Climate Change Influences Potential Distribution of Infected Aedes aegypti Co-Occurrence with Dengue Epidemics Risk Areas in Tanzania

Background Dengue is the second most important vector-borne disease of humans globally after malaria. Incidence of dengue infections has dramatically increased recently, potentially due to changing climate. Climate projections models predict increases in average annual temperature, precipitation and extreme events in the future. The objective of this study was to assess the effect of changing climate on distribution of dengue vectors in relation to epidemic risk areas in Tanzania. Methods/Findings We used ecological niche models that incorporated presence-only infected Aedes aegypti data co-occurrence with dengue virus to estimate potential distribution of epidemic risk areas. Model input data on infected Ae. aegypti was collected during the May to June 2014 epidemic in Dar es Salaam. Bioclimatic predictors for current and future projections were also used as model inputs. Model predictions indicated that habitat suitability for infected Ae. aegypti co-occurrence with dengue virus in current scenarios is highly localized in the coastal areas, including Dar es Salaam, Pwani, Morogoro, Tanga and Zanzibar. Models indicate that areas of Kigoma, Ruvuma, Lindi, and those around Lake Victoria are also at risk. Projecting to 2020, we show that risk emerges in Mara, Arusha, Kagera and Manyara regions, but disappears in parts of Morogoro, Ruvuma and near Lake Nyasa. In 2050 climate scenario, the predicted habitat suitability of infected Ae. aegypti co-occurrence with dengue shifted towards the central and north-eastern parts with intensification in areas around all major lakes. Generally, model findings indicated that the coastal regions would remain at high risk for dengue epidemic through 2050. Conclusion/Significance Models incorporating climate change scenarios to predict emerging risk areas for dengue epidemics in Tanzania show that the anticipated risk is immense and results help guiding public health policy decisions on surveillance and control of dengue epidemics. A collaborative approach is recommended to develop and adapt control and prevention strategies.


Introduction
Dengue is the most important arboviral infection in the world [1,2]. The disease is endemic in more than 100 countries [3,4] and continue to spread worldwide due to emerging ecological distribution areas of disease vectors and the global movement of humans [5]. Climate has a significant role in mosquito distribution due to emerging favourable conditions especially rising temperatures [6][7][8]. Globally, climate change projections indicate an increase in mean temperatures ranging between 1.8°C and 4°C by the end of the 21st century [9]. In Tanzania, temperatures are expected to rise between 1°C to 3°C above baseline, and precipitation and extreme events are expected to increase in frequency by the 2050s [10]. Climate change is likely to affect human health from impacting food insecurity and malnutrition to increasing in risk areas for vector-borne diseases [11,12]. Climatic change creates new ecological niches for vectors hence altering temporal and spatial distribution of vector-borne disease [13][14][15][16][17][18]. Climate change has been implicated as a contributing factor in dengue globally [18][19][20][21] and the first outbreak of Chikungunya virus in temperate climate [8]. Studies indicate that climate change and variability influence dynamics and potential spatio-temporal distribution of dengue vectors and hence potential of disease endemicity [22][23][24][25][26][27].
In Tanzania, dengue outbreaks have been reported repeatedly in 2010, 2012 [28][29][30], 2013 and 2014 [31]. The most important vector in Tanzania is Ae. aegypti which is found in urban environments and prefers feeding on humans and the virus circulation is maintained through transovarial transmission [24,32]. Ae. aegypti usually bites in shady areas during the day or when the weather is cloudy, but biting is significantly high two hours after sunrise and before sunset [33]. Transmission of DENV to humans by mosquitoes involves complex processes influenced by mosquito genetics, viral genetics and bioclimatic factors [34]. The climatic conditions in which the immature mosquito larvae develop have a particularly large influence on the viral susceptibility and transmission capability of adult female mosquitoes [35][36][37].
Ecological Niche Models (ENMs) use environmental-climatic factors such as temperature, precipitation, elevation and derived-normalized difference vegetation index to predict climate change effects on disease vectors distribution [38,39]. There is inadequate information on linkages of current environmental conditions with dengue epidemics and the role of future climatic conditions in determination of potential epidemic risk areas in Tanzania. Hence, we use ENMs such as Maximum Entropy Species Distribution Modelling (MaxEnt) [40] for Ae. aegypti distribution to explore the geographic distribution of Ae. aegypti in relation to dengue fever epidemics. We also aim to identify bioclimatic conditions correlated with dengue fever epidemics in Tanzania.

