São Paulo is located in Southeastern Brazil (23°32′51″ S/46°38′10″ W), has an area of 1572km2, a population of over 11,000,000 inhabitants, and is divided into 96 Administrative Districts (ADs). The city has a subtropical climate, with rain in the summer and autumn months (December–May), and dry weather in the winter and spring months (June–November).

We received approval from the Institutional Review Board of the São Paulo Secretary of Health (Comitê de Ética em Pesquisa da Secretaria Municipal de Saúde do Município de São Paulo) to use de-identified data from blood samples that were collected by public and private health facilities. All animal experiments were performed in accordance to the guidelines of the Institutional Ethics Review Committee (Brazilian Society of Science in Laboratory Animals-SBCAL/COBEA) and the Animal Care Committee of the Institute of Biomedical Sciences (Ethics Committee in Animal Experimentation – CEEA) – University of São Paulo (protocol #014/10).

Data regarding dengue cases in São Paulo between 2010 and 2011 were obtained from the Health Surveillance Coordination of São Paulo (COVISA). The majority of cases (99.5%) were classic dengue fever (no complications related to plasma leakage, fluid accumulation, respiratory distress, severe bleeding, or organ impairment). Only cases in which epidemiological investigation indicated the place of residence as the most likely place of infection were included in this study. During 2010–2011, viruses were isolated from 136 patients at the Adolfo Lutz Institute, São Paulo, Brazil, as described previously (serotype 1: 92%; serotype 2: 4%, serotype 3: 4%).10,11

Dengue cases were geocoded according to the patient's home address, using the digital cartographic database of São Paulo (GEOLOG, LOC MSP) and Terraview 4.1.0 software (available at http://www.dpi.inpe.br/terraview/index.php).

Land surface temperature (LST) was calculated using thermal remote sensing images from the LANDSAT-5 thematic mapper (TM) to identify temperature zones and UHI in São Paulo. LANDSAT-5 TM multispectral images were selected for their lack of cloud cover. Three satellite images were obtained in 2010 (05/04/2010, 08/24/2010, and 11/28/2010), and three additional images in 2011 (05/23/2011, 07/26/2011, and 11/28/2011). Thermal band (Band 6) was used to calculate the LST in the selected images. Using the image obtained on 07/26/2011, bands 2 and 3 were used to establish the limits for soil and water, respectively, while band 4 was used to determine the vegetation cover. Spectral radiance was calculated, and temperature conversion was performed as described previously.12 Ten different temperature zones (increments of 2°C between 20°C and 40°C) were identified. The average LST for 2010–2011 was obtained using SPRING 5.1.7 software (http://www.dpi.inpe.br/spring/portugues/download.php).13

The supervised classification of images was performed using SPRING 5.1.7 and known São Paulo landmarks (parks, rivers, and urban zones), by applying the Bhattacharya algorithm.14,15 LST and vegetation maps were constructed using vector data (scale, 1/10,000). Vegetation areas were classified as low, moderate, or high cover.

A classification of “slum-like” was based on the lack of basic services, substandard housing, and illegal/inadequate building structures. Information regarding slum-like areas was obtained from the 2010 census database of the Brazilian Institute of Geography and Statistics (IBGE). The map of slum-like areas was generated using vector data at 1/10,000 scale using the SPRING 5.1.7.

The Kernel estimation method16 was applied to identify hotspots with a high concentration of reported dengue cases. A spatial point distribution of dengue cases was constructed using the Kernel method and Terraview 4.1.0 to classify the case density in São Paulo.

Population data from census tracts (scale 1/10,000) was obtained from the 2010 census database of the Brazilian Institute of Geography and Statistics. Population density was classified as 0–7000, 7001–10,000, 10,001–20,000, 20,001–30,000, 30,001–40,000, and >40,000 inhabitants/km2. Household income was categorized as <510, 511–1275, 1276–2550, 2551–5100, and >5100US$/month. Land use was categorized as Res-H: predominantly residential houses, Res-B: predominantly residential buildings, NotRes: minimal residential use (typically commercial or industrial), and Wasteland: vacant lots.

Census tract data were converted in a rectangular grid with 5-m resolution (raster data), and the number of inhabitants for each cell was calculated by multiplying the population density by the cell area (25m2) using SPRING 5.1.7. The use of a 5-m resolution grid allowed us to fit all pixels inside census tracts with different sizes and shapes.

