows for estimation of the mean and percentiles of the dispersal distance of Cx. eraticus. Specificaly, 2!!is an estimate of mean dispersal distance (Clark et al. 1998, Cousens et al. 2008). Percentiles of dispersal distance were estimated by first converting the dispersal kernel g(d) to the distance pdf f(d) (Cousens et al 2008): f(d)= ! 2 exp("( d ).! We then calculated the median, 95 th , and 99 th percentiles of f(d) through integration using the rectangle method with subintervals that were 1 meter in length. We used an exponential form for 9! the dispersal kernel because it has received the most support from empirical studies of insect dispersal (Graton and Vander Zanden 2009). Finaly, counts of Cx. eraticus at adult sampling sites were often zero, so that the Negative Binomial was a more appropriate distributional asumption for these data than the Poison (Chi-Square Test of Goodnes-of-Fit ? Poison: ! 2 = 304.36 , df = 12, p < 0.001; Chi- Square Test of Goodnes-of-Fit ? Negative Binomial: = 5.90 , df = 11, p = 0.88). As such, we modeled the number of females collected at each adult sampling site as following a Negative Binomial (p,k) distribution, whereby k is a dispersion parameter and p is a function of the mean. Formaly, the models we used were of the form:! where In these models, !#$%! !are the two parameters to be estimated, d ij is the distance in meters betwen adult sampling site i and larval pond j, b j is the variable serving as the proxy measure of larval pond productivity, the index i runs over al sampling sites 1,..n = 26, and the index j runs over al larval ponds 1,,,m= 15. Y i ~NB (p i ,k) ln(? i )=lA i +ln!b j 1 2"# j= m $ exp(%( d ij ) =ln(A)+l" 1 2#$ ),A i = for al sapling st i=1,,,n 10! We conducted modeling in a Bayesian framework, whereby inference about parameters is based upon examination of their posterior distributions. We used Markov Chain Monte Carlo (MCMC) algorithms to sample from the posterior distribution of the unknown parameters, circumventing the need for their explicit derivation (Gilks et al. 1996). We ran these algorithms using R2WinBUGS, the implementation of WinBUGS in R. We set three chains to run for 2000 iterations with a burn-in period of 1000 iterations. The prior distributions were uniform (- 1000,1000) for !, uniform(0,5000) for !, and gama (0.001,0.001) for k. Initial values for the parameters were drawn from Uniform (-10,10) and Uniform (0,500) distributions for ! and , respectively, while the initial value for k was set to 1, 2, or 3. Convergence diagnostics were asesed using the Coda package. We based model selection on minimization of the Deviance Information Criterion (DIC). After selecting a final model for the 2006 data, we validated the model by applying it to 2007 and 2008 adult sampling sites. We used the 2007 and 2008 adult sampling sites for validation because they difered in location from the ones used in model development. Because the counts of Cx. eraticus at the adult sampling sites were summed over a diferent number of visits to each sampling site in 2007 and 2008 compared to 2006, we used Spearman?s Rank Correlation Coeficient to ases model performance (Guisan and Zimermann 2000). Additionaly, we were interested in the performance of the predicted relative density of mosquitoes from the model compared to another possible correlate that could be used to rank areas in terms of their relative densities. Specificaly, we considered the number of overlapping larval buffers at a sampling site, using the mean dispersal distance derived from the model as the buffer radius-length, as this other possible correlate. We estimated the Spearman?s Rank correlation coeficient betwen the relative density of Cx. eraticus and this variable, and then 11! compared both correlation coeficients using the Test of Two Correlated Correlation Coeficients (Meng et al. 1992). Finaly, we were interested in the relative performance of variables representing the total number of overlapping buffers of any radius-length in predicting the ranking of areas in terms of relative density of Cx. eraticus. As such, we created a set of variables similar to the last one described above, with buffer radius-lengths difering betwen variables by increments of 100 meters. We then estimated Spearman?s Rank Correlation Coeficient betwen each of these variables and relative density of Cx. eraticus in 2007 and 2008. Results We collected a total of 2900 resting female Cx. eraticus from the adult sampling sites throughout TNF betwen 2006 and 2008. Seven hundred and seven (24. 4%) of these individuals were excluded from our analyses because they were collected during sampling intervals when not al traps were visited, yielding a total of 205 individuals used in dispersal distance models developed using data from 2006 and 1988 individuals used in model validation and correlation analyses using the 2007-2008 data. In 2006, when the adult sampling sites were distributed throughout an area radiating out 3.0 km from the core site (Figure 1), we collected an average 7.88 (SD=11.73, n=26) females/sampling site over the seven two-wek intervals used in the analysis (Figure 3). The distances betwen adult sampling sites where resting females were collected and the nearest larval pond ranged from 0.107 km to 1.946 km. In 2007 and 2008, when the adult mosquito collections were focused within the circular area radiating out only 1.5 km from the core site (Figure 2), we collected an average of 48.49 12! females/sampling site (SD=56.08, n=41) over the 15 weks used in our analyses (Figure 3). All adult sampling sites in the study area had individuals present during at least one visit, and were distributed at distances ranging from 0.05 to 1.04 km from the nearest larval pond. In total, 787 Cx. eraticus larvae were collected, with an average of 3.47 larvae/30-dip sample (SD=6.64, n=227). Of the 21 ponds we sampled, 15 were found to have Cx. eraticus larvae present during the first sixty dips, and were thus classified as larval ponds (Figure 1). Chains used in the model development al indicated a satisfactory degree of convergence after 2000 iterations, as the Gelman-Rubin statistics for parameter values in each model were al les than 1.10. Average values of samples from the posterior distributions of ! ranged from - 8.61 to 2.16, from 483.46 to 1682.38 for!, and from 0.44 to 0.82 for k. Overal, model deviances ranged from 136.65 to 148.59 (Table 1). DIC of the models ranged from 137.33 to 145.03, with the area model, the one that used the area of the larval pond of origin as a measure of productivity, selected as the best-fiting model based on its lower DIC compared to the other five models (Table 1). The estimate of ! based on the mean of samples from its posterior distribution in the area model was -6.37 with a 95% credible interval of (-8.08,-4.80). The dispersal parameter, !, was estimated as 483.46 with a 95% credible interval of (258.53,1058.95). The mean value of the k for the area model was 0.82, with a 95% credible interval of (0.34,1.70) (Table 1, Figure 4). The estimated mean dispersal distance of Cx. eraticus (2!) ! was 0.966 km. The median, 95 th , and 99 th percentiles for dispersal distance were estimated as 0.811 km, 2.291 km, and 3.206 km, respectively. Application of the model to the validation dataset gave strong support to the model, as Cx. eraticus relative density at 2007 and 2008 adult sampling sites was significantly asociated 13! with the total height at the sampling site of the overlapping dispersal kernels centered on each larval pond and weighted by pond area (r S =0.689, 95% CI = (0.477, 0.820), n=41, p < 0.0001, Figure 5). The rank correlation betwen Cx. eraticus relative density and the number of overlapping larval buffers of radius-length equal to the mean dispersal distance were also significantly asociated (r S =0.597, 95% CI= (0.347, 0.761) , p < 0.0001, n =41) (Figure 6). Overal, there was no evidence to suggest that the strengths of asociation betwen these two variables?overlapping dispersal kernel height weighted by pond area and the number of overlapping buffers?and Cx. eraticus relative density were significantly diferent from one another (Z= 1.04, p > 0.10, n=41). Asociations betwen ranks of Cx. eraticus relative density and the number of overlapping larval buffers at a site were positive for al other radius-lengths considered other than mean dispersal distance. However, these asociations were only significant (p < 0.05) for buffers of radius-lengths les than 2500 m. Al asociations were weaker than that betwen the rank of Cx. eraticus relative density and predicted relative density based on the best-fiting model parameterized by 2006 data (overlapping dispersal kernel height weighted by pond area) (Figure 6). Discussion The distance that a mosquito can disperse from its site of emergence is of critical importance in studies of arboviruses. The conventional mark-release-recapture approach to the study of mosquito dispersal has the potential to bias estimates of dispersal distance because animals are often captive reared and/or released in an unfamiliar area that is not necesarily suitable habitat (Silver 2008). In this study we used a modeling approach based on sampling of wild Cx. eraticus to estimate the dispersal distances of mosquitoes emerging from natural 14! wetlands during a gonotrophic cycle. Despite concerns about overestimation of dispersal distances based on mark-release-recapture of lab-bred mosquitoes, our estimate of the average dispersal distance of 0.967 km is close to the mean dispersal distance of 0.73 (+-0.61) km for Cx. eraticus based on mark recapture (Morris et al. 1991). Moreover, our results confirm that Cx. eraticus is a strong flier, given that most mosquito species studied typicaly disperse a no more than few hundred meters during appetential flight (Service 1997). Knowledge of dispersal paterns can be used to predict the relative abundance of Cx. eraticus anywhere in the area of the study, as demonstrated by the strong asociation betwen the two correlates that we derived based on modeling results, total height of overlapping dispersal kernels weighted by pond area and number of overlapping buffers with radius length equal to the mean dispersal distance. For vector control, acurate estimates of mosquito abundance have obvious utility. Knowledge of mosquito abundance across a landscape could help formulate a strategy for adulticide application, and it could also be used to preddict changes in relative density of Cx. eraticus under various scenarios of larvicide applications to larval ponds. Estimates of dispersal distance resulting from the current study could also potentialy be used to delineate boundary zones around the TNF focus outside of which a potentialy infected bridge vector?i.e. a female ovipositing in a pond that overlaps an area of enzootic transmision betwen Cs. melanura and avian hosts?has a low probability of dispersing. For example, based our results of the current study, the probability that a Cx. eraticus female wil disperse further than 3.21 km from an oviposition site is 0.01. However, exposure to the bite of a bridge vector such as Cx. eraticus wil be a function of not only of the probability of dispersal to a given distance of a single vector, but also the total number of vectors dispersing such that a proposed 15! radius ~3 km for the TNF EEV boundary zones could stil involve an unaceptable exposure risk. Risk asesment from a human health perspective is beyond the scope of this paper, but we have provided important information for the development of such asesments. Whether or not Cx. eraticus proves to be an important component in EEV transmision in the southeastern United States, in this study we have demonstrated that models originaly developed to model sed dispersal are adaptable to the problem of estimating mosquito dispersal. Mark-release-recapture studies are subject to biases resulting from the use of captive bred animals, and studies of wild mosquitoes emerging from natural wetlands should be preferable for estimating the movement paterns of mosquitoes. While we applied the sed dispersal modeling approach to a species of mosquito that breeds in relatively discrete areas that are easily identifiable, it could potentialy be applied to mosquito species that have a more continuous distribution of breeding habitat, by representing sites of emergence as the centroids of pixels clasified as suitable breeding habitats in raster images and using asociated atribute data on productivity of the habitats represented by those pixels (Brown et al. 2008, Jacob et al. 2009). The acuracy of our estimates of the mean and upper percentiles of the dispersal distance of Cx. eraticus is contingent upon three asumptions. First, a critical asumption is that al potential larval ponds of Cx. eraticus that could contribute adults to the population were identified. While we fel confident in our inventory of source ponds, some of the adults that we captured could have emerged from ponds outside of our sampling area or from smal aquatic habitats. Minimaly, we identified al of the larval ponds near each sampling site. Thus, we are confident that our model included the primary sources of, and al of the sources that make a significant contribution to, adults for each sampling site. 16! The second asumption is that the measure of pond productivity used in the final model (larval pond area) is proportional to the number of adults that originate from that pond. Factors such as availability of aquatic vegetation and predator densities could afect productivity. However, as ponds in TNF are al relatively homogenous, being relatively shalow with vegetation scatered across their surface, the use of pond area as a proxy for mosquito production sems reasonable. Without more information on habitat needs of larval Cx. eraticus with which to ases pond quality and more data on the characteristics of each pond in the study area, there is no simple means to improve upon the use of pond area as a proxy for mosquito production. Our final asumption stems from the reduction of the areal extent of each larval pond down to representation by its centroid alone, so the dispersal distance may have been overestimated given dispersal of adults from any point betwen a pond centroid and its edge. However, the average distance betwen each larval pond?s centroid and larval sampling points along its edge was only 0.063 km, a negligible distance relative to the estimated mean dispersal distance. In conclusion, use of models originaly developed to estimate sed dispersal distances appears to be a powerful approach to characterizing the dispersal of Cx. eraticus given knowledge solely of the location of larval habitats and adult sampling sites. Evidence of the utility of this approach is the strong asociation betwen overlapping dispersal kernel heights predicted from the model and relative density when the model was applied to new sampling site locations. Our approach could be used in characterizing dispersal for other mosquito vector species, leading to more acurate predictions of their spatial distribution and thus the efective areas where vertebrate hosts are at risk of exposure to the pathogens they transmit. 17! References Alig, R.J., Kline, J.D., and M. Lichtenstein. 2004. Urbanization on the US landscape: looking ahead in the 21 st century. Landscape Urban Plan. 69: 219-234. Brown, H.E., and M. A. Diuk-Waser, Y. Guan, K. Caskey, and D. Fish. 2008. Comparison of three satelite sensors at three spatial scales to predict larval mosquito presence in Connecticut wetlands. Remote Sens Environ. 122: 2301-2308. Burket-Cadena, N.D., M.D. Eubanks, and T.R. Unnasch. 2008. Preference of female mosquitoes for natural and artificial resting sites. J Am Mosq Contr Asoc. 24: 228-235. Chamberlain, R.W., R.K. Sikes, D.B. Nelson, and W.D. Sudia. 1954. Studies on the North American arthropod-borne encephalitides: VI. Quantitative determinants of virus-vector relationships. Am J Hyg. 60:278-285. Clark, J.S. 1998. Why trees migrate so fast: Confronting theory with dispersal biology and the paleorecord. Am Nat. 152: 204-224. Clark J.S., E. Macklin, and L. Wood. 1998. Stages and spatial scales of recruitment limitation in southern Appalachian forests. Ecol Monogr. 