Dengue fever epidemic in Tanzania
Following 2014 dengue epidemic in Tanzania, epidemiological and entomological investigations was carried out from May to June 2014 in Dar es Salaam [31] (Fig 1). Data on health facility-based confirmed human cases of dengue were recorded from January 2014 until end of May 2014. During that period, the disease continued spreading to different regions in the country. According to the unofficial report by the Ministry of Health and Social Welfare, a total of 961 confirmed cases were recorded in nine districts, of the seven regions in Tanzania mainland. 99% (N = 951) of the cases were from Dar es Salaam and the remaining 1% (N = 10) were reported from Mbeya (N = 2), Kigoma (N = 3), Mwanza (N = 2), Dodoma (N = 1), Kilimanjaro (N = 1) and Njombe (N = 1) regions but all with travel history to Dar es Salaam (Fig 2).

Occurrence data for infected Ae. aegypti
Ae. aegypti mosquitoes occurrence data (S1 File) for both infected and uninfected with DENV was collected during a cross-sectional study conducted in Dar es Salaam from May to June 2014 [31]. During this study, adult Aedes mosquitoes were trapped using Mosquito Magnets (Mosquito Magnet Cordless Liberty Plus) [41] fixed in a total of twenty seven sentinel sites for three consecutive days in each site (Fig 1). Pre-mature stages of Aedes mosquitoes were collected from water containers using standard plastic dipper and then reared to adult stage. Adult mosquitoes were identified to genus or species using morphological keys [42,43]. All sampling sites were geo-referenced for the purpose of this study. Both field collected and emerged adults were killed and stored in liquid nitrogen for later screening for DENV using quantitative real time reverse transcription polymerase chain reaction (qRT-PCR).
A total of 368 mosquito pools containing ten Ae. aegypti each separated due to site of collection were ground for extraction of ribonucleic acid (RNA) using PureLink 1 viral RNA kit (Invitrogen, Carlsbad, CA) according to manufacturer's instructions. A qRT-PCR for the universal detection of DENV-1 to DENV-4 serotypes was performed using RealStar 1 Dengue qRT-PCR kit version 1.0 (Altona Diagnostics GmbH, Hamburg, Germany), developed based on previous reports [44]. Of these, 330 pools were subjected to qRT-PCR for DENV detection. An overall 8.18% (N = 27) of the Ae. aegypti mosquito pools were infected with DENV [31]. Due to nature of spread of the disease, limited resources and time to conduct research to whole country, this data on infected Ae. aegypti was extrapolated to other un-sampled areas in whole country in order to identify other high risk areas under the current and predicted future climate scenarios.

Bioclimatic variables
Environmental conditions such as temperature and rainfall influence distribution of mosquitoes and affect the ecological niche for infectious Ae. aegypti mosquitoes. Three datasets of nineteen bioclimatic variables at a spatial resolution of about 1 square kilometer were downloaded from http://www.worldclim.org/. For the current scenario, bioclimatic conditions were interpolated from observed data representative of 1950-2000. For 2020 and 2050 scenarios, downscaled global climate model (GCM) data from Coupled Model Inter-comparison Project Phase 5 (CMIP5) was used [45].

Ecological-niche modelling
Presence record for infected Ae. aegypti co-occurrence with dengue were used together with bioclimatic data layers to develop ecological niche models. We used MaxEnt (version 3.3.1) to develop models for potential distribution of infected mosquito species in relation to reported disease cases. The dataset for all scenarios was split in the ratio of 3:1 for the training and testing respectively as previously done [40,50,51]. We used default settings for MaxEnt except that we specified a random seed with 50% of points set aside for model evaluation together with the regularization multiplier factor to reduce over-fitting due to many bioclimatic variables. Modifying the regularization multiplier helped to generate risk maps that can be extrapolated to a larger countrywide scale.
Because presence-only infected Ae. aegypti data originated from Dar es Saalam where most dengue cases were recorded, we adjusted the regularization multiplier such that predicted models results could be extrapolated to the whole country to identify other high risk un-sampled area as to the main purpose of ENMs. We used a minimum training threshold to convert raw model outputs into actual distributional estimates. The predicted distributions of risk areas are assessed by estimating the probability at maximum entropy based on assumption of uniform probability [40,52]. Twenty-seven single time occurrence records for infected Ae. aegypti tested by site of collection were used. These points were few compared to the number of total human cases records in the whole country. Despite records of few points, MaxEnt has shown to be successful in generating biologically meaningful models with occurrence records as few as six [48].
Predicted areas were identified as high risk due to probability estimates on potential occurrence of infected Ae. aegypti in the area.

