Effectiveness of Wolbachia-infected mosquito deployments in reducing the incidence of dengue and other Aedes-borne diseases in Niterói, Brazil: A quasi-experimental study

Background The introduction of the bacterium Wolbachia (wMel strain) into Aedes aegypti mosquitoes reduces their capacity to transmit dengue and other arboviruses. Evidence of a reduction in dengue case incidence following field releases of wMel-infected Ae. aegypti has been reported previously from a cluster randomised controlled trial in Indonesia, and quasi-experimental studies in Indonesia and northern Australia. Methodology/Principal findings Following pilot releases in 2015–2016 and a period of intensive community engagement, deployments of adult wMel-infected Ae. aegypti mosquitoes were conducted in Niterói, Brazil during 2017–2019. Deployments were phased across four release zones, with a total area of 83 km2 and a residential population of approximately 373,000. A quasi-experimental design was used to evaluate the effectiveness of wMel deployments in reducing dengue, chikungunya and Zika incidence. An untreated control zone was pre-defined, which was comparable to the intervention area in historical dengue trends. The wMel intervention effect was estimated by controlled interrupted time series analysis of monthly dengue, chikungunya and Zika case notifications to the public health surveillance system before, during and after releases, from release zones and the control zone. Three years after commencement of releases, wMel introgression into local Ae. aegypti populations was heterogeneous throughout Niterói, reaching a high prevalence (>80%) in the earliest release zone, and more moderate levels (prevalence 40–70%) elsewhere. Despite this spatial heterogeneity in entomological outcomes, the wMel intervention was associated with a 69% reduction in dengue incidence (95% confidence interval 54%, 79%), a 56% reduction in chikungunya incidence (95%CI 16%, 77%) and a 37% reduction in Zika incidence (95%CI 1%, 60%), in the aggregate release area compared with the pre-defined control area. This significant intervention effect on dengue was replicated across all four release zones, and in three of four zones for chikungunya, though not in individual release zones for Zika. Conclusions/Significance We demonstrate that wMel Wolbachia can be successfully introgressed into Ae. aegypti populations in a large and complex urban setting, and that a significant public health benefit from reduced incidence of Aedes-borne disease accrues even where the prevalence of wMel in local mosquito populations is moderate and spatially heterogeneous. These findings are consistent with the results of randomised and non-randomised field trials in Indonesia and northern Australia, and are supportive of the Wolbachia biocontrol method as a multivalent intervention against dengue, chikungunya and Zika.


Introduction
Dengue is a mosquito-borne disease transmitted primarily by the Aedes aegypti mosquito, which has increased globally in both case burden and geographic footprint over the past 50 years. Approximately 40% of the world's population are at risk of dengue transmission, with an estimated 400 million infections per year resulting in 50-100 million clinical cases and 3.6 million hospitalisations [1,2]. The economic cost to health systems and communities has been PLOS NEGLECTED TROPICAL DISEASES estimated at $8.9 billion per annum [3]. In Brazil, more than 1.5 million dengue cases and 782 deaths were reported nationally in 2019, with in excess of 1300 cases per 100,000 population in the worst affected Central-West region. In the same year 132,000 cases of chikungunya-also transmitted by Ae. aegypti mosquitoes-were reported, including 92 deaths. Current strategies for dengue control are limited to efforts to suppress immature and adult mosquito numbers, through spraying of insecticides and community campaigns to reduce breeding sites. Even where considerable resources are invested in these activities, sustained suppression of mosquito densities has been elusive, and seasonal outbreaks continue to occur [4,5]. There is a well-recognised need for new, affordable and effective tools for control of dengue and other Aedes-borne arboviruses, including chikungunya and Zika [4,6].
Stable introduction of the common insect bacterium Wolbachia (wMel strain) into Ae. aegypti has been shown in the laboratory to result in Ae. aegypti having reduced transmission potential for dengue and other Aedes-borne arboviruses including chikungunya, Zika, Yellow Fever and Mayaro virus [7][8][9][10][11][12][13][14]. Female Ae. aegypti mosquitoes infected with wMel transmit the bacterium with high fidelity to their offspring via infected eggs and wMel manipulates mosquito reproductive outcomes via a process called cytoplasmic incompatibility, which favours introgression of wMel into a wild-type population [13]. Accumulating evidence from field sites in Australia and Indonesia has demonstrated large reductions in dengue incidence in areas where short-term releases of wMel-infected mosquitoes have resulted in introgression and sustained high prevalence of wMel in local Ae. aegypti populations [15][16][17]. A recently completed cluster randomised trial of wMel Wolbachia deployments in Yogyakarta, Indonesia, conclusively demonstrated the efficacy of the method, with a 77% reduction in dengue incidence in Wolbachia-treated neighbourhoods compared to untreated areas [18]. The Yogyakarta CRT included chikungunya and Zika as secondary endpoints, but insufficient cases were detected to permit an evaluation of efficacy against these arboviruses. Acquiring field evidence for the effectiveness of Wolbachia in reducing transmission of these arboviruses is a priority, as is the accumulation of real-world evidence for public health impact from large-scale implementations of wMel-infected Ae. aegypti in the complex urban environments common throughout dengue-endemic areas.
Pilot releases of Wolbachia-infected mosquitoes started in 2014 in Rio de Janeiro and in 2015 in Niterói, Brazil, and achieved successful establishment of Wolbachia throughout the two small pilot site communities, each with a population of 2500-2800 people [19,20]. In 2017 Niterói became the first site in Brazil to move to scaled deployments across a large urban area. The intervention involved a phased approach including engagement with and acceptance by the community, communication strategies to ensure the communities were informed and supportive, releases of Wolbachia-infected Ae. aegypti mosquitoes, and monitoring of the levels of Wolbachia in Ae. aegypti in the field.
We report here the entomological and epidemiological outcomes of a large-scale non-randomised deployment of Wolbachia-infected Ae. aegypti mosquitoes in the Brazilian city of Niterói, for the control of dengue and other Aedes-borne diseases. The impact of Wolbachia deployment on dengue, chikungunya and Zika incidence was evaluated via a quasi-experimental study, using controlled interrupted time series analysis of routine notifiable disease surveillance data, in accordance with a pre-defined protocol [21].

