Large-scale longitudinal cohort studies are necessary to characterize temporal and geographic variation in Aedes aegypti (L.) (Diptera: Culicidae) production patterns and to develop targeted dengue control strategies that will reduce disease. We carried out pupal/demographic surveys in a circuit of ≈6,000 houses, 10 separate times, between January 1999 and August 2002 in the Amazonian city of Iquitos, Peru. We quantified the number of containers positive for Ae. aegypti larvae and/or pupae, containers holding pupae, and the absolute number of pupae by 4-mo sampling circuits and spatially by geographic area by using a geographic information system developed for the city. A total of 289,941 water-holding containers were characterized, of which 7.3% were positive for Ae. aegypti. Temporal and geographic variations were detected for all variables examined, and the relative importance of different container types for production of Ae. aegypti was calculated. Ae. aegypti larvae and pupae were detected in 64 types of containers. Consistent production patterns were observed for the lid status (lids: 32% wet containers, 2% pupal production), container location (outdoor: 43% wet containers, 85% pupal production), and method by which the container was filled with water (rain filled: 15% wet containers, 88.3% pupal production); these patterns were consistent temporally and geographically. We describe a new container category (nontraditional) that includes transient puddles, which were rare but capable of producing large numbers of pupae. Because of high variable pupal counts, four container categories (large tank, medium storage, miscellaneous, and nontraditional) should be targeted in addition to outdoor rain-filled containers that are not covered by a lid. The utility of targeted Ae. aegypti control is discussed, as well as the ability to achieve control objectives based on published but untested threshold values.
Contemporary dengue control programs are increasing the emphasis on community participation in suppression of Aedes aegypti (L.) to manage disease by elimination (source reduction) or treatment (larvicide) of Ae. aegypti-infested containers located in or around households (Gubler 1989, Chan et al. 1971, Gubler and Casta-Velez 1991, PAHO 1994). Because resources for vector control continue to diminish worldwide, dengue control programs, whether run vertically by government health departments or directly by community members, are being forced to consider the relative productivity of distinct larval habitats to 1) provide more precise estimates of entomological risk for human dengue infection and 2) devise streamlined surveillance programs and customized treatment regimes for unique situations in each location or region.
Several investigators have noted the poor relationship between traditional Stegomyia indices with dengue transmission because they do not account for differences in the production of adult Ae. aegypti from individual containers (Tun-Lin et al. 1995, 1996; Focks and Chadee 1997; Reiter and Gubler 1997). Early efforts to quantify these differences involved direct quantification of immature forms (larvae and pupae) by house or per 1000 persons (Chan et al. 1971, Bang et al. 1981, Chan 1985) and later by the adult productivity index (API), which represents the sum of different container types weighted by the mean number of larvae associated with each individual container type (Tun-Lin et al. 1996). Logistic problems counting individual immature stages and in some cases poor correlation to adult mosquito densities limited the application of these indices on an operational basis (Tun-Lin et al. 1996, Reiter and Gubler 1997). Tun-Lin et al. (1995) also defined the concepts of key-containers and key-container types as individual containers of any type or class that contribute significantly to an urban Aedes problem. They cite an example in Fiji where tires and drums comprise only 10–20% of the total water-holding containers but account for 83–99% of the Ae. aegypti produced. The implicit assumption of this approach to dengue control is that identification of key container types will lead to site specific and cost-effective control programs if treatment is effectively focused on key container types that produce most of the adult Ae. aegypti.
The pupal and demographic survey methodology quantifies pupae rather than larvae (Focks et al. 1993a, b; Focks and Chadee 1997) because in theory it is more practical to count the absolute number of Ae. aegypti pupae than other life stages (Southwood et al. 1972, Focks et al. 1981), and because pupal mortality is slight and well-characterized, the number of pupae per person is highly correlated with the number of adults per person (Focks et al. 1981, 1995). Using this methodology, the relative importance of a container type is defined as the product of the container abundance multiplied by the average standing crop of pupae (pupae per wet container). Proportions of pupal production can then be calculated directly from the resulting estimates of pupae per hectare for each individual container type, identifying key containers that account for high proportions of total production. Focks (2003) has proposed that estimates of average standing crop of pupae for different container types could be extrapolated to larger areas so that risk assessment and control strategies could be streamlined to include only counts of containers by type and numbers of humans. The primary concern with this approach is the consistent observations that the distribution of Ae. aegypti-infested containers and households can be highly clustered through time and space within communities, making results from one-time small scale surveys sensitive to sampling error and variation (Tun-Lin et al. 1995, 1996; Focks and Chadee 1997; Getis et al. 2003; Morrison et al. 2004).
The history of dengue virus transmission has been well documented in Iquitos by the U.S. Naval Medical Research Unit (NAMRU) infectious disease field surveillance program (Watts et al. 1999). Dengue-1 and dengue-2 (American strain) were first detected in Iquitos during 1990 and 1995, respectively. At the end of 1995, the overall dengue seroprevalence rate, for dengue-1 and dengue-2 combined, was 58% (D.W., unpublished data). During December 2001, dengue-3 and in February 2002 the Asian strain of dengue-2 was isolated for the first time in Iquitos (M.S., unpublished data).
The entomological surveys we report here were carried out as part of a 5-yr prospective cohort study that monitored dengue virus transmission and Ae. aegypti population densities in the homes and neighborhoods of ≈2,400 people, representing a sample of >20% of the city blocks, in the most populated areas of Iquitos. The data set includes 3.7 yr of daily mosquito surveys carried out by highly skilled entomological technicians under constant supervision for assurance of quality control. The unprecedented size and scope of our data set allow detailed evaluation of container productivity concepts summarized above.