Model performance evaluation
Partial receiver operating characteristic/area under the curve (ROC/AUC) approach was used to test model predictive performance [53]. This approach does not require absence data to characterize commission errors (sensitivity). Previous models used ROC approaches that require both absence and presence data but present numerous problems [53,54]. During each prediction in this procedure, infected Ae. aegypti occurrence data was re-sampled with replacement by bootstrapping method to determine AUC ratios generated from proportion area predicted present and sensitivity. AUC ratios were calculated from observed data and random prediction. It was assumed that a good model would give AUC ratio above 1. Significance testing of model performance was done by plotting the AUC ratios replicates that would indicate a normal distribution.

Variable importance
Of the nineteen-bioclimatic variables tested, twelve generally contributed importantly to the best model and were used as predictors for infected Ae. aegypti distributions for all climatic scenarios. Relative contributions of each variable to our prediction results by iteration of the algorithm during regularization and by random permutation in the jackknifing procedure showed that precipitation of driest month (BIO-14) and temperature annual range (BIO-7) contributed more to the model output for the current scenario whereas precipitation of warmest quarter (BIO-18) and precipitation of driest month (BIO-14) had higher permutation significance for the current scenario. For 2020 climate scenario, precipitation of driest month (BIO-14) and mean diurnal range (BIO-2) contributed more whereas precipitation of driest month (BIO-14) and precipitation of warmest quarter (BIO-18) had higher percentage on permutation importance. For 2050 climate scenario, precipitation of driest month (BIO-14) and mean diurnal range (BIO-2) contributed more whereas permutation importance was only contributed by precipitation of driest month (BIO-14), precipitation of coldest quarter (BIO-19) and precipitation seasonality (BIO-15) only. Temperature annual range (BIO-7), mean temperature of warmest quarter (BIO-10) and temperature seasonality (BIO-4) did not indicate any contribution to model output in 2050 climate scenario (Table 1).

Model performance evaluation
Model performance evaluation using commission and omission error rates based on infectious Ae. aegypti vectors occurrence points showed that it was statistically significant better than random prediction (p < 0.05) for the current and future climate scenarios. The partial ROC/AUC program generated AUC ratios ranging from 1.046 to 1.741 at the given 1-omission threshold of 0.95 using accepted omission error of 5% to the AUC at 50% for random prediction to specify the percentage of testing points which included in each of the random subsets. The models showed that the probability of the presence of infectious Ae. aegypti in relation to disease appears to be localized in the coastal areas while spreading towards the central areas of the country at high success rate according to the dengue epidemic records of 2014 (Fig 2).

Predicted current risk areas
Model prediction results for the current scenario indicate high habitat suitability for infectious Ae. aegypti in relation to dengue epidemic to be highly localized in the coastal areas such as Dar es Salaam, Pwani, Morogoro, Tanga and Zanzibar. Risk areas also appear in Lindi and Ruvuma regions. Other areas with low probability of dengue epidemics are indicated to appear in areas around major lakes such as Lake Victoria (Mwanza and Musoma), Lake Tanganyika (Kigoma) and Lake Nyasa (Iringa and Njombe) (Fig 2). The models indicate high risk in areas previously recorded disease [28,30,[55][56][57][58]. Despite records of dengue cases in Mbeya, Dodoma and Kilimanjaro, our models could not identify the area being suitable habitat for infected Ae. aegypti co-occurrence with dengue virus given that input bioclimatic variables. Prediction of low probability risk areas in Kigoma and Mwanza suggests the possibility that the disease could have occurred in the areas during the same period when epidemic was recorded in Dar es Salaam but the disease was not detected to a large scale due to lack of proper diagnosis.

Predicted distribution of risk area at future climate scenarios
Predicted risk maps for 2020 and 2050 climate scenarios show risk intensification in dengue epidemic risks areas as previously identified in the current scenario. However, for 2020, the risk seems to disappear in parts of Iringa near Lake Nyasa and Ruvuma. Dengue risk indicated broad-scale potential for change and shift in the distribution towards the central part of the country especially for 2050 projections (Figs 3 and 4). Models show anticipated high-risk areas southern parts of Lake Victoria spreading to eastern parts of Lake Tanganyika while leaving higher emerging risks in many parts of the country. Predicted suitability probability for 2050 indicated risk intensification nearly in all parts of the country (Fig 4). In general, findings indicated Dar es Salaam, Tanga, Pwani and Zanzibar would remain at high risk through 2050.