Ethics statement
Approval to release Wolbachia-carrying Ae. aegypti mosquitoes into urban areas was obtained from three Brazilian governmental bodies: the National Agency of Sanitary Surveillance (ANVISA); the Ministry of Agriculture, Livestock and Supply (MAPA); and the Brazilian Institute of Environment and Renewable Natural Resources (IBAMA), which issued a Temporary Special Registry (Registro Especial Temporário (RET), nr. 0551716178/2017). Ethical approval was also obtained from the National Commission for Research Ethics (CONEP-CAAE 59175616.2.0000.0008).

Study setting
Niterói, a municipality of the state of Rio de Janeiro is situated in the Guanabara Bay across from Rio de Janeiro city (22˚52 0 58@S 43˚06 0 14@W)). According to the last national census in 2010 it had a population of 484,918 living in an area of 135 km 2 . The city is divided into 7 health districts for administrative planning. For the evaluation of the impact of Wolbachia mosquito deployments, Niteroi was divided into four release zones and 1 control zone, which are aligned with neighbourhood administrative boundaries (Fig 1). Table 1 shows the baseline characteristics and release summary of each zone.

Community engagement
WMP Brazil's Communication and Engagement (C&E) strategy was developed prior to mosquito releases, following a thorough analysis of geographical, social, political, economic and cultural factors in the proposed release areas as previously described [22].
In Niterói the C&E plan was focused on three key areas: public schools, primary health care units and social leadership, due to their reach and influence within the release area, including into vulnerable communities. Community Reference Groups (CRGs) were also created, to serve as advisory committees populated by representatives of the planned release areas, to inform the activities of WMP Brazil. This group was also responsible for providing feedback on all communication materials and C&E strategies that were proposed throughout the WMP's activities in their areas.
Prior to the release of wMel-infected mosquitoes in each area, a survey of awareness and acceptance of the method was conducted by an independent company. In order to reach a wide range of people living and working in the release areas, time-location sampling was used to survey passers-by in busy public locations in each neighbourhood. Respondents (n = 3485 in total) were 18 years and over, and lived or worked in the neighbourhood where the survey was conducted. The questionnaire was developed with the CRG, and included questions on awareness ("Have you heard about the Wolbachia method?"), understanding after explanation of the method ("Do you understand that this method replaces the population of Aedes aegypti mosquitoes with Aedes aegypti mosquitoes carrying Wolbachia, which have a reduced capacity to transmit dengue, Zika and chikungunya?") and acceptance of the proposed wMel releases ("Do you agree with Fiocruz releasing these mosquitoes with Wolbachia here in your neighbourhood?").