68:213-235. Clark J.S., M. Silman, R. Kern, E. Macklin, and J. HileRisLambers. 1999. Sed dispersal near and far: Paterns across temperate and tropical forests. Ecology. 80:1475-1494. Cohen, S.B., K. Lewoczko, D.B. Huddleston, E. Moody, S. Mukherje, J.R. Dunn, T.F. Jones, R. Wilson, and A.C. Moncayo. 2009. Host feding paterns of potential vectors of eastern equine encephalitis virus at an epizootic focus in Tennese. Am J Trop Med Hyg. 81: 452- 456. Cousens, R., Dytham, C., and R. Law. 2008. Dispersal in plants: a population perspective. Oxford: Oxford University Pres. 18! Cupp, E.W., K. Klingler, H.K. Hasan, L.M. Vlguers, and T.R. Unnasch, 2003. Transmision of eastern equine encephalomyelitis virus in Central Alabama. Am J Trop Med Hyg. 68: 495- 500. Cupp, E.W., K.J. Tennesen, W.K. Oldland, H.K. Hasan, G.E. Hil, C.R. Katholi, and T.R. Unnasch. 2004. Mosquito and arbovirus activity during 1997-2002 in a wetland in northeastern Misisippi. J Med Entomol. 41:495-501. Gilks, W.R., S. Richardson, and D. Spiegelhalter. 1995. Markov Chain Monte Carlo in Practice (Interdisciplinary Statistics). Chapman & Hal/CRC, London. Graton, C. and J.M. Vander Zanden. 2009. Flux of aquatic insect productivity to land: comparison of lentic and lotic ecosystems. Ecology 90: 2689-2699. Gray, K.M., Burket-Cadena, N.D., Eubanks, M.D., and T.R. Unnasch. 2010. Crepuscular flight activity of Culex eraticus (Diptera: Culicidae). J Med Entomol. (in pres) Guisan, A. and N.E. Zimermann. 2000. Predictive habitat distributions in ecology. Ecol Model. 135: 147-186. Hasan, H.K., E.W.Cupp, G.E. Hil C.R. Katholi, K. Klingler, and T.R. Unnasch. 2003. Avian host preference by vectors of eastern equine encephalomyelitis virus. Am J Trop Med Hyg. 69: 641-647. Horsfal, W.R. 1955. Mosquitoes: Their Bionomics and Relation to Disease. The Ronald Pres Company, New York. Jacob, B.G., K. L. Arhert, D. A. Grifith, C. M. Mbogo, A. K. Githeko, James L. Regens, J. I. Githure, R. Novak, and J. C. Beier. 2009. Evaluation of Environmental Data for Identification of Anopheles (Diptera: Culicidae) Aquatic Larval Habitats in Kisumu and Malindi, Kenya. J Med Entomol. 42:751-755. 19! Meng, X.L., R. Rosenthal, D.B. Rubin, 1992. Comparing correlated correlation-coeficients. Psychol Bull. 111(1): 172-175. Morris, C.D., V.L. Larson, L.P. Lounibos. 1991. Measuring mosquito dispersal for control programs. J Am Mosq Contr. 7: 608-615. Nathan, R. and H.C. Muller-Landau. 2000. Spatial paterns of sed dispersal, their determinants and consequences for recruitment. Trends Ecol. Evol. 15: 278-285. Norris, D.E. 2004. Mosquito-borne diseases as a consequence of land use change. EcoHealth. 1:19-24. Patz, J.A., S.H. Olson, C.K. Uejio, and H.K. Gibbs. 2008. Disease emergence from global climate and land use change. Med Clin N Am 92: 1473-1491. Pavlovsky, E.N. 1966. Natural nidality of transmisable diseases: With special reference to landscape epidemiology of zooanthroponoses. University of Illinois Pres, Urbana, IL. Pimentel, D., S. Cooperstein, H. Randel, D. Filberto, S. Sorrentino, B. Kaye, C. Nicklin, J. Yagi, J. Brian. J. O?Hern, A. Habas, and C. Weinstein. 2007. Ecology of increasing diseases: Population growth and environmental degradation. Hum Ecol. 35: 653-668. Scott, T.W., and S.C. Weaver, 1989. Eastern equine encephalomyelitis virus: epidemiology and evolution of mosquito transmision. Adv Vir Res. 37: 277-328. Service, M.W. 1997. Mosquito (Diptera : Culicidae) dispersal - The long and short of it. J Med Entomol. 34: 579-588. Silver, J.B. 2008. Mosquito Ecology: Field Sampling Methods. Springer, Dordrecht, Netherlands. Vilari, P., A. Spielman, N. Komar, M. McDowel, and R.J. Timperi. 1995. The economic burden imposed by a residual case of eastern equine encephalitis. Am J Trop Med Hyg. 52: 20! 8-13. Wear, D. N., and J. G. Greis. 2001. The southern forest resource asesment: draft summary report. U.S. Forest Service, Ashevile, North Carolina, USA. Whitley, R.J., and J.W. Gnann, 2002. Viral encephalitis: familiar infections and emerging pathogens. Lancet. 359: 507-514. 21! Table 1. Mean (95% credible intervals) for parameters and deviance of the models developed for Cx. eraticus abundance at 2006 adult sampling sites in Tuskegee National Forest, Alabama, with DIC of the models also presented. Model names correspond to the variable used as a metric for larval pond productivity in each model, as explained in the Methods section. 22! Figure 1. Map of the EEV study area in the TNF study area during 2006. Forest boundaries are shown in green, and the political boundary of the city of Tuskegee is shown in black. Black circles represent the locations of the adult sampling sites in 2006, and the pink circles mark the centroids of al larval ponds of Cx. eraticus. 23! Figure 2. Map of the EEV study area the TNF study area during 2007 and 2008. The dotted green line delineates the area beyond which a Cx. eraticus female that emerges from any of the outer larval sites in the TNF has les than a 0.05 probability of dispersing. Circles represent the locations of adult sampling sites in 2007 and 2008, with sites with a low rate of capture of Cx. eraticus coded as light pink, and sites with high rates of capture coded as red. Clasification of sampling sites is based on whether the total number of Cx. eraticus adult females was below or above the median count among al sampling sites during 2007-2008. The total height of dispersal kernels as parameterized in the area model centered on each larval pond, weighted by pond area, and overlapping in each pixel is shown with equal interval symbology ranging from lowest (white) to highest (dark green). 24! Figure 3. Histograms of the total number of female Cx. eraticus collected at adult sampling sites in the TNF study area in either 2006 or 2007 and 2008 cumulatively. 25! Figure 4. Density and trace plots of MCMC samples from the posterior densities of parameters in the best-fiting model for Cx. eraticus abundance at 2006 adult sampling sites, Model area. Three chains were run for 2000 iterations with a burn-in period of 1000. A thinning interval of three was used, so that the number of samples shown is 334. 26! Figure 5. Scaterplot of the total number of Cx. eraticus females collected at 41 adult sampling sites betwen 2007 and 2008 in the TNF study area versus the total height of dispersal kernels as parameterized in the area model centered on each larval pond, weighted by pond area, and overlapping at a site. 27! Figure 6. Scaterplot of Spearman?s Rank Correlation Coeficient (r S ) for asociations betwen the relative density of Cx. eraticus at adult sampling sites in the TNF study area in 2007 and 2008 and the total number of overlapping larval buffers at the site. The dotted line across the top of the figure indicates the r S (0.69) for the asociation betwen the relative density of Cx. eraticus at an adult sampling site in 2007 and 2008 and the total height of dispersal kernels as parameterized in the area model centered on each larval pond, weighted by pond area, and overlapping at a site. ! ! ! 28! ! II. A MUTLI-YEAR STUDY OF MOSQUITO FEDING PATERNS ON AVIAN HOSTS IN A SOUTHEASTERN FOCUS OF EASTERN EQUINE ENCEPHALITIS VIRUS Abstract Eastern equine encephalitis virus (EEV) is a mosquito-borne pathogen that cycles in birds, but also causes severe disease in humans and horses. We examined paterns of avian host utilization by vectors of EEV in Alabama from 2001 to 2009 using blood-meal analysis of field-collected mosquitoes and avian abundance surveys. Northern cardinal (Cardinalis cardinalis) was the only preferred host (fed upon significantly more than expected, based on their abundance) of Culiseta melanura, the enzootic vector of EEV. Preferred hosts of Cx. eraticus, a putative bridge- vector of EEV, were American robin (Turdus migratorius), Carolina chickadee (Poecile carolinensis), barred owl (Strix varia), and northern mockingbird (Mimus polyglottis). Our results provide insight into the relationships betwen vectors of EEV and their avian hosts in the Southeast, and suggest that northern cardinal may be important in the ecology of EEV in this region. Introduction Birds are implicated as the primary reservoir hosts in transmision cycles of many mosquito- borne pathogens. 1 Recent research has emphasized that some bird species are over-utilized relative to their local availability by vector mosquitoes as bloodmeal sources, 2,3,4,5,6,7 and that 29! overutilization of certain avian species could influence the transmision of pathogens. 8 For example, temporal and spatial paterns of American robin (Turdus migratorius) abundance have been asociated with variability of the rate of human cases of West Nile virus (WNV) and prevalence of WNV in mosquito populations. 9,6 Such paterns have been atributed to a higher rate of feding upon American robins by foraging mosquitoes relative to other avian species, 10 which leads to a higher probability of infection of robins, potentialy causing robins to function as a superspreader of the virus. 3 Studies of avian host utilization by mosquitoes are vital components to elucidating the ecology of arbovirus transmision. To date, however, such studies have been largely restricted to potential vectors of WNV. 3,4,5,6,7 The extent to which observations from studies of WNV can be applied to other mosquito-borne pathogen systems for which birds serve as reservoir hosts is uncertain, as these pathogens vary in the ecological factors that influence transmision. 11 The ecology of the transmision dynamics of eastern equine encephalitis virus (EEV) is distinct from that of WNV. WNV is a periurban disease for which the primary enzootic vectors in the U.S. are Culex pipiens in the Northeast and Cx. quinquefasciatus in the Southeast. 12 In contrast, EEV is endemic to bottomland hardwood swamps, and the primary enzootic vector over most of its range is the ornithophilic mosquito Culiseta melanura. 13 Given the very diferent habitats in which the two viruses occur, as wel as the diferent vectors that transmit them, it sems likely that other aspects of the ecology of these viruses also difer. In this study, we asesed paterns of avian host species utilization by mosquitoes by employing data collected over nearly a decade on the East Gulf Coastal Plain in Alabama. We determined the identity of avian blood meals in eight mosquito species: Culiseta melanura, Cx. restuans, Aedes vexans, Coquiletidia perturbans, Cx. eraticus, Culex pecator, Cx. teritans, 30! and Ochlerotatus sticticus. Five of these species have been implicated in transmision of EEV, but the avian host preferences of these mosquitoes have not been characterized in detail. Culiseta melanura is widely recognized as the primary enzootic vector of EEEV, 14 and Cx. restuans has recently been proposed to function as an enzootic vector as wel. 15 Aedes vexans, Cq. perturbans, and Cx. eraticus have been proposed to play roles as bridge vectors. 14,16 Culex pecator, Cx. teritans, and Oc. sticticus were also found to have bird blood meals during this study but their roles in transmision of EEV are uncertain. In a previous study, we found that, collectively, mosquitoes at a study area in Tuskegee National Forest (TNF) fed on available vertebrate clases, with degree of catholiscism varying by mosquito species. 17 In the present study, we focus specificaly on characterizing paterns of avian host use by mosquitoes in TNF. A preliminary analysis of the paterns of mosquito feding on avian species in the same study area was published elsewhere. 2 That study examined paterns of host use based on bird abundances estimated from a smal number of point counts made within a limited portion of the study area. Moreover, the avian abundances used in the earlier analysis were based on estimates that did not acount for imperfect detectability of avian species from point count surveys. Here, we analyze multiple years of blood meal data using an improved null model for mosquito utilization of avian hosts relative to availability that acounts for species detectability. 18 Our goal was to produce comprehensive and acurate estimates of forage ratios of avian hosts for mosquito vectors of EEV. Materials and Methods Mosquito Surveys and Blood Meal Source Identification We studied the blood-feding paterns of mosquitoes over a nine-year period in a study area in Tuskegee National Forest (TNF) in Macon County, Alabama by regularly collecting 31! blood-engorged mosquitoes and using DNA analysis to identify the sources of their bloodmeals. This site has been the center of an ongoing study of the ecology of mosquitoes and their interactions with avifauna and herpetofauna in this focus of EEV since 2001, and it is described more fully elsewhere. 16,17,19,20 Briefly, the study site is a circular area encompasing 28 km 2 predominated by a complex of forest, ponds and wetlands. Much of the land is part of TNF, but also extends into adjacent private lands. Mosquitoes were collected annualy betwen March and October betwen 2001 and 2009, except in 2005, when no mosquito sampling occurred. Mosquito collection entailed aspiration of individuals from natural resting sites and a variety of container types serving as artificial resting sites imediately surrounding each sampling location, with the container types used varying over the course of the study. 17,19,20 Following collection, mosquitoes were transported to the laboratory at Auburn University, sorted on a chil table by species, sex, and engorgement status, and stored at -70?C until procesing for blood meal identification. The identity of host blood meals at the study site was determined by specific amplification of a portion of the vertebrate cytochrome B gene, as previously described. 2,17,21,22 The identity of the amplicons was determined using a combination of heteroduplex analysis and direct DNA sequencing, as previously described. 2, 5,17 Avian Surveys and Modeling We conducted surveys of avian populations at 338 locations spread uniformly throughout the study area from 15 May through June in 2008. We used a systematic sampling design for the surveys, whereby survey points were located 250 m apart on vertices of a grid that covered the study area. Surveys were conducted during two non-overlapping rounds of sampling so that each point was visited twice during the summer. Estimating the probability of detection for each 32! species requires repeated sampling of the same location, so during each of the two visits to a point count location, three consecutive 4-min counts were conducted. The species identification of al birds sen or heard within 100 m of the point-count location was recorded during each 4- min count period. Al sampling sesions occurred betwen 0500 and 1100 CDT. Nocturnal bird surveys were also conducted at a subset of 50 of the bird point count locations spaced 500 m apart. Nocturnal surveys were conducted using a combination of silent point counts and audio playback of the target species. Upon arriving at the survey location the observer conducted three consecutive three-minute counts of al individual birds detected within 200 m. The observer then played 20-seconds of chuck-will?s-widow (Caprimulgus carolinensis) cals followed by a one-minute count period. Next the observer played 20-seconds of whip-poor- will (Caprimulgus vociferous) cals followed by a one-minute count period. This procedure was then repeated for eastern screech-owls (Megascops asio), barred owls (Strix varia), and great horned owls (Bubo virginianus) in that order, with each species? cal being played followed by a one-minute count period. Care was taken to make sure that the audio recordings were not audible more than 200 m away from the observer. Nocturnal surveys were conducted by a single observer betwen 15 June and 3 July, 2009 betwen the hours of 2000 - 2400 and 0400 ? 0500 CDT. We estimated densities of avian hosts at mosquito-sampling sites by applying predictive models of density for each bird species recorded during 2008 and 2009 point counts. 