The study reported here had three objectives. First, we characterized the temporal and geographic patterns of potential Ae. aegypti larval habitats (water-holding containers), containers positive for Ae. aegypti larvae, pupae, or both, and absolute pupal abundance. Second, we identified the most important container types and container characteristics associated with adult Ae. aegypti production in Iquitos. Third, we discuss the implications of our findings for both estimating entomological risk for dengue virus transmission and the potential for container specific target control programs.
Materials and Methods
Our study was conducted in Iquitos City located in the Amazon forest (73.2′ W, 3.7° S, 120 m above sea level) in the Department of Loreto, northeastern Peru (Fig. 1). Iquitos has been described in detail in earlier studies in the city (Hayes et al. 1996, Watts et al. 1999, Getis et al. 2003, Morrison et al. 2004, Schneider et al. 2004). Iquitos is comprised of four districts: San Juan, Maynas, and Punchana running from south to north and Belen on the east (Fig. 1). We restricted our activities to an area of ≈16 km2 in the districts of Maynas, Punchana, and small portions of Belen and San Juan.
Our study area was divided into eight similarly sized geographic zones based on known neighborhoods served by distinct health centers (Fig. 1). In the three most northern zones—Punchana (PU), Maynas (MY), and San Antonio (SA)—more than one-half of the houses belong to the district of Punchana, which has its own local government and services. The five remaining zones—Putumayo (PT), Iquitos (IQ), Morona Cocha (MC), Bagazan (BG), and Tupac Amaru (TA)—belong predominantly to the district of Maynas. The zones of Tupac Amaru (TA), Bagazan (BG), and Putumayo (PT) each have areas where houses are seasonally flooded. Homes in those areas are either constructed on stilts, float on the water as flooding occurs, or the first story is abandoned and the second story is occupied during flooding. Access to flooded houses is limited to canoe or a series of wooded walkways that are constructed between homes. Piped water is available throughout the city. Water availability seems to be correlated with proximity to the city water treatment plant, which is located in the southwestern part of the zone of Tupac Amaru.
Entomological data were collected between January 1999 and August 2002 as part of a prospective longitudinal cohort study serologically monitoring dengue virus infections in a cohort of ≈2,400 students and their families. Serological results will be presented in future publications, but the design of the study was linked directly to the selection of the human participants in the study. An entomological survey circuit was established that ensured that we sampled each serological participant’s household as well as neighboring households (same block or opposing side of street). Each circuit contained 17 short (1–2-wk) surveys or sampling groups that alternated between zones in the north, central, and southern parts of the city. The duration of a complete circuit was ≈4 mo, with the exception of the first year of the study (Table 1). Selection of study participants was designed to obtain a geographically stratified sample of households. Briefly, each of the eight zones described previously was subdivided into a grid of approximately five equal areas. Three blocks from each of these areas was selected randomly, yielding a total of 15 sample blocks in each of the eight geographic zones. During the first 9 mo of the project (January-September 1999), areas that included blocks with study participants were identified, mapped in the field, and information was captured in our geographic information system (GIS) for the city, and a unique code was painted on the front of each house that matched the house code in our GIS database. Subsequently, entomological surveys were carried out in households designated by the lead author. On blocks or opposing sides of streets where participants lived, all accessible houses were sampled. In a few cases where only one household on a block contained a study participant, only a single house was sampled on that block. Thus, each circuit represents replicates of ≈6,000 households (range 5,721–6,466) from the same areas of the city, divided into 17 groups (Table 1). Within a circuit each house was sampled once. Circuits were not exact replicates primarily because of differences in access to houses (if someone was home at the time of the survey) and differences in the composition of the study cohort over time (drop out of study participants and recruitment of new participants). As participants dropped out or entered the study, some blocks were dropped or added to the survey circuit.
Entomological surveys were carried out daily, Monday through Friday, between the hours of 0700 and 1300 by six two-person collection teams by using a survey methodology described in detail in previous publications (Morrison et al. 2004, Schneider et al. 2004). Briefly, each survey consists of a brief questionnaire, inspection of household for water-holding containers, and collection of adult mosquitoes using a back-pack aspirator. The survey duration averaged 9 min (range, 2–170 min) and varied with the number of adult mosquitoes and containers detected with immature Aedes.
All water-filled containers were examined for immature stages of Ae. aegypti. To do this containers were measured (diameter, length, width, and height), categorized (64 container types collapsed into 14 broad categories; see Tables 2 and 3), scored for solar exposure (proportion of day with direct sunlight, 0–1), fill method (manually filled/frequency, rain-filled, rain-filled with aid of rain gutter or roof), and whether the container had a lid. For Ae. aegypti-positive containers, the number larvae were estimated (1–10, 11–100, and >100), and the total number of pupae were counted. All pupae and a sample of larvae were placed in a whirl pack plastic bag (Nasco, Fort Atkinson, WI) and labeled with a house and container code so that specimens could be linked to the exact container and household from which they came. All larvae, pupae, and adults were transported the same day they were collected to the field laboratory in Iquitos for processing as described in Morrison et al. (2004).
A geographic information system, using ARC/INFO and ARC/VIEW (ESRI, Redlands, CA) software, was developed for the city of Iquitos (Getis et al. 2003). All entomological data collected in our surveys could be joined to their geographic coordinates via a spatially specific house code.
