This investigation received Institutional Review Board approval (Aedes aegypti density threshold preventing human arboviral infections; #1910024340R004; Ponce Research Institute, Ponce Medical School Foundation, Inc.). Participants in the serology study were 1 – 51 years old. Participation in this study was voluntary and participants were informed that they can withdraw at any time. Written informed consent was obtained from all study participants aged ≥ 21 years. Caregivers provided written informed consent for participants aged < 21 years. Oral assent was obtained from participants aged between 7 and 13 years and written assent from participants aged between 14–20 years.

The study was conducted in Caguas City, Caguas Municipality (18.23412 N, −66.0485 W), Puerto Rico, from January 2021 to June 2024 (Figure 1). Averages (± Std. Error) of weekly rainfall, daily temperature, and daily relative humidity during the study were 26.99 ± 1.52 mm (n= 751), 26.53 ± 0.025 °C (n= 5262), and 80.28 ± 0.08 % (n= 5262), respectively. It was designed as a four-arm, cluster-randomized trial with three replicates per arm, involving approximately 300 houses per cluster (3610 houses across 12 clusters; Supplementary Table 1). The number of houses per cluster was chosen given available resources. The four treatments included the deployment of zero, one, two, or three AGO traps per house across most houses within each cluster (Figure 1). AGO traps were used to control Ae. aegypti based on previous effectiveness studies conducted in Puerto Rico (Barrera et al. 2014a, b; 2017; 2019a, b). Throughout the manuscript we will refer to AGO traps as IAGO or SAGO traps. Although intervention AGO traps (IAGO) and surveillance AGO traps (SAGO) are identical in design, they served different purposes: mosquito counts were conducted exclusively in SAGO traps, while the more numerous IAGO traps were used solely for intervention.

The primary outcome was the number of female Ae. aegypti per surveillance trap per week. The hypothesis suggested that an increase in the number of traps per house would lead to a significant reduction in the average number of female Ae. aegypti mosquitoes and consequently result in lower dengue virus prevalence among mosquitoes and dengue antibodies in residents. Secondary outcomes included the prevalence of infected female Ae. aegypti mosquitoes and changes in dengue seroprevalence among a sample of residents from a first serosurvey in 2022 to a second serosurvey in 2023. To assess whether Ae. aegypti populations rebounded after trap removal, mosquito abundance was monitored in eight randomly selected clusters (out of the original 12). Trap removal occurred over two months (January–February 2024), with post-intervention data collected from March to June 2024 (Figure 2).

Before initiating the study, we engaged with participating communities and the Caguas municipality to explain the trial’s purpose and seek their collaboration. Study clusters were selected based on similar historical averages of female Ae. aegypti per trap per week, as well as comparable land use characteristics, and human population densities (Supplemental Table 1). Mosquito abundance data were sourced from a previous surveillance program conducted by the Puerto Rico Vector Control Unit in Caguas City between June and November 2018, which used AGO traps. Human population and land use information were derived from the 2012 Territorial Plan of Caguas City. The 12 clusters were randomly assigned to one of the four treatments using Python’s “random” library (https://python.readthedocs.io/en/stable/library/random.html).

Traps were deployed in the treatment clusters between January and April 2021 (Figure 2). Surveillance traps were installed in late April 2021, with the first weekly report on female Ae. aegypti abundance per trap generated in May 2021. In each cluster, 20 SAGO traps were evenly distributed, fixed at consistent locations, and monitored weekly. Mosquitoes collected in SAGO traps were counted, and pooled samples were prepared for virus identification. The average distance between SAGO traps within clusters was about 60.7 meters (1.3 traps/Ha), calculated using the Average Nearest Neighbor tool in ArcGIS Pro (ESRI, Redlands, CA). The number of IAGO traps varied by treatment, with 226 (12.8 traps/Ha), 446 (29.8 traps/Ha), and 655 (46.6 traps/Ha) traps per cluster in treatments with one, two, and three traps per house, respectively. The percentage of houses with traps ranged from 72–78%, 74–75%, and 68–74% for clusters with one, two, and three traps per house respectively.

AGO traps use sticky glue to capture gravid female mosquitoes attracted to the trap for oviposition. Each trap consists of a black plastic bucket containing ten liters of water and a 30-gram packet of hay grass, which decomposes to produce odors that lure female mosquitoes. The capture chamber includes a fine mesh screen that prevents mosquitoes from reaching the water. Every week, trapped mosquitoes were removed from the sticky glue, placed on paper towels, and identified by sex and species in the field by trained technicians. Both SAGO and IAGO traps were installed outdoors and serviced every two months, during which the traps were cleaned, and water, hay packets, and sticky boards were replaced (Barrera et al. 2014b).