18 We used N-mixture models to incorporate heterogeneity in detectability of individual species into models of occupancy and abundance. We modeled mean density of each species as a linear combination of covariates describing the relative abundances of habitat types in 100-m buffers around bird point count locations derived from Alabama Gap Analysis Project (GAP) land cover map 23 and 33! the National Land Cover Database Tree Canopy Cover Map, 24 asuming a Poison error distribution. While such count data may follow a Negative Binomial distribution, the Poison distribution has been found to be appropriate for many species detected during our surveys. 25 Al modeling was carried out using the program PRESENCE 26 with further details of model development given elsewhere. 27 Predicted densities of nocturnal species were standardized to a 100-m buffer, the area sampled for birds during diurnal avian surveys. Because of practical considerations, we were forced to exclude a smal number of avian species a priori from forage ratio calculations. First, house finches (Carpodacus mexicanus) were excluded from these analyses because individuals of this species were housed in sentinel cages at the center of the study area from 2002-2004. Avian species from Orders Ciconiiformes (herons) and Pelecaniformes (anhinga) were also excluded because point count methodologies do not provide acurate estimates of their densities 28 and we had no means to acurately census for these species. Inadequate numbers of wood ducks (Aix sponsa), chickens (Gallus gallus), and red-tailed hawks (Buteo jamaicensis) were detected to model abundance, and thus these three species were also excluded from forage ratio calculations. Statistical Analysis We estimated the rate of use of avian host species identified in blood meals collected betwen 2001 and 2009 relative to their availability by diferent mosquito species using the forage-, or selection-ratio approach described elsewhere. 29,30 With this approach, the ratio of the relative abundance of a host species in the bloodmeal sample to its relative abundance in the avian community is a forage- or selection-ratio. In the current study, the relative abundance of an avian host species in the bloodmeal sample was calculated separately for Cx. eraticus, Cx. 34! restuans, and Cs.melanura, using bloodmeal abundances summed across al study years. The relative abundance of a host species in the avian community was calculated separately for each of the three focal mosquito species using average estimated densities of avian hosts at al sites where individuals of each mosquito species were collected, respectively (Table 1). Statistical significance of the forage ratio estimate for an avian species was based on overlap of the 95% confidence intervals of the estimate with the value one. 30 An avian species was considered to have been prefered if it was overutilized relative to its rate availability to a mosquito species, such that the lower 95% confidence limit for the forage ratio estimate was greater than one. A species was inferred to have been avoided if it was underutilized relative to its rate of availability, such that the upper 95% confidence limit for the forage ratio estimate was les than one. An avian species for which the 95% confidence interval for its forage ratio included one was considered to have been fed upon opportunistically. 30 We additionaly estimated forage ratios for each avian host species using blood meals collected strictly betwen May 1st and August 15th to determine whether or not forage ratio estimates were biased by potentialy non-constant relative abundances of avian species betwen March and October of each year. The study area in and around TNF represents a rural environment undergoing no wide- scale alteration of habitats and with stable bird populations betwen years. We thus asumed that the composition of the avian community had been stable over the course of the study period, such that it was reasonable to use point-count surveys in the TNF study area conducted during 2008 and 2009 as representative of the relative abundances of each species in the avian community over the course of the study. To formaly test the validity of this asumption, we compared the avian community structure in and around TNF betwen 2001 and 2009 with data 35! from Breeding Bird Survey 31 along the Warrior Stand Route. The Warrior Stand Route runs through the TNF study area and was conducted across the same years as mosquito surveys. This BBS route has been surveyed within five days of the same date under nearly identical weather conditions and by a single observer (GEH) since 1998, so that comparisons of abundances betwen years are not biased by heterogeneity in detectability of species due to season, weather, time of day, or observer efects. We created a joint (2001+2009) data set, with record entries indicating the species identification of individual birds observed during the 2001 and 2009 Warrior Stand Breeding Bird Surveys, and the total number of records in the dataset equal to the total number of individuals observed in 2001 and 2009 (n tot = 1707, n 2001 = 856, n 2009 = 851). 32 We randomly sampled 856 individuals from this joint data set and asigned them to the first simulated 2001 community, and asigned al remaining 851 individuals in the joint community dataset to the first simulated 2009 community. We then calculated the diferences betwen these two simulated communities of the Shannon Index (H) and the Simpson Index (D), two common diversity indices used to ases community structure. 33 We repeated this randomization and index calculation procedure to yield 10,000 estimates each of the diferences in H and D betwen 2009 and 2001 simulated community-pairs. We then calculated the proportions of the simulated community-pairs for which the absolute value of the diferences in D and H were greater than the absolute value of the observed diferences in D and H betwen 2001 and 2009, respectively. We used these proportions as estimates of the p-values for two-tailed tests of the null hypothesis that the avian community structure had not changed along the Warrior Stand Route betwen 2001 and 2009. 32 36! Results A total of 42 avian species were identified as the sources for 528 blood meals from nine species of mosquito in the TNF study area betwen 2001 and 2009 (Table 2). Culex restuans and Culiseta melanura fed primarily on perching birds (Order Paseriformes), taking 72.0 and 77.4 % of blood meals from perching birds, respectively and secondarily on herons (Family Ardeidae, Order Ciconiiformes), taking 24.0 and 11.3 % of blood meals from herons, respectively. Other avian hosts of these mosquitoes included yelow-biled cuckoos (Order Cuculiformes), representing 4.0 and 5.66 % of blood meals from Culex restuans and Culiseta melanura, respectively; and owls (Order Strigiformes), representing 5.66 % of Culiseta melanura blood meals. Neither Culex restuans nor Culiseta melanura were found to fed upon chickens or wild turkeys (Order Galiformes), anhinga (order Pelecaniformes), raptors (Order Falconiformes), ducks (Family Anatidae; order Anseriformes) or hummingbirds (order Apodiformes). Culex eraticus fed primarily on herons (64.3%), followed by birds from a wide variety of avian orders, including by perching birds (24.7%), ducks (5.8%), owls (2.4%), galinaceous birds (1.2%), cuckoos (1.0%), anhinga (0.5%) and hummingbirds (0.2%). The majority of avian bloodmeals were derived from birds that have established breeding populations in central Alabama (Table 2). Those species that do not have breeding populations in central Alabama either over-winter, migrate through (e.g., American bitern Botaurus lentiginous), or are domesticated (chicken). Culiseta melanura significantly overutilized northern cardinal (Cardinalis cardinalis) relative to its rate of availability in the avian community, and thus the northern cardinal was inferred to have been a preferred host of Cs. melanura. Ninety-five percent confidence intervals of forage ratios for the 14 other bird species that were identified in Cs. melanura blood meals included a value of one, suggesting that these bird species were fed upon opportunisticaly 37! (Table 3; Figure 1). Avian species inferred to have been preferred by Cx. eraticus included American robin, orchard oriole (Icterus spurious), northern mockingbird (Mimus polyglottis), wild turkey (Meleagris gallapavo), Carolina chickadee (Poecile carolinensis), barred owl, and northern cardinal. Carolina wren (Thryothorus ludovivianus) and hooded warbler (Wilsonia citrina) were both inferred to have been avoided by Cx. eraticus. Forage ratios of the remaining 13 species that Cx. eraticus fed upon had 95% confidence intervals that included a ratio of one, suggesting that those species were fed upon opportunisticaly (Table 3; Figure 1). The 95% confidence intervals around the forage ratios of the seven other bird species in blood meal sample from Cx. restuans included a ratio of one, indicating that these bird species were also fed upon opportunisticaly by this mosquito species (Table 3; Figure 1). In general, forage ratio estimates based on expected frequencies of les than five blood meals under the null model of opportunistic feding should be viewed with caution. 30 Avian species with expected frequencies greater than or equal to five in the bloodmeal sample from Cx. eraticus were Carolin wren, yelow-biled cuckoo (Coccyzus americanus), tufted titmouse (Baeolophus bicolor), and northern cardinal, and northern cardinal in the bloodmeal samples from Cs. melanura. Comparisons of the forage ratios estimated from bloodmeal data covering the entire March-to-October mosquito sampling period and bloodmeal data from May 1st and August 15 th , when birds are not migrating in east-central Alabama, revealed a high degree of consistency for estimates betwen the two periods for both Cs. melanura and Cx. eraticus (Table 4). However, the confidence intervals of orchard oriole, wild turkey, and northern cardinal included one when forage ratios and asociated standard errors for these three species were based on the bloodmeals collected strictly betwen May 1 st and August 15 th . Comparison betwen the Cs. restuans 38! samples were not made because only 4 individuals of this species yielding avian-derived bloodmeals were collected betwen May 1 st and August 15 th over the eight years of sampling. The observed diference in the Shannon Index betwen the 2001 and 2009 data from the BBS Warrior Stand Route was 0.022; the proportion of the community-pairs for which the absolute value of the diference in H exceded the absolute value of this observed value was 0.624. The observed diference in the Simpson Index betwen the 2001 and 2009 data from the BBS Warrior Stand Route was 0.003; the proportion of the community-pairs for which the absolute value of the diference in H exceded the absolute value of this observed value was 0.436. There was thus no evidence to reject the null hypothesis of a stable avian community structure betwen 2001 and 2009 along the Warrior Stand Route, using either the Shannon or Simpson Index as a measure of avian community structure and an alpha-cutoff of 0.05. Moreover, the rank of species abundances in the 2001 and 2009 samples were positively correlated (r S(49) = 0.90, p <0.001). As such, there was strong support for the validity of our assumption that the relative abundances of avian species in the TNF study area estimated from point count surveys during 2008 and 2009 were representative of their relative abundances over the course of the study period. Discussion Through comparisons of the sources of mosquito blood meals to the local avian community, we found that putative vectors of EEEV in the Southeast do not fed upon birds opportunisticaly; rather, these mosquito species use some species of birds more or les than expected based on their relative abundance in the environment. While a number of studies have previously demonstrated similar heterogeneity in mosquito feding paterns, 2,3,4,5,6,7,34 our study is the first to demonstrate such heterogeneity at the host-species level for Cs. melanura, the 39! primary enzootic vector of EEV. Our results provide evidence that the northern cardinal is a preferred host of Cs. melanura. As such, the northern cardinal wil likely be exposed more frequently to EEV than other avian species and thus we predict that it plays an important role in ecology of EEV in the Southeast. In addition to northern cardinal, ten avian species?common yelowthroat (Geothlypis trichas), gray catbird (Dumetela carolinensis), eastern towhee (Pipilo erythrophthalmus), Louisiana waterthrush (Parkesia motacila), yelow-throated vireo (Vireo flavifrons), barred owl (Strix varia), hooded warbler (Wilsonia citrina), Acadian flycatcher (Empidonax virescens), red- eyed vireo (Vireo olivaceus), and blue-gray gnatcatcher (Polioptila caerulea)? had forage ratio estimates for Cs. melanura that were greater than one, suggesting that these species may also be preferred hosts. Three of these species?barred owl, common yelowthroat, gray catbird?were fed upon much more than expected, based on their relative abundances. While the confidence interval for the forage ratios of these ten species included one?making their over-representation in blood meals not statisticaly significant?standard error calculations were based on sample sizes that were too smal to provide reliable estimates of confidence intervals. 30 Despite low sample sizes and large standard errors, we suggest that northern cardinal, barred owl, common yelowthroat, and gray catbird have the highest probabilities for playing important roles in EEV transmision among the avian species for which forage ratios were estimated in the current study. The established model for EEV transmision in the northeastern United States, which implicates Cs. melanura as the primary enzootic vector of the virus, is commonly extrapolated as an appropriate model for transmision of EEV throughout North America. Recent studies, however, have suggested that this ?northeastern model? may not acurately depict transmision of EEV in southeastern foci, and that other mosquito species, especialy Cx. eraticus and Cx. 40! restuans, may be important to enzootic transmision in the southeastern region. 15,35 If Cx. eraticus or Cx. restuans play prominent roles in EEV transmision in the Southeast, inference about avian host preferences of these mosquito species become important. Culex eraticus and Cx. restuans both had high forage ratios for northern cardinal, and northern cardinal was inferred to be a preferred host species of Cx. eraticus when forage ratios were calculated using the entire sample of bloodmeals collected betwen March and October. We asumed that the avian community structure is most stable betwen 1 May and 15 August, the period after spring migration and before late summer dispersal and migration of birds. When forage ratios were based on the restricted samples of bloodmeals collected during this period, neither Cx. eraticus nor Cx. restuans exhibited significant feding preferences for northern cardinal. The few bloodmeal samples available during this restricted period for Cx. restuans limits our ability to make inferences regarding significant rates of over-and-under utilization of avian hosts by this mosquito species. For Cx. eraticus, however, samples sizes of blood meals from betwen 1 May and 15 August were adequate to make inferences, and we found that the American robin, Carolina chickadee, barred owl, and northern mockingbird are the preferred hosts of Cx. eraticus. As such, the northern cardinal may be les important in EEV transmision compared to these species if Cx. eraticus is more important enzootic vector of the virus in the Southeast. These results underscore the need for further research of the relative contribution of diferent mosquito species to EEV in this region. Al of our conclusions regarding forage ratios must be considered in light of the fact that we had to exclude some species from our forage ratio analyses because we were unable to acurately census these birds. Notable among these were herons, which comprised a large proportion of the bloodmeals from Cx. eraticus. The necesity of excluding herons from our 41! analysis was unfortunate because herons may also play an important role in the ecology of EEV in the southeastern United States, and herons comprised a large percentage of avian blood meals in ours study. The fraction of total avian blood meals from herons varied by mosquito species, comprising 64.3% percent of the avian bloodmeals of Cx. eraticus but just 11.3% percent of Cs. melanura blood meals. This diference could reflect contact rates of mosquitoes and herons, given diferences in the ecologies of the mosquitoes and birds. Culex eraticus breeds in permanent ponds and the densities of Cx. eraticus females decline with distance from these breeding sites. 