Introduction There are three main patterns of dengue virus transmission that merit a better understanding of the involvement of Aedes aegypti: 1- Major dengue epidemics occur every few years [1], [2]; 2- Dengue viruses have become endemic or hyperendemic (co-circulation of two or more serotypes) in many countries [3], [4], and 3- There is intra-annual, seasonal dengue transmission with peak incidence during the second half of the year in the northern hemisphere and during the first half of the year in the southern hemisphere, each one associated with elevated temperature and precipitation [5]–[7]. There is a lack of long-term, longitudinal studies on the temporal dynamics of Ae. aegypti that would, otherwise, allow us to understand whether mosquito outbreaks are responsible for inter-annual dengue epidemics. It is common to observe relatively large densities of Ae. aegypti that do not result in major outbreaks, particularly after major epidemics [8], suggesting that other factors such as temporal changes in population immunity or the introduction of new serotypes can significantly influence inter-annual epidemic patterns [9]–[11]. Climate variability, particularly El Niño Southern Oscillation (ENSO) teleconnections with local weather, has been associated with inter-annual dengue epidemics [2], although it would seem that the relationship is complex, non-linear, and perhaps non-stationary [1], [12], [13]. Amarakoon et al. [6] found significant effects of temperature on dengue epidemics in the Caribbean, particularly one year after the onset of an ENSO event. Because dengue viruses are transmitted by the bite of infected mosquitoes, dengue virus endemicity or hyperendemicity requires the existence of sufficient vectors to produce uninterrupted transmission in spite of adverse, seasonal weather conditions (e.g., lack of rain), such as that observed in urban areas with long dry seasons [14]. Recurrent virus introductions facilitate dengue endemicity. In dengue endemic/hyperendemic countries, dengue viruses are disseminated among regions so that virus re-introductions are frequent and do not depend solely on virus import [4], [15]. There is evidence showing that even in relatively small countries such as Puerto Rico, some dengue genotypes can circulate uninterruptedly for prolonged periods of time [16]. Vertical transmission of dengue virus in Ae. aegypti could play a role in the maintenance of endemicity but more evidence is required to understand its role in nature [17]. Perhaps, the single, most important factor determining dengue endemicity is the habit of people of adding water to containers, which can be for drinking, cooking or bathing (water-storage) and for other purposes, such as ornamentation (fountains), watering plants, etc. Production of Ae. aegypti in those containers can be so important as to trigger dengue outbreaks during the dry season [18]. Additionally, the existence of cryptic containers with water producing large numbers of Ae. aegypti has been more frequently reported [19]–[22], and in at least one occasion, those recondite containers have been linked to local dengue virus transmission [23]. Some cryptic containers, such as septic tanks in Puerto Rico, can produce Ae. aegypti throughout the year [24]. There is evidence showing that the intra-annual cycle of dengue transmission is driven by weather and mosquitoes [7]–[9], [25]–[29]. However, there seems to be spatial variability in the temporal dynamics of Ae. aegypti. For example, detailed studies of the temporal change in adult and immature stages of Ae. aegypti in a temple compound in Bangkok, Thailand in 1966 failed to show an association between weather, mosquitoes, and dengue incidence [30], [31]. The lack of seasonal fluctuation in the Ae. aegypti population was explained by the prevalence of containers that were manually filled with water by people, such as water-storage jars and ant-traps. Assuming that there were enough mosquitoes throughout the year, dengue incidence could have increased due to temperature-induced rises in biting rates, and shortened mosquito gonotrophic cycles and virus extrinsic incubations periods that can significantly increase vectorial capacity [32], [33]. On the other hand, entomological surveys conducted in five places in Bangkok in 1962 showed a sharp increase in the number of Ae. aegypti at the beginning of the rainy season, followed by a peak in cases of dengue hemorrhagic fever two months later [34]. Similar observations were reported for Koh Samui Island, Thailand [35]. Contrasting observations on the dynamics of Ae. aegypti have also been reported in Puerto Rico. Moore et al. [36] found positive correlations between rainfall, immature mosquito populations (Breteau Index), and dengue, and that the relationship between rainfall and mosquitoes was more marked in the drier, southern parts of the island. The authors explained that most larval habitats of Ae. aegypti occurred outdoors and were filled at least partly by rain; a result that agrees with more recent surveys [37]. In contrast, Scott et al. [38] did not find any significant associations between rainfall or temperature and female adult