Mosquito production
The Rio wMel-infected Ae. aegypti line described in Garcia et al 2019 [23] was used for releases. The wMel-infected lines were maintained in controlled laboratory conditions, in 900 PLOS NEGLECTED TROPICAL DISEASES cm 2 mesh-sided rearing cages. Each cage contained 2500-2750 adults, and was fed using donated non-transfusional usable human blood (agreement FIOCRUZ/ Hemominas OF. GPO/CCO-Nr224/16), once per week for two to three gonotrophic cycles. As a quality assurance procedure each blood bag was tested for dengue, Zika, chikungunya, Mayaro and yellow fever viruses, as described previously [9,11,24]. Two separate colonies were maintained, a broodstock (kept in Belo Horizonte) and a release-production colony (kept in Rio de Janeiro). Male Ae. aegypti adults (from F0-F1 field collected material) were introduced into the broodstock cages at a rate of 10-20% every 5 generations. This outcrossing frequency was sufficient to maintain kdr resistant genotypes within the broodstock colony throughout its maintenance (S1 Text and S1 Fig). Material from the broodstock colony was then transferred to the releaseproduction colony where it was amplified through 2 amplifications without the addition of field collected males. A minimum sample of 168 mosquitoes from the release-production colony was screened for wMel infection on a weekly basis, using quantitative polymerase chain reaction (qPCR) as described below. wMel prevalence was 100% in all but three weekly screening events, and was never below 97%. Quantitative analysis of wMel in these samples detected a fairly constant wsp:rps17 copy number between 4 to 6 (S2 Fig).
From April 2017 until April 2018 immature stages for adult releases were reared at a density of approximately 1.0 larvae/ml and fed a diet of ground Tetramin Tropical Flakes (Tetra Holding [US] Inc. Germany, Product number 77101). From May 2018, immature stages for adult releases were reared at a density of approximately 2.75 larvae/ml and fed a diet of fish food: liver powder: yeast extract (4:3:1). We found no detrimental effects on outcomes, including development time, size, egg output or wMel density, with increases in larval density up to 2.75/ ml. In both rearing regimes, when approximately 10-30% of larvae had pupated, the larvae/ pupae were sieved and between 180-220 larvae/pupae were placed in a release device. The release device was a cylindrical PVC crystal tube approximately 28 mm in diameter and 250 mm in length, covered with a fixed mesh on one side and a removable mesh on the other side. Adults were allowed to emerge for 5-6 days and were maintained on a 10% sugar solution for 12-36 hours prior to releases. We estimated that the releases were slightly male biased with an average female:male ratio within the devices of 3:4. The release devices were then stacked, sugar-free into boxes for transport to the release site.

Wolbachia deployments
Mosquito deployments took place over a release area of 40 km 2 during a period of 35 months (February 2017-December 2019). Adult wMel-infected mosquitoes were released weekly from a moving vehicle. In zones 1-3 mosquito release points were initially determined using a 50 meter grid overlaid on the release areas, with one release point per grid square. In zone 4 the density of release points was adjusted for the residential population in each neighbourhood, with the aim of releasing a cumulative total of 100 mosquitoes per resident (average distance between release points on a regular grid was 41 meters). In all areas, the initial release points determined on the grids were then distributed to the nearest vehicle-accessible road for vehicle releases (S3 Fig). Releases were staged throughout the urban constructed areas in each release zone. Green non-constructed areas were excluded from releases as they provide less favourable habitats for Ae. aegupti and had few or no human residents. Initial release periods were 10-16 weeks duration, with subsequent re-releases conducted in local areas where wMel prevalence was <40% in 3 consecutive monitoring events as measured at least 4 weeks after the conclusion of releases. This 40% threshold was based on previous estimates of the unstable equilibrium point for wMel, above which invasion can occur [25]. This resulted in re-releases being conducted in approximately 30% of the initial release areas. Most areas of zones 1 and 2 had two periods of releases, zone 3 had three periods of releases and zone 4 only 1 release period.

Wolbachia monitoring
Mosquitoes were collected weekly during and after releases using a network of BG Sentinel traps (Biogents AG, Regensburg, Germany, Product number NR10030) at an average density of 16 BG traps/km 2 throughout release areas (S4 Fig). Once wMel prevalence was detected at >60% in 3 consecutive monitoring events measured at least 4 weeks after the conclusion of releases, trap numbers were reduced to 50% within a neighbourhood (S4 Fig). Mosquitoes were sent to the laboratory for sorting, morphological identification and counting. The number of mosquitoes caught in each BG trap was recorded by species, sex, and in total. Mosquito samples were stored in 70% ethanol until screening for wMel-strain Wolbachia. Screening was performed weekly until week ending 8 April 2018 and fortnightly thereafter.

