Browse the corpus

Efficacy of Wolbachia-Infected Mosquito Deployments for the Control of Dengue. BACKGROUND: Aedes aegypti mosquitoes infected with the wMel strain of Wolbachia pipientis are less susceptible than wild-type A. aegypti to dengue virus infection. METHODS: We conducted a cluster-randomized trial involving releases of wMel-infected A. aegypti mosquitoes for the control of dengue in Yogyakarta, Indonesia. We randomly assigned 12 geographic clusters to receive deployments of wMel-infected A. aegypti (intervention clusters) and 12 clusters to receive no deployments (control clusters). All clusters practiced local mosquito-control measures as usual. A test-negative design was used to assess the efficacy of the intervention. Patients with acute undifferentiated fever who presented to local primary care clinics and were 3 to 45 years of age were recruited. Laboratory testing was used to identify participants who had virologically confirmed dengue (VCD) and those who were test-negative controls. The primary end point was symptomatic VCD of any severity caused by any dengue virus serotype. RESULTS: After successful introgression of wMel into the intervention clusters, 8144 participants were enrolled; 3721 lived in intervention clusters, and 4423 lived in control clusters. In the intention-to-treat analysis, VCD occurred in 67 of 2905 participants (2.3%) in the intervention clusters and in 318 of 3401 (9.4%) in the control clusters (aggregate odds ratio for VCD, 0.23; 95% confidence interval [CI], 0.15 to 0.35; P = 0.004). The protective efficacy of the intervention was 77.1% (95% CI, 65.3 to 84.9) and was similar against the four dengue virus serotypes. The incidence of hospitalization for VCD was lower among participants who lived in intervention clusters (13 of 2905 participants [0.4%]) than among those who lived in control clusters (102 of 3401 [3.0%]) (protective efficacy, 86.2%; 95% CI, 66.2 to 94.3). CONCLUSIONS: Introgression of wMel into A. aegypti populations was effective in reducing the incidence of symptomatic dengue and resulted in fewer hospitalizations for dengue among the participants. (Funded by the Tahija Foundation and others; AWED ClinicalTrials.gov number, NCT03055585; Indonesia Registry number, INA-A7OB6TW.).

The “Applying Wolbachia to Eliminate Dengue” (AWED) trial was financially supported by the Tahija Foundation and the trial sponsor was the Universitas Gadjah Mada, Indonesia. The protocol was published20,21 and is provided in the Supplementary Appendix at nejm.org. Community support for randomised wMel releases was obtained from leaders of 37 urban villages following a community engagement and mass communications campaign. For enrolment into the clinical cohort in primary health care facilities, written informed consent was obtained from all participants or their guardian where the participant was a minor. In addition, participants aged between 13 and 17 years gave written informed assent to participate. The trial was conducted in accordance with the principles of the Good Clinical Practice guidelines of the International Conference on Harmonisation and was approved by the Universitas Gadjah Mada and the Monash University Human Research Ethics Committees. The trial data was analysed by the independent trial statisticians (NPJ and SMD). The funders had no role in the analysis of the data, in the preparation or approval of the manuscript, or in the decision to submit the manuscript for publication.

as Gadjah Mada and the Monash University Human Research Ethics Committees. The trial data was analysed by the independent trial statisticians (NPJ and SMD). The funders had no role in the analysis of the data, in the preparation or approval of the manuscript, or in the decision to submit the manuscript for publication. The baseline characteristics of the study site are described in Table S1. Briefly, the study site was a continuous urban area of 26 km2 and with a population of approximately 311,700. The study site was subdivided into twenty-four contiguous clusters, each approximately 1km2 in size and where possible having borders that would slow the dispersal of mosquitoes between clusters. Of the 24 clusters, 12 were randomly allocated to receive open label Wolbachia deployments and 12 left untreated (Figure 1 and Figure S1). In treated clusters most community members will have been unaware of treatment assignment because mosquito release containers were discreetly placed in a minority of residential properties for a time-limited period in 2017 (Table S2). No placebo was used in the untreated arm. Constrained randomisation was used to prevent a chance imbalance in the baseline characteristics or spatial distribution of treated and untreated clusters (described in Supplementary Appendix).