Our analysis focused on differences in relative abundance of immature forms of Ae. aegypti. Dependent variables we examined were water-holding containers independent of their infestation status with Ae. aegypti (hereafter, referred to as containers [CT]); containers positive for larvae and/or pupae of Ae. aegypti (hereafter referred to as positive containers [PC]), containers containing Ae. aegypti pupae, a subset of positive containers (hereafter referred to as pupae-positive containers [PPC]); and the absolute number of Ae. aegypti pupae [PU] collected from containers. We summarized these variables by hectare (HA), household (HSE), or person (PER) surveyed for each of the 17 sampling groups within each circuit (N = 170). The total hectares surveyed and persons are the sum of the area of each housing lot and the total number of residents living in each of the households, respectively. Thus, the averages we present represent the mean CT/HA, PC/HA, PPC/HA, PU/HA, CT/HSE, PC/HSE, PPC/HSE, PU/HSE, or PU/PER of the sampling groups over container classification variables (container type, fill method, location, lid, and volume), space (by zone and sampling group), and time (circuit).
To identify differences in mean abundance of water-holding, positive, and pupae positive containers and absolute numbers of pupae, we used analysis of variance (ANOVA) with the general linear models procedure (PROC GLM) of SAS (SAS Institute 1989). Models were constructed for each transformed-dependent variable (i.e., LnCT/HA, LnPC/HA, LnPPC/HA, LnPU/HA, and SqPU/PER) with the following independent variables: sampling circuit (J-S99, JL99-J00, J-AP00, …, MY-AG02), geographic zone (BG, IQ, MC, MY, PT, PU, SA, TA; Fig. 1), and sampling group nested within zones (17 groups: BG01, BG02, IQ01, MC01, …, TA01, TA02), container type (see categories in Table 4), fill method (Table 6), lid status (lid, no lid), location (indoors, outdoors), volume, and a combined container categories based on the three previous characteristics (Fig. 7). Least square means (LSMeans) were used to test differences among mean rates within main effects and interactions terms; when appropriate, main effect variables were then classified into groups and differences were tested using contrast statements in PROC GLM (SAS Institute 1989). A more detailed description of ANOVA model construction and results is presented in the Appendix. Because of the extensive size of the data set, many interaction terms were significant; in this article, we focus on significant differences that have operational implications. We refer readers who have interest in the finer details of the geographic and temporal interaction terms to the appendix.
During our entomological surveys, 289,941 water-filled (wet) containers were observed and characterized in an area with an average of 63 houses and 420 persons per hectare sampled. We detected an average of 4.85 wet containers per household surveyed (308 containers per hectare). Of those, 7.3% contained Ae. aegypti immature stages and 3.6% were positive for pupae. The majority of wet containers was located inside houses (57.4%), did not have lids (68.0%), and was filled manually by occupants of the household with water from a pipe or well (85.0%). In contrast, only 17.9% of positive containers were located inside houses, 97.5% did not have a lid, and 19.5% were manually filled. Wet containers were generally infested only with Ae. aegypti, but 5.5% of the positive containers had larvae or pupae from the genera Culex.
Temporal Patterns in Container Abundance and Infestation Rates.
Overall, container abundance showed little variation over time with the exception of the first year of the study (circuits 1 and 2), results from survey circuits ranged from 283 to 340 containers per hectare (Fig. 2 A) and 4.6–5.6 containers per house. After 1999, container abundance never exceeded 306 containers per hectare.
There were significant differences (P < 0.0001) in temporal abundance patterns for positive containers (Fig. 2B) and pupae-positive containers (Fig. 2C), which fell into three groups: high (22–27 PC/HA; 11–14 PPC/HA), medium (18–20 PC/HA; 10 PPC/HA), and low (14–16 PC/HA; 6–7 PPC/HA). Similarly, pupae per hectare (Fig. 2D), and pupae per person (Fig. 2E) fell into three slightly different groups: high (215–219 PU/HA, 0.50–0.53 PU/PER), medium (155–182 PU/HA, 0.37–0.44 PU/PER), and low (83–119 PU/HA, 0.21–0.28 PU/PER). Before 2002, abundance patterns seemed seasonal, with highest abundance occurring between January and April 1999 (only for PC and PPC) and 2000. In 2001, however, there was an unusual, perhaps early increase in Ae. aegypti densities during the September-December trimester, which immediately preceded a dengue-3 epidemic in 2002. There was a subsequent decrease in container and pupal abundance in 2002 that corresponded to intensified efforts by the Ministry of Health to control Ae. aegypti in response to the dengue-3 epidemic.
Geographic Patterns of Container Abundance and Infestation Rates.
Wet container, positive container, pupae-positive container, pupal abundance, and house density varied significantly among the eight geographic zones in Iquitos (Fig. 3, A-C). The number of containers per hectare adjusted for differences in household density fell into four groups: highest (382 CT/HA), high (330 CT/HA), medium (282–312 CT/HA), and low (251–260 CT/HA) abundance (Fig. 3A). This pattern was distinct from that observed for infested containers.
The zone Maynas had significantly more Ae. aegypti-positive containers (Fig. 3B), pupae-positive containers, and pupae (Fig. 3C) than other zones. Zonal means for positive containers per hectare formed three groups: high (24–29 PC/HA), medium (19–21 PC/HA), and low (16–18 PC/HA) abundance (Fig. 3B). The low abundance zones (Bagazan, Putumayo, and Tupac Amaru) each included houses in proximity to a major river and were flooded seasonally.