Wolbachia molecular detection
A maximum of 10 adult Ae. aegypti per BG trap per collection were screened for the presence of wMel using either quantitative polymerase chain reaction (qPCR), or a colorimetric loopmediated isothermal amplification (LAMP) assay. Taqman qPCR was performed on a Roche LightCycler 480 as described previously [16,26]. Briefly, the qPCR cycling program consisted of a denaturation at 95˚C for 5 min followed by 40 cycles of PCR (denaturation at 95˚C for 10 min, annealing at 60˚C for 30 sec, and extension at 72˚C for 1 sec with single acquisition) followed by a cooling down step at 40˚C for 30 sec. LAMP reactions were performed in a Bio-Rad C1000 96-well PCR thermocycler with a 30min incubation at 65˚C as previously described [16]. Individual reactions consisted of 2X WarmStartR Colorimetric LAMP Master Mix (New England BioLabs, Cat# M1800S), primers and 1 μL of target DNA from a 50μl single mosquito squash buffer extraction assay, in a total reaction volume of 17 μL. An individual mosquito was scored as positive for Wolbachia if the Cp (crossing point) value in qPCR was below 28, or if the well in the LAMP assay was yellow upon visual inspection. Equivocal results were counted as negative. Details of primer and probe nucleotide sequences are included in S1 Text.

Epidemiological data
Data on dengue and chikungunya cases notified to the Brazilian national disease surveillance system (SINAN) were used to evaluate the epidemiological impact of Wolbachia releases. Reporting of both diseases is mandatory in Brazil. Dengue notification data for Niterói is available from SINAN since 2007 and chikungunya since 2015. Notified dengue and chikungunya cases reported to SINAN are predominantly suspected cases based on a clinical case definition [27].
Between 2007-2014, approximately 15% of notified dengue cases had supportive laboratory test results, usually from IgM serology. Since the Zika epidemic in Brazil in 2015, laboratory confirmation of dengue has relied on PCR only due to cross-reactive serological responses, and only one dengue case notified in 2015-2020 included laboratory confirmation. For chikungunya, 24% of cases notified in 2015-2020 had supportive IgM serology results. For the purpose of this analysis, we include all notified dengue and chikungunya cases (suspected and laboratory confirmed).
Anonymized disaggregate data on notified suspected and laboratory-confirmed dengue, severe dengue, chikungunya and Zika cases were obtained from the SINAN system through the Health Secretariat of Niterói, for the period from January 2007 (January 2015 for chikungunya and Zika) to June 2020. Population data by neighbourhood of residence from the Brazilian 2010 census (IBGE) was used to estimate the population in each Wolbachia release zone.

Measurement of epidemiological impact
The wMel intervention effect was estimated using controlled interrupted time series analysis performed separately for each release zone compared with the pre-defined control area, and for the aggregate release area compared with the control area, as described in a published study protocol [21]. The primary analysis included data from January 2007 (dengue) or January 2015 (chikungunya and Zika), until June 2020, encompassing 8-37 months of post-intervention observations. For zone-level analyses, negative binomial regression was used to model monthly dengue, chikungunya and Zika case counts in the intervention and control areas, with an offset for population size. Seasonal variability in disease incidence was controlled using flexible cubic splines with knots placed at 6-monthly intervals. For the primary analysis, a binary 'group' variable indicated the study arm (intervention or control). A binary 'treatment' variable distinguished the pre-intervention period and the post-intervention period. The zone-level post-intervention period was defined as four weeks after wMel releases had commenced throughout the whole zone; the corresponding post-intervention period was also applied to the control area for each zone-level analysis. The intervention effect was estimated from the interaction between the 'group' and 'treatment' variables, which allows explicitly for a level change in the outcome (dengue/chikungunya/Zika case incidence) in both intervention and control areas in the post-intervention period. Robust standard errors were used to account for autocorrelation and heteroskedasticity. A mixed-effects negative binomial regression was used to model monthly dengue, chikungunya or Zika case counts in the aggregate release area compared with the control area, with an offset for population size and controlling for seasonal variability in incidence using flexible cubic splines with knots placed at 6-monthly intervals. Clustering of dengue/chikungunya/Zika cases by release zone was modelled as a random effect by including a random intercept at the zone level and allowing for a random slope on the intervention. A binary 'treatment' variable distinguished the pre-intervention period and the post-intervention period, with the control area classified as 'pre-intervention' throughout. Robust standard errors were used to account for autocorrelation and heteroskedasticity. The zone-level and aggregate release area analyses included the pilot release area of Jurujuba within zone 1.
To account for within-zone heterogeneity in wMel establishment and dengue incidence, a secondary neighbourhood-level analysis was also performed in which Wolbachia exposure was determined by the measured wMel prevalence in Ae. aegypti collected from each neighbourhood, and a three-month moving average calculated to smooth the variability in monthly wMel prevalence, categorised into quintiles of exposure. In zone 1 we excluded the neighbourhood of Jurujuba where pilot wMel releases were staggered across seven sectors over a period of 16 months and wMel monitoring was initially done only in small pockets of the neighbourhood where releases had already occurred, because the wMel time-series during this staged release period was not representative of the whole of Jurujuba neighbourhood (whereas the dengue cases data was aggregate for the whole neighbourhood). This analysis included data to March 2020 only, as no Wolbachia monitoring was possible April-June 2020 due to restrictions on movement in response to the Covid-19 pandemic. Mixed-effects negative binomial regression was used to model monthly dengue case notifications by neighbourhood, in each of the four release zones individually and in all zones combined, compared with the pre-specified control zone. The model included population size as an offset and neighbourhood as a random effect. Given the large number of zero dengue case counts (zero-inflation) at the neighbourhood level, an alternative analysis using a zero-inflated negative-binomial model with robust standard errors to account for clustering was considered. Model fit was not improved by accounting for zero-inflation, as assessed using the Akaike Information Criterion (AIC), and was thus not used in the analyses. This secondary analysis was not performed for chikungunya or Zika due to the sparsity of case data at the neighbourhood level.