n a minority of residential properties for a time-limited period in 2017 (Table S2). No placebo was used in the untreated arm. Constrained randomisation was used to prevent a chance imbalance in the baseline characteristics or spatial distribution of treated and untreated clusters (described in Supplementary Appendix). wMel-infected Ae. aegypti were sourced from an outcrossed colony described previously.14 As expected, these wMel-infected mosquitoes had reduced transmission potential for DENV compared to wild-type Ae. aegypti, as described in the Supplementary Appendix and Figures S2 and S3. Mosquitoes were released as eggs into intervention clusters between March and December 2017. Each cluster received between 9–14 rounds of releases (see Table S2 for details). Mosquito releases, and the monitoring of wMel frequencies via a network of 348 BG-Sentinel adult mosquito traps (BioGents), are described in the Supplementary Appendix. Participant enrolment to measure the efficacy endpoint was performed at a network of 18 government-run primary care clinics in Yogyakarta City and adjacent Bantul District.

wMel-infected Ae. aegypti were sourced from an outcrossed colony described previously.14 As expected, these wMel-infected mosquitoes had reduced transmission potential for DENV compared to wild-type Ae. aegypti, as described in the Supplementary Appendix and Figures S2 and S3. Mosquitoes were released as eggs into intervention clusters between March and December 2017. Each cluster received between 9–14 rounds of releases (see Table S2 for details). Mosquito releases, and the monitoring of wMel frequencies via a network of 348 BG-Sentinel adult mosquito traps (BioGents), are described in the Supplementary Appendix. Participant enrolment to measure the efficacy endpoint was performed at a network of 18 government-run primary care clinics in Yogyakarta City and adjacent Bantul District. Patients presenting to the clinics were eligible for the study if they met the inclusion criteria, a) fever (either self-reported or objectively measured, defined as forehead or axillary temperature >37.5°C) with a date of onset between 1–4 days prior to the day of presentation, b) aged between 3–45 years old and c) resided in the study area every night for the 10 days preceding illness onset. Participants were not eligible if they a) had localising features suggestive of a specific diagnosis other than an arboviral infection, e.g. severe diarrhea, otitis, or pneumonia, or b) were enrolled in the study within the previous 4 weeks.

resided in the study area every night for the 10 days preceding illness onset. Participants were not eligible if they a) had localising features suggestive of a specific diagnosis other than an arboviral infection, e.g. severe diarrhea, otitis, or pneumonia, or b) were enrolled in the study within the previous 4 weeks. Enrolled participants provided demographic information, geolocated residential address and a detailed travel history (durations and geolocations) for the past 10 days. A 3 ml venous blood sample was collected for arbovirus diagnostic tests. No other diagnostic investigations were performed. Participants were followed up 14–21 days later to determine whether they a) were alive, and b) had been hospitalised since their enrolment in the study. No information on the clinical severity of virologically-confirmed dengue (VCD) cases was acquired No further information on disease severity or clinical diagnoses was acquired.

ts were followed up 14–21 days later to determine whether they a) were alive, and b) had been hospitalised since their enrolment in the study. No information on the clinical severity of virologically-confirmed dengue (VCD) cases was acquired No further information on disease severity or clinical diagnoses was acquired. Study participants were classified as VCD cases if their enrolment plasma sample was DENV test-positive in a multiplex (DENV, chikungunya and Zika virus) reverse-transcriptase polymerase chain reaction (RT-PCR) and/or in an enzyme-linked immunosorbent assay (ELISA) for dengue NS1 (BioRad Platelia). Study participants were classified as test-negative controls if their enrolment plasma sample was test-negative by RT-PCR for DENV, chikungunya and Zika viruses, and also test-negative for DENV NS1 and negative in dengue IgM and IgG capture ELISAs. The diagnostic algorithm is shown in Figure S4. Note that the DENV serotype was determined via a 2nd independent RT-PCR test (Simplexa) by an independent laboratory at the Eijkman Institute, Jakarta. Details of the diagnostic methods are provided in the Supplementary Appendix. The primary endpoint was the efficacy of community-based deployments of wMel-infected Ae. aegypti mosquitoes in reducing the incidence of symptomatic, virologically-confirmed dengue cases of any severity in Yogyakarta residents aged 3–45 years in release (intervention) areas, relative to non-release (untreated) areas. Secondary endpoints reported here include the efficacy against each of the four DENV serotypes.

