Declining Prevalence of Disease Vectors Under Climate Change

More than half of the world population is at risk of vector-borne diseases including dengue fever, chikungunya, zika, yellow fever, leishmaniasis, chagas disease, and malaria, with highest incidences in tropical regions. In Ecuador, vector-borne diseases are present from coastal and Amazonian regions to the Andes Mountains; however, a detailed characterization of the distribution of their vectors has never been carried out. We estimate the distribution of 14 vectors of the above vector-borne diseases under present-day and future climates. Our results consistently suggest that climate warming is likely threatening some vector species with extinction, locally or completely. These results suggest that climate change could reduce the burden of specific vector species. Other vector species are likely to shift and constrain their geographic range to the highlands in Ecuador potentially affecting novel areas and populations. These forecasts show the need for development of early prevention strategies for vector species currently absent in areas projected as suitable under future climate conditions. Informed interventions could reduce the risk of human exposure to vector species with distributional shifts, in response to current and future climate changes. Based on the mixed effects of future climate on human exposure to disease vectors, we argue that research on vector-borne diseases should be cross-scale and include climatic, demographic, and landscape factors, as well as forces facilitating disease transmission at fine scales.


Leishmaniasis
Leishmaniasis is a neglected disease in Ecuador reported since the beginning of the 20 th century and has been recorded from 21 of the 24 Ecuadorian provinces 9 . To date, at least eight Leishmania spp. have been identified to infect humans and non-human mammals 10,11 . Leishmaniasis reports in rural areas occur from sea level to ~2,700 m 10 elevation 11 . More than 60% of all Leishmania species of Ecuador are reported in the subtropical and tropical lowlands of the Pacific region (Fig. S5), where Le. panamensis and Le. guyanensis are the most common. In the highlands of the Andes region in central Ecuador (Fig. S5), the main leishmaniasis agents are Le. mexicana (>80%) and Le. major-like [11][12][13][14]

Chikungunya
Chikungunya is an arboviral disease with symptoms somewhat similar to dengue fever, characterized by sudden onset of fever, rash, and severe joint pain lasting until 14 days; some cases result in persistent arthritis even 25 days after initial symptoms 19 . At the end of 2014, the first imported case of chikungunya virus was reported in Ecuador and less than a year later, the number of locally transmitted cases raised to ~30,000 since the start of the epidemic to October 2015 7 . Strikingly, the number of chikungunya cases is likely underreported in the country 20 with 7.5 % of cases established by epidemiological link. Additionally, the eminent risk of the introduction of other arboviruses to Ecuador is an emergent public health concern; for example, by September 2016 zika virus has been detected in at least 2,000 human patients in several coastal provinces of the country 21 . Yellow fever is nowadays a disease of less concern, but still present in Ecuador 22 .

Malaria
Malaria is still the most important vector-borne disease worldwide transmitted by Anopheles mosquitoes 23 with reports of up to 80,000 cases/year in Ecuador.
Interestingly, the country has been reporting most cases of dengue and few cases of malaria in the last five years 24 (Fig. S4), a pattern that changes the current public health priority from malaria to dengue 25 . This trend has been poorly addressed and might be associated with efficient vector control programs, reduction in epidemiological surveillance, or other factors including climate variability reducing vector abundance as fewer cases of malaria have been reported even in regions lacking Anopheline mosquitoes control. However, malaria is still of importance in Ecuador given the potential for re-emergence, the persistence of Anopheline mosquitoes in rural areas where the disease was endemic, the mortality associated with infection, and DALYs 23 .
By May 2016, 229 cases of malaria were reported in Ecuador 15 .

Chagas
Chagas disease is another vector-borne disease historically endemic in the Andean region and in Ecuador where it is still a cause of morbidity and mortality ( The framework of our modeling approach included a detailed characterization of the study area, occurrences, and model fit for each vector species under present-day environmental conditions derived from satellite imagery and future climate. A summary of the methodology is found in Figure S6 and is explained in detail below. We removed the duplicate occurrences by species to obtain single occurrence points by site and to reduce model overfitting due to oversampled areas 43 (Supplementary Dataset). The vector occurrence records were then used to develop ecological niche estimations, under the assumption that each record originated from a stable population that can persist without need of immigration.

Areas for model calibration
A critical step in the niche modeling process is the selection of areas for model calibration 43 ; the extent of such areas dramatically impacts the area predicted as suitable for vector occurrence 44 . The study design of ecological niche models should be based on biogeographic features for each vector species 44 . For this, study designs should follow the BAM framework (sensu Soberón and Peterson 45 ). This framework is used in modern ecology during the study design and interpretation of ecological niche modeling. The BAM framework identifies three factors that interact in the species' ecological niche delimitation: B: biotic; A: abiotic; and M: dispersal capacity. Under this framework and in view of the considerable effects of the study area extent in model results 44 , modelers should establish specific geographic areas for model calibration for each species according to the species' dispersal capacity M.
Considering the impact of the study area extent on ecological niche model predictions and the need for biological realism in the study design 43,44 , models should be calibrated in areas representing a proxy of M 44 . We assumed that each species' M could be estimated based on the average geographic distance among vector occurrences 46 .
Briefly, for each vector species we estimated a centroid point from all the occurrences of the vectors' distribution and measured the distance between the centroid to all the occurrences (Table S1). We focused on the vectors' native range to avoid over estimation in transoceanic dispersal of the invasive vectors of the Aedes genus. The average geographic distance between the centroid and occurrences was used to generate a buffer zone where models were calibrated (Fig. S7). This method provides impartiality from the modeler to establishing the extent of the calibration area, furthermore it is based on the potential dispersal of the species providing biological meaning to model results, and allows the characterization of the environments occupied by species across their distribution 44 . For the estimation of the average distance between occurrences of Ae. aegypti and Ae. albopictus, we used occurrences in the native ranges of Africa and South Asia respectively 47 . For all other species, we estimated the average distance restricted to reports of their native distribution in the Americas. albopictus models using remote sensing data were developed using a regularization coefficient of 1.  65 . Final models were converted to binary using a threshold based on minimum training presence (Fig. S6). These maps were used to estimate suitable areas by vector on present-day climates and people living in these suitable areas. We estimated overall patterns of people density exposed to vectors under current climate using 1 km resolution of human population in Ecuador from LandScan 18 . We employed this data set also to assess the population at risk under future conditions considering the agreement of LandScan with future human population models from Ecuador 71 (r 2 = 0.43, p < 2 x 10 -16 ). We then estimated the area predicted suitable for the vectors.