Sensitivity analyses
As a sensitivity analysis, we excluded pre-intervention observations prior to 2012 to achieve greater balance between pre-intervention and post-intervention period lengths while maintaining sufficient data to inform on pre-intervention trends [28].

Power estimation
Power was estimated for the ITS analysis using 1000 simulated datasets drawn from a negative binomial distribution fitted to a ten-year time series (2007-2016) prior to Wolbachia deployment, of monthly dengue case notifications from release and control zones in Niterói and Rio de Janeiro. The simulated time series of dengue case numbers in the control zones as well as the pre-Wolbachia release dengue case numbers in the treated zones were drawn directly from this model-generated distribution. Post-Wolbachia release dengue case numbers in the treated zones were drawn from the same model-generated distribution, modified by an additional parameter for an intervention effect of Relative Risks = 0.6, 0.5, 0.4, 0.3. For each of these four 'true' effect sizes and a null effect (RR = 1), applied to each of the 1000 simulated time series, the 'observed' effect size was calculated from a negative binomial regression model of monthly case counts in the treated and untreated zones, as described above. Post-intervention time periods of 1, 2 or 3 years were simulated, with the pre-intervention period fixed at 7 years. The estimated power to detect a given effect size was determined as the proportion of the 1000 simulated scenarios in which a significant intervention effect (p<0.05) was observed. These simulations indicate 80% power to detect a reduction in dengue incidence of 50% or greater after three years of post-intervention observations, and a reduction of 60% or greater after two years.

Wolbachia establishment in Niterói
Awareness (prior knowledge of the Wolbachia method) ranged from 36 to 50% and acceptance (agreement with the proposed wMel releases in the neighbourhood) ranged from 65 to 92%, in the public survey conducted prior to releases in Niterói. No negative media nor negative community incidents were registered, and the Community Reference Group endorsed the start of releases.
Heterogeneity in wMel Wolbachia establishment was observed in three of the four release zones (Fig 2). In the initial release area of zone 1, Wolbachia prevalence was greater than 80% in the first quarter of 2020 (up to 11 months post-release) and there was low variability across the neighbourhoods. Local wMel introgression has been more variable in zones 2 and 3, with a median wMel prevalence of 40-70% among neighbourhoods during the post-release period (11 months and 9 months, respectively). In zone 4, a longer post-intervention observation period is required to evaluate the trajectory of wMel establishment. Aedes albopictus is present throughout the city and was detected at a similar abundance in our monitoring network during and after releases (S5 Fig).

Arboviral disease trends pre-and post-Wolbachia intervention
During the ten years prior to the start of scaled Wolbachia mosquito releases in Niterói in early 2017, seasonal peaks in dengue case notifications occurred each year (Fig 3A), usually in March and April (Fig 3D). A median of 2,818 dengue cases were notified each year 2007-2016 (per capita incidence 581/100,000 population), with a minimum of 366 cases in 2014 (75/ 100,000) following a maximum of 11,618 in 2013 (2,396/100,000). In the three years following the start of phased Wolbachia releases, annual city-wide dengue case notifications were 895, 1,729 and 378 in 2017, 2018 and 2019 respectively, and the seasonal peaks in dengue incidence

PLOS NEGLECTED TROPICAL DISEASES
occurred predominantly in the areas of Niterói that had not yet received Wolbachia deployments (Fig 4).
Chikungunya surveillance commenced in January 2015. Between 44 and 533 chikungunya cases were notified annually in Niterói in 2015-2019, with the exception of 2018 when an explosive outbreak resulted in 3091 reported cases; 95% of those occurred in the six months January to June. The highest per capita incidence of chikungunya during the 2018 outbreak was in the untreated control zone (1,413 cases/100,000 population; Fig 5), followed by zone 4 where Wolbachia deployments had not yet commenced (958/100,000). In zones 1, 2, and 3 where deployments were underway and zone-level Wolbachia prevalence was between 20-55%, the incidence of chikungunya case notifications during the 2018 outbreak was 106/ 100,000, 244/100,000 and 201/100,000, respectively.
There were 8,247 Zika cases reported in Niterói between 2015 and June 2020, 91% (n = 7,532) of which were reported in 2015-2016 when Brazil experienced an unprecedented Zika outbreak (Fig 6). From 2017, when phased wMel deployments began in Niterói, until June 2020 a total of 715 Zika cases were notified in Niterói, of which of 95 were reported from areas where wMel deployments had already occurred: 12 in zone 1, 28 in zone 2, 48 in zone 3, and 7 in zone 4.