20 Herons are waterbirds and thus are also more likely to be found at permanent ponds. Culiseta melanura, in contrast, breeds in water pockets asociated with buttresed trees and temporary puddles of water created by uprooted trees that occur in swamp habitats, 36 but not necesarily near permanent water, and thus may encounter waterbirds les frequently than Cx. eraticus. Given the high proportion of blood meals derived from herons, we would conclude that herons are likely to be frequently exposed to EEV. Interestingly, in a study of exposure of diferent avian species to EEV in Louisiana, 54.8% of heron species tested positive for EEV neutralizing antibodies, whereas only 26.2% of paserine species were seropositive. 37 Notable among the exposure rates of individual species was the high seroprevalence of yelow-crowned night-herons (Nyctanassa violacea, 86.1%), 37 the second most common avian host of Cx. eraticus in the current study. Several studies of defensive behaviors of birds to foraging mosquitoes found that some ciconiiforms, due to their stand-and-wait foraging technique, are highly susceptible to questing mosquitoes. 38,39,40 Despite the high proportion of blood meals derived from herons, the high seroprevalence of EEV in wild herons, and evidence that herons are important hosts for many of medicaly-important mosquitoes, relatively litle efort has been 42! directed towards quantifying the role of ciconiiform birds in the amplification of arboviruses, relative to paserines. Results of the current study underscore the need for further research investigating the role of ciconiiform species in EEV transmision. The role that a preferred species play in EEV transmision follows directly from its forage ratio, i.e. the likelihood that an individual of that species wil be fed upon. Because they are more likely to be fed upon in general, highly ranked (preferred) hosts have a higher likelihood of being fed upon by an infected mosquito than lower ranked (les-preferred) hosts. It also follows that such highly ranked hosts, simply because they are a more probable hosts, are more likely to be fed upon by a second uninfected mosquito that subsequently becomes infected. Consequently, asuming uniform reservoir competence (i.e., magnitude and duration of viremia), an individual of a more preferred avian host species should be a more important virus amplifier than an individual from a les-preferred species. Al else equal, preferred host species wil play a more important role in transmision dynamics than those that are les preferred. One of the factors that could confound the relationship betwen vector feding preference and host importance in virus amplification is reservoir competence. Preferred host species wil have a more important role in transmision dynamics only if al avian host species are equivalent in terms of their reservoir competence. Conversely, if a preferred avian species is a poor reservoir host, then that species wil act as a dilution host, reducing contact rates betwen vectors and competent reservoirs. 41 Reservoir competence has been reported for a number of avian species, 42 but it was not possible for us to infer interspecific diferences in reservoir competence of birds based on such estimates. Further research is needed in this area to more acurately ases the influence of variability in reservoir competence of avian hosts on the transmision dynamics of EEV. 43! The shift in inference in terms of preferred hosts of Cx. eraticus that occurred when forage ratio calculations were based on samples collected either from March through October or betwen 1 May and 15 August is not surprising given that the month of peak intensity of blood- feding in this species is August (Fig 2). Individuals of avian host species with a high relative abundance in the Cx. eraticus bloodmeal sample collected betwen March and October may be more preferable hosts compared to individuals of other species, or they may simply have inflated forage ratios caused by changes in the avian community betwen May and the later weks of August, when post-breeding dispersal and migration begins. The later possibility sems highly plausible, because resident birds make up an increasing proportion of the avian community as migratory species leave the study area beginning in August. Thus, northern cardinal and wild turkey, which are not long-distance migrants, are likely to have a greater rate of availability to Cx. eraticus from mid August through October than reflected in their relative abundances based on point count surveys conducted during the breeding season. This confounding influence of an underestimated rate of availability of avian hosts was les likely to be present in the calculation of the forage ratio estimates of Cs. melanura, as this species has a peak intensity of blood- feding in TNF in May (Fig 2), when the structure of the avian host community as estimated from point count surveys should be highly stable. Our goal in conducting this study was to estimate forage ratios for avian species for putative vectors of EEV. Such forage ratios represent potential proxy measures of the level of exposure to EEV that individuals of diferent host species experience. An alternative approach to measuring EEV exposure is to estimate the seroprevalence or seroconversion rate of EEV directly in birds or to asay for antibodies of the virus. In studies in Michigan 43 and New Jersey 4 , northern cardinals and gray catbirds had high EEV antibody seroprevalences, when 44! present in the sample of surveyed birds. Barred owls and common yelowthroats, the species with the highest Cs. melanura forage ratios in this study, were not present in samples from either of these studies. Overal, these studies confirm our asertion that northern cardinal, gray catbirds, and potentialy barred owls and common yelowthroats, are exposed at a high rate to EEV, in regions where Cs. melanura is the primary enzootic vector of the virus, given evidence from this study of the high Cs. melanura forage ratios for these species. References 1. Gubler DJ, 2002. The global emergence/resurgence of arboviral diseases as public health problems. Arch Med Res 33: 330-342. 2. Hasan HK, Cupp EW, Hil GE, Katholi CR, Klingler K, Unnasch TR, 2003. Avian host preference by vectors of eastern equine encephalomyelitis virus. Am J Trop Med Hyg 69: 641-647. 3. Kilpatrick AM, Daszak P, Jones MJ, Marra PP, Kramer LD, 2006. Host heterogeneity dominates West Nile virus transmision. Proc R Soc Lond B Biol Sci 273: 2327? 2333. 4. Patrican LA, Hacket LE, Briggs JE, McGowan JW, Unnasch TR, Le JH, 2007. Host- feding paterns of Culex mosquitoes in relation to trap habitat. Emerg Infect Dis 13: 1921- 1923. 5. Savage HM, Aggarwal D, Apperson CS, Katholi CR, Gordon E, Hasan HK, Anderson M, Charnetzky D, McMilen L, Unnasch EA, Unnasch TR, 2007. Host choice and West Nile virus Infection rates in blood fed mosquitoes, including members of The Culex pipiens complex, from Memphis and Shelby County, Tennese 2002? 2003. Vector Borne Zoonotic Dis 7: 365?386. 6. Diuk-Waser MA, Molaei G, Simpson JE, Folsom-O'Kefe CM, Armstrong PM, Andreadis 45! TG, 2010. Avian communal roosts as amplification foci for West Nile virus in urban areas in northeastern United States. Am J Trop Med Hyg 82: 337- 343. 7. Mackay AJ, Kramer WL, Mece JK, Brumfield RT, Foil LD, 2010. Host feding paterns of Culex mosquitoes (Diptera: Culicidae) in East Baton Rouge Parish, Louisiana. J Med Entomol. 47: 238-248. 8. Carver S, Bestal A, Jardine A, Ostfeld RS, 2009. Influence of hosts on the ecology of arboviral transmision: Potential mechanisms influencing dengue, Murray Valey encephalitis, and Ross River virus in Australia. Vector-borne Zoonot 9: 51-64. 9. Kilpatrick AM, Kramer LD, Jones MJ, Marra PP, Daszak P, 2006. West Nile virus epidemics in North America are driven by shifts in mosquito feding behavior. Plos Biol 4: 606- 610. 10. Simpson JE, Folsom-O'Kefe CM, Childs JE, Simons LE, Andreadis TG, Diuk-Waser MA, 2009. Avian host-selection by Culex pipiens in experimental trials. PLoS ONE 4: e7861. doi:10.1371/journal.pone.0007861. 11. Kuno G, Chang, GJ, 2005. Biological transmision of arboviruses: reexamination of and new insights into components, mechanisms, and unique traits as wel as their evolutionary trends. Clin Microbiol Rev 18: 608-637. 12. Turrel MJ, Dohm DJ, Sardelis MR, O'Guinn ML, Andreadis TG, Blow JA, 2005. An update on the potential of North American mosquitoes (Diptera : Culicidae) to transmit West Nile virus. J Med Entomol 42: 57-62. 13. Weaver SC, 2001. Eastern equine encephalitis. Service MW, ed. The Encyclopedia of Arthropod-transmited Infections. New York: CABI Publishing, 151?159. 14. Scott TW, Weaver SC, 1989. Eastern equine encephalomyelitis virus: epidemiology and 46! evolution of mosquito transmision. Adv Virus Res 37: 277?328. 15. Cohen, SB Lewoczko K, Huddleston DB, Moody E, Mukherje S, Dunn JR, Jones TF, Wilson R, Moncayo AC, 2009. Host feding paterns of potential vectors of eastern equine encephalitis virus at an epizootic focus in Tennese. Am J!"#$%!&'(!)*+!,-.! ! &'()&'*/! 16. Cupp EW, Klinger K, Hasan HK, Viguers LM, Unnasch TR, 2003. Eastern equine encephalomyelitis virus transmision in central Alabama. Am J Trop Med Hyg 68: 495?500. 17. Burket-Cadena ND, Graham SP, Hasan HK, Guyer C, Eubanks MD, Katholi CR, Unnasch TR, 2008. Blood feding paterns of potential arbovirus vectors of the genus Culex targeting ectothermic hosts. Am J Trop Med Hyg 79: 809-815. 18. Royle JA, 2004. N-mixture models for estimating population size from spatialy replicated counts. Biometrics 60: 108-115. 19. Burket-Cadena ND, Eubanks MD, Unnasch TR, 2008. Preference of female mosquitoes for natural and artificial resting sites. J Am Mosq Control Asoc 24: 228-235. 20. Estep LK, Burket-Cadena ND, Hil GE, Unnasch RS, Unnasch TR. Estimation of dispersal distances of Culex eraticus in a focus of eastern equine encephalitis virus in the southeastern United States. J Med Entomol : in pres. 21. Apperson CS, Harrison BA, Unnasch TR, Hasan HK, Irby WS, Savage HM, Aspen SE, Watson DW, Rueda LM, Engber BR, Nasci RS, 2002. Host-feding habits of Culex and other mosquitoes (Diptera: Culicidae) in the borough of Queens in New York City, with characters and techniques for identification of Culex mosquitoes. J Med Entomol 39: 777?785. 47! 22. Le, JH, Hasan H, Hil G, Cupp EW, Higazi TB, Mitchel CJ, M.S. Godsey MS, Unnasch TR, 2002. Identification of mosquito avian-derived blood meals by polymerase chain reaction-heteroduplex analysis. Am J Trop Med Hyg 66: 599-604. 23. Kleiner KJ, Mackenzie MD, Silvano AL, Grand JA, Grand JB, Hogland J, Irwin ER, Mitchel MS, Taylor BD, Earnhardt T, Kramer E, Le J, McKerrow AJ, Rubino MJ, Samples K, Terando A, Wiliams SG, 2007. GAP Land Cover Map of Ecological Systems for the State of Alabama (Provisional). Alabama Gap Analysis Project. Acesed January 28 th , 2008 from ww.auburn.edu\gap 24. Homer C, Huang C, Yang L, Wylie B, Coan M, 2004. Development of a 2001 National Landcover Database for the United States. Photogramm Eng Rem S 70: 829- 840. 25. Chandler RB, King DI, Destefano S, 2009. Scrub-shrub bird habitat asociations at multiple spatial scales in beaver meadows in Masachusets. Auk 126: 186-197. 26. Hines JE, 2006. Presence 2.1. Software to estimate patch occupancy and related parameters. USGS-PWRC. Available at: ww.mbr- pwrc.usgs.gov/software/presence.html. Acesed December 1, 2009. 27. McClure CJW, Estep LK, Hil GE. Using public land cover data to determine habitat asociations of breeding birds in Tuskegee National Forest, Alabama. South J Appl For: in pres. 28. Johnson DH, Gibbs JP, Herzog M, Lor S, Niemuth ND, Ribic CA, Seamans M, Shafer TL, Shriver WG, Stehman SV, Thompson WL, 2009. A sampling design framework for monitoring secretive marshbirds. Waterbirds. 32: 203-215. 29. Hes AD, Hayes RO, Tempelis CH, 1968. The use of the foraging ratio technique in 48! mosquito host preference studies. Mosq News 28: 386?389. 30. Manly BF, McDonald L, Thomas DL, McDonald TL, Erickson WP, 2002. Resource Selection by Animals: Statistical Design and Analysis for Field Studies. Dordrecht, The Netherlands: Kluwer Academic Publishers, 50-62. 31. Sauer, JR, Hines, JE, Falon, J, 2003. The North American Breding Bird Survey, Results and Analysis 1966?2002. Laurel, MD: USGS Patuxent Wildlife Research Center. 32. Solow AR, 1993. A simple test for change in community structure. J Anim Ecol: 62: 191- 193. 33. Peet RK, 1974. The measurement of species diversity. Ann Rev Ecol Syst 5: 285-307.! 34. Molaei G, Oliver J, Andreadis TA, Armstrong PM, Howard JJ, 2006. Molecular identification of blood-meal sources in Culiseta melanura and Culiseta morsitans from an endemic focus of eastern equine encephalitis virus in New York. Am J Trop Med Hyg 75: 1140-1147. 35. Cupp EW, Tennesen KJ, Oldland WK, Hasan HK, Hil GE, Katholi CR, Unnasch TR, 2004. Mosquito and arbovirus activity during 1997-2002 in a wetland in northeastern Misisippi. J Med Entomol 41: 495-501. 36. Hayes RO, Maxfield HK, 1967. Interruption of diapause and rearing larvae of Culiseta melanura (Coq.). Mosq News 27: 458-461. 37. Stam DD, 1958. Studies on the ecology of equine encephalomyelitis. Am J Public Health 48: 328-335. 38. Edman JD, Kale HW, 1971. Host behavior: its influence on the feding succes of mosquitoes. Ann Entomol Soc Am 64: 513-516. 49! 39. Webber LA, Edman JD, 1972. Anti-mosquito behavior of ciconiiform birds. Anim Behav 20: 228-232. 40. Edman JD, Day JF, Walker ED, 1984. Field confirmation on the diferent antimosquito behavior of herons. Condor 86: 91-21. 41. LoGiudice K, Ostfeld RS, Schmidt KA, Kesing F. 2003. The ecology of infectious disease: Efects of host diversity and community composition on Lyme disease risk. PNAS 100 (2) 567-57. doi:10.1073/pnas.0233733100 42. Komar N, Dohm DJ, Turel MJ, Spielman A, 1999. Eastern equine encephalitis virus in birds: Relative competence of European starlings (Sturnus vulgaris). Am J Trop Med Hyg 60: 387-391. 43. McLean RG, Frier G, Parham GL, Francy DB, Monath TP, Campos EG, Therrien A, Kerschner J, Calisher CH, 1985. Investigations of the vertebrate hosts of eastern equine encephalitis during an epizootic in Michigan, 1980. Am J Trop Med Hyg 34: 1190- 1202. 44. Crans WJ, Cacamise DF, McNely JR, 1994. Eastern equine encephalomyelitis virus in relation to the avian community of a coastal cedar swamp. J Med Entomol 31 : 711- 728. 50! Table 1. Predicted relative abundances of avian species observed during point-count surveys in TNF and used in forage ratio calculations. 51! Table 2. Total number of blood meals derived from avian species for mosquitoes collected in TNF betwen March and October from 2001 through 2009. 52! Table 3. Forage ratios (95% CI) of the avian species from which blood meals were derived for Cx. eraticus, Cx. restuans, and Cs. melanura betwen March and October from 2001 through 2009. 53! Table 4. Forage ratios of avian species using al bloodmeals collected betwen March and October, or alternatively, strictly betwen May and August 15 th . 54! Figure 1. Forage ratios for avian species present in at least two of the total bloodmeal samples collected Cs. melanura, Cx. eraticus, and Cx. restuans in TNF betwen March and October from 2001 through 2009. Bars show estimated standard errors, and the numbers above bars are sample sizes. 55! Figure 2. Relative frequencies of blood-engorged Cs. melanura (N=70), Cx. eraticus (N=1457), and Cx. restuans (N=28) collected in TNF each month betwen 2001 and 2009. ! ! ! ! 