Although similar to positive containers, pupae-positive containers, pupae per hectare, and pupae per person showed a slightly different pattern. The abundance patterns were similar for both pupae per hectare and pupae per person: Maynas had significantly more pupae than other zones (Fig. 3C); Bagazan, Tupac Amaru, and Iquitos had the lowest average pupal densities per hectare, and Bagazan and Tupac Amaru had significantly fewer pupae per person than other zones.
There was considerable variation in water-holding containers in Iquitos. We identified 64 container categories during entomological surveys that we subsequently collapsed into 14 broad categories that are described in Tables 2 and 3. Plastic containers (243 CT/HA) were far more abundant than any other category; they were observed 11–19 times more often than cooking containers (22 CT/HA) and medium storage containers (13 CT/HA), which were the next most abundant container types. Large tanks, bottles, and tires were found at 4.3–7.5 containers per hectare. The remaining container categories (miscellaneous, wells, nontraditional, pet dishes, cans, flower vases, and natural containers) were relatively rare, ranging from 0.6 to 3.4 containers per hectare.
Nontraditional containers (as defined in Table 2), tires, and natural containers were most likely to be infested with immature Ae. aegypti; 64.2, 59.6, and 58.4% were infested, respectively (Tables 2 and 3, see container indices [CI]). Infestation rates were somewhat lower in cans, miscellaneous, flower vases, and bath containers, ranging from 32.4 to 45.1%. Despite their high relative abundance, plastic containers had the lowest infestation rate of all container categories (3.6%). Indices for remaining container types ranged from 6.5 to 28.7%.
Table 4 summarizes the relative abundances of Ae. aegypti-infested containers during the entire study period based on backtransformed (geometric mean) adjusted least square means from our ANOVA analysis, in contrast to arithmetic means presented in Tables 2 and 3. Furthermore, the seven least productive categories defined in Table 3 have been collapsed into an "other" category for statistical analysis. Containers showing large differences between the arithmetic and geometric means are most likely to be "key containers" (Tun-Lin et al. 1995) because of the high variability in pupal counts. The mean number of positive containers per hectare was highest for plastic containers, followed by the "other" category, and then by tires, medium storage, and cooking containers, whose mean values were not statistically different from one another (letters indicate statistically significant differences by LSMeans, Table 4).
Variation in the number of pupae per hectare was very high, reflected by the large range of values among the sampling groups and the impact that variable transformation had on means for different container types. Large tanks and nontraditional containers showed the highest variation in Ae. aegypti production, whereas geometric means (Table 4) underestimated overall production by four- to six-fold (compare PU/HA between Tables 2 and 4). Medium storage and the miscellaneous container categories also exhibited significant variation in pupae per hectare.
If we define a key container as one containing ≥500 pupae, ≈2% of the pupae-positive large tanks and nontraditional containers would be classified as key containers. Of the key large tanks, three of seven had >1000 pupae on the day of the survey, and four of seven contained Culex pupae in the same container (Table 5). Of the seven nontraditional containers where ≥500 pupae were collected, only one had >750 pupae; a puddle on the floor of the house (Table 5). In fact, five of these containers were described as puddles of relatively low volume (3–60 liters), whereas one was a large puddle (≈180 liters) and one was a large hole in the ground (≈1,500 liters). The puddles in these cases occurred on cement floors. As observed with large tanks, five of seven of these nontraditional key containers also contained Culex pupae. A total of four medium storage containers and two large (286–740-liter) miscellaneous containers had ≥500 pupae.
Ae. aegypti pupae are produced and adults emerge from all types of water-holding containers (Table 4). The most productive container type was the broad category of plastic containers (buckets, basins, tubs, and food storage containers), which comprised 79% of all water-holding containers but constituted only 37% of the positive and pupae-positive containers and 27% of the total pupae collected. Plastic containers accounted on average for 40% of all positive container and pupal production, significantly more than any other category (LSMeans adjusted P < 0.05). Relative abundance patterns of plastic containers were consistent among circuits and across geographic zones, despite small but statistically significant temporal and spatial difference in magnitude.
Of the other categories, only tires comprised >10% of the total positive containers in Iquitos. In contrast to plastic containers, tires account for 1% of all containers independent of the circuit or zone sampled, but 11% of positive containers, 13% of pupae-positive containers, and 12% of pupal production in Iquitos. Tires account for 9–13% of pupae produced from positive containers across circuits and 8–14% across zones. Pupae production was variable, ranging from 8 to 21% and 6 to 22% across circuits (Fig. 4 A) and zones (Fig. 4B), respectively.
If we define the relative importance (Focks and Chadee 1997) of a container solely on the absolute number of pupae, medium storage and large tanks follow plastic containers, accounting for 16 and 14% of pupal production, respectively (Tables 2 and 4); these container types comprise nearly 6% of all water-holding containers, 9% of positive containers, and 13% of pupae-positive containers. The number of medium storage and large tanks per hectare, as well as positive containers in each category was remarkably consistent overtime, but exhibited significant geographic variation. For example, Maynas (14 CT/HA) and Punchana (10 CT/HA) have significantly more large tanks than other zones (3–8 CT/HA). Overall, the contribution of these two container types combined to positive container production was highest in the three northern zones of Maynas, Punchana, and San Antonio (17–18%) compared with the five remaining zones (7–11%). The same pattern was observed for pupae, with medium storage containers and large tanks accounting for 24–33% of production in the three northern zones compared with only 8–15% in the five other zones (Fig. 4B, significant LSMeans). These differences further illustrate that using positive container production patterns can underestimate the importance of certain types of containers. Rounding out the seven most productive container types (Table 2) were the cooking, miscellaneous, and nontraditional container categories.