Ae. aegypti in the wetter, northern San Juan city area. Given that intra-annual dengue transmission is sharply seasonal in Puerto Rico, with maximum transmission during the hot and rainy seasons [39], [40], we decided to investigate the role of weather and human influence on the temporal dynamics of Ae. aegypti and dengue in the San Juan metropolitan area. Efficient Ae. aegypti adult trap devices (BG-Sentinel) and CDC enhanced ovitraps were used to monitor the mosquito populations. Our observations were structured to minimize temporal and spatial dependence, which tend to undermine p-values in statistical analyses [41]. Also, weather and dengue variables were designed to represent direct mechanistic relationships with the number of mosquitoes. Our results evidenced significant effects of rainfall, temperature, and human behavior on the temporal dynamics of Ae. aegypti and dengue in northern Puerto Rico. Materials and Methods Study sites The study was carried out in two separate neighborhoods, each consisting of two contiguous US census tracts of the Metro Area of San Juan, Puerto Rico (Figure 1). One pair of bordering census tracts was located in urbanization “Villa Carolina”, Carolina municipality (VC; 9240 persons and 1996 buildings; 18°23′52″N 65°57′26″W; US Census 2000).We studied a second pair of adjacent census tracts approximately 3 km from VC: urbanization “Extension El Comandante” in Carolina municipality and urbanization “El Comandante” in San Juan municipality (EC; 6951 persons and 1979 buildings; 18°24′02″N 65°59′30″W; US Census 2000). We will be referring to these latter two census tracts as “El Comandante” in most of the report. 10.1371/journal.pntd.0001378.g001 Figure 1 Map of the study areas. The map shows the municipalities of San Juan city, Puerto Rico and the location of the airport in relation to the two neighborhoods investigated. Each neighborhood is composed of two adjacent census tracts. Carolina Municipality has a spatial, insecticide spraying program (truck-mounted Ultra Low Volume equipment) that is active throughout the year, whereas the San Juan Municipality uses a similar insecticide spraying technique but mostly around notified cases of dengue. Thus, both census tracts in VC were subjected to frequent ULV insecticide treatments, whereas in EC only the census tract that belongs to the Carolina Municipality may have been regularly sprayed. Establishing the frequency and coverage of insecticide spraying was attempted but unsuccessful. Weather variables Total annual rainfall and mean daily temperature were 1,388 mm and 27.0°C, respectively at the nearby (4–7 km) Muñoz-Marin International Airport in 2008. Rainfall in the San Juan area occurs year round, with a short, relatively dry season ( 0.05) between EC (3.13±0.90) and VC (3.17±0.77). By contrast, the mean number of adult Ae. aegypti captured per BG trap per day by the end of June 2008 was larger in EC (10.38±1.10) than in VC (6.39±0.44), though the difference was not statistically significant at α = 0.05 (log10+1 transformed, unequal variances assumed, t = 1.79; df = 302; P>0.05). The pupal sex ratios (male: female) were 0.9∶1 in EC and 1∶1 in VC. Adult Aedes aegypti dynamics in BG-Sentinel traps A total of 20 mosquito species was captured in the study sites from November 2007 to December 2008 in BG traps (Table 1). The most abundant mosquito species were Culex quinquefasciatus and Aedes aegypti. The number of adult Cx. quinquefasciatus collected in EC was several times larger than in VC (Table 1). We observed that several houses in the eastern part of EC had septic tanks, but we could not sample them. The number of adult males of Ae. aegypti captured in VC was half of that captured in EC (Table 1). The overall mean numbers of Ae. aegypti females and males per trap per day were 4.76±0.22 (±95% CI) and 4.06±0.29, respectively in EC (n = 3059 trap days). There were 3.80±0.14 females and 2.13±0.11 males per trap per day in VC (n = 3048). The results of a linear mixed model comparing the average number of females per trap per day (log10-trasformed) between the two study sites for the 20 temporal observations every three weeks, with ADJRAIN and ADJTEMP as covariates, indicated significant differences between sites (F = 15.9; P 0.05). 10.1371/journal.pntd.0001378.g004 Figure 4 Temporal changes in rainfall, mosquitoes, and dengue. Panel A shows changes in adjusted rainfall (mm), number of Aedes aegypti females per BG-Sentinel trap per day, number of eggs per CDC ovitrap per day, and adjusted dengue incidence (cases per 100000 inhabitants) in “El Comandante” (EC) and Panel B shows these parameters in “Villa Carolina” (VC), San Juan city, Puerto Rico from October 2007 to December 2008. 