Reduction in dengue, chikungunya and Zika incidence post-Wolbachia intervention
Using interrupted time series (ITS) analysis to account for underlying temporal trends in case incidence and staggered implementation of the intervention, we found that wMel Wolbachia deployments were associated with a significant reduction in dengue incidence in each of the four release zones (Fig 6A). The magnitude of this reduction ranged from 46.0% (95%CI 21.0, 63.0) in zone 3 to 75.9% (95%CI 62.1, 84.7) in zone 2. Overall, Wolbachia deployments were associated with a 69.4% (95%CI 54.4, 79.4) reduction in dengue incidence in Niterói (Fig 7  and S1 Table). the untreated control zone (dashed line), January 2007-June 2020. Light blue shading indicates the beginning of the epidemiological monitoring period in each zone, one month after initial releases were completed in each respective zone. Darker blue shading indicates the aggregate wMel infection prevalence (right-hand Y axis) in each zone in each calendar month from the start of the epidemiological monitoring period until March 2020 (no wMel monitoring April-June 2020). Panels B,D,F,H show the same data but zoomed into the period from May 2017 -March 2020 and with the dengue incidence axis rescaled, to show more clearly the trends in release and control zones in the post-intervention period. https://doi.org/10.1371/journal.pntd.0009556.g004

PLOS NEGLECTED TROPICAL DISEASES
This Wolbachia intervention effect against dengue was also apparent overall, and in each zone, in the neighbourhood-level analysis that considered quintiles of wMel prevalence in local Ae. aegypti populations, although we found evidence of only marginal additional reductions in dengue incidence at higher levels of Wolbachia beyond 20-40% wMel prevalence ( S6  Fig and S2 Table). There was substantial month-to-month variation in wMel quintiles within neighbourhoods (S7 Fig), which was reduced but not removed by taking a three-month moving average of wMel prevalence. The results were little changed in the sensitivity analysis, which excluded pre-intervention observations prior to 2012 (S8 Fig and S3 Table).
A total of 897 severe dengue cases were reported in Niterói between 2007 and early 2020, 691 of which were from one of the four intervention zones and 206 from the control zone. Only three of these cases occurred in the post-intervention period, two in zone 2 and one in zone 3. The control zone has not had any severe dengue cases reported since 2016. These numbers were too sparse to be analysed using our ITS model, even when allowing for zeroinflation.
We found in ITS analysis that chikungunya incidence was also significantly reduced following Wolbachia deployments in Niterói as a whole (56.3% reduction in incidence; 95%CI 15.9, 77.3) and in three of the four individual release zones (Fig 8 and S1 Table). Zika incidence was PLOS NEGLECTED TROPICAL DISEASES reduced by 37% (95%CI 1.5, 59.5) following Wolbachia deployments in Niterói as a whole, though not in individual release zones (Fig 9 and S1 Table).

Discussion
Large-scale phased deployments of wMel strain Wolbachia-infected Aedes aegypti mosquitoes in Niterói, Brazil during 2017-2019, resulted in wMel establishment in local Ae. aegypti populations at an infection frequency of 33-90% by March 2020, when field monitoring was paused due to the emergence of SARS-CoV-2 in Brazil. More than one-quarter of the total 373,000 residents of the intervention area were living in neighbourhoods where local wMel prevalence was 60% or greater by March 2020, predominantly in zones 1 and 2 where releases commenced earliest. In the remaining intervention areas, wMel prevalence was more heterogeneous and a resumption of entomological monitoring is planned in order to evaluate the longterm trajectory of wMel introgression into the local Ae. aegypti population.
Despite this heterogeneity in Wolbachia establishment, a significant reduction in the incidence of dengue, chikungunya and Zika case notifications was observed in Wolbachia-treated areas of Niterói, compared with a pre-defined untreated control area. This epidemiological impact on dengue was replicated across all four release zones, and in three of the four zones for chikungunya. Aggregate across the whole intervention area, the wMel deployments were associated with a 69% reduction in dengue incidence, a 56% reduction in chikungunya incidence and a 37% reduction in Zika incidence. Given the recognised lack of evidence for efficacy of routinely available approaches to arboviral disease control [4] based on elimination of