Table 3 summarizes the characteristics of seven container categories that contributed between 0.5 and 2.2% of total pupae collected during the study. Included are containers that are often recommended for targeted control by public information campaigns such as bottles, cans, flower pots (vases), and pet dishes as well as others that also should be considered, such as parts of toilets (bath), natural items, and wells. Although none of these containers would be considered a key container type individually their combined contribution is second to plastic containers. Collectively, they account for 6% of all containers, 19% of positive containers, 17% of pupae-positive containers, and 8% of all pupae detected during the study period. This relationship was consistent over time and space.
During the course of our study, we observed Ae. aegypti larval habitats in sites where they were not expected and often in sites where they coexisted with Culex larvae. Some merit detailed description. First, nontraditional category, which only contributed 0.4% of the total containers, produced nearly 10% of the pupae. Containers in this category were generally characterized by an unusual shape or lack of form, most notably puddles accumulating on floors inside or outside houses (usually cement) or in random accumulations of water on plastic tarps or sheet metal. Also included were rain gutters, PVC tubing, drains, and catch basins, which have often been overlooked in surveys previously. As discussed above, the quantity of pupae collected from this container class was highly variable; when observed this category had a high infestation rate (CI = 64%). Second, within the bath category, a larval habitat commonly observed was the water inlet hole, a small depression where the tank usually connects to the toilet. This small cylindrical hole would often accumulate rainwater or splash. Of a total of 1,881 of these sites, 523 (30%) and 181 (9.6%) were positive for Ae. aegypti larvae and pupae, respectively. We also inspected a number of catch basins and open flowing drains where water had become stranded, which often contained foul-smelling water. Of the 112 inspected, 70 (63%) were positive for larvae and 39 (34%) for pupae only.
Container management was an extremely important factor affecting production of Ae. aegypti. Containers reported by residents as manually filled (264 CT/HA) were 9 times and 13 times more abundant than those filled by rainwater, either passively (28.4 CT/HA) or actively (19.5 CT/HA), respectively (P < 0.0001). Although only 14% of wet containers were rain-filled, they accounted for 81% of positive container, 85% of pupae-positive container, and 88% of pupal production (Table 6). On average, 49% of rain-filled containers were infested with immature Ae. aegypti compared with 1.7% of manually filled containers. Rain-filled containers were divided into two groups: passively filled (unmanaged) and those filled with rainwater collected actively from roof runoff, rain gutters, or roof leaks. The latter category shows the highest degree of variation in Ae. aegypti production patterns.
The relative importance of passively filled containers was consistent across the circuits (see Appendix for exception). Within each circuit, however, the abundance of positive and pupae-positive containers filled by rain passively was always significantly higher than those actively rain-filled or those filled manually (LSMeans adjusted α < 0.05). The same trend (passively filled > actively filled > manually filled) was observed within each geographic zone, but the magnitude of this relationship showed significant variation (P = 0.0001).
Pupal production by fill method was consistent overtime (no significant circuit*fill method interaction term), but it did show significant geographic variation, where pupal production from passively rain-filled containers ranged from 56 to 79%; and three zones, Tupac Amaru (79% of pupae), Bagazan (71%), and Iquitos (75%), had relatively more pupal production from passively rain-filled containers than the remaining zones (56–65%). This difference corresponded to a higher contribution of pupae from actively rain-filled containers.
Ae. aegypti larval habitats were observed both inside and outside homes in Iquitos, but only 2.4% of indoor containers compared with 14.6% of outdoor containers were infested. Although the majority of wet containers (58%) are located indoors, 82% of positive containers, 85% of pupae-positive containers, and 86% of pupae were located outdoors (Table 6).
The overall abundance patterns of indoor and outdoor containers, positive containers, pupae-positive containers, and pupae per hectare were consistent across sampling circuits and geographic zones, where positive and pupae-positive containers and pupae located outdoors were always more abundant than those located indoors (see Appendix for details).
In Iquitos, covering potential larval habitats seems to be an effective method of preventing Ae. aegypti infestation. Unlidded containers were twice as abundant as lidded containers (P < 0.0001; Table 6), but >97% of positive containers, pupae-positive containers, and pupae came from unlidded containers. There was no significant variation in this trend among circuits or zone. Only 0.6% of lidded containers were found infested compared with 11% of unlidded containers.
To determine the impact container vol-ume has on Ae. aegypti production in Iquitos, we classified containers into the following broad volume categories: ≤3, 3–15, 15–50, 50–500, and >500 liters. Containers in the 15–50-liter category (185 CT/HA), which include most plastic buckets and cooking pots, were significantly more abundant than all other categories (Fig. 5 A; P < 0.0001 LSMean adjusted α < 0.05). The container index (percentage of containers infested) was highest in small size categories: 34 and 19% in containers ≤3 and 3–15 liters, respectively, and lowest in the most abundant 15–50-liter category (2%). Between 7% (50–500 liters) and 10% (>500 liters) of large containers were infested. Similarly, positive container abundance was highest in containers <15 liters (Fig. 5B; P < 0.0001), whereas pupae-positive container and absolute pupal abundance were highest in the 3–15-liter category (Fig. 5C and D; P < 0.0001). Pupal production was not significantly different between the 50–500, 15–50, and ≤3 liter categories. Containers >500 liters, ≈2% of the total containers, produced 13% of the total pupae per hectare (27 pupae/HA). The standing crop of pupae was 1.5, 0.2, 0.9, and 3.8 pupae per container for ≤15, 15–50, 50–500, and >500-liter containers, respectively. Relative abundance patterns of wet containers, positive containers, pupae-positive containers, and pupae per HA were consistent both over circuits and zones sampled (no significant interactions).