Several previous studies have shown that viral RNA of DENV, CHIKV, and ZIKV can be detected in dead mosquitoes that have been exposed at room temperature or in the field for over one week (Barrera et al. 2019a). Female Ae. aegypti mosquitoes collected from SAGO traps were pooled weekly into groups of 1–20 mosquitoes per vial and preserved at −20°C until processed for dengue, chikungunya, and Zika virus detection. To prevent contamination, dissecting needles were sterilized between transfers from individual traps to vials. Viral RNA was identified using the CDC’s trioplex RT-PCR assay (Santiago et al. 2018). To optimize testing efficiency, we created super pools by combining 50μl aliquots from each of ten individual pools into a single 500μl sample. If a super pool tested positive for dengue virus, each constituent pool was individually tested by RT-PCR to determine the presence of the virus and to identify the dengue serotype (DENV-1, DENV-2, DENV-3, or DENV-4) (Santiago et al. 2013). If a super pool tested negative, no further testing was performed.

To assess whether residents in the study areas were differentially infected with dengue viruses according to treatment during part of the investigation, we carried out two serosurveys. The first took place from June to September 2022 and the second from October to December 2023. Phlebotomists and interviewers collected 50 finger-prick blood samples from each of the 12 clusters, resulting in a total of 600 tests per serosurvey. When the 50 blood samples were obtained in each study cluster, we stop recruiting participants. These samples were tested on-site for immunoglobulin M (IgM) antibodies using the CTK Biotech OnSite Dengue IgG/IgM Combo Rapid Test and for immunoglobulin G (IgG) antibodies using the CTK Biotech OnSite Dengue IgG Rapid Test (https://ctkbiotech.com/onsite-dengue-igg-igm-combo-rapid-test-ce/). During household visits to the clusters, eligible participants were identified after the purpose of the study was explained and their consent to participate was obtained. Participants were eligible if they met the following criteria: residence in one of the study areas, age 1–50 years for the first serosurvey or 1–51 years for the second, provision of informed consent, completion of a short questionnaire, and agreement to provide a blood sample.

Georeferenced data of confirmed dengue cases and circulating virus serotypes for Caguas municipality were obtained from the Puerto Rico Health Department.

Differences in the average number of female Ae. aegypti per trap per week among treatment arms were analyzed using a Generalized Linear Mixed Model, with a negative binomial distribution. The model included an autoregressive order-one covariance structure to account for repeated measures and incorporated fixed effects for accumulated rainfall (two and three weeks prior to sampling) and the average daily air temperature and relative humidity over the previous three weeks. Random effects were specified for cluster and trap ID. The analyses were done using the `glmmTMB` package in R (https://cran.r-project.org/web/packages/glmmTMB/index.html).

To test if there were significant differences in the average number of Ae. aegypti females per trap per week after trap removal between control and treatment clusters, we extended the GLMM described above by including an interaction term between treatment levels (0, 1, 2, or 3 traps per house) and trapping period (pre-trap removal, trap removal, and post-trap removal). The model evaluated whether changes in Ae. aegypti density during and after trap removal differed across treatment groups. Data for the pre-trap removal (September–December 2023) and post-trap removal (March–June 2024) periods were analyzed over four months, while data from the transitional trap removal period (January–February 2024) spanned two months.

Minimum dengue infection rates of pooled Ae. aegypti females were calculated using the “PooledInfRate”1.6 package in R (https://github.com/CDCgov/PooledInfRate). The minimum infection rate assumes that if a pool is positive for the virus, then at least one mosquito in the pool was positive. It is calculated by dividing the number of positive pools by the total number of mosquitoes tested. The Vector Index (VI), or number of expected infected mosquitoes per thousand, was calculated as the product of the infection rate times the average number of Ae. aegypti per trap per week (Jones et al. 2011). To calculate confidence intervals, we used non-parametric bootstrapping with 10,000 iterations and resampling data with replacement. The 95% CIs were obtained using the percentile method, capturing the middle 95% of the resampled differences. The package used in R for estimation was boot 1.3–31 (https://cran.rstudio.com/web/packages/boot/index.html).

We also used GLMMs to explore potential statistical differences in IgM and IgG antibody prevalence across treatments and demographics. These models assessed whether seroprevalence significantly varied by treatment, survey year, age group, sex, the interaction between age group and year, and average weekly hours spent in the household. The GLMMs were specified with a binomial distribution and a logit link function. Cluster-level variability was addressed by including cluster as a random effect to control for intra-cluster correlation. Models were fit using the glmer function in the R lme4 package (https://github.com/lme4/lme4/blob/master/man/glmer.Rd).

To estimate confidence intervals for the difference in IgG seroprevalence between 2022 and 2023 within each treatment group, we used non-parametric bootstrapping with 10,000 iterations. For each iteration, we resampled the observed data with replacement and computed the difference in seroprevalence proportions. The 95% CIs were obtained using the percentile method, capturing the middle 95% of the resampled differences. This approach avoids normality assumptions and is more robust for small sample sizes. We used the boot() function from the boot package for bootstrapping. Statistical analyses were conducted using R software version 4.4.0 (R Foundation for Statistical Computing, Vienna, Austria).