ti mosquitoes in reducing the incidence of symptomatic, virologically-confirmed dengue cases of any severity in Yogyakarta residents aged 3–45 years in release (intervention) areas, relative to non-release (untreated) areas. Secondary endpoints reported here include the efficacy against each of the four DENV serotypes. Reflective of the novel design, the sample size requirements to demonstrate a 50% reduction in dengue incidence, which was considered the minimum effect size for public health value, evolved over time. The full sample size narrative is provided in the Supplementary Appendix. Briefly, 400 VCD cases and four times as many controls was determined to be sufficient to detect a 50% reduction in VCD case incidence with 80% power. The emergence of SARS-CoV-2 in Yogyakarta in March 2020 prevented the continued enrolment of participants in clinics, with enrolment stopping on 18th March 2020. On 5th May 2020, the trial steering committee endorsed the recommendation from the trial investigators to terminate the trial having recruited 385 VCD cases.

The emergence of SARS-CoV-2 in Yogyakarta in March 2020 prevented the continued enrolment of participants in clinics, with enrolment stopping on 18th March 2020. On 5th May 2020, the trial steering committee endorsed the recommendation from the trial investigators to terminate the trial having recruited 385 VCD cases. The statistical analysis plan was published22 and is available in the Supplementary Appendix. The dataset for analysis included all enrolled VCD cases and all test-negative controls, excluding participants enrolled prior to Wolbachia establishment throughout intervention clusters (defined as one month after completion of releases in the last cluster) and excluding test-negative controls enrolled in a calendar month with no enrolled dengue cases. The primary intention-to-treat (ITT) analysis considered Wolbachia exposure as a binary classification based on residence in a cluster allocated to Wolbachia deployment or not. Residence was defined as the primary place of residence during the 10 days prior to illness onset. The intervention effect was estimated from an aggregate odds ratio (OR) comparing the exposure odds (residence in a Wolbachia-treated cluster) among VCD cases versus test-negative controls, using the constrained permutation distribution as the foundation for inference. The null hypothesis was that the odds of residence in a Wolbachia-treated cluster was the same among VCD cases as test-negative controls. Efficacy of the intervention was calculated as 100*(1-aggregate OR). A predefined exploratory analysis evaluated the efficacy of the intervention in preventing hospitalised virologically-confirmed dengue cases.

the odds of residence in a Wolbachia-treated cluster was the same among VCD cases as test-negative controls. Efficacy of the intervention was calculated as 100*(1-aggregate OR). A predefined exploratory analysis evaluated the efficacy of the intervention in preventing hospitalised virologically-confirmed dengue cases. An additional pre-defined cluster-level ITT analysis was performed by calculating the VCD case proportion in each cluster. The difference in the average proportion of VCD cases between the intervention clusters and untreated clusters was used to test the null hypothesis of no intervention effect (a t-test statistic) and to derive an estimate of the cluster-specific relative risk, with inference based on the constrained permutation distribution.23,24 The same intention-to-treat analyses described above were applied for the secondary endpoint of serotype-specific efficacy, with case populations restricted to each of the DENV serotypes in turn, and with the same test-negative control population as for the primary analysis. Per protocol analyses considered exposure contamination by assigning a Wolbachia exposure index to each participant based on the wMel frequency in their cluster of residence only, or by combining this frequency with the participant’s recent travel history. A generalized linear model was fitted, with balanced bootstrap resampling based on cluster membership, to estimate the relative risk of VCD and associated confidence interval in each quintile of Wolbachia exposure, relative to baseline. Detailed methods are provided in the Supplementary Appendix.

The baseline characteristics of the study site are described in Table S1. Briefly, the study site was a continuous urban area of 26 km2 and with a population of approximately 311,700. The study site was subdivided into twenty-four contiguous clusters, each approximately 1km2 in size and where possible having borders that would slow the dispersal of mosquitoes between clusters. Of the 24 clusters, 12 were randomly allocated to receive open label Wolbachia deployments and 12 left untreated (Figure 1 and Figure S1). In treated clusters most community members will have been unaware of treatment assignment because mosquito release containers were discreetly placed in a minority of residential properties for a time-limited period in 2017 (Table S2). No placebo was used in the untreated arm. Constrained randomisation was used to prevent a chance imbalance in the baseline characteristics or spatial distribution of treated and untreated clusters (described in Supplementary Appendix).

wMel-infected Ae. aegypti were sourced from an outcrossed colony described previously.14 As expected, these wMel-infected mosquitoes had reduced transmission potential for DENV compared to wild-type Ae. aegypti, as described in the Supplementary Appendix and Figures S2 and S3. Mosquitoes were released as eggs into intervention clusters between March and December 2017. Each cluster received between 9–14 rounds of releases (see Table S2 for details). Mosquito releases, and the monitoring of wMel frequencies via a network of 348 BG-Sentinel adult mosquito traps (BioGents), are described in the Supplementary Appendix.