10.1371/journal.pntd.0001378.g005 Figure 5 Relationships between mosquitoes and rainfall. Panel A presents the number of female Ae. aegypti per BG-Sentinel trap per day versus adjusted rainfall (mm) for each sampling date in “El Comandante” (EC) and “Villa Carolina” (VC), San Juan city, Puerto Rico, and Panel B shows the number of eggs per CDC ovitrap per day versus adjusted rainfall in each neighborhood. The corresponding correlation coefficients and Type I error probabilities are presented next to the location. Aedes aegypti dynamics in ovitraps Overall average number of eggs/ovitrap/day was similar in EC (31.94±1.07; n = 1745; Total eggs = 55743) and VC (29.76±1.02; n = 1705; Total eggs = 50737). A linear mixed model did not find significant differences in the average number of eggs per ovitrap between neighborhoods (F = 0.19, P>0.05). The number of eggs in the ovitraps was correlated with the number of females captured per BG trap in EC (r = 0.61; P 0.05). Dengue dynamics Dengue prevalence during the period of study was similar in San Juan (71 cases per 100000 inh.) and Carolina (77 cases per 100000 inh.) municipalities. Dengue incidence reached maximum values in San Juan municipality during the beginning of 2008 and in September 2008; in both cases, peak dengue incidence followed peaks in female mosquito density in EC (Figure 4A). The peaks of mosquito density observed in June 2008 were not associated with a large increase in dengue incidence (Figure 4). There were also two maxima of dengue incidence in Villa Carolina municipality; one at the end of December 2007 and the other in November 2008 (Figure 4B). Dengue incidence reached the lowest levels by the end of the “drier season” (San Juan) or beginning of the rainy season (Carolina; Figure 2, 4). Dengue incidence was positively and significantly correlated with the number of female Ae. aegypti in both neighborhoods (Figure 6A). Similarly, dengue incidence was positively correlated with the number of eggs per ovitrap in EC but did not reach statistical significance in VC (Figure 6B). It can be noted that mosquito density never reached values below two in BG traps or below ten in ovitraps during the study (Figure 6). With the exception of one sample in VC (May 2008), all vector density values observed in this study were associated with dengue incidence. 10.1371/journal.pntd.0001378.g006 Figure 6 Relationships between dengue incidence and mosquitoes. Panel A presents dengue incidence (cases per 100000 inhabitants) as a function of the number of female Ae. aegypti per BG-Sentinel trap per day for each sampling date in “El Comandante” (EC) and “Villa Carolina” (VC), San Juan city, Puerto Rico, and Panel B shows dengue incidence as a function of the number of eggs per CDC ovitrap per day in each neighborhood. The corresponding correlation coefficients and Type I error probabilities are presented next to the location. Discussion Ae. aegypti dynamics and dengue endemism The pupal surveys conducted during the drier and wetter parts of the year in San Juan city during the end of 2007 and 2008 revealed that most pupae were produced in containers whose water content was managed by humans: water storage vessels (5-gal pails, barrels), leaking-water meters, and plant pots. This is the first report on water meters as productive aquatic habitats for Ae. aegypti. Containers that were mainly filled with water by rains increased during the rainy season, most notably used tires (Figure 3). Therefore, it would seem that the reason why the Ae. aegypti population did not reach very low levels at a time when rainfall was scant (Figure 4) was due to the production of mosquitoes in containers managed by people. This phenomenon was remarkably similar in both neighborhoods. A longitudinal study that investigated the temporal changes in aquatic habitats and oviposition of Ae. aegypti in northern Venezuela showed similar dynamics, although in such study the main habitats producing mosquitoes during the prolonged dry seasons were 55-gal barrels used for water storage [14]. Lambdin et al. [45] found that buckets, barrels, and tires were the most productive containers during the dry and wet seasons in American Samoa. It is perplexing that Ae. aegypti mosquitoes were being produced in water storage containers and tires in Puerto Rico because the country has a reliable supply of tap water and a tire-recycling program. Our results seem to add support to the hypothesis that dengue endemicity (uninterrupted transmission) is favored by the persistent production of mosquitoes in containers whose water is managed by people. Weather effects The present longitudinal study of Aedes aegypti in neighborhoods of San Juan City revealed significant changes in the number of adult mosquitoes in BG traps that were positively associated with rainfall and temperature (Figures 4, 5). The typical bimodal rainfall pattern of northern Puerto Rico [46] was associated with two main peaks in mosquito abundance, particularly in EC (Figure 4). The first peak in mosquito density was associated with a small hump in the figure of dengue cases in June and the second peak in mosquito density preceded maximum dengue incidence later in the year (Figure 4). It could be observed that dengue incidence decreased after reaching its peak, in association with concomitant reductions in rainfall and Ae. aegypti density (Figure 4). Thus, the present study is in agreement with a previous longitudinal study of Ae. aegypti in Puerto Rico that showed significant effects of rainfall on larval indices and dengue incidence [36]. These authors suggested that such a relationship