56! III. DEVELOPING MODELS FOR THE FORAGE RATIOS OF AVIAN HOSTS FOR CULISETA MELANURA AND CULEX ERRATICUS USING HOST CHARACTERISTICS Abstract The relative rates of contact betwen bird species and mosquito vectors, as measured by forage ratios, suggest that some bird species are used as hosts more than would be expected by chance. While interspecific variation in the rates of utilization of diferent avian hosts is potentialy one of the factors driving spatial and temporal paterns of the occurrence of mosquito-borne pathogens, the ecological factors that might make some birds more or les susceptible to questing mosquitoes have been litle studied. We developed linear regresion models for two mosquito species, Culiseta melanura and Culex eraticus, and used multimodel inference to identify avian host characteristics that could be used to predict forage ratios of these two species. We found nesting stratum, body mas, and length of nestling stage of avian host species to be useful for predicting Cx. eraticus forage ratios. Nestling stage length received strong support as a predictor in our model of Cs. melanura forage ratios. Our results suggest that characteristics of avian hosts may predict relative rates of contact of avian host species with mosquito vectors. Introduction Many vector-borne pathogens require both an arthropod host and a vertebrate host to complete their life cycles (Marquardt et al. 2004). For many such pathogens, however, the identity and importance of arthropod vectors is relatively wel known, while the suite of vertebrate hosts is 57! large, variable, and often poorly characterized. Specificaly for mosquito-borne pathogens, transmision dynamics may be influenced by heterogeneities amongst vertebrates in their relative rates of contact with mosquito vectors (Kilpatrick et al. 2006). Many mosquito species demonstrate distinct preferences for specific groups of vertebrates, ranging from vertebrate clas- to species-level specificity (Boakye et al. 1999; Hasan et al. 2003; Burket-Cadena et al. 2008). As a consequence of these preferences, groups of vertebrates vary in their individual-specific rates of contact with mosquito vectors. Variability in individual-specific contact rates operates in conjunction with variability in reservoir competence to produce heterogeneities amongst vertebrate species in their relative contribution to pathogen transmision (Kilpatrick et al. 2006; Kent et al 2009). Heterogeneities amongst avian species in terms of their relative contact rates with mosquitoes may be particularly important to predicting and controlling outbreaks of mosquito-borne disease in the United States, as birds are implicated as reservoir hosts of the majority of mosquito-borne pathogens of public health concern in this country (Gubler et al. 2001). The rarest, but most severe in terms of morbidity and mortality, of these viruses is eastern equine encephalitis virus (EEV; Scott and Weaver 1989, Vilari et al. 1995). Understanding of the transmision dynamics of EEEV is for the most part restricted to circumstances in which Cs. melanura is the primary enzootic vector, as typicaly occurs in the northeastern US, and where birds are reservoir hosts (Scott and Weaver 1989). In studies of EEV life cycle, no consideration is generaly given to the influences of specific avian host species in transmision; rather, birds as are considered as a homogenous clas of reservoir hosts. Interspecific variation among avian hosts in contact rates with vector species is ignored despite evidence for heterogeneities among species in contact rate (Hasan et al. 2003; Molaei et al. 2006; Estep et al. 58! in pres) and in reservoir competence (Komar et al. 1999). Thus, studies of EEV would benefit from a more thorough understanding of contact rates betwen vectors and paserine species and heterogeneity amongst those species in reservoir competence. To understand heterogeneity in contact rates betwen mosquitoes and birds researchers need to know the sources of blood meals from field-collected mosquitoes and the relative abundances of birds in the area. With such data, the researcher can then compare the rate of utilization of avian hosts (relative abundance in the blood meal sample) to their rates of availability (relative abundance in avian community). More formaly, a forage ratio (Hes 1968) can be calculated as the ratio of the proportion of blood meals from the hosts to the proportion of the avian community comprised of that species (Manly et al. 2002). Collection of data necesary to acurately estimate forage ratios for al individual species poses a logistical chalenge. Amongst other chalenges, the rarer or les-preferred a species is, the fewer wil be the blood-engorged mosquitoes that have fed on it. and the lower wil be the acuracy of the estimate of its forage ratio. In the extreme case, forage ratios for species that are not present in the study area, but that may be important in transmision in nearby areas, cannot be estimated. Thus, barring intensive sampling over sites spread across a broad geographic area, we currently lack the means to estimate the relative contact rates betwen vectors of EEV and al species of avian hosts potentialy involved in its transmision. One possible, but previously unexplored, solution to this problem is the development of a statistical model of forage ratios based on characteristics of the host species. Estimates of characteristics for North American birds, including life-history traits and those relating to habitat utilization are readily available from the ornithological literature. As such, models of forage ratios based on characteristics of avian host species could potentialy be applied to almost any 59! species to estimate its contact rate with vector species without the need for collection and analysis of blood-engorged mosquitoes and avian community surveys. In the current study, we explored the development of such models. We sought to identify those avian host characteristics for which data are available that are asociated with forage ratios of two EEV vector species that fed regularly on avian hosts in the southeastern United States: Cs. melanura and Cx. eraticus. Culiseta melanura is widely recognized as the primary enzootic vector of EEV (Scott and Weaver 1989). Cx. eraticus has been proposed to play a role as a bridge vector in the southeastern United States (Cupp et al. 2003). As such, insights gained from modeling avian characteristics and feding preferences of these two mosquito species could potentialy be used for predictive model development of EEV transmision in this region. Materials and Methods Inferential Approach and Data Sources We used a multi-model inferential approach (Burnham and Anderson 2002) to identifying host characteristics asociated with forage ratios for each of two mosquito vector species, Cs. melanura and Cx. eraticus. We first developed a set of candidate general linear models to predict the forage ratios of avian host species using characteristics of those species. To determine whether a host characteristic would be useful for predicting forage ratios, we then used model-averaged estimates of the coeficients of the predictor variables and 85% confidence intervals around those estimates (Arnold 2010). We used unconditional weighted standard errors in calculating confidence intervals, such that predictor variables with confidence intervals that excluded zero were concluded to be useful for prediction (Arnold 2010). The strength of evidence for asociations betwen predictor variables and forage ratios was also considered 60! through examination of importance weights asociated with each predictor variable (Burnham and Anderson 2002). Forage ratios for each avian species used in model development were estimated from a long-term study of feding paterns of mosquitoes in Tuskegee National Forest in Alabama (Estep et al. in pres). A high proportion of bird species that were recorded during censuses were not detected in any blood meal samples and inclusion of such zero-valued forage ratios would have caused significant violations of linear regresion modeling. Thus, we restricted our analysis to those avian species with forage ratios > 0, and an asumption of this analysis is that the species that we included were representative of the entire avian community. Such species could either have high contact rates with mosquitoes and low relative abundances, or they could have low contact rates with mosquitoes and high relative abundances. Also, we excluded barred owl, a species with a high Cs. melanura and Cx. erraticus forage ratio estimates, from our analysis. Barred owls are nocturnal sit-and-wait predators, and it sems plausible that they have high forage ratios because they are highly susceptible mosquito hematophagy because of their foraging technique, much like some wading wading birds (Edman and Kale 1971, Edman et al. 1984), rather than due to the factors considered in our analysis. In total, we used 14 observed forage ratios for development of the Cs. melanura model, and 21 observed forage ratios for development of the Cx. eraticus model (Estep et al. in pres, Appendix I). We identified seven host characteristics that could influence the contact rate betwen mosquitoes and avian hosts (Table 1). We based this set of predictor variables partly on their potential asociation with host atractivenes, defensive behavior, or probability of encounter with mosquitoes, and partly on the availability of information in the ornithological literature. We imputed the number of broods per season for brown-headed cowbird (Molothrus ater), a brood 61! parasite with a wide range of hosts, using the average brood size across al avian species detected during point count surveys in TNF. The beginning and end of the host-seking seasons of Cs.melanura and Cx. eraticus were estimated as the quarter-months during which the 2.5th and 97.5 th percentiles of the dates of capture of engorged individuals of each species were captured (Table 1). The beginning and end of the host-seking season were the first quarter of May and the second quarter of August, respectively, for Cs. melanura, and the first quarter of April and the third quarter of September, respectively for Cx. eraticus. Models that included al possible predictor variable combinations comprised the candidate model set for each species. The communal roosting variable was not included in any models in the Cs. melanura candidate set because no species for which forage ratio estimates were available roosted communaly except in winter. We specified models using the lm function in R and used a weighted least squares approach to parameter estimation (R Core Development Team 2008, Kutner et al. 2005). Specificaly, we weighted each estimated forage ratio for an avian species by the inverse of the variance of the forage ratio estimate (Estep et al. in pres). As such, species with smaler variances about forage ratio estimates received a higher weight of influence in parameter estimation than those with large variances (Kutner et al. 2005). Based on initial model diagnostics, we natural-log transformed the forage ratio variable (response) and the body mas variable (predictor). Variance Inflation Factors were < 10 for predictor variables in al models considered, such that multicollinearity was unlikely to bias parameter estimates (Kutner et al. 2005). We compared the bias-corrected AIC (AICc) betwen al models in the candidate set for each species and used this criterion in calculating importance weights of models. We performed 62! model averaging of parameter estimates over al models in the 95% confidence set, i.e. the most highly-ranked models that together comprise 95% of the total AIC weights of al models in the candidates set (Burnham and Anderson 2002). Importance weights for predictor variables in the Cs. melanura or Cx. eraticus forage ratio models were calculated using al models in the candidate sets for either species Cs. melanura or Cx. eraticus. As such, al predictor variables had an equal probability of inclusion in models in the set used for calculating importance weights. Results Cx. eraticus model The top-ranked model in the Cx. eraticus candidate set included stratum, body mas, nestling, and habitat-edge asociation as predictor variables and had an importance weight of 0.31 (Table 2). Another model in the Cx. eraticus candidate set was competitive with the top- ranked model, having an AICc value that was within two units of the AICc of the top-ranked model and an importance weight of 0.26; this second-ranked model included the four covariates in the top-ranked model, and additionaly, availability (Table 2). Al models other than the two top-ranked ones had litle support as the true model of Cx. eraticus forage ratios, given large diferences in AICc values from the top-ranked models and low importance weights (Table 2, Burnham and Anderson 2002). Mid-story and canopy-nesting avian host species had higher Cx. eraticus forage ratios than those nesting in lower strata (? s = 1.95, LCL= 0.72, UCL = 3.18; Figure 1; Table 3), and stratum had the highest important weight amongst al predictor variable considered for the Cx. eraticus models (0.91; Table 3). The direction of asociation betwen Cx. eraticus forage ratios and body mas was positive (? m = 0.88, LCL= 0.18, UCL = 1.01; Figure 2; Table 3), and 63! the body mas variable had an importance weight of 0.88. The direction of asociation betwen Cx. eraticus forage ratios and nestling, the variable with the third-highest importance weight of 0.83, was negative (? n = -0.07, LCL= -0.12, UCL = -0.01; Figure 3; Table 3). Model-averaged coeficient estimates and asociated 85% unconditional confidence intervals of these three top- ranked variables, stratum, body mas, and nestling, suggest that they al could al be used to develop predictive models of the Cx. eraticus forage ratios (Table 3). Habitat-edge asociation and availability, the two variables included in top-ranked models but for which 85% confidence intervals included zero, had coeficient estimates that were positive (? e = 0.98, LCL= -0.06, UCL = 2.02, importance weight = 0.75) and negative (? a = -0.27, LCL= -0.88, UCL = 0.33, importance weight = 0.38), respectively (Table 3). The two other predictor variables considered in modeling Cx. eraticus forage ratios, cavity and roost, had relatively low importance weights (! 0.16) (Table 3). Cs. melanura model Nestling was the only predictor variable included in the most highly ranked Cs. melanura forage ratio model (importance weight = 0.31, Table 4). Al other models in the Cs. melanura candidate set had AICc values that were either greater than two units of the AICc of the top- ranked model, or within two units of the AICc of the top-ranked models but with litle diference in the log-likelihood from the top-ranked model (Table 4). As such, al models other than the top-ranked one had litle support as the true model of Cs. melanura forage ratios (Burnham and Anderson 2002, Arnold 2010). The relationship betwen Cs. melanura forage ratios and nestling was negative (? a = - 0.05, LCL= -0.09, UCL = -0.01), such that species with shorter total periods of availability of nestlings had higher forage ratios (Figure 3). Moreover, the exclusion of zero from its 85% 64! confidence interval suggests that this variable may be useful for predicting Cs. melanura forage ratios. The 85% unconditional confidence intervals for al other variables except the nestling considered in model development included zero (Table 5). The importance weight asociated with the nestling variable was 0.83, while the importance weights asociated with al other variables were relatively low (! 