Ultimately, we must identify container types based on a combination of container characteristics described above that are responsible for a high proportion of overall Ae. aegypti production, but only a small proportion of the total wet containers. The flow chart in Fig. 6 summarizes relative abundance and production of water-holding containers, positive containers, pupae-positive containers, and pupae by the three most important container characteristics described above. Figure 6 clearly shows that by focusing on 1) unlidded, 2) outdoor, and 3) passively rain-filled containers, we can theoretically identify >57% of Ae. aegypti production by examining only 8.9% of all wet containers. If we also included actively rain-filled containers from the same branch of the flow chart and additional 13.7% of positive containers and 20.7% of pupal production would be identified by examining only 13.3% of all wet containers. To evaluate the temporal and geographic variation in these overall production patterns, we created the following combined categories: lidded, indoor rain-filled (includes actively and passively filled), outdoor manually filled, indoor manually filled, outdoor actively rain-filled, and outdoor passively rain-filled. For ANOVA, positive container, pupae-positive container, and pupal abundance in these six categories was highest in passively rain-filled outdoor containers with no lids (11 PC/HA, 6.4 PPC/HA, 85 PU/HA), followed by actively rain-filled outdoor containers with no lids (2.6 PC/HA, 1.29 PPC/HA, 20 PU/HA) (P < 0.0001, LSMeans P < 0.05). Manually filled containers with no lids were equally abundant outside (1.6 PC/HA, 0.6 PPC/HA, 4.6 PU/HA) as inside (1.8 PC/HA, 0.7 PPC/HA, 4.4 PU/HA) (LSMeans adjusted α > 0.05). Positive containers with lids were the least abundant (0.5 PC/HA, 0.2 PPC/HA). Despite the significance of the container category*circuit and container category*zone interaction terms, the relative abundance patterns were very consistent over time and across geographic zones. In all circuits, passively rain-filled outdoor containers with no lids were significantly more abundant than other container types accounting for between 57–65% of all positive containers, and 61–72% of all pupae-positive containers. Geographically, the patterns were nearly identical, with outdoor no lid passively rain filled > outdoor no lid actively filled > other categories > lidded (P < 0.0001; Fig. 7).
Our study represents an initial step in the process of characterizing quantitative relationships between Ae. aegypti population densities and dengue virus transmission in the Amazonian region of Peru. We characterized aquatic sites where immature stages of Ae. aegypti were found by quantifying site-specific abundance and productivity patterns over 3.7 yr and sampling the same sites ≤10 separate times. Although some larviciding campaigns were carried out by the Ministry of Health, these efforts were intermittent and overall coverage was poor. Our data were collected in a city with active dengue virus transmission and represents the largest data set in size and scope for Ae. aegypti of which we are aware. Our primary objective was to identify the most productive containers for Ae. aegypti, so we could develop a targeted control strategy for the city of Iquitos. We first had to establish a practical classification scheme for containers found in Iquitos, and then determine whether the production patterns observed are sufficiently consistent overtime and space to be of practical utility. In the paragraphs below, we address each of these issues and then show how a targeted control strategy based on our data can be used in concert with entomological thresholds to achieve different control goals.
Previously published classification schemes include purely descriptive (e.g., pot, bucket), functional (e.g., disposable, permanent, storage), size-related, or a combination (e.g., outdoor storage drum). Only four of our broad categories accounted for ≥10% of Ae. aegypti production; plastic containers accounted for 37% of all positive and pupae-positive containers and 25% of all pupae. The plastic container category, however, is inappropriate for a targeted campaign because of its low infestation rate (3.6%) and extremely high abundance. For example, only one of every 26 plastic containers is positive; in a city-wide program treatment of negative containers would constitute a waste of public health resources. In Iquitos, larvicides used in permanent containers such as medium storage containers and large tanks, the current intervention strategy, only targets 30% of Ae. aegypti production. This observation illustrates why control efforts have been insufficient in the city. Based on our data, we conclude that other criteria need to be considered, especially those that relate to the location and water management strategies for containers.
Three characteristics, two directly observable and one requiring that occupants of the household be questioned, could be used to identify high-risk containers in Iquitos. For example, targeted control programs that ignored containers with lids would theoretically require treatment or elimination of only 68% of existing containers and would only miss 2.0% of the pupae produced. More than 95% of lidded containers are in the plastic or cooking category where manufactured tight-fitting lids are protective against Ae. aegypti infestation.
Container location also could be used to streamline control efforts. In Iquitos, 42.6% of all wet containers are located outdoors, but they account for >82% of Ae. aegypti production. Eliminating surveys inside houses could have important logistical advantages, because indoor collections require additional time and can cause inconvenience or embarrassment to household occupants. During studies from Australia, routine surveys could not be carried out indoors because of objections from the public (Tun-Lin et al. 1996). If the two Iquitos criteria were combined so that only outdoor containers without lids were targeted, 38% of wet containers would require treatment or removal, with the potential to reduce adult mosquito production by up to 83%.
