Comparison of capture methods for the diagnosis of adult anopheline populations from State of Mato Grosso , Brazil

Introduction: The present study compares human landing catches of primary malaria vectors with two alternative methods of capture: the Shannon trap and the Mosquito magnet. Methods: This study used regression models to adjust capture data to a negative binominal distribution. Results: Capture numbers and relative percentages obtained from the three methods vary strongly between species. The highest overall captures were obtained for Anopheles triannulatus with captures for the Shannon trap and the Mosquito magnet measuring more than 330% higher than captures obtained by human landings. For Anopheles darlingi, captures by the Shannon trap and the Mosquito magnet were about 14% and 26% of human landing catches, respectively. Another species with malaria transmission potential that was not sampled by human landing captures weascaptured by the Shannon trap and the Mosquito magnet (Anopheles oswaldoi). Both alternative sampling techniques can predict the human landing of Anopheles triannulatus, but without proportionality. Models for Anopheles darlingi counts, after totaling daily captures, are significant and proportional, but prediction models are more reliable when using the Shannon trap compared with the Mosquito magnet captures. Conclusions: These alternative capture methods can be partially recommended for the substitution of human landing captures or, at least, as complementary forms of monitoring for malarial mosquitoes.

Mosquito sampling is a requirement for malaria vector population studies and, therefore, is an important component in malaria disease control.A reduction in the malaria incidence in the Amazon of about 45% between 2000 and 2002 and the decrease in high-risk municipalities from 160 to 76 were both partially attributed to an improvement in and the decentralization of entomological surveys 1 .However, the reduction in areas of social instability, including areas with mining, migration and irregular occupation have also led to an eventual improvement in the treatment of the disease.
Count results, as well as achievable spatial and temporal coverage of epidemiological studies of anopheline populations are heavily influenced by sampling methods.Currently, the collection of malaria mosquitoes landing on human baits is considered to be the most representative method for determining human landing activity because female mosquitoes are subsequently collected as they attempt to feed on the human collectors 6 .Frequent reports of malaria transmission to collectors, even in regions with low prevalence, have led to in serious ethical considerations and reduce the readiness of technicians in health organizations to apply these techniques 7 .

METHODS
Missawa NA et al -Capture methods for the diagnosis of adult anopheline populations In addition, the all-night collection method is labor intensive and imposes bias because of variation in the host-attractiveness of human collectors 8,9 .Under the financial and human resources available, achievable temporal and spatial coverage is limited, particularly for the surveillance of a territory with continental dimensions, such as Brazil.Furthermore, the use of alternative traps allows for standardizing the attractant, thereby improving the comparability of capture results 10 .
The efficiency levels of alternative captures methods for malaria transmitting anopheline species, however, have exhibited varying results 11 .It has been widely recognized that both light and odorbaited traps result in inferior counts of A. darlingi than do human landing captures [12][13][14][15] , but there are no systematic studies available for the transitional forest region in the Southern Amazon.Thus, it is not known if alternative capture techniques can be applied to reliably predict human landing catches.
In this context, the Shannon trap and the Mosquito magnet were evaluated for extra-domiciliary sampling in a malaria endemic area in Central Brazil.We used generalized linear regressions based on a negative binominal distribution to adjust the catching data.
Capture campaigns were performed at two rural locations in the municipality of Sinop, which is located in the Central Northern region of the State Mato Grosso, Central Brazil.The sites were located alongside the Teles Pires River inside semi-decidual riparian stands of the transitional forest ecotones at latitude S11º41'27,1" and W55º42'48,9" and S11º41'47,1" and W55º42'75,6", respectively, at an altitude of approximately 380m (Figure 1).
Sampling locations were about 30km from the urban area of Sinop and were, therefore, distant from a potential focus of urban malaria.Neither study area had ever been treated with insecticides.
In the present study, human landing (HL) collections of the anopheline species were compared with two alternative entomologic sampling techniques: the Mosquito magnet (Defender Model, American Biophysics Corp, East Greenwich RI) (MM) and a white cotton Shannon trap (ST).
In both sampling locations, the three capture methods were implanted at a distance of 100m in opposing directions from farm  domiciles to avoid interference among the experiments.One person was responsible for every catching method.To limit bias, capturers alternated their sampling method after one hour of the experiment.A total of 48 sampling campaigns (252 hs of capture) were conducted bimonthly between 2007 and 2008 during three-and four-day campaigns during the rainy season and three campaigns during the dry season.Samplings were conducted without interruption between the hours of 17:00 and 24:00.Human landing collection techniques followed protocol 34 6 .The malaria transmission risk to researchers during an entomological field survey can be considered low because no malaria case has been reported for the study area.The ST had a dimension of 1.7m x 1.4m x 1.4m and was constructed with a white cloth.White light and a protected collector were used as the primary attractions.Specimens were removed hourly using a manual aspirator.The MM (Mosquito Magnet™ model MM4100) uses a counter flow technology to capture insects.Propane gas is catalytically converted into carbon dioxide (CO 2 ), heat, and moisture to simulate a human presence.Voucher specimens were then taken to the laboratory of entomology at the Federal University of Mato Grosso.
Relevant findings in entomological studies are represented by count data, such as the non-negative integer number of specimens captured.The comparison of capture results through pair wise statistical tests are commonly performed after logarithmic transformation of the count data, which demands the addition of one to zero counts.Smith 16 , however, pointed out that this procedure may produce misleading results and that regression models may be more appropriate for comparing capture results 11,17 .
The random sampling of a homogeneous population of mosquitoes is expected to result in a Poisson distribution of counts.In addition to this random sampling variation, mosquito counts vary as a result of differences in underlying densities and, therefore, the observed variance in recorded mosquito numbers is typically greater than the mean (over dispersion) 11 , particularly if there are elevated numbers of zero-counts 18 .Therefore, count data were fitted to a two-parameter, negative binomial distribution 19 .Captures were intended to be generated from an underlying distribution of densities by taking a distinct value for each matched set of capture methods.The second parameter was used to adjust for the variance independent of the mean.For the regression adjustment, we applied the Generalized Linear Model module of the SPSS 17 statistical software package and used a negative binomial model with a log link.Model performance was evaluated by I) the deviance, which is defined as two times the difference of the log-likelihood for the maximum achievable model (i.e., each subject's response serves as a unique estimate of the negative binomial parameter) and the log likelihood under the fit model; II) standard error of the regression coefficients and the Wald 95% confidence limits calculated as an estimate ± (zα/2)*(Standard Error) where zα/2 is a critical value from the standard normal distribution; and III) chi-Square and p-values for testing the null hypothesis that an individual predictor's regression coefficient is zero if the rest of the predictors are in the model 20 .1).