The average numbers of female Ae. aegypti captured per trap per week decreased as the number of IAGO traps per house increased, from 9.68 ± 0.09 (95% CI; N= 8279 trap-weeks) in clusters without IAGO traps to 5.48 ± 0.06 (N= 8274), 3.92 ± 0.04 (N= 8295), and 2.15 ± 0.03 (N= 8272) in clusters with one, two, and three IAGO traps per house, respectively (Figure 3). GLMM analysis indicated significant effects of treatment (number of IAGO traps per house), rainfall, temperature, and relative humidity (p-values< 0.001; Table 1). Compared with control clusters, reductions in the average number of female Ae. aegypti were 43.6% (95%CI: 23.5 – 58.5), 60.3% (46.2 – 70.8), and 79.6% (72.3 – 85.0) in clusters with one, two, and three traps per house, respectively. Weekly fluctuations in Ae. aegypti counts were smaller in clusters with higher trap-house densities, with short-term increases primarily associated with increases in rainfall (Figure 3).

After the removal of IAGO traps in January and February of 2024, Ae. aegypti populations increased in the experimental clusters compared to the control clusters (Figure 4). Average numbers of Ae. aegypti per trap per week changed from 13.4 to 11.2 in control clusters, 6.8 to 10.1 in clusters with one trap/house, 4.3 to 9.6 in clusters with two traps/house, and 2.5 to 8.1 in clusters with three traps/house (Table 2). GLMM analysis revealed significant interactions (p< 0.001) between treatment intensity (0, 1, 2, or 3 traps per house) and trapping period (pre-removal, during removal, and post-removal (Supplemental Table 2). The interaction effects were driven by a slight decrease in female Ae. aegypti numbers in control clusters as compared to marked increases in treated clusters after trap removal (Figure 4).

During the study, 39 dengue virus detections were made in female Ae. aegypti captured in SAGO traps, comprising four instances of DENV-1 and 35 instances of DENV-2. Dengue-positive pools were identified across clusters, with 17 positive pools in control clusters and 14, 5, and 3 positive pools in clusters with one, two, and three traps per house, respectively (Table 3). Overlapping 95% confidence intervals for minimum infection rates between treatments indicated no statistically significant differences (p> 0.05; Table 3). The corresponding Vector Index values (95% C.I.) were 2.01 (1.01–2.92), 1.70 (0.71–2.54), 0.60 (0.11–1.09), and 0.36 (0 – 0.73) for control clusters and clusters with one, two, and three traps per house, respectively. These findings suggest lower numbers of infected mosquitoes (Vector Index) in clusters with three traps compared to controls and those with only one trap per house.

There were 126 confirmed dengue cases in Caguas municipality during the study period, of which 31 were DENV-1, 42 DENV-2, 18 DENV-3, and 36 unidentified. The number of reported dengue cases in the 12 study clusters was small (3 – 5 cases/treatment clusters) and similar across treatments.

Participation of residents was uniform across both serosurvey years, with 50 participants per cluster and 150 participants per treatment group. The sex distribution was consistent between the two serosurveys, with 53.5% female and 46.5% male participants in the first serosurvey, and 53.7% female and 46.3% male participants in the second serosurvey. The age distribution also showed similar patterns across both serosurveys. IgM test positivity was low across clusters during the first (2.7 – 6.0%) and second (0 – 2%) year. IgG test positivity was similar and high during the first (67 – 74%) and second (68 – 73%) year.

Distinct differences in IgG seroprevalence were observed across age groups, indicating a trend of increasing seroprevalence with age (Figure 5). The results of the GLMM analysis revealed no significant differences (p > 0.05; Supplemental Table 3) in overall IgG seroprevalence between intervention groups (one, two, or three traps/house compared to controls). Additionally, serosurvey year did not have a significant impact, as the estimated overall seroprevalence remained consistent in both years. Sex and average weekly hours spent in the household were also not significant predictors (Supplemental Table 3). The primary factor associated with past dengue infection in our study population was age group.

The statistical interaction of the age group and serosurvey year was significant (p< 0.05), reflecting that increases in IgG prevalence was observed only among participants 1–10 years old (Figure 5; Supplemental Table 3). Bootstrapped 95% confidence intervals for the difference in IgG seroprevalence between 2022 and 2023 were calculated for each treatment group. The largest increase (95% C.I.) was observed in the control group (25%; 0.00 – 0.50), followed by clusters with one trap per house (17.8%; −0.13 – 0.49), two traps (5.7%; −0.19 – 0.30), and three traps (4.9%; −0.14 – 0.33). The overlapping confidence intervals across treatment groups suggest that changes in seroprevalence were not significantly different between treatment groups (Figure 5).