PLOS NEGLECTED TROPICAL DISEASES
breeding sites and insecticide-based suppression of adult mosquito populations, and considering the magnitude of the historical burden of Aedes-borne disease in Niterói, an intervention effect of this magnitude represents a substantial public health benefit.
Results from a recent cluster randomised trial of wMel-infected Ae aegypti deployments in Yogyakarta Indonesia demonstrated 77% efficacy in preventing virologically confirmed dengue cases [18], with comparable efficacy against all four dengue virus serotypes. Previous nonrandomised controlled field trials in Indonesia [15] and northern Australia [16,17] demonstrated 76% and 96% effectiveness, respectively, in reducing the incidence of dengue cases notified to routine disease surveillance systems. In each of those sites the trajectory of wMel establishment was more rapid and more homogeneous across the release area than observed in Niterói. In the present study, the epidemiological impact in the area of Niterói where wMel introgression occurred most rapidly and homogeneously (zone 1) was highly comparable with the Indonesian studies: 77% (95%CI 64, 86). Another Wolbachia strain, wAlbB, has been successful introgressed into Ae. aegypti field populations in Kuala Lumpur, Malaysia [29], although with instability in wAlbB frequencies in some release areas after cessation of releases, which the authors attributed to immigration of wild-type mosquitoes into the small release sites (area 0.05-0.73 km 2 ) from surrounding untreated areas.
The reasons for slower and more heterogeneous wMel introgression here, compared to Indonesia and Australia, are not fully understood. A likely contributing factor is that these scaled deployments have largely used adult mosquitoes released from vehicles, which does not deliver as spatially homogeneous a deployment as occurred previously in Indonesia and