Supplementary
Finally, we estimated the percent of change of area suitable for the vector species and population exposed to vectors between present-day and future climate scenarios.

Supplementary Material: Results and Discussion
Occurrences

Present-day models
The selected uncorrelated variables for our present-day models were mean and standard deviation of the monthly EVI time series data; long-term precipitation for two periods including i) November, December, and January; and ii) May, June, and July; The M estimation for Ae. aegypti represented the broader study area, with occurrences in five continents (Fig. S7) (Fig. S7).
According to our AIC evaluations, the best regularization coefficients based on remote sensing data and precipitation ranged between 0.5 and 3, the most frequent values were 1 and 2, while for models based climate for future conditions, parameters ranged between 0.5 and 5, the most frequent was 0.5 (Table S2). There was a negative 26 association between the number of parameters and AIC values (r 2 range = 0.42 -0.93; p<0.05), with the exception of R. ecuadoriensis when calibrated using climate. We did not find associations between the number of occurrences and the regularization coefficient value of the best Maxent models calibrated using remote sensing (r 2 = 0.31; p=0.07) nor climate (r 2 = 0.18; p=0.15). We found associations between the lowest AIC values and lowest regularization coefficients in 67% of the vector species modeled using remote sensing and in 64% of the species modeled using climate (Table S2). The AUC evaluations showed a good discriminatory capacity of models to predict occurrences better than a random model (Table S3).
Supplementary When the final models were developed and transferred to Ecuador, broad suitable areas were found across the country for Ae. albopictus and Ae. aegypti; highlands in central Ecuador limit the potential distribution of these vectors, but Ae. aegypti appear to be a more generalist species, tolerating environments available in highest zones (Fig.   S1). In contrast to countries located in temperate regions, tropical countries possess Chinchipe. Under the 2100 scenario, some models failed to find suitable conditions in Ecuador for some vectors (Fig. S2), for example, T. dispar may be extirpated due to unsuitable conditions in the future considering the absence of tolerable climate for the species (Fig. S1). However, the comparison between present-day and future climate in Ecuador via Mobility-Oriented Parity test revealed areas with future climate not available in the present-day distribution of this species (Fig. S2). The presence of non-analogous environments, where no prediction was allowed, was particularly evident for species with narrow distributions with maximum temperatures above those available currently in the species range ( Fig. S1 and S2).
Our future climate models offer coarse spatial resolution variables (i.e., ~20km) compared with the fine resolution from remote sensing data we employed (i.e., ~1km; communities that may result in unexpected ecological surprises not captured or forecasted by our models 77 . We are sympathetic with these limitations of climate based models, but also recognize that future climate scenarios provide opportunities to anticipate and adapt to the effect of climate change on human health 78 . Additionally, we employed data of human populations across Ecuador under current climate conditions and explored the risk of this population under future climate. We avoided the use of future human population scenarios considering that the main gain of future population models would be a temporal match between future climate and future population, but at the cost of amplifying uncertainty. In other words, future climate models have considerable uncertainty, thus, using future population models will result in risk estimations including the uncertainties of both future climate models and future population models, thereby amplifying the uncertainties in the system. We found that patterns of current and future population in Ecuador will remain stable. However, future research exploring the accuracy of essambling future climate with future population models is warranted.
Occurrence data may also contribute to uncertainty from sampling bias 79 . Bias could be generated from oversampling in areas of easy access or from countries with active epidemiological surveillance systems, thus, biased data will result in biased models 79 .
To mitigate the effect of bias in model calibration, we utilized binary models instead of continuous models resembling sampling bias effort 43 . Additionally, we employed all the occurrences available to us for each vector species, this allowed us to capture a representative sample of the environmental signature required by each vector species across its known geographic distribution. Models for some species were calibrated from a low number of occurrences, likely reflecting the low reporting effort and data digitalization of Ecuador or the low abundances of such species. Previous niche models of vector species have considered vector reports in the study area only, neglecting vector data from areas outside the areas of interest (e.g., [80][81][82]  hartmanni has not been reported as a proven vector of leishmaniasis; althought its role in disease transmission is debatable, the species is widely distributed in areas where the disease is endemic and has been found infected with Endotrypanum parasites 9 .
Other sandfly species including Lu. tortura and Lu. serrana have also been found infected with parasites, but their role as vectors of the disease is not well understood.
We found that tuning Maxent parameters, specifically the regularization coefficient, provides better performance than using default parameters. This step appears to be important when calibrating models for species of public health concern considering that model fit appears to be sensitive to the regularization coefficient selected 69  Finally, anticipating potential areas and populations at risk of exposure to disease vectors is a priority for effective disease control interventions 89 . While a previous study aimed to describe the occurrence of Rhodnius ecuadoriensis in two Ecuadorian provinces 90 , to our knowledge, this is the first effort to model a complex ensemble of vector species under present-day and future climate conditions in Ecuador using remote sensing and future climate models (Fig. S3).