Discussion
Despite the fact that dengue epidemics being driven by many factors including rapid urbanization and lack of adequate sanitary or mosquito control measures, climate change is a major driver of vector-borne diseases epidemics [18]. In this study, we used ecological niche models to develop risk maps of the current distribution of suitable infected Ae. aegypti habitat in Tanzania. We then use climate change predictions to determine how the change in bioclimatic factors influence potential distribution of infected Ae. aegypti, the distribution of the principal vector of dengue in relation to disease epidemic risk areas. Our results show that changing climate will expand the range of Ae. aegypti and potentially intensify risks and expand current distributions of dengue. Previous study show that climate change is happening and it is likely to expand the geographical distribution of several mosquito-borne diseases [59]. There is sufficiency evidence incriminating dengue epidemics with variations in temperature and rainfall [19,59,60]. In Tanzania, no previous studies have been conducted to predict the role of these climatic conditions towards contribution in accelerated emergence of high-risk areas for dengue epidemics.
According to the history of dengue epidemics or dengue infections in Tanzania have been reported in Dar es Salaam and Zanzibar [29,57,58]. But our models were able to indicate that during the current climate scenario, most of the eastern coastal areas were at high risk during the same period. These results suggest that dengue epidemic might have occurred in other coastal areas such as Tanga, Pwani and Morogoro but the disease was not noticed due to lack of rapid diagnostic tools and due to feverish clinical symptom of the disease that can clinically be misdiagnosed with malaria. Records of misdiagnosis of fevers associated with arboviral infections to malaria cannot be under-estimated in Tanzania [55,61].
Incorporating 2020 and 2050 climate change predictions in the niche models of infected A. aegypti, we estimate an intensification and shift of risk areas estimated from probability of occurrence of infected mosquitoes. Despite minor risk detected at the current scenario, Lake Victoria zone and northern areas of Tanzania have recorded detection of arboviruses such as Chikungunya virus and Rift Valley Fever among hospital based patients [30,62,63]. The arboviruses use nearly similar climatic drivers similar to the ones causing dengue epidemics. Trends in change for temperature, rainfall and other environmental conditions influence the length of the immature stages by providing necessary resources within the larval habitats. Model projections into 2050 conditions predict increases in intensification and distribution of high risks area nearly in all parts of the country. Future climate models scenarios show spread of disease risk from coastal areas to towards the central zones of the country while intensification in areas surrounding lake zones continues. Therefore, the identified areas should be considered when monitoring for potential future epidemics.
Despite that fewer bioclimatic variables were important predictors of the information generated by our models, their predictive performance and accuracy allows for inherited certain level of uncertainty from the modelled climate dataset that was used since the climatic conditions for the future are themselves predictions from the models [45]. Also, our ecological niche modeling approach use only one time infected mosquito occurrence records without considering other entomological parameters such as container index, Breteau index, house index and pupal index. We urge for a careful interpretation and use of results as our finding are only limited to climate change as the main driver of dengue epidemics, while other factors also present potential increase for future disease risk areas such as ecological imbalanced state as a result of habitat fragmentation, urbanization, land-use changes, and human-imposed species disequilibria, making some other areas especially susceptible to the uncertain effects of global change. These activities create favourable environmental conditions for survival of mosquito vectors hence may worsen the situation in the future. Despite these limitations, recommend surveillance for dengue fever epidemics originating in these predicted areas in the upcoming decades especially in the new suitable habitat surrounding the major lakes.

Conclusion
Climate change represents a threat to emergence of more risk areas for dengue epidemics in Tanzania. Predicted distributions of risk areas present a cause of concern among disease managers. Results show that the influence of future climate scenarios on anticipated potential distribution of risk areas for dengue epidemic risk areas is immense and results will help in guiding future public health policy decisions on surveillance and control of dengue epidemics. A collaborative approach is recommended to develop and adapt control and prevention strategies that will help manage the anticipated risk. Concerted efforts are required towards adaptation to impact of climate change on vector-borne infectious diseases.
Supporting Information S1 File. File containing dengue occurrence and bioclimatic data therein attached denguebioclimatic-data. (ZIP)