0.30, Table 5). Directions of asociations betwen forage ratios and host trait variables were consistent with those observed for the Cx. eraticus forage ratio models, except in the case of the habitat-edge asociation and availability variables. Discussion ! Our observations offer the first evidence that the relative rate of contacts betwen avian hosts and mosquitoes can be predicted from ecological variables. Asociations betwen susceptibility to hematophagy and host characteristics are to be expected because the ecology of both the vector and host wil determine whether a mosquito succesfully acquires a blood meal from a particular bird species (LoGiudice et al. 2003). The implications of demonstrating specific ecological variables that predict the relative rates of contact betwen avian hosts and both Cs. melanura and Cx. eraticus, however, are significant, because forage ratios could be used in elucidating paterns of the occurrence of EEV in the southeastern United States. Our succes at predicting host-vector interactions from environmental variables suggests that pathogen transmision might also be predictable from relatively simple ecological variables. Mosquitoes often exhibit vertical niche partitioning within habitats (Snow 1955, Swanson et al. 2010), and we expected mosquitoes to encounter bird species that nest within these individual vertical niches more often than species that nest in strata outside of the vector niche. In other words, we expected birds whose vertical niches overlapped those of vector mosquito 65! species to have inflated forage ratios, such that nesting stratum would be useful in predictive model development. Consistent with this prediction, our results show that nesting stratum can be used to predict Cx. eraticus forage ratios of avian hosts. Nesting stratum also received moderate support as an influential variable in the Cs. melanura forge ratio models, whereby species nesting in higher strata have higher forage ratios. Although the weight of evidence for an influence of nesting stratum of Cs. melanura forage ratios was weak in the current study, the direction of the trend suggests avian hosts species that nest higher up in the forest canopy may be at a greater risk of exposure to both Cx. eraticus and Cs. melanura hematophagy, and by extension, EEV. Moreover, our results suggest that the weight of the influence of nesting stratum on forage ratios for avian hosts is diferent for Cs. melanura and Cx. eraticus, as would be expected given diferences in vertical niche partitioning betwen these mosquito species. Variability in vertical distribution of mosquitoes amongst habitats types precluded formation of specific predictions about the direction of asociation betwen nesting stratum and forage ratios for avian host species. For example, studies of the vertical distribution of Cs. melanura demonstrate a high degree of inter-site variability, with either no asociation betwen abundance and trap height detected, or the direction of the detected asociation dependent upon habitat type (reviewed in Nasci and Edman 1981). In a study in South Carolina with traps suspended at heights of 1.5, 5, and 10 meters proportions of mosquito samples comprised of Cx. eraticus increased substantialy with lowered trap height (Swanson et al. 2010). However, results of such studies were based on captures of al mosquitoes, not necesarily the subset of females actively seking hosts. In a study of host-seking Cs. melanura, results were much more clear-cut, wherein traps baited with birds suspended farther from the forest floor atracted more mosquitoes than lower traps, corroborating the direction of the asociation betwen stratum 66! and forage ratio observed in our study (Nasci and Edman 1981). Moreover, a positive asociation betwen average hematoparasite load of avian species and nesting stratum was found in surveys of wild-caught birds in Louisiana (Garvin et al. 1997). The explanation for this asociation was that certain ornithophilic vectors are more common in forest canopies than at lower levels, so species nesting high are exposed to hematophagy thus blood-borne parasites. The positive asociation betwen forage ratio and the body mas of birds in both the Cx. eraticus and Cs. melanura models was not surprising given that birds with larger body mases have greater rates of output of carbon dioxide (Grubb 1983). Carbon dioxide is one of the primary biochemical atractants for mosquitoes (reviewed in Nicolas and Silans 1989). Thus, larger bird would be expected to atract questing mosquitoes at a higher rate compared to smaler birds. This basic asumption was supported by a recent study in which the body mas of birds mosquito traps was positively asociated with the number of mosquitoes captured at the trap (Suom et al. 2010). However, birds with greater body mases also occupy a greater volume of space, so simply by Brownian motion, mosquitoes are more likely to encounter larger birds. Larger birds may also have higher tolerances to hematophagy (Edman and Scott 1987). Any one of these factors might explain the asociation betwen forage ratios and body mas observed in the current study. The total number of days that nestlings are available, calculated as product of the average number of broods and average nestling stage length, was the variable that received the highest support in the Cs. melanura models. Moreover, this variable is likely useful for predicting Cs. melanura forage ratios of avian host species. It also received strong support as a variable useful for prediction in the Cx. eraticus models. The direction of this asociation was negative, however? the opposite direction predicted by the hypothesis that nestlings may be particularly 67! vulnerable to hematophagy by mosquitoes (Blackmore et al. 1958; Kale et al 1952; Grifing et al. 2007; Burket-Cadena et al. 2010). Reasons for the negative direction of asociation are unclear. One possibility is that the result is confounded with average nestling length of individuals. Average nestling stage length and the total availability of nestlings were highly correlated (species used for Cs. melanura forage ratio models: r S (14) = 0.73, p < 0.01; species used for Cx. eraticus forage ratio models: r S (21) = 0.85 p < 0.01). Average nestling stage length is one of the life-history characteristics asociated with average nest predation rate, whereby species at a high risk of nest predation tend to have shorter nestling stage lengths (Martin 1995). As such, a negative asociation betwen average nestling and forage ratios would support the inference of predation risk influencing the defensive behavior of avian species, and thus their contact rates with mosquitoes. Alternatively, in species with shorter nestling stages, the risk of hematophagy to recently fledged birds may be particularly high, because such young birds are independent and receive no protection from patents (Burket-Cadena et al. 2010). If recently fledged birds are preferentialy targeted by questing mosquitoes for bloodmeals, as has been suggested previously (Loss et al. 2009), then those species with short nestling stages would be likely to be fed upon by mosquitoes at a high rate. Habitat is an obvious environmental factor that could link vectors and hosts, so it was not surprising that an asociation with habitat edge was included as a predictor variable in the top- ranking Cx. eraticus forage ratio model. We had predicted that an asociation with habitat-edge would relate to forage ratio positively in the Cx. eraticus model and negatively in the Cs. melanura model. The basis for these predictions was a previous study of mosquito microhabitat asociations (Bidlingmayer 1971), which showed diferent strengths of habitat-edge asociation in wooded swamp habitats betwen Cs. melanura and species in the Culex (Melaconion) 68! subgenus, of which Cx. eraticus is a member (Darsie and Ward 2005). Specificaly, the average number of Cs. melanura captured per trap night was higher in traps located in swamps compared to traps located at swamp edges, field edges, or in fields. As such, Bidlingmayer (1971) suggested that Cs. melanura is a swamp interior species that avoids swamp edges and the higher levels of ilumination asociated with them. Conversely, the highest average trap counts of Culex (Melanoconion) species were at field edges, followed by traps in fields, at swamp edges, or in swamps (Bidlingmayer 1971). We predicted that if there were similar microhabitat asociations of Cs. melanura and Cx. eraticus in the wetland habitat of TNF, then Cs. melanura would have a higher rate of encounter with swamp interior species, resulting in an inflated forage ratio for those species. In contrast, Cx. eraticus was expected to encounter swamp- and field- edge avian species at a higher rate than swamp interior species, such that the forage ratio of habitat-edge asociated avian species would be inflated. Our results confirmed our predictions, as the direction of influence of habitat-edge asociation of forage ratios was positive in the Cx. eraticus model and negative in the Cs. melanura model. Overal, however, the influence of this variable may be relatively smal compared to other host trait variables and thus may not be useful for prediction. Duration of availability of an avian species during the breeding season of Cs. melanura or Cx. eraticus had relatively low importance weights as a variable useful for predicting forage ratio models for either species. While this variable was included in the second-highest ranking model in the Cx. eraticus model set, the 85% confidence intervals included zero for the estimates of the coeficient of this variable, and the direction of asociation was in the opposite direction as predicted. Moreover, the direction of asociation of this variable with forage ratios was not consistent betwen mosquito species models. As such, whether or not an avian host 69! species was present at the TNF site for the entire mosquito breeding season appears to have had minimal influence on their forage ratio estimates for either mosquito species compared to the influence of other host traits useful for predicting the relative rates of contact betwen birds and mosquitoes described in this study. References Anderson D.R., K.P. Burnham, and W. L. Thompson. 2000. Null hypothesis testing: Problems, prevalence, and an alternative. J Wildlife Manage. 64: 912-923. Andreadis T.G., and P. M.Armstrong. 2007. A two-year evaluation of elevated canopy trapping for Culex mosquitoes and West Nile virus in an operational surveilance program in the northeastern United States. J Am Mosquito Contr. 23: 137-148. Arnold T.W. 2010. Uninformative Parameters and Model Selection Using Akaike's Information Criterion. J Wildlife Manage. 74: 1175-1178. Bender, D.J., T.A. Contreras, and L. Fahrig. 1998. Habitat loss and population decline: A meta- analysis of the patch size efect. Ecology. 79: 517-533. Boakye D.A., J. Tang , P. Truc, A. Merriweather, T. R. Unnasch. 1999. Identification of bloodmeals in haematophagous Diptera by cytochrome B heteroduplex analysis. Med Vet Entomol. 13: 282-287. Bidlingmayer W.L. 1971. Mosquito flight Paths in Relation to the Environment. I. Illumination Levels, Orientation, and Resting Areas. Annals of the Entomological Society of America. 64: 1121-1131. 70! Blackmore J.S., and R.P. Dow. 1958. Diferential feding of Culex tarsalis on nesting and adult birds. Mosq News. 18:15-17. Burket-Cadena N.D., S.P. Graham, H.K. Hasan, C. Guyer, M.D. Eubanks, C.R. Katholi, and T.R. Unnasch. 2008. Blood Feding Paterns of Potential Arbovirus Vectors of the Genus Culex Targeting Ectothermic Hosts. Am J Trop Med Hyg. 79: 809- 815. Burket-Cadena N.D., R.A. Ligon, M. Liu, H.K. Hasan, G.E. Hil, M.D. Eubanks, and T.R. Unnasch. 2010. Vector-Host Interactions in Avian Nests: Do Mosquitoes Prefer Nestlings over Adults? Am J Trop Med Hyg. 83: 395-399. Burnham K.P. and D.R. Anderson. 2002. Model Selection and Multi-Model Inference: A Practical Information-Theoretic Approach. Springer-Verlag, New York. Cupp E.W., K. Klingler, H.K. Hasan, L.M. Vlguers, and T.R. Unnasch. 2003. Transmision of eastern equine encephalomyelitis virus in Central Alabama. Am J Trop Med Hyg. 68: 495- 500. Darsie R.F., and R.A. Ward. Identification and Geographical Distribution of the Mosquitoes of North America, North of Mexico. University Pres of Florida, Gainesvile, Florida. Dunning J.B. Jr., 2008. CRC Handbook of avian body mases, 2 nd edition. CRC Pres/Taylor and Francis Group, Boca Raton, Florida. Edman J.D., Day, J.F., and E.D. Walker. 1984. Field confirmation of laboratory observations on the diferential antimosquit behavior of herons. Condor 86: 91-92. Edman J.D., and H.K. Kale II. 1971. Host behavior: Its influence on the feding succes of mosquitoes. Ann Entomol Soc Am 64: 513-516. Estep, L.K., C. J. W. McClure, N.D. Burket-Cadena, H. K. Hasan, T.L. Hicks, T. R. 71! Unnasch, and G.E. Hil. A multi-year study of mosquito feding paterns on avian hosts in a southeastern focus of eastern equine encephalitis virus. Am J Trop Med Hyg. in pres. Garvin M.C., and J.V. 1997. An alternative hypothesis for heavier parasite loads of brightly colored birds: Exposure at the nest. Auk. 114: 179-191. Gough, G.A., J.R. Sauer, and M. Ilif. 1998. Patuxent Bird Identification Infocenter. Version 97.1. Patuxent Wildlife Research Center, Laurel, MD. http:/ww.mbr- pwrc.usgs.gov/id/framlst/infocenter.html Grifing S.M., A.M. Kilpatrick, L. Clark, and P.P. Marra. 2007. Mosquito landing rates on nesting American robins (Turdus migratorius). Vector-borne Zoonot. 7: 437-443. Grubb B.R. 1983. Alometric relations of cardiovascular function in birds. Am J Physiol. 245: H567-H572. Gubler D.J., P. Reiter, K.L. Ebi, W. Yap, R. Nasci R, and J.A. Patz. 2001. Climate variability and change in the United States: Potential impacts on vector- and rodent-borne diseases. Environ Health Persp. 109: 223-233. Hasan H.K., E.W. Cupp, G.E. Hil, C.R. Katholi, K. Klingler, and T.R. Unnasch. 2003. Avian host preference by vectors of eastern equine encephalomyelitis virus. Am J Trop Med Hyg. 69: 641-647. Hes A.D., R.O. Hayes, and C.H. Tempelis. 1968. The use of the foraging ratio technique in mosquito host preference studies. Mosq News. 28: 386-398. Kent, R., L. Juliusson, M. Weismann, S. Evans, and N. Komar. 2009. J Med Entomol. 46: 380- 72! 390. Kilpatrick A.M., P. Daszak, M.J. Jones, P.P. Marra, L.D. Kramer. 2006. Host heterogeneity dominates West Nile virus transmision. P R Soc B. 273: 2327-2333. Komar N., D.J. Dohm, M.J.Turel, A. Spielman. 1999. Eastern equine encephalitis virus in birds: Relative competence of European starlings (Sturnus vulgaris). Am J Trop Med Hyg. 60: 387-391. Kutner, M.H., C.J. Nachtsheim, J. Neter, and W. Li. 2005. Applied linear statistical models, Fifth edition. McGram-Hil/Irwin, New York. LoGiudice K, Ostfeld RS, Schmidt KA, and F. Kesing. 2003. The ecology of infectious disease: Efects of host diversity and community composition on Lyme disease risk. P Natl Acad Sci Usa. 100: 567-571. Loss, S.R., G.L. Hamer, T.L. Goldberg, M.O. Ruiz, U.D. Kitron, E.D. Walker, and J.D. Brawn. 2009. Neslting paserines are not important hosts for amplification of West Nile virus in Chicago, Illinois. Vector-borne Zoonot 9: 13-17. Manly B.F., L. McDonald, D.L. Thomas, T.L. McDonald, and W. P. Erickson. 2002. Resource Selection by Animals: Statistical Design and Analysis for Field Studies. Springer, Dordrecht, The Netherlands. Marquardt W.H. (ed). 2004. Biology of Disease Vectors, Second Edition. Elsevier Academic Pres, Burlington, Masachusets. Martin, T.E. 1995. Avian life history evolution in relation to nest sites, nest predation, and food. 73! Ecol Monogr 65: 101-127. McIntyre N.E. 1995. Efects of forest patch size on avian diversity. Landscape Ecol. 10: 85-99. Molaei G., T. G. Andreadis, P.M. Armstrong, J.F. Anderson, and C.R. Vossbrinck. 2006. Host feding paterns of Culex mosquitoes and West Nile virus transmision, northeastern United States. Emerg Infect Dis. 12: 468-474. Nasci R.S., and J.D. Edman. 1981. Vertical and temporal flight activity of the mosquito Culiseta melanura (Diptera, Culicidae) in southeastern Masachusets. J Med Entomol. 18: 501-504. Nicolas G., and D. Silans. 1989. Immediate and latent efects of carbon-dioxide on insects. Annu Rev Entomol. 34: 97-116. Poole, A.. (ed) 2005. The Birds of North America Online: http:/bna.birds.cornel.edu/BNA/. Cornel Laboratory of Ornithology, Ithaca, NY. R Development Core Team (2008). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3- 900051-07-0, URL: http:/ww.R-project.org. Scott, T.W., and S.C. Weaver. 