Species captured by one or more methods included Anopheles (Anopheles) benarrochi
Anopheles triannulatus was by far the most frequently captured species, representing 87% (HB), 98% (MM) and 97% (ST) of all captured individuals.Captures for A. triannulatus were 334% and 348% of HL catches for MM and ST, respectively.Relative capture frequency of A. darlingi, the principal malaria vector in Brazil, exhibited more variation between the compared methods and was much higher for HL captures (10%) than for MM (0.005%) and ST captures (0.009%).This result corresponds to relative capture of 14% (MM) and 26% (ST) of HL catches. A. oswaldoi, a species relevant for malaria transmission in the Amazon region according several authors 19,21 , was captured at almost equal rates by the MM and ST.The relative capture rates of A. mediopunctatus, a species not considered relevant for malaria transmission, were lower for MM than for ST captures, representing 43% and 126% of HL catches, respectively.
Temporal inter-day capture patterns for the three methods were similar for A. triannulatus with an expressive peak at dusk (18-19 pm) (Figure 2).Only ST captures varied less throughout the sampling periods with a slight decrease during the following hours.Capture curves for A. darlingi revealed a slight hematophagic activity cycle during the early evening hours 22 for HL and MM sampling.Average maximum capture rates for ST were delayed.Results, however, were influenced by the low density of the species during most sampling campaigns.The peak HL captures between 22:00 and 23:00 originated from a single sampling day during the rainy season in which eight specimens were captured.
Differences in absolute numbers obtained from MM and ST captures do not independently invalidate these alternative catching techniques for the prediction of HL densities.If catch numbers for each capture technique are proportional to each other, it can be assumed that similar fractions of the mosquito population are captured, and HL can be predicted by simple linear models.If each hourly capture period is used as a sample, highly significant negative binomial (NB) regression models (p < 0.001) are obtained for A. triannulatus for both capture methods.The scatters of ST and MM catches against the matched HL catches suggest, however, that there is considerable variation within the fit relationships for the two trapping methods (Figure 3) and that there is no proportionality between the captures (coefficient intervals not including 1); both methods underestimate HL in most cases, particularly for higher catch numbers using ST.The limited performances of the models of both capture methods are reflected in the deviances of 544.51 and 339.84 and standard errors of 0.0195 and 0.0155, respectively.For A. triannulatus, MM catches were slightly better fit to HL than to ST captures.
At only 4.8% of the hourly MM captures and 6.7% of ST captures, specimens of A. darlingi were caught, whereas this accounted for 16.3% for HL captures.If an intercept is included, the NB regression model identifies a highly significant coefficient for ST captures (p < 0.001) and a significant one for MM captures (p < 0.025).Negative binomial regression performance is slightly better for ST, but both models must be considered notably poor due to coefficient standard errors of 0.25 and 0.44, respectively, and Wald chi-square values for B of only 21.56 and 5.03, respectively.
In both capture methods, HL counts of A. darlingi were proportional, but the model performance and the prediction of nonzero cases was poor (Figure 3).Because catch numbers were much lower, a catch with ST statistically corresponds to higher catches of two specimens with HL.
As seen in the average hourly captures, catches for A. darlingi did not coincide temporally.If the daily captures are summarized, regression models for both alternative capture methods were strongly improved (Figure 4).
The performance of the ST captures was superior to that obtained for MM catches, resulting in coefficient standard errors of 0.16 vs. 0.27 and Wald chi-square values for B of 23.57 vs. 18.07, respectively.In the case of ST captures, the model revealed proportionality; however, in the case of MM captures, it was only due to the high standard error (0.37).In contrast to A. darlingi, models for A. triannulatus did not improve for both alternative capture techniques if captures were summarized (not shown). .

FIGURE 1 -
FIGURE 1 -Mosquito sampling sites in the Sinop municipality, State of Mato Grosso, Brazil.

FIGURE 2 -
FIGURE 2 -Inter-day curves for total captures of matched Mosquito magnet, Shannon trap and Human landing catches (April 2007 through February 2008) for Anopheles triannulatus and Anopheles darlingi.

FIGURE 3 -
FIGURE 3 -Numbers of anopheline mosquitoes caught using Shannon trap (A, B) and Mosquito magnet collection methods (C, D) versus those caught with matched outdoor Human landing catches.Lines show the negative binomial regression models.