Participant enrolment to measure the efficacy endpoint was performed at a network of 18 government-run primary care clinics in Yogyakarta City and adjacent Bantul District. Patients presenting to the clinics were eligible for the study if they met the inclusion criteria, a) fever (either self-reported or objectively measured, defined as forehead or axillary temperature >37.5°C) with a date of onset between 1–4 days prior to the day of presentation, b) aged between 3–45 years old and c) resided in the study area every night for the 10 days preceding illness onset. Participants were not eligible if they a) had localising features suggestive of a specific diagnosis other than an arboviral infection, e.g. severe diarrhea, otitis, or pneumonia, or b) were enrolled in the study within the previous 4 weeks.

Enrolled participants provided demographic information, geolocated residential address and a detailed travel history (durations and geolocations) for the past 10 days. A 3 ml venous blood sample was collected for arbovirus diagnostic tests. No other diagnostic investigations were performed. Participants were followed up 14–21 days later to determine whether they a) were alive, and b) had been hospitalised since their enrolment in the study. No information on the clinical severity of virologically-confirmed dengue (VCD) cases was acquired No further information on disease severity or clinical diagnoses was acquired.

Study participants were classified as VCD cases if their enrolment plasma sample was DENV test-positive in a multiplex (DENV, chikungunya and Zika virus) reverse-transcriptase polymerase chain reaction (RT-PCR) and/or in an enzyme-linked immunosorbent assay (ELISA) for dengue NS1 (BioRad Platelia). Study participants were classified as test-negative controls if their enrolment plasma sample was test-negative by RT-PCR for DENV, chikungunya and Zika viruses, and also test-negative for DENV NS1 and negative in dengue IgM and IgG capture ELISAs. The diagnostic algorithm is shown in Figure S4. Note that the DENV serotype was determined via a 2nd independent RT-PCR test (Simplexa) by an independent laboratory at the Eijkman Institute, Jakarta. Details of the diagnostic methods are provided in the Supplementary Appendix.

The primary endpoint was the efficacy of community-based deployments of wMel-infected Ae. aegypti mosquitoes in reducing the incidence of symptomatic, virologically-confirmed dengue cases of any severity in Yogyakarta residents aged 3–45 years in release (intervention) areas, relative to non-release (untreated) areas. Secondary endpoints reported here include the efficacy against each of the four DENV serotypes.

Reflective of the novel design, the sample size requirements to demonstrate a 50% reduction in dengue incidence, which was considered the minimum effect size for public health value, evolved over time. The full sample size narrative is provided in the Supplementary Appendix. Briefly, 400 VCD cases and four times as many controls was determined to be sufficient to detect a 50% reduction in VCD case incidence with 80% power. The emergence of SARS-CoV-2 in Yogyakarta in March 2020 prevented the continued enrolment of participants in clinics, with enrolment stopping on 18th March 2020. On 5th May 2020, the trial steering committee endorsed the recommendation from the trial investigators to terminate the trial having recruited 385 VCD cases.

The statistical analysis plan was published22 and is available in the Supplementary Appendix. The dataset for analysis included all enrolled VCD cases and all test-negative controls, excluding participants enrolled prior to Wolbachia establishment throughout intervention clusters (defined as one month after completion of releases in the last cluster) and excluding test-negative controls enrolled in a calendar month with no enrolled dengue cases. The primary intention-to-treat (ITT) analysis considered Wolbachia exposure as a binary classification based on residence in a cluster allocated to Wolbachia deployment or not. Residence was defined as the primary place of residence during the 10 days prior to illness onset. The intervention effect was estimated from an aggregate odds ratio (OR) comparing the exposure odds (residence in a Wolbachia-treated cluster) among VCD cases versus test-negative controls, using the constrained permutation distribution as the foundation for inference. The null hypothesis was that the odds of residence in a Wolbachia-treated cluster was the same among VCD cases as test-negative controls. Efficacy of the intervention was calculated as 100*(1-aggregate OR). A predefined exploratory analysis evaluated the efficacy of the intervention in preventing hospitalised virologically-confirmed dengue cases.

