Studies were conducted in the mesoregion of Araçatuba (21.2089S; 50.4328W; ca. 11,250 km² and ca. 700,000 inhabitants) in northwest São Paulo State, Brazil. A total of 280 households in 42 sRCT clusters were included in the Araçatuba region (Fig. 1; Additional file 1: Table S1). The climate in this region is the Aw type (tropical sub-warm and sub-dry) according to the Köppen-Geiger classification  with two distinct seasons: a dry and cool season from April to September (autumn through to winter), and a hot and wet season from October to March (spring through to summer). The mean annual temperature was 23.8 ºC (min: 17.0, max: 30.6), total annual rainfall was 1309 mm, and the wettest months were January, February and December in decreasing order of rainfall (2014-2016). Climate data (rainfall and temperatures) were also obtained from a weather station located at Araçatuba city from July-2015 to April-2016 . This station was selected to be representative for the 42-clusters studied (the farthest cluster was located 110 Km away from the station in straight line).
All experiments were carried out within private households and within their yards either at the front or back of the house. A household was defined as the house (where the owners sleep, have a kitchen and living room), the yard containing a dog sleeping site and a chicken roost (either a purpose built structure and/or a roosting tree). The average number of hosts per household was (Min, Max; X ± SD): dogs (1, 12; 2.65 ± 1.80), chickens (1, 125; 24.51 ± 21.80) and humans (0, 10; 3.50 ± 1.83). Other poultry (geese, guinea fowls, ducks) and other animals (pigs, goats) were common and kept within the yard which may also have contained fruit trees, flowers or shrubs.
Study design and trapping
The study design followed that of the previously described sRCT [18,19] and collections of mosquitoes and biting midges were concurrently made when collecting sand flies. Clusters, households, and dogs were recruited in a three-step procedure (recruitment, cluster stratification and randomisation, and treatment allocation) .
The collections were made in each of the three arms of the trial; (a) synthetic pheromone + insecticide co-located in chicken roosting sites including chicken sheds (PI-arm); (b) deltamethrin-impregnated collars fitted to dogs (DC-arm); and (c) a placebo control (C-arm). In this study the sand fly pheromone has no effect on the behaviour of the Culicoides or biting midges.
Within the PI-arm, microencapsulated lambda-cyhalothrin was sprayed using a hand-compression sprayer (GUARANY 441-10 compression sprayer, Guarany Industria e Comercio Ltda, SP) according to the guidelines of the Brazilian Ministry of Health of São Paulo State . The pheromone lure containing 10 mg of synthetic pheromone for sand fly attraction was placed near the insecticide sprayed area. The pheromone is known to be highly specific, with no attraction even to other subspecies of Lu. longipalpis sand flies , therefore we excluded any effect on mosquitoes and biting midges. Sprayed sites were mostly (i) variable size (open, close, semi-close) chicken sheds, (ii) roosting trees from ground level to 3 m up the roosting tree particularly focussed on roosting branches, and to a lesser extent (iii) on walls adjacent to ground roosting chickens or similar unusual sites (3 m2 area).
Within the DC-arm, each dog living in the dwelling was provided with a collar impregnated with 1.0 g of deltamethrin (Scalibor® Dog Collar, Intervet Productions S.A., France). Collars were replaced every 5-6 months across the study as needed according to the manufacturer instructions.
Control-arm (C), chicken shelters were sprayed with pure water (in the same manner as PI-arm) rather than insecticide, and dogs received a placebo collar. Households selected for the C-arm were described as insecticide-free by the householders as they had no previous residual insecticide application.
The whole trial lasted for 42-months and was divided into rounds that coincided with the timing of the insecticide application in the PI arm and the water spraying in the C arm. The applications of sprayed insecticide and water were carried out every three months between January 2012 and March 2016 giving a total of 17 applications (four rounds per year).
Thus, in this current study we evaluated the impact of the residual insecticide lambda-cyhalothrin biting midges in the chicken roosts, dog sleeping places, and the interior of people’s houses (inside dwellings).
