Anthropogenic landscape decreases mosquito biodiversity and drives malaria vector proliferation in the Amazon rainforest

significance level of 5%). To compare the distribution of the Ny. darlingi abundance among the different landscape categories and five vegetation classes found in our field collection areas in the Amazon: Anthropogenic (Atp), Open Tropical Moist Forest (OTMF), Open Tropical Moist Forest with Arecaceae dominance (OTMFA), Tropical Moist Forest (TMF) and Campi- narana (Cpn) (Brazilian Institute of Geography and Statistics—IBGE classification) [55], we applied the Kruskal-Wallis with Mann-Whitney-Wilcoxon’s multiple comparison test with Bonferroni correction. The analysis of variance (oneway ANOVA) with Tukey’s corrections was applied to verify Shannon-index variation among landscapes categories, and to test the null hypothesis that landscape structure does not influence species abundance, considering 5% as the significance level. Tukey’s correction was employed to compare the landscape structure categories in five classes of vegetation found in our collection localities. Variation in species composition among sampling units in each landscape structure category was addressed employing Canonical correspondence analysis -CCA in R-project, vegan package [66]. In the step 2, we used a simple linear regression and generalized linear mixed models (GLMM) to verify the relationship between Ny. darlingi abundance and the Shannon index, PLAND, ED and DW. We employed the GLMM statistical analyses because both edge density and landscape categories were random variables. A table with all models can be found in S2

Table in S1 File. In all models we employed PLAND and ED measured at each sampling unit (3.14 km2), as predictor variables, while ED measured at 0.5 km radius (both from peridomes- tic and forest fringe habitats) and classified as categorical variables, was used as a random effect. The two mosquito sampling points (ST and HLC) and their mosquito diversity assem- blages were assessed together as fixed variables. Four simple regression models were con- structed—Shannon index, PLAND, ED and DW. For GLMM, three models were constructed, all of them using Shannon index, PLAND, and DW as fixed effects, and ED categories (low, moderate, high) at peridomestic or forest fringe habitat, or landscape structure categories as random effects. The PLAND and ED metrics were calculated at the 1 km radius and used to categorize landscapes A–D only. For the models (1) and (2), the ED was calculated at the 0.5 km radius for both peridomestic and forest fringe habitats and used as a random effect. The center of the 0.5 km circle of both peridomestic and forest fringe habitats were inside a 1 km radius circle. Mosquito data from peridomestic and forest fringe habitats were combined for the GLMM analysis. The choice of using ED as a random effect were based on the results of the descriptive analysis, which showed a statistically significant difference in Shannon-index (S3 Fig in S1 File) to Culicidae biodiversity. The categorical approach to PLAND analysis did not show differences for Shannon-index variance (P > 0.05).

We used path analysis to test the prediction that changes in Ny. darlingi abundance with increasing Shannon index are the cause of changes in richness (number of different taxonomic mosquito species morphologically and/or molecularly distinct) [58, 64, 67, 68], and to quantify the direct effects on Ny. darlingi abundance caused by variation in the Shannon index, PLAND, ED, landscape structure categories (as dummy variable), vegetation classes (as dummy variable) and DW, while removing the effects of other independent variables [69, 70]. We inserted the vegetation class to verify whether it represents a confusion variable [71]. We employed the Structural Equation Model (SEM) to visualize the model using a path diagram with “semPlot” package of R-project v. 3.6. The GLMM models were adjusted by the distance between the sampling units and drainage networks because of the bionomic characteristics of Ny. darlingi as an ecotone species that lay eggs in partially shaded habitats, including forest fringe [20, 34]. All statistical analysis was car- ried out using RStudio v.1.1.456 software (R Development Core Team, R Foundation for Sta- tistical Computing, Austria) R-project (available at http://www.r-project.org), and “BiodiversityR”, “wordcloud” and “nlme” packages. PLOS ONE Anthropogenic landscape and mosquito biodiversity drives malaria vector proliferation PLOS ONE | https://doi.org/10.1371/journal.pone.0245087 January 14, 2021 7 / 25