By simply asking a household member at the time of a survey how a container is filled, high-risk containers can be identified and important information can be provided for community-based source reduction campaigns. Manually filled containers, a surrogate for managed containers, are abundant and account for 85% of all containers but only 20% are positive and 12% produce pupae. The importance of ignoring manually filled containers would be two-fold. First, the cost of larvicide would be lowered significantly. Second, public health messages to the community would not discourage water storage, a necessary practice in Iquitos where 77% of the houses have piped water that is unreliable. A program based on these characteristic alone would have the potential to reduce Ae. aegypti production in Iquitos by 80–89%. If control were focused only on outdoor rain-filled containers without lids, a potential reduction of 72% of positive containers and 78% of pupae could be expected.
In addition to the consistent patterns described above, large tanks, nontraditional, medium storage, and miscellaneous containers require special attention in targeted control programs because of their highly variable production patterns and potential to be key producers (>500 pupae found in a single collection). For example, nontraditional container category, which included transient larval habitats without a well defined size or shape, such as puddles, collections of water in plastic tarps, holes in floors, catch basins, drains, PVC tubing, and rain gutters only contributed 9.5% to the total pupal production but had the highest average standing crop of pupae and thus represents one of the most dangerous development sites. In other geographic locations rain gutters (Montgomery and Ritchie 2002) and subterranean sites (Russell et al. 1997) were similarly found to be important producers of adult Ae. aegypti. Thus, by targeting of the four categories independent of fill method, location, or lid status, an additional 14.7% of all pupal production could be eliminated, increasing an overall decrease pupal production by nearly 92%.
In other regions of the world, Ae. aegypti production patterns can be distinct from those observed in Iquitos. For example, in Thailand Ae. aegypti production in highest in indoor containers with lids (Kittayapong and Strickman 1993). The basic methodology of characterizing production patterns based on container characteristics, however, is extremely informative, providing concrete information on the most efficient ways to reduce mosquito populations. Understanding what kinds of containers are producing Ae. aegypti will provide the information necessary to select containers to treat with larvicides or biocontrol agents (e.g., Mesocyclops;Nam et al. 1998) As many have suggested in previous publications, we endorse the identification of Ae. aegypti production patterns to develop targeted control strategies as an alternative to traditional eradication-based programs (Tun-Lin et al. 1996, Focks and Chadee 1997, Focks et al. 2000, Focks 2003).
An important assumption of a targeted approach, which merits verification, is that container production patterns are sufficiently consistent over time and space that the broad-scale application will be successful. We observed a high degree of variability in Ae. aegypti production patterns among individual surveys; measures of pupae per hectare showed the most variability of the three production estimates (PC/HA, PPC/HA, PU/HA). We identified seasonal and geographic trends in Ae. aegypti abundance; however, the biological significance of these differences were not always apparent. The higher variability in measures pupae per hectare obscured some of the seasonal variation observed in positive container abundance, but clearly identified September-December 2001, which corresponded to a marked increase in adult mosquito populations (A.C.M., unpublished data) and immediately preceded the initiation of dengue-3 epidemic (M.S., unpublished data). Geographically, the zone Maynas had consistently high levels of Ae. aegypti infestation, corresponding to the highest seroprevalence rates for dengue-1 and dengue-2 at the initiation of our study in 1999 (A.C.M., unpublished data). Despite differences in immature Ae. aegypti abundance, productivity patterns based on fill method, lid status, and container location were consistent both temporally and for the most part geographically. The only significant interaction was between location and geographic zone, where zones in the northern part of the city had more pupal production from indoor containers associated with a higher average standing crop of pupae from the nontraditional category. This highlights the importance of rare but highly productive containers and how the extrapolation of production patterns estimated from a single survey could provide erroneous information.
Geographic differences in production patterns based on fill method may reflect important differences in the water management practices of the human population. For example, in zones with lower pupal production overall (Bagazan, Iquitos, and Tupac Amaru) had significantly more pupae passively rain-filled containers than in zones with more Ae. aegypti production. We interpret this in the following manner: there is a city-wide problem associated with unmanaged containers that are responsible for >60% of the Ae. aegypti production, but a strong geographic trend between northern and southern zones where additional production in the north is associated with water storage practices. In the three southern zones and Iquitos, 85–93% of the households surveyed had piped water compared with 58–77% in Putumayo and the three northern zones of the city. In Maynas and Putumayo, more production was observed from actively rain filled than manually filled containers, indicating that in areas where water is less available water storage is the source of excess pupae production. The temporal and geographic differences observed were not sufficient to alter the overall target control strategy proposed above. Instead, they highlight the need to carry out rigorous detailed surveys over both time and space, by using pupae per hectare to devise targeted control strategies. Furthermore, we conclude that unless estimates of standing crop of pupae are based are a large, replicate surveys the presence or absence of larvae or pupae per container is a more appropriate estimate of adult production than absolute pupal counts.