Two strains of A. aegypti were used in our laboratory experiments. The white-eyed, Puerto Rico Rexville D (Higgs) strain17 was from our laboratory stocks, while the University of São Paulo (USP) strain was established in 2012 using eggs collected using ovitraps at 23°33′48.50″ S/46°44′35.72″ W and 23°33′49.09″ S/46°44′36.66″ W. Both strains were maintained in a secure insectary (Biosafety Level II) at the Department of Parasitology, Institute of Biomedical Sciences, University of São Paulo, Brazil. Experiments involving the USP strain used F1 progeny from eggs collected in the ovitraps. The mosquitoes were bred and maintained in a controlled atmosphere (27°C, 80% relative humidity, and a 12:12h light:dark cycle). Larvae were fed powdered fish food (Sera, Heinsberg, Germany), and adult mosquitoes had access to 10% sucrose solution ad libitum.

To study the effects of temperature on oviposition, Higgs strain female mosquitoes were blood-fed on mice for 10min (27°C) and divided into four groups of 10 mosquitoes, which were maintained at 20°C, 24°C, 28°C, or 32°C for 96h. A wet surface was provided for oviposition at 72h, and the number of eggs laid in the final 24h was recorded. We also studied the blood-feeding behavior of female Higgs mosquitoes, using groups of 10 mosquitoes that were blood-fed on mice for 10min at 20°C, 24°C, 28°C, or 32°C. Individual weights before and after feeding were recorded, and the group averages were calculated. Although we attempted similar experiments using the USP strain, they did not feed under the experimental conditions.

Multivariate cluster analysis (MCA) was used to examine the associations between the incidence of dengue and the socioeconomic and environmental factors.18 Standard analyses (e.g., multiple and logistic regression) could not be applied to our dataset, as it consisted of continuous and categorical variables.19 Although regression models might be used after converting the data (e.g., discretization of continuous data, or creation of dummy variables using categorical variables), they were not used to avoid the introduction of bias.

Twenty-five variables (ten LST ranges, five ranges of household income, four categories of land use pattern, three vegetation cover categories, territorial extension, population density, and proportion of slum-like areas) and 96 ADs were considered. AD data (rather than census tract data) was chosen, given the wide range of census tracts areas (1580–5,172,060m2), which would have jeopardized the evaluation of the tract features’ influence on the incidence of dengue. Given the distinct numerical ranges of the different variables, data were standardized prior to the analysis as follows: standard score=(raw score−mean)/standard deviation.

A dissimilarity matrix of Euclidean distances within each pair of variables was constructed, and a phenogram was built (using complete linkage as the amalgamation rule) to illustrate the calculated distances between the groups. Statistica 10 (Statsoft, Oklahoma, USA) was used to perform multivariate analysis and for graphical outputs.

Data from the laboratory experiments were analyzed using one-way analysis of variance, and the Bonferroni test was used to evaluate inter-group differences. Analyses were performed using BioEstat 5.0 (available at http://www.mamiraua.org.br/pt-br/downloads/programas/).20p-Values <0.05 were considered statistically significant.

During 2010–2011, a total of 7,415 dengue cases were recorded in the existing surveillance systems. Dengue cases were typically concentrated in areas with high LST, and the majority of cases (93%) occurred in areas with LST >28°C; fewer cases occurred in slum-like areas (Fig. 1). Kernel estimation method showed that the highest densities of dengue cases occurred in the north, east, northeast, southeast, and southwest areas (Fig. 2), with LST ranging 30–34°C (Fig. 1A).

Fig. 1.

The distribution of dengue cases by temperature zones and slum-like areas. Land surface temperature (A), and slum-like areas (B) were geocoded using vector data (scale, 1/10,000). The area outlined in black is the main commercial and financial zone of São Paulo.

(0.77MB).

Fig. 2.

Kernel estimation of the distribution of dengue cases in São Paulo during 2010–211. A kernel map was built using the spatial point distribution of the 7,415 dengue cases reported during 2010–2011. The area outlined in black is the main commercial and financial zone of São Paulo.

(0.15MB).

The dengue incidence was higher in high LST areas in the overall city area (Fig. 3, upper panel), and in slum-like areas (Fig. 3, middle panel). A similar result was obtained when the incidence was re-calculated after excluding the slum-like areas from overall São Paulo area. However, in the 20–24°C temperature range a relatively high incidence was calculated because the population density in these areas was low (Fig. 3, bottom panel).

Fig. 3.

Dengue incidence in urban heat islands. Geocoded cases were divided according to the land surface temperature, and the incidence rate was calculated by dividing the number of cases by the population of that area, and multiplying the quotient by 100,000.

(0.12MB).

The analysis of dengue occurrence based on population density demonstrated that most cases were clustered in highly populated zones (Fig. 4A), although the incidence (in cases per 100,000 individuals) was low in densely populated areas and relatively high in less-populated areas (Fig. 4B).

Fig. 4.

Dengue incidence by population density. (A) Population densities were calculated using census tracts data (inhabitants/km2), and (B) dengue incidence (cases per 100,000 inhabitants) in the population density zones was calculated. The area outlined in black is the main commercial and financial zone of São Paulo.