1989. Eastern equine encephalomyelitis virus: epidemiology and evolution of mosquito transmision. Adv Vir Res. 37: 277-328. Snow W.E. 1955. Feding activities of some blood-sucking diptera with reference to vertical distribution in bottomland forest. Ann Entomol Soc Am. 48: 512-521. Suom C., H.S. Ginsberg, A. Bernick, C. Klein, P.A. Buckley, C. Salvatore, R. A. LeBrun. 2010. Host-seking activity and avian host preferences of mosquitoes asociated with West Nile virus transmision in the northeastern USA. J Vector Ecol. 35: 69- 74! 74. Swanson D.A., and P.H. Adler 2010. Vertical distribution of haematophagous Diptera in temperate forests of the southeastern USA. Med Vet Entomol. 24: 182-188. Vilari P., A. Spielman, N. Komar, M. McDowel, R.J. Timperi. 1995. The economic burden imposed by a residual case of eastern encephalitis. Am J Trop Med Hyg. 52: 8-13. Whitcomb, R.F., C.S. Robbins, J.F. Lynch, B.L. Whitcomb, M.K. Klimkiewicz, and D. Bystrak. 1981. Efects of forest fragmentation on avifauna of the eastern deciduous forest. In: Burges, R.L. and D.M. Sharpe (eds) Forest dynamics in man-dominated landscapes. Springer-Verlag, New York. 75! Table 1. Predictor variables used in for Cs. melanura forage ratio and Cx. eraticus forage ratio candidate models. 76! Table 2. Bias-corrected AIC (AICc) table for Cx. eraticus forage ratio models with moderate or strong support. AICc is the bias-corrected Akaike Information Criterion, K is the number of parameters estimated, " i is the diference in AICc from the model that minimized the AICc, and w i is the importance weight. Al models with " i <7 are represented. 77! Table 3. Results of model averaging for Cx. eraticus forage ratio models and importance weights of predictor variables (weight). MAE is the model-averaged estimate of the variable coeficients and SE is the standard error of that estimate. 78! Table 4. Bias-corrected AIC (AICc) table for Cs. melanura forage ratio models with moderate or strong support. AICc is the bias-corrected Akaike Information Criterion, K is the number of parameters estimated, " i is the diference in AICc from the model that minimized the AICc, and w i is the importance weight. Al models with " i <7 are represented. ! ! ! ! ! ! 79! Table 5. Results of model averaging for Cs. melanura forage ratio models and importance weights of predictor variables (weight). MAE is the model-averaged estimate of the variable coeficients and SE is the standard error of that estimate.! 80! Figure 1. Cs. melanura(a) and Cx. eraticus (b) forage ratios of avian host species nesting either at ground/low-levels or in the mid-story/canopy. Circle sizes are proportional to observation weights. 81! Figure 2. Scaterplot of Cs. melanura(a) and Cx. eraticus (b) forage ratios for avian host species versus body mas, with best-fit line from simple linear regresions overlain. Circle sizes are proportional to observation weights. 82! Figure 3. Scaterplot of Cs. melanura(a) and Cx. eraticus (b) forage ratios for avian host species versus nestling stage length, with best-fit line from simple linear regresions overlain. Circle sizes are proportional to observation weights. 83! Appendix I. Forage ratios and host-trait data used in model development. ! ! ! ! ! ! ! ! ! ! ! 84! IV. USING ATRIBUTES OF AVIAN COMUNITIES TO PREDICT LOCAL ENZOTIC TRANSMISSION OF EASTERN EQUINE ENCEPHALITIS VIRUS Abstract Because eastern equine encephalitis virus (EEV) is a dangerous pathogen to both humans and horses, there is urgent need to develop models to predict paterns of transmision on a local spatial scale. Here, we develop such a predictive model of EEV transmision using data on the avian host community. We used data on EEV seroconversion collected during 2009 from chicken sentinel flocks distributed across Walton County, Florida to develop a logistic regresion model of the log-odds of a site having a high versus low rate of EEV transmision. Covariates considered in model development included linear and quadratic efects for avian community size and the relative abundances of avian host species regarded as preferred hosts of Culiseta melanura. The weights of variables asociated with relative abundances of focal avian species exhibited a perfect rank-correlation with estimates of how preferred the bird species were as hosts of Cs. melanura. While preliminary, these results suggest that presence of hosts that are preferred by Cs. melanura and overal reservoir competence for EEV are factors that figure prominently into local-scale EEV transmision. Introduction Eastern equine encephalitis virus (EEV) is a highly virulent pathogen with a complex life cycle, involving both a mosquito vector and an avian reservoir host (Morris 1988, Scott and 85! Weaver 1989). Despite high mortality rates (60-80%), human cases of EE are quite rare, relative to other arboviruses that occur in the United States (Scott and Weaver, 1989, Vilari et al. 1995). An important consequence of such rarity is that human and horse infections are dificult to predict (Letson et al. 1993). Given the severity of cases of EE in both humans and horses there is an urgent need for the development of models that can predict where and when EEV wil occur. Despite stil incomplete knowledge of the factors that determine occurrence of the virus on a regional scale, progres has been made in the development of predictive models of the temporal occurrence of EEV at known transmision locales. Specificaly, elevations in the wekly minimum infection rates (MIRs) and in the number of infected primary enzootic vectors (Culiseta melanura) have been shown to be asociated with the number of cases of human eastern equine encephalitis virus in Masachusets (Hachiya et al. 2007). MIR and detection of positive Cs. melanura may be useful for developing an Early Warning System for EEV, but these variables have not been used to predict where transmision wil occur. While the use of vector MIR?s may be useful in forecasting the timing of EEV transmision, acurate prediction of the spatial distribution wil likely depend on atributes not just of vector populations but also of avian communities. Birds are the primary reservoir hosts of EEV, and thus are important to EEV transmision (Scott and Weaver 1989). Individuals of diferent avian host species vary in their rates of contact with mosquitoes, as estimated by forage ratios (Hasan et al. 2003, Estep et al. in pres). They also difer in their capacities to replicate the virus following exposure (reservoir competency) and thus in their ability to infect subsequent mosquitoes that fed on them (Komar et al. 1999). Such heterogeneity across hosts with respect to reservoir competences and relative contact rates with vectors can result in variability amongst 86! sites in rates of EEEV transmision when hosts are have a variable distribution across a landscape. One means to describe virus transmision is to use a model based on frequency- dependence. Under frequency-dependent transmision, the rate of transmision is directly proportional to I/N, the proportion of the host community comprised of infective individuals, where I = number of infective hosts and N = total number of hosts in the community (Anderson and May 1979). The proportion, or frequency, of infectives in the host population is just an estimate of the probability that a given host that a vector contacts is infective when al hosts have the same probability of contact with a vector. Asuming that it is this probability of a given host that a vector contacts being infective, rather than the proportion of infectives in the host community per se, that drives frequency-dependent transmision, the rate of transmision wil be most heavily influenced by those species that have high forage rations or extreme reservoir competences (Kilpatrick et al. 2006). The mode of virus transmision wil vary betwen vector-borne pathogen systems, depending on the behavior of the vector (Antonovics et al. 1995). At one extreme where vectors have short search times for hosts, the rate of transmision wil be density-dependent; in other words, it wil be proportional to the number of infectives in the host population, rather than the frequency of infectives, as in frequency-dependence. Under density-dependence, the rate of transmision is expected to increase with total size of the host community for a given frequency of infectives (Antonovics et al. 1995). As such, the rate of transmision should increase with total host community size for a given probability of a host that a vector contacts being infective. At the other extreme of vector behavior where vectors have long search times for hosts, the rate of transmision wil be proportional to I/N 2 . In this case, the rate of transmision is 87! expected to increase initialy with increasing host community size. It then plateaus and thereafter exhibits ?inverse density dependence?, such that the rate of transmision decreases with increasing host community size for larger communities (Antonovics et al. 1995, Antonovics and Alexander 1992, de Castro and Bolker 2005). Given that avian species are heterogeneous for the rate of vector contact and reservoir competency, it sems plausible that composition of the local avian community could influence the transmision of EEEV across locations. More specificaly, we expect to find support in spatialy-explicit models of EEV transmision for the inclusion of variables representing the relative abundances of species that are highly influential on the probability of a host that a vector contacts being infective, i.e. those species that have high forage ratios and extreme reservoir competences. Additionaly, after controlling for the relative abundances of highly influential species, we expect to find support for inclusion of variables of host community size. To date, no models for predicting local EEV transmision have incorporated data on avian communities. In this study, we sought to determine whether models based solely on the composition and size of avian communities would be able to predict the likelihood that mosquito-to-bird transmision of EEV would occur. We made the simplistic asumption that Cs. melanura is the primary enzootic vector of the virus and that birds are the only reservoir hosts involved in its transmision. We tested this hypothesis using data derived from seroconversion of sentinel chickens from arbovirus surveilance sites, avian point-counts, and estimates of the relative rates of contact of avian host species with Cs. melanura (Estep et al. in pres). 88! Materials and Methods Data Sources Data on the enzootic transmision of EEEV used in model development originated from Walton County, FL mosquito control districts. These two districts collectively monitored 24 sentinel chicken flocks in 2009, located throughout the county as part of a long-term arbovirus surveilance program (Moore 1993, Figure 1). Sentinel flocks were comprised of either 2-3 (South Walton district) or 6 (North Walton district) chickens held in outdoor sentinel cages. Mosquito control personnel collected blood samples from al individuals in each sentinel flock wekly and shipped them to the Florida Department of Health for testing for the presence of EEEV neutralizing antibodies via hemaglutinnin inhibition and serum neutralization asays (Florida Interagency Arboviral Task Force 2006). Individual chickens with evidence of serconversion from an EEEV antibody-negative to positive status were sacrificed and replaced with na?ve individuals following reporting of test results. Observational data used as predictor variables in model development originated from point-count surveys of the avian communities around sentinel sites. We asigned four avian survey sites to each sentinel flock, for a total of 96 survey sites (Figure 3). Individual survey sites in the set of four sites surrounding each flock were asigned to points at regular intervals along the perimeter of a buffer with radius-length 250 meters centered on the flock, such that the 2 line segments connecting survey site on opposites sides of the flock intersected and formed right angles at the flock locations. Buffers were rotated by a random angle betwen 0 and 90 degrees, such that the surveys site locations varied amongst flocks but were consistently equidistant from each other within a flock. 89! Avian survey sites were visited once during June of 2010 by a single observer trained in the vocal and visual identification of avian species that breed in southeastern United States (Bibby et al. 1992). Surveys were conducted betwen 0500 and 1000 EDT. Each visit was divided into five three-minute sesions during which the identity and location of al birds sen or heard were recorded (Mackenzie and Royle 2005). From these data, the average number of individuals of each species at each avian survey site was calculated. The abundance of each species at each sentinel flock was estimated as the mean of its averaged abundance at each of its four surrounding survey sites. Analytical Approach Our analytical approach focused on developing a model of the probability of a site having a high rate EEV transmision. Sentinel sites where at least one chicken seroconverted in 2009 had higher rates of infection of the susceptible hosts (sentinel chickens) than sites where there were zero seroconversions. By extension, sites where at least one chicken seroconverted had higher rates of EEV transmision than sites where al chickens remained na?ve, as the rate of transmision of a pathogen is directly proportional to the rate at which susceptible hosts becoming infected (Anderson and May 1991). We thus designated sites as either having a low- or high- rate of EEV transmision based on whether or not at least one chicken seroconverted there. We used the indicator variable for a site having versus low rate of transmision as the response variable in logistic regresion models (1=high, 0=low). We used a multi-model inference approach to our analysis (Anderson et al. 2000, Burnham and Anderson 2002). We specified a set of thirten candidate models. 90! The first two models in the candidate set described the log-odds of the sentinel site having a high rate of EEV transmision as constant (model for the mean; intercept only and no covariates) or dependent on a single covariate, avian community size (total number of birds, averaged across the four point count survey areas surrounding the sentinel site). Each of the next five models in the candidate set were elaborations of the second model that had avian community size as the single predictor variable, whereby the log-odds of the response was modeled as a linear function of avian community size plus the relative abundance of one of five focal avian species. Focal avian species were those that satisfied two requirements (1) avian species inferred to be preferred hosts of Cs. melanura, the primary enzootic vector of EEV, based on forage ratios>1 from published field studies (Estep et al. in pres) (2) species that were observed during point count surveys around sentinel flocks in 2010. One species inferred to be a preferred host of Cs. melanura, yelow-throated vireo, was not considered in model development because it was present at point count locations surrounding only one sentinel site. As such, models three through seven included the relative abundance of either blue-gray gnatcatcher, common yelowthroat, eastern towhee, northern cardinal, or red-eyed vireo, as these species were present at point count location surrounding at least two sentinel sites and were found to be preferred hosts of Cs. melanura (Estep et al. in pres)(Table 1). The eighth model specified the efect of avian community size as quadratic, such that it included the linear and quadratic terms for avian community size to keep the model hierarchicaly correct. Following the same protocol used for the model that included only a linear efect of avian community size, the last five models were constructed as elaborations as the eighth model. They included the linear and quadratic efects for avian community size and, 91! individualy, the relative abundance of either blue-gray gnatcatcher, common yelowthroat, eastern towhee, northern cardinal, or red-eyed vireo. Models were specified using the glm function in R software using the binomial error distribution (R Development Core Team 2008). The avian community size variable was natural log-transformed, and relative abundance variables were arcsine-square-root-transformed. Al variables were centered about their means. Model averaging was performed over the 95% confidence set of models, i.e. those models that had the greatest model weights and that together comprised 95% of the model weight in the final candidate set (Burnham and Anderson 2002). We used 85% confidence intervals about parameter estimate as a basis for inferring whether or not a variable was potentialy useful for prediction (Arnold et al. 2010). QAICc values were used in calculating importance weights for variable weights in models in the final candidate set (Burnham and Anderson 2002). Results A total of 67 chickens distributed over 15 of the 24 sites monitored seroconverted from a status of naive to a status of positive for EEv antibodies in 2009, such that 15 sites were designated as having a high rate of EEV transmision in 2009 and 9 were clasified as having a low rate. The average number of chickens that seroconverted over the entire season, calculated strictly over the 15 sites with high rates of transmision, was 4.47 (SD = 4.54 , min = 1, max = 17). High transmision-rate sites occurred throughout Walton County, with 6 of 8 sites clasified as high- rate sites in the North Walton County district (north of the Choctawhatchee Bay), and 9 of 16 sites clasified as high-rate sites in the South Walton County district (south of the Choctawhatchee Bay; Fig 1). 