This trial was performed in Yogyakarta, Indonesia (Figure 1). wMel was durably established in the Ae. aegypti populations in each of the 12 intervention clusters (Figure 2). The monthly median (interquartile range) cluster level wMel prevalence was 95.8% (91.5–97.8%) during the 27 months of clinical surveillance.

This trial was performed in Yogyakarta, Indonesia (Figure 1). wMel was durably established in the Ae. aegypti populations in each of the 12 intervention clusters (Figure 2). The monthly median (interquartile range) cluster level wMel prevalence was 95.8% (91.5–97.8%) during the 27 months of clinical surveillance. 53,924 patients were screened for study eligibility at 18 primary care clinics between January 8th 2018 and March 18th 2020 and 8144 persons were enrolled. Of these, 6306 participants met the requirements for the primary analysis dataset; 2905 participants were resident in the wMel intervention arm and 3401 in the untreated arm (Figure 3). Four virologically-confirmed chikungunya cases (1 in the wMel-treated arm and 3 in the untreated arm) were excluded from the primary analysis dataset. No Zika cases were detected. The median age (interquartile range) of participants was 11.6 years (6.7, 20.9) and 48.8% of participants were female (Table S3). 295 (4.7%) of the 6306 participants in the analysis dataset were hospitalised in the time between their enrolment and follow-up 14–21 days later. Hospitalisation was significantly less frequent for participants resident in the wMel-treated arm (81/2905, 2.8%) compared to the untreated arm (214/3401, 6.3%) (OR 0.43, 95% confidence interval [CI], 0.32 to 0.58; P=0.004) (Table S4). This lower probability of hospitalisation was evident across the clinic network (Figure S5). 385 (6.1%) of 6306 participants in the analysis dataset were VCD cases and 5921 (93.8%) were test-negative controls. VCD cases and test-negative controls were well-matched by age and gender (Table S3).

.58; P=0.004) (Table S4). This lower probability of hospitalisation was evident across the clinic network (Figure S5). 385 (6.1%) of 6306 participants in the analysis dataset were VCD cases and 5921 (93.8%) were test-negative controls. VCD cases and test-negative controls were well-matched by age and gender (Table S3). The incidence of VCD cases was significantly lower in the wMel-treated arm (67 VCDs amongst 2905 participants (2.3%)) than in the untreated arm (318/3401 (9.4%)) (OR 0.23, 95% CI, 0.15 to 0.35; P=0.004). This represented a protective efficacy of 77.1% (95% CI, 65.3 to 84.9) (Figure 4). The intervention effect was evident by 12 months after wMel-establishment (Figure S6). Protective efficacy was similar across serotypes, being highest for DENV-2 (83.8%; 95% CI, 72.1 to 90.6) and lowest for DENV-1 (71.0%; 95% CI, 18.2 to 89.7) (Figure 4). For all four serotypes the lower bound of the 95% CI for protective efficacy was greater than 0. There were 13 hospitalisations for VCD amongst 2905 participants (0.4%) from the wMel treated arm compared to 102 hospitalisations for VCD amongst 3401 participants (3%) from the untreated arm, for a protective efficacy of 86.2% (95% CI, 66.2 to 94.3) (Figure 4 and Table S5).