was patent in the drier cities of the island because rainfall is more seasonal. The lack of significant effects of weather on adult Ae. aegypti in a previous study in San Juan city in northern Puerto Rico, where rainfall is more uniformly distributed throughout the year [38], seemed to confirm such hypothesis. However, in the present study Ae. aegypti dynamics was strongly driven by weather and human activity. The only previous study of Ae. aegypti dynamics using BG traps in relation to weather variables was done in tropical Cairns, Australia [47]. The authors used weekly values of temperature, rainfall, and relative humidity and explored varying time lag effects using multiple regression analyses. They did not find significant effects of rainfall at any time lags, but significant effects of relative humidity with a lag of two weeks and mean daytime temperature at lag 0. Oviposition (eggs/ovitrap/day) was also influenced by rainfall and this variable was correlated with the abundance of Ae. aegypti females in BG traps. Both female adults and oviposition were correlated with dengue incidence. Such an association could be more easily seen in EC (Figures 4, 6). Mogi et al. [48] reported marked seasonal changes in Ae. aegypti oviposition associated with the rainy season, with maximum numbers occurring one month before the peak of dengue cases in northern Thailand. It would appear that ovitraps could be used as inexpensive indicators of the risk of dengue transmission, although perhaps these traps may not consistently reflect overall mosquito population abundance. BG traps capture Ae. aegypti in various physiological states but underestimate some groups, such as nulliparous females [49], whereas ovitraps reflect the number of ovipositing or gravid females [50], [51]. Ovitraps have been used to monitor temporal changes in Ae. aegypti populations and to assess the impact of insecticide treatments [52]. The use of ovitraps for surveillance purposes can be made easier because the number of eggs per ovitrap can be significantly related to the percentage of positive traps using an empirical model, so once the model is developed there is no need to count the eggs collected in the traps [53]. We propose a closer examination of the value of standard CDC ovitraps or similar devices that monitor gravid females as indicators of dengue transmission. The results obtained in this investigation suggest that the levels of Ae. aegypti females per BG trap or the number of eggs per ovitrap should be reduced well below two and ten, respectively to prevent dengue transmission (Figure 6). Mogi et al. [48] did not observe dengue hemorrhagic fever cases in Chiang Mai, Thailand when the number of eggs in ovitraps baited with water was two or less. Variation between neighborhoods Pupal surveys indicated rather similar entomological indices and numbers of pupae of Ae. aegypti in both study areas in June, 2008. However, the number of adult Ae. aegypti captured in BG traps was significantly greater in EC than in VC. Among the main differences observed between the two areas was the fact that there was a deficit of male Ae. aegypti mosquitoes captured in BG traps in VC (Table 1). This unbalance in the proportion of males was not observed in the pupal surveys, suggesting that it resulted later on during the adult life of the mosquitoes. Additionally, an outstanding difference in dynamics between the two sites was observed in August and September, when it appeared like the second peak in mosquito abundance in VC was trimmed in comparison with EC (Figure 4). A hypothesis explaining this difference between neighborhoods is that there was more frequent application of truck-mounted, Ultra Low Volume (ULV) spraying of insecticide in VC, particularly during the dengue season. Unfortunately, we could not gather enough data to document when and where insecticides were applied in each neighborhood. Yet, of the 78 municipalities of Puerto Rico, Carolina municipality was the one that spent the most in the chemical control of mosquitoes [54]. Focks et al. [55] reported that truck-mounted ULV spraying killed an average of 88% of the males and 30% of the females in New Orleans, Louisiana. This greater impact of insecticide spraying on Ae. aegypti males is consistent with our observations in VC. However, if insecticide spraying actually reduced the number of mosquitoes captured in BG traps, it did not significantly affect the number of eggs in ovitraps since there were no differences between neighborhoods. The Carolina municipality also carries out insecticide spraying to eliminate other biting mosquito species that come from nearby marshes and mangroves (Table 1). An additional difference between neighborhoods was the comparatively larger density of Culex quinquefasciatus observed in EC (Table 1), which was probably due to the presence of septic tanks that were observed in some of the less urbanized sectors of EC. It would appear that current levels of mosquito population control were insufficient to make a difference in terms of dengue transmission between municipalities because dengue prevalence during the study was similar in both administrative areas.