(0.41MB).

The majority of dengue cases occurred in areas with low vegetation cover (Fig. 5). Zones with high vegetation cover had the lowest incidence rate (3.2 cases per 100,000 inhabitants), while the moderate and low vegetation zones had incidence rates of 50.2 and 72.3, respectively. Areas with high vegetation cover were cooler (26±2°C) compared to areas with moderate or low vegetation cover (both, 29±2°C).

Fig. 5.

Dengue incidence by vegetation cover. (A) Dengue cases and vegetation cover areas were geocoded using geographical information systems. (B) The dengue incidence (cases per 100,000 inhabitants) and (C) land surface temperatures in the low, moderate, or high vegetation cover zones are shown.

(0.46MB).

MCA of the 96 ADs (Supplementary Information: Fig. S1, Tables S2 and S3) revealed a complex relationship between the incidence of dengue and the socioeconomic and environmental factors (Fig. 6). In the phenogram, dengue incidence was clustered in high-LST areas (32–36°C), while areas with LST <30°C showed a relatively low similarity with dengue occurrence despite the presence of socioeconomic features in the low LST areas that potentially promote disease transmission, like low household income, high population density, and slum-like areas. As would be expected, the slum-like areas clustered with the lowest household income ranges. Areas with low vegetation cover clustered with higher LST (28–30°C), while high vegetation cover clustered with lower LST (22–24°C) (Fig. 6), in agreement with results shown in Fig. 5C.

Fig. 6.

Multivariate cluster analysis of socioeconomic status, environmental conditions, and dengue incidence (cases per 100,000 inhabitants) for the Administrative Districts of São Paulo. A complete linkage phenogram of Euclidean dissimilarity distances among these variables was plotted. Branches distances are mutually proportional.

Abbreviations: Area, territorial extension (km2); Dengue, dengue incidence; LST, land surface temperature; MHI, monthly household income; HVC, high vegetation cover; MVC, moderate vegetation cover; LVC, low vegetation cover; Res-H, predominantly residential houses; Res-B, predominantly residential buildings; NotRes, minimal residential use (typically commercial or industrial); PopDens, population density.

(0.11MB).

Supplementary material related to this article can be found, in the online version, at http://dx.doi.org/10.1016/j.bjid.2014.10.004.

Dengue Incidence and socioeconomic aspects from ADs of São Paulo.

Dengue Incidence and environmental aspects from ADs of São Paulo.

The larval development of the Higgs and USP strains was studied at four different temperatures for seven days after the eggs hatched (Fig. 7). At 20°C, the larvae did not reach the pupal stage, while approximately half reached the pupal stage at 24°C (the remaining half were L4 and L3 larval stages). At 28°C and 32°C, the majority of the larvae reached adulthood by day 7.

Fig. 7.

Temperature influence on A. aegypti life cycle. Four groups of fifty larvae each were separated after the eggs hatched, and maintained at 20°C, 24°C, 28°C, or 32°C. On day 7, the number of individuals in the different stages of development was recorded (larval stages L1, L2, L3, L4, or pupae or adults). Experiments were performed in triplicate. The mean (±standard deviation) percentage of viable individuals on day 7 is presented for the Higgs and University of São Paulo (USP) strains.

(0.11MB).

In another set of experiments to determine the effect of temperature on oviposition, Higgs females that were maintained at 32°C laid a greater number of eggs compared to those maintained at lower the temperatures (Fig. 8A). At 28°C or 32°C, a higher percentage of female mosquitoes fed on mice blood (60% and 76%, respectively) (ANOVA: F=19.1, p=0.0009) compared to those at 20°C and 24°C (Fig. 8B). In addition, the amount of blood taken per mosquito was higher at 28°C and 32°C, compared to that at 20°C and 24°C (ANOVA: F=4.6, p=0.0068) (Fig. 8C).

Fig. 8.

Temperature effects on oviposition and blood-feeding behavior of A. aegypti (Higgs strain) female mosquitoes. Four groups of ten female mosquitoes each were blood fed on mice, and maintained at 20°C, 24°C, 28°C, or 32°C for 96h. The number of eggs laid in the last 24h during the temperature exposure was recorded (A). Female mosquitoes were individually weighed (3.3±0.4mg), and divided into groups of 10 mosquitoes each, and the groups were allowed to feed on mice for 10min at 20°C, 24°C, 28°C, or 32°C. The number of blood fed mosquitoes (B) and their individual weight after blood meal (C) was recorded. Experiments were performed in triplicate. The boxes represent the range between the 25th and 75th percentiles. The horizontal line in each box represents the mean value, the circles represent outlier values, and * indicates p<0.01.

(0.13MB).