for protective efficacy was greater than 0. There were 13 hospitalisations for VCD amongst 2905 participants (0.4%) from the wMel treated arm compared to 102 hospitalisations for VCD amongst 3401 participants (3%) from the untreated arm, for a protective efficacy of 86.2% (95% CI, 66.2 to 94.3) (Figure 4 and Table S5). An additional prespecified ITT analysis compared VCD cases as a proportion of total participants in each cluster, between study arms. In all but one of the wMel-treated clusters the VCD proportion was lower than untreated clusters, yielding a relative risk of 0.23 (95% CI, 0.06 to 0.47; P=0.004) (Figure 5). Figure S7 shows the VCD proportion and wMel prevalence over time in individual clusters. When stratified by serotype, the relative risk of VCD caused by the two most prevalent serotypes, DENV-2 and DENV-4, was significantly lower in the wMel-treated arm (Figure S8). Per protocol analyses assigned a Wolbachia exposure index to each participant based on the wMel frequency in their cluster of residence only, or by accounting also for wMel frequencies and time spent in other locations. Protective efficacy against VCD increased with incremental increases in participants’ Wolbachia exposure index when cluster of residence and recent travel history were considered (Figure S9A). When only the wMel frequency in the cluster of residence was considered a threshold effect was observed in that only cluster-level wMel frequencies >80% were protective (Figure S9B).

53,924 patients were screened for study eligibility at 18 primary care clinics between January 8th 2018 and March 18th 2020 and 8144 persons were enrolled. Of these, 6306 participants met the requirements for the primary analysis dataset; 2905 participants were resident in the wMel intervention arm and 3401 in the untreated arm (Figure 3). Four virologically-confirmed chikungunya cases (1 in the wMel-treated arm and 3 in the untreated arm) were excluded from the primary analysis dataset. No Zika cases were detected. The median age (interquartile range) of participants was 11.6 years (6.7, 20.9) and 48.8% of participants were female (Table S3). 295 (4.7%) of the 6306 participants in the analysis dataset were hospitalised in the time between their enrolment and follow-up 14–21 days later. Hospitalisation was significantly less frequent for participants resident in the wMel-treated arm (81/2905, 2.8%) compared to the untreated arm (214/3401, 6.3%) (OR 0.43, 95% confidence interval [CI], 0.32 to 0.58; P=0.004) (Table S4). This lower probability of hospitalisation was evident across the clinic network (Figure S5). 385 (6.1%) of 6306 participants in the analysis dataset were VCD cases and 5921 (93.8%) were test-negative controls. VCD cases and test-negative controls were well-matched by age and gender (Table S3).

The incidence of VCD cases was significantly lower in the wMel-treated arm (67 VCDs amongst 2905 participants (2.3%)) than in the untreated arm (318/3401 (9.4%)) (OR 0.23, 95% CI, 0.15 to 0.35; P=0.004). This represented a protective efficacy of 77.1% (95% CI, 65.3 to 84.9) (Figure 4). The intervention effect was evident by 12 months after wMel-establishment (Figure S6). Protective efficacy was similar across serotypes, being highest for DENV-2 (83.8%; 95% CI, 72.1 to 90.6) and lowest for DENV-1 (71.0%; 95% CI, 18.2 to 89.7) (Figure 4). For all four serotypes the lower bound of the 95% CI for protective efficacy was greater than 0. There were 13 hospitalisations for VCD amongst 2905 participants (0.4%) from the wMel treated arm compared to 102 hospitalisations for VCD amongst 3401 participants (3%) from the untreated arm, for a protective efficacy of 86.2% (95% CI, 66.2 to 94.3) (Figure 4 and Table S5). An additional prespecified ITT analysis compared VCD cases as a proportion of total participants in each cluster, between study arms. In all but one of the wMel-treated clusters the VCD proportion was lower than untreated clusters, yielding a relative risk of 0.23 (95% CI, 0.06 to 0.47; P=0.004) (Figure 5). Figure S7 shows the VCD proportion and wMel prevalence over time in individual clusters. When stratified by serotype, the relative risk of VCD caused by the two most prevalent serotypes, DENV-2 and DENV-4, was significantly lower in the wMel-treated arm (Figure S8).

Per protocol analyses assigned a Wolbachia exposure index to each participant based on the wMel frequency in their cluster of residence only, or by accounting also for wMel frequencies and time spent in other locations. Protective efficacy against VCD increased with incremental increases in participants’ Wolbachia exposure index when cluster of residence and recent travel history were considered (Figure S9A). When only the wMel frequency in the cluster of residence was considered a threshold effect was observed in that only cluster-level wMel frequencies >80% were protective (Figure S9B).