PLOS NEGLECTED TROPICAL DISEASES
Australia. In contrast, the small-scale pilot releases in the Jurujuba neighbourhood of Niterói in 2015 achieved rapid and sustained introgression of wMel after 8-31 weeks of egg-based releases [20]. Additionally, Niterói release areas were complex urban environments with high rise areas and large informal settlements, where field activities were frequently interrupted by security issues and where physical barriers to spread [30], spatial heterogeneity in mosquito abundance [31], and limited mosquito dispersal [32] could have contributed to slower wMel introgression. The wild-type egg bank in such a setting is likely to be large and spatially heterogeneous, and would take time to be depleted, which is likely to have contributed to the heterogeneity in wMel introgression and intermediate frequencies of Wolbachia observed in our study. Regular monitoring of the wMel-Ae. aegypti broodstock has demonstrated insecticide susceptibility profiles comparable with wild-type material, so a concern of increased susceptibility to insecticide is not considered to be an issue here. Impaired maternal transmission by wMel -infected females [33,34] and loss of induction of cytoplasmic incompatibility by wMel -infected males [35] has been observed by others at high, but field relevant, temperatures. Exposure of immature Ae. aegypti to very high temperatures in small water containers cannot be excluded as a contributing factor to the wMel introgression patterns observed in Niteroi, especially in the more informal settlements where the urban landscape is more vulnerable to temperature variations. Entomological monitoring in future years will help clarify the longterm trajectory of wMel introgression in Niterói.
In large and complex urban environments, a homogeneous high level of introgression of wMel may prove operationally challenging and slow to achieve, even with optimised release PLOS NEGLECTED TROPICAL DISEASES methods and longer post-release monitoring. This poses the question of what minimum threshold of wMel prevalence is needed to achieve interruption of local arbovirus transmission, and whether a dose-response relationship is observed between wMel prevalence and disease reduction. Predictions from mathematical models have suggested that even in conservative scenarios where scaled Wolbachia deployments only reduce the reproduction number (R0) of dengue by 50%, this could lead to reductions in global case incidence of 70% [1], although the impact is predicted to be highly spatially heterogeneous, with smaller relative reductions in areas with highest transmission intensity. Our findings support this prediction of epidemiological impact with imperfect wMel-mediated transmission blocking, by demonstrating that measurable reductions in dengue, chikungunya and Zika disease accrue even at a moderate prevalence of wMel in local Ae. aegypti populations. A secondary analysis based on measured wMel prevalence and dengue case notifications at the neighbourhood-level found only a marginal increase in the wMel intervention effect beyond 20-40% prevalence, which was unexpected. This analysis also indicated substantial variability in wMel prevalence over time (within neighbourhoods). This may be attributable in part to sampling variability due to small Ae. aegypti catch numbers in some areas but may also indicate true local instability in Wolbachia levels. When combined with people's mobility and risk of acquiring dengue outside their neighbourhood of residence these factors may help explain the non-linear association between measured neighbourhood-level monthly wMel infection prevalence and dengue risk. The absolute abundance of wild-type Ae. aegypti, independent of wMel prevalence, is also relevant to understanding local dengue risk and could not be accounted for in our time series analyses because of a lack of baseline (pre-intervention) mosquito collection data. We cannot exclude that incompatibility within the population of wMel positive and negative Ae. aegypti could impact overall population size and contribute to the observed epidemiological outcomes. Overall, a contribution of indirect effects, confounding by mosquito population size, and imperfect wMel exposure measurement to the observation of an epidemiological impact even at moderate wMel prevalence cannot be excluded, and this observation needs replication in other settings.
This study has some limitations. Deployments of wMel-infected Ae. aegypti were not randomised, so there is the potential for measurement of the intervention effect to be confounded by other factors that differ between the release areas and the pre-defined control area. Routine disease surveillance data is imperfect both in specificity (not all notified cases are true dengue/ chikungunya/Zika cases) and in sensitivity (not all dengue/chikungunya/Zika cases are notified). However, the risk of these factors influencing the measurement of the epidemiological endpoint here is reduced by the inclusion of a parallel control with a historical dengue time series that is highly synchronous with each of the release areas for ten years pre-intervention. The replication of the dengue intervention effect in each of the four release zones, and for chikungunya in three zones, also mitigates the possibility that any parallel change in vector control practices or healthcare seeking behaviour in intervention areas could have confounded the observed result. For chikungunya and Zika, there is substantial uncertainty around the point estimate of the intervention effect because case notifications for these two diseases were very sparse in both release and control areas outside of a single large outbreak in 2018 and 2015-16.
We have demonstrated that wMel introgression can be achieved across a large and complex urban environment over a period of three years to a prevalence in local Ae. aegypti which, while still heterogeneous, is sufficient to result in a measurable reduction in dengue, chikungunya and Zika case incidence. Ongoing entomological and epidemiological monitoring will provide additional information on the trajectory of wMel establishment in areas where releases have occurred more recently, or introgression has been slower, and on the full magnitude and durability of the public health benefit. average of wMel% and excludes the Zone 1 pilot release area of Jurujuba. Point estimates (markers) and 95% confidence intervals (horizontal bars) are from controlled interrupted time series analysis of monthly dengue case notifications to the Brazilian national disease surveillance system (Jan 2007 -March 2020), by neighbourhood, in each release zone and in the aggregate release area. wMel prevalence was calculated as the percentage of trapped Ae. aegypti positive for wMel, in each neighbourhood each month, grouped by quintile. The lowest quintile (wMel 0-20%) served as the reference category for calculation of the incidence rate ratio (IRR) and included the monthly observations within that quintile from the respective release zone, as well as all observations from the untreated control zone (n = 3,021 neighbourhoodmonths observed and n = 11,278 notified dengue cases). This sensitivity analysis excludes all observations prior to 2012, five years prior to the start of releases. Point estimates (markers) and 95% confidence intervals (horizontal bars) are from controlled interrupted time series analysis of monthly dengue case notifications to the Brazilian national disease surveillance system (Jan 2012 -March 2020), by neighbourhood, in each release zone and in the aggregate release area. wMel prevalence was calculated as the percentage of trapped Ae. aegypti positive for wMel, in each neighbourhood in a moving three-month window, grouped by quintile. The lowest quintile (wMel 0-20%) served as the reference category for calculation of the incidence rate ratio (IRR) and included the observations within that quintile from the respective release zone, as well as all observations from the untreated control zone (n = 1,881 neighbourhood-months observed and n = 6,996 notified dengue cases). (JPG) S1 Table. Dengue, chikungunya and Zika incidence rate ratios in Wolbachia-release zones compared to the control zone. IRRs are from negative binomial regression models of monthly case counts (Jan 2007 -June 2020 for dengue; Jan 2015 -June 2020 for chikungunya and Zika), with an offset for population size and 6-monthly flexible cubic splines to account for seasonal effects. The mixed effects model for the aggregate Niteroi release area included a random effect for release zone. (DOCX) S2 Table. Dengue incidence rate ratios with increasing wMel prevalence in Aedes aegypti populations in Niteroi neighbourhoods. IRRs are from mixed effects negative binomial regression models of monthly dengue case counts (Jan 2007 -March 2020) by neighbourhood, with an offset for population size, 6-monthly flexible cubic splines to account for seasonal effects, and a random effect for neighbourhood. (DOCX) S3 Table. Dengue incidence rate ratios with increasing wMel prevalence in Aedes aegypti populations in Niteroi neighbourhoods-sensitivity analysis excluding pre-intervention observations prior to 2012 to achieve greater balance in the length of pre-intervention and post-intervention observation periods. IRRs are from mixed effects negative binomial regression models of monthly dengue case counts (Jan 2012 -March 2020) by neighbourhood, with an offset for population size, 6-monthly flexible cubic splines to account for seasonal effects, and a random effect for neighbourhood. (DOCX)