If combined with information on entomological thresholds, targeted control strategies could be specifically designed to reduce Ae. aegypti populations to levels that predictably diminish or prevent virus transmission. Entomological thresholds for dengue in Iquitos, based on our entire data set, are not yet available, but preliminary thresholds have been estimated previously by Focks et al. (2000) based on temperature during the warmest months and known seroprevalence rates. If we compare the Ae. aegypti density (0.57 PU/PER) from observed in September-December 2001 surveys where mean and maximum temperatures averaged 26 and 33°C, respectively, to the Focks et al. (2000) threshold estimates, we can identify what targeted control strategies would have theoretically prevented the 2002 dengue-3 epidemic in Iquitos. Threshold estimates ranged from 0.06 pupae per person at 32°C to 1.08 pupae per person at 26°C, assuming dengue-3 seroprevalence rates of 0% before September 2001. If we consider the Focks et al. (2000) threshold estimates for 28°C (i.e., 0.47 pupae per person), the ratio of observed pupae per person to the threshold value was 1.2, indicating that only a 20% reduction in pupae would have been sufficient to prevent transmission. If, however, we consider a more realistic estimate at 32°C (0.06 pupae per person), the ratio would be 8.5, requiring an 89% reduction in pupae. If these scenarios were to withstand empirical testing, two different control strategies could be used. In the case when only 20% reduction was required, a simple larviciding program targeting large tanks and medium storage containers, which are easily identifiable and amenable to chemical control, would theoretically eliminate 30% of pupal production and achieve our control goal by >10%. The second alternative would require elimination or treatment of all rain-filled outdoor containers without lids (78% reduction) plus large tanks, nontraditional, medium storage, and miscellaneous containers (additional 14% reduction).
Although accounting for productivity seems promising in theory, additional work must be done to validate this approach. First, targeted control programs such as those suggested above, need to be tested rigorously and directly in the field. It must be confirmed that removal of a portion of the productive containers will result in the predicted reduction in mosquito density. The possibility must be empirically ruled out that mosquitoes will have a compensatory response to reduced productivity from or removal of certain development sites. Second, although pupae and adult densities have been correlated under field conditions, in many studies correlations were weak (Tun-Lin et al. 1996, Getis et al. 2003) and clearly did not account for temporal differences in mosquito developmental stages. We have strong evidence from Iquitos that entomological risk is appropriately measured at the household level. We have not yet defined the appropriate spatial scale for measuring transmission risk, something that will need to account for human movement and behavior in relation to the blood-feeding behavior of daytime-biting Ae. aegypti (Getis et al. 2003). Despite the limitations of carrying out pupal and demographic surveys, the information gained from them on productivity can have immediate benefits to Ae. aegypti control programs, especially for developing site specific control strategies and developing appropriate health education messages for local populations.
The first model (M01) was constructed to characterize overall temporal and geographic patterns in container, positive container, pupae-positive container, and absolute pupae abundance. Containers or pupae per hectare (or house) surveyed for each sampling group were transformed either by the lne(rate + 1) or square root (rate) to reduce heterosecdasticity and to ensure a normal distribution of the residuals. Independent variables were sampling circuit (J-S99, JL99-J00, J-AP00, …, MY-AG02), geographic zone (BG, IQ, MC, MY, PT, PU, SA, TA; Fig. 1), and sampling group nested within zones (17 groups: BG01, BG02, IQ01, MC01, …, TA01, TA02); and for containers per hectare, the number of households per hectare. We attempted to construct the most parsimonious models, including only main effects and second order interactions whenever possible, while observing normally distributed residuals. LSMeans were used to test differences among mean rates within main effects (circuit, zone, group) and interactions terms; main effect variables were then classified into groups and differences were tested using contrast statements in PROC GLM (SAS Institute 1989).
To examine Ae. aegypti productivity differences among the various characteristics of containers, including container type (M02), fill method (M03), location (M04), lid status (M05), and volume (M06), container and pupae totals per hectare were summarized for each variable within each of the 17 sampling groups and separate ANOVA models were carried out for each characteristic. In addition, a combined container classification (M07)) based on fill method, lid status, and location was developed and tested in the same way. As described above, the most parsimonious model was selected where normality assumptions were still met. As described previously, LSMeans were used to detect significant differences among main effects and interaction terms. Significant differences were considered at a α = 0.05/(total number of comparisons). In the case of positive interaction (e.g., fill method*circuit or lid*zone), we only examined differences among the container characteristic within each circuit, zone, or sampling group to describe any temporal or geographic variation for that particular characteristic.
We thank the residents of Iquitos, Peru, for allowing us to undertake this study in and around the homes. We greatly appreciate support of the Loreto Regional Health Department, including Drs. Carlos Calampa, Jorge Reyes, Ruben Naupay, Carlos Vidal, and Martin Casapia, who all facilitated our work in Iquitos. Gerson Perez Rodriguez supervised the collection and processing of mosquitoes. Entomological surveys were carried out by Jimmy Maykol Castillo Pizango, Rusbel Inapi Tamani, Juan Luiz Sifuentes Rios, Nestor Jose Nonato Lancha, Federico Reategui Viena, Victor Elespuru Hidalgo, Edson Pilco Mermao, Abner Enrique Varzallo Lachi, Fernando Chota Ruiz, Angel Puertas Lozano, Guillermo Inapi Huaman, and Manuel Ruiz Rioja. Jimmy Roberto Espinoza Benevides, Fernando Espinoza Benevides, and Jose Elespuro Bastos carried out data entry and verification. We thank Karla Block for help with the logistics and coordination for this project. Drs. Tadeus Kochel, Kevin Russell, Truman Sharp, Mike Zyzak, George Schoeler, and Lucy Rubio and Roxana Lescano of the U.S. Naval Medical Research Center in Lima, Peru, were instrumental in facilitating these studies. We thank Jason Rasgon and Chris Oman for the work on the Geographic Information System for the city of Iquitos; Sharon Minnick and Linda Styer for simulating conversations, statistical advice, and support; and Ronald Knight for a thorough review of the manuscript. This research was supported by a grant (AI-42332) from the National Institute of Allergy and Infectious Disease.