Establishment of wMel in Ae. aegypti mosquitoes in Yogyakarta reduced the incidence of symptomatic VCD amongst 3–45 year olds by 77%. Reassuringly, protective efficacy was observed against all four DENV serotypes, with greatest confidence for DENV-2 and -4 as these were the most prevalent serotypes. Efficacy against VCD requiring hospitalisation, a pragmatic proxy of clinical severity, was 86%. Eleven of the twelve wMel treated clusters had a lower proportion of VCD cases than untreated clusters, demonstrating consistent biological replication of the intervention effect. The conceptual underpinnings of the test negative design used in this trial, and the statistical framework for population inference, have been described.23 Acute undifferentiated fever of one to four days duration was set as the clinical basis for participant eligibility to avoid selection bias at the point of recruitment and to enable virological detection of dengue cases. Trial operational procedures, particularly blinding of research staff, aimed to prevent bias in follow-up, laboratory testing and outcome classification. The intervention (mosquito releases) was delivered openly, and not placebo controlled, for several months in each cluster during 2017. There was no evidence this changed the health care seeking behaviour of community members in subsequent years because similar numbers of participants meeting the eligibility criteria were enrolled from each arm of the trial.

s delivered openly, and not placebo controlled, for several months in each cluster during 2017. There was no evidence this changed the health care seeking behaviour of community members in subsequent years because similar numbers of participants meeting the eligibility criteria were enrolled from each arm of the trial. wMel-infected mosquito populations were not static and spatially heterogenous wMel contamination was measured at the edges of untreated clusters in year two of the trial. Nonetheless the efficacy estimates from per protocol analyses, which accounted for individual participants’ recent exposure to wMel via changes in cluster-level wMel frequencies and/or human movement, did not exceed that measured in the ITT analysis. We plan more nuanced exploratory analyses, outside the scope of the current protocol, to explore the fine spatial and temporal connections between wMel prevalence and risk of VCD.

cent exposure to wMel via changes in cluster-level wMel frequencies and/or human movement, did not exceed that measured in the ITT analysis. We plan more nuanced exploratory analyses, outside the scope of the current protocol, to explore the fine spatial and temporal connections between wMel prevalence and risk of VCD. The efficacy results reported are consistent with a body of laboratory and field observations. Predictions from an ensemble of mathematical models have suggested that the reduced infectiousness observed in wMel-infected Ae. aegypti could be sufficient to reduce R0 (the basic reproductive number) to below one in many dengue endemic settings, which could result in local elimination of disease.3,25,26 Previous non-randomised field studies in Australia16,17 and Indonesia14 provided evidence of large epidemiological impacts after wMel was introgressed. A quasi-experimental study of Wolbachia deployments in seven urban villages on the northwestern border of Yogyakarta, compared to three untreated control villages on the southeastern border of the city, reported a 76% reduction in the incidence of hospitalised dengue hemorrhagic fever during 30 months post-deployment.14 Together with the trial reported here, this body of work suggests that when wMel is established at high prevalence in local Ae. aegypti populations then reductions in dengue incidence follow. Another Wolbachia strain, wAlbB, also has pathogen-blocking properties and can be introgressed into Ae. aegypti field populations.15 This suggests the possibility of a portfolio of Wolbachia strains, each with different strengths and weaknesses, for application as public health interventions in dengue endemic areas.

w. Another Wolbachia strain, wAlbB, also has pathogen-blocking properties and can be introgressed into Ae. aegypti field populations.15 This suggests the possibility of a portfolio of Wolbachia strains, each with different strengths and weaknesses, for application as public health interventions in dengue endemic areas. Stable wMel transinfection imparts a viral resistant state in Ae. aegypti mosquitoes that attenuates superinfection by several medically-important Flaviviruses and Alphaviruses. Multiple mechanisms have been proposed to explain this phenotype, including Wolbachia-induced triggering of innate immune effectors27,28 and changes in intracellular cholesterol transport.29 DENV could plausibly evolve resistance to wMel however the requirement for alternating infection of human and mosquito hosts, together with what appears to be a complex mode of action, could be a constraint on the adaptive emergence of resistant virus populations. Future research should survey arbovirus populations for signals of Wolbachia-associated selective pressure. The wMel introgression approach represents a novel product class for the control of dengue.30 An attractive aspect of this strategy is that it maintains itself in the mosquito population and does not need re-application.31 Future trials should explore the multivalency of the intervention, since laboratory studies12,32–35 suggest wMel should also attenuate transmission of Zika, chikungunya, Yellow Fever and Mayaro viruses by Ae. aegypti.