92! The average size of the 24 avian communities across sentinel sites based on these average species abundance estimates was 6.02 individuals (SD = 2.48 , min = 3.25, max = 12.40). Individuals from sixty diferent avian species were detected during point-count surveys. Two of the thirten models in the model set had strong support. The model that included linear and quadratic efects for avian community size was the top-ranking model. The model that included linear and quadratic efects for avian community size and the relative abundance of common yelowthroat ranked second with a "QAICc of 1.26. The weights asociated with these two top-ranking models were 0.36, 0.19, respectively (Table 2). Overal, the six models that included the quadratic efect of avian community size had higher weights than the models that included only a linear efect for avian community size when weights were calculated over the set of models that included al models except for the model that was found to have a poor fit to the data (le Cesie and van Houwelingen, 1991, Table 2). Eighty-five percent confidence intervals of the model-averaged parameter estimate over models in the ninety-five percent confidence set for al covariates considered in model development included zero, except for the quadratic efect of avian population size. (Table 3; Figure 3). While the eighty-five percent confidence interval for the efect of the relative abundance of common yelowthroat on the log-odds of the response included zero, its parameter estimate exhibited a strong negative skew (Table 3; Figure 4). Ranks of importance weights asociated with the variables representing the relative abundances of preferred avian host species of Cs. melanura were perfectly correlated with the ranks of the forage ratios of those species, such that the null hypothesis of zero correlation betwen ranks was rejected (r S =1.00, n=5, p=0.01677; Figure 5). 93! Discussion Understanding the transmision dynamics of EEV and developing a model to predict its occurrence on a local scale is chalenging given the potential involvement of multiple mosquito species and a suite of avian hosts that vary in capacity to serve as a reservoir hosts. Our results reflect the complexity of this vector/host/viral system. The relative abundance of no single avian species was found to be useful for prediction of EEV transmision, but our models suggest that heterogeneities amongst avian species do indeed impact the transmision dynamics of EEV. This insight provides an important step forward in our understanding of the contribution of the avian community to EEV transmision and provides a foundation for future modeling eforts. The top-ranked model in our candidate model set included linear and quadratic terms for avian community size. However, the model that included linear and quadratic efects of avian community size, as wel as the relative abundance of common yelowthroat, had a QAICc value within 2 units of the top-ranked model, suggestive of strong support for this model. Models that are competitive with the top-ranked model based on "QAICc values, but that are simply elaborations of the top-ranked model with one variable added, require that the log-likelihood values of either model be additionaly examined to determine if the more elaborate model has esentialy the same log-likelihood as the top-ranked model. In this case of equivalent log- likelihoods betwen models, the more elaborate model is typicaly considered to be les competitive (Burnham and Anderson 2002, Arnold 2010). In our analysis, however, the second- ranked model with the common yelowthroat variable has a log-likelihood clearly diferent from the top-ranked model such that we infer presence of common yelowthroat has a biological influence on EEV transmision. As such, the model that included linear and quadratic efects for avian community size, as wel a the relative abundance of common yelowthroats, is 94! competitive with the top-ranked model for being the one amongst al models we considered that most closely approximates the true model of EEV transmision. It is interesting and probably not coincidental that the common yelowthroat was the species of bird with the highest forage ratio estimate for Cs. melanura among al avian species in a study conducted 200 km north in Macon County, Alabama, that were detected in Walton County surveys (Estep et al. in pres), as wel as the bird species whose relative abundance was inferred to influence EEV transmision based on our results. The higher the forage ratio estimate of a species, the greater the proportion of Cs. melanura bites we would expected to be directed towards that host and away from other hosts. A species with higher forager ratio estimates should have the greatest influence on the probability that vectors are infective and thus on virus transmision. This influence results simply because individuals of species with high feding index values come into contact more often with Cs. melanura than les-preferred hosts. The common yelowthroat has the highest forage ratio among birds in this region (Estep et al. in pres) and is inferred from our results to have a biological influence on transmision. Perhaps just as importantly, however, the ranks of the importance weights for the relative abundances of the four other species considered in our models correlated perfectly with the ranks of their forage ratios. The importance weight of a variable is the probability that the variable would be in the top-ranking model if the same set of models were run on a diferent dataset (Burnham and Anderson 2002). Overal, the relationship betwen importance in predicting viral transmision and preference as a host for Cs. melanura is strong evidence that heterogeneity amongst avian species in terms of their relative rates of contact has a strong influence on transmision. 95! Our model also suggests that diferences there are diferences among avian species in their competence as virus amplifiers. The direction of asociation betwen the relative abundance of common yelowthroat and transmision was negative, such that there appears to be a lower rate of transmision at sites where common yelowthroats have a greater relative abundance. This patern of negative asociation was also evident for eastern towhees. Such a negative asociation suggests that common yelowthroats and eastern towhees act as dilution hosts (LoGiudice et al. 2003), diverting bites away from more competent avian species and efectively lowering the probability that a vector feds upon an infected bird. In contrast, our model suggests that northern cardinal, red-eyed vireo, and blue-gray gnatcatcher may act to increase the rate of transmision, possibly because they have an above- average reservoir competence. Overal, these results suggest that variation among avian hosts in terms of their relative rates of contact with Cs. melanura in combination with their reservoir competence influence EEV transmision and spatial paterns of variability in rates of transmision. However, it appears that it is the variability amongst avian host species in relative rates of contact with Cs. melanura that is more important in determining the magnitude of their influence on transmision than variability in their reservoir competences. While preliminary, our results provide the strongest evidence to date that heterogeneity amongst avian host species in terms of their rates of contact with Cs. melanura and reservoir competence for EEV influences transmision of the virus. Therefore, any model considers diferential contributions of individual host species to transmision and variability in those contributions across geographic locations due to diferences in avian community composition has the potential to elucidate paterns and to predict of local-scale variation in EEV transmision. 96! References Anderson, D.R., K.P. Burnham, and W. L. Thompson. 2000. Null hypothesis testing: Problems, prevalence, and an alternative. J Wildlife Manage. 64: 912-923. Anderson, R.M., and R.M. May. 1979. Population biology of infectious-diseases.1. Nature. 280: 361-367. Arnold, T.W. 2010. Uninformative Parameters and Model Selection Using Akaike's Information Criterion. J Wildlife Manage. 74: 1175-1178. Antonovics, J., Y. Iwasa, and M.P. Hasel. 1995. A generalized-model of parasitoid,venereal, and vector-based transmision proceses. Am Nat. 145: 661-675. Antonovics, J. and H.M. Alexander. 1992. Epidemiology of anther-smut infection of Silene alba (= S-Latifolia) caused by Ustilago alba ? paterns of spore deposition in experimental population. P Roy Soc Lond B Bio. 250: 157-163. Bibby, C.J., N.D. Burges, and D.A. 1992. Bird Census Techniques. Academic Pres, New York. Burnham, K.P. and D.R. Anderson. 2002. Model Selection and Multi-Model Inference: A Practical Information-Theoretic Approach. Springer-Verlag, New York. de Castro, F., and B. Bolker. 2005. Mechanisms of disease-induced extinction. Ecol Let. 8: 117- 126. Estep, L.K., C.J.W. McClure, N.D. Burket-Cadena, T.R. Unnasch, and G.E. Hil. A multi-year study of mosquito feding paterns on avian hosts in a southeastern focus of eastern equine encephalitis virus. Am J Trop Med Hyg. in pres. 97! Florida Interagency Arbovirus Task Force. (2006). Surveilance and Control of Selected Arthropod-borne Diseases in Florida (3-6 & D-1). Talahase: Florida Department of Health. Hasan, H.K., E.W. Cupp, G.E. Hil, C.R. Katholi, K. Klingler, and T.R. Unnasch. 2003. Avian host preference by vectors of eastern equine encephalomyelitis virus. Am J Trop Med Hyg. 69: 641-647. Hachiya, M., M. Osborne, C. Stinson, and B.G. Werner. 2007. Human eastern equine encephalitis in Masachusets: Predictive indicators from mosquitoes collected at 10 long- term trap sites, 1979-2004. Am J Trop Med Hyg. 76: 285-292. Kilpatrick, A.M., P. Daszak, M.J. Jones, P.P Marra, and L.D. Kramer. 2006. Host heterogeneity dominates West Nile virus transmision. P R Soc B. 273: 2327-2333. le Cesie, S. and H. van Houwelingen. 1991. A goodnes-of-fit test for binary regresion models, based on smoothing methods. Biometrics. 47: 1267-1282. Letson, G.W., R.E. Bailey, J. Pearson, and T.F. Tsai. 1993. Eastern equine encephalitis (EE) ? A description of the 1989 outbreak, recent epidemiological trends, and the asociation of rainfal with EE occurrence. Am J Trop Med Hyg. 49: 677-685. LoGiudice K, Ostfeld RS, Schmidt KA, Kesing F. The ecology of infectious disease: Efects of host diversity and community composition on Lyme disease risk. P Natl Acad Sci Usa. 2003;100:567-571. Mackenzie, D.I. and J.A. Royle. 2005. Designing occupancy studies: general advice and alocating survey efort. J Appl Ecol. 42: 1105-1114. 98! Moore, C.G., R.G. McLean, C.J. Mitchel, R.S. Nasci, T.F. Tsai, C.H. Calisher, A.A. Marfin, P.S. Moore, and D.J. Gubler. 1993. Guidelines for arbovirus surveilance programs in the United States. U.S. Department of Health and Human Services, Fort Collins, CO. Morris, C.D. 1998. Eastern equine encephalomyelitis. In: Morris, T.P. (ed) The Arboviruses: Epidemiology and Ecology. CRC Pres, Boca Raton, FL. R Development Core Team (2008). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3- 900051-07-0, URL: http:/ww.R-project.org. Scott, T.W., and S.C. Weaver. 1989. Eastern equine encephalomyelitis virus: epidemiology and evolution of mosquito transmision. Adv Vir Res. 37: 277-328. Vilari, P., A. Spielman, N. Komar, M. McDowel, and R.J. Timperi. 1995. The economic burden imposed by a residual case of eastern encephalitis. Am J Trop Med Hyg. 52: 8- 13. 99! Table 1. Descriptive statistics of the relative abundances of avian species at sentinel sites in Walton County, Florida in 2009. Species represented are those that were inferred to be preferred hosts of Cs. melanura in Tuskegee National Forest based on forage ratio estimates > 1 (Estep et al. in pres). Alpha codes follow guidelines of the American Ornithological Union?s Checklist of North American Birds. 100! Table 2. Summary of atributes of candidate models considered in developing a logistic regresion model of the log-odds of a sentinel site having a high- versus low-level of virus activity (>1 chicken seroconverted to EEV antibodies) in Walton County in 2009. QAICc is the bias-corrected quasi-Akaike Information Criterion, K is the number of parameters estimated, " i is the diference in QAICc from the model that minimized the QAICc, and w i is the QAICc weight. Goodnes-of-fit (GOF) of each models to the data was asesed with the le Cesie and Houwelingen test (le Cesie and Houwelingen 1991). 101! Table 3. Variable weights and model-averaged estimates of intercept and variable coeficients for EEV models in Walton County, Florida. 102! Figure 1. Sentinel site locations in Walton County, Florida used in model development. EEV- positive sites are those where at least one chicken seroconverted to EEV antibodies in 2009. 103! Figure 2. Image of 2009 avian point-count-locations centered on individual sentinel sites in Walton County, Florida. 104! Figure 3. Inferred rates of EEV transmision among sentinel sites of variable avian community sizes in Walton County, Florida, with estimated function of the probability of a site having a high level of virus activity overlain. EEV-negative sites were those where at least chicken seroconverted to EEV antibodies in 2009, whereas sites where no chickens seroconverted were designated as EEV-negative sites. The model-averaged estimate and asociated 85% weighted unconditional confidence interval was -2.194 (-6.077, 1.688) for the linear efect of avian community size (N; ln-transformed, centered about mean) and 18.857 5.472, 32.242) for the quadratic efect of avian community size (N 2 ; ln-transformed, centered about mean). 105! Figure 4. Relative abundances of common yelowthroat at EEV-positive and EEV-negative sentinel sites in 2009 in Walton County, Florida. Plots show 1.5*inter-quartile range (whiskers), the interquartile range (box edges), and the median (horizontal line). EEV-negative sites were those where at least chicken seroconverted to EEV antibodies in 2009, whereas sites where no chickens seroconverted were designated as EEV-negative sites. Model-averaged estimates of coeficients for relative abundances are shown in plots, with 85% weighted unconditional confidence intervals. 106! Figure 5. Asociation betwen weights asociated with variables of relative abundances of avian species vs. estimated forage ratios of those species (r S (5) = 1.000, p= 0.0167). Species shown are those inferred to be preferred hosts of Cs. melanura based on forage ratio estimates (Estep et al. in pres). ! !