Adult mosquitoes and Culicoides biting midges were collected using CDC suction traps (HP Biomédica, Minas Gerais, Brazil) employing a standard incandescent bulb, and adapted to be powered by a rechargeable 6V-battery . Traps were attached to a fine mesh collecting bag suitable for Culicoides. Trapping rounds were implemented for one day per round per household during a period of 18 h (set up in the afternoon and retrieved the following morning) approximately every three months after the lambda-cyhalothrin or deltamethrin-dog collar application. Each new round where trapping took place was referred to as a trapping round. After 13 rounds of insecticide intervention, we started four trapping rounds for both biting Diptera groups (round 14: 20 July - 10 August 2015, round 15: 15 October - 05 November 2015, round 16: 12 January - 27 February 2016 and round 17: 11 April - 03 May 2016), keeping insecticide interventions unaltered until the end of April 2016. Thus, the final dataset was generated from 123, 110 and 112 trapping days in 103, 88 and 89 households in 14, 12 and 13 intervention clusters in C-arm, PI-arm and DC-arm respectively, for each Dipteran group.
The three CDC traps per household were one located close to a chicken roosting site (e.g. chicken shed or roosting tree), one at the dog sleeping site (e.g. a dog pen or kennel), and one within the house (e.g. a living room, kitchen or bathroom, to minimise disturbance of the residents). In the infrequent event of heavy rain or strong wind, the days’ collections were discarded, and trapping was repeated the following day.
Sample processing and species identification
The live collected insects were placed in a -20º C freezer for 20 min to kill them prior to being placed in 70% alcohol. They were stored until the culicids were separated from the Culicoides, sorted by sex and counted under a binocular stereomicroscope (Quimis Ltda., São Paulo) at x4 magnification.
In Culicidae, female morphological features were not conclusive because of their preservation in alcohol. Male culicids were identified to species level based on male genitalia morphology. Because of the large numbers of Culex specimens, only a subsample (ca. 30% of the total catches) were randomly selected from the three household locations and slide-mounted for determination of species. Heavy-sclerotized male genitalia was first cleared (10% potassium hydroxide for 24 h), then dehydrated (ethanol series from 70% to 100%) and finally immersed in a clearing agent (eugenol) before being mounted in balsam and allowed to dry at room temperature for several days (adapted from Consoli and Lourenço-Oliveira et al.) . Specimens were identified in the Laboratorio de Transmissores de Hematozoários of the Institute Oswaldo Cruz (IOC, Rio de Janeiro, Brazil) using taxonomic keys [3,24–27]. Culicoides species identification was based initially on wing pattern and then confirmed by mounting the specimens directly in Canada Balsam on glass slides, allowed to dry at room temperature for several days, and identified with the appropriate taxonomic keys [28–30] and with access to the reference collection of Neotropical Culicoides housed at the Museo de La Plata, Buenos Aires, Argentina. Voucher specimens of both Diptera groups are available upon request.
Data were statistically analysed for impact of insecticide intervention (abundance and distribution) and climatic variables (temperature and rainfall). Household covariate data, the abundance of people, dogs, and chickens were collected separately from Diptera abundance, being recorded once per round as part of routine trial activities . As such, the per household covariate data recorded on the date closest to that of Diptera capture was assumed to be representative for each household. Data were matched to Dipteran counts by household ID and date. To assess the impact of the insecticide interventions, we compared changes in the total numbers (as well as numbers of males + females separately) of mosquitoes and biting midges captured per household, and at each of the described house, dog, and chicken capture sites. The abundance and distribution of both Dipteran groups inside houses, dog, and chicken sites were compared with the C-arm as it is a better representative of the natural dispersion compared to the treatment arms.
Daily Diptera trapping records per household were excluded from the analysis if the record was incomplete e.g. where any data such as trap location within households was missing. Similarly, data were also excluded if household covariate data was missing. Outliers, such as households associated with unusually high host abundance (>1000 chickens such as chicken farms) are also excluded from analyses.
Being highly over-dispersed, Diptera counts were analysed by negative binomial regression. Household host abundances of humans, dogs, and chickens plus seasonal variation between rounds were expected to confound capture rates of biting Diptera, thus, we adjusted for these by inclusion trapping round and host abundance as fixed a priori predictors in all Diptera count analyses. Finally, repeated sampling across municipalities and within some households led to important structuring in the data. This was accounted for in all multivariate models by clustering on the highest level of structuring, municipality .
Raw monthly data of the control arm were used to plot monthly pattern of capture rate over climatic variables as without the intervention effects it was considered to be the most indicative of seasonal trends. The 3-day average temperature/rainfall associated with each Diptera trapping day were used for the statistical analysis, confirming any association between Diptera count on a given day and local climate variables. Climatic plots were constructed using Geometric-Williams (GW) means plus 95% CI to make a fairer comparison due to overdispersion over daily Diptera capture rates. All data were analysed in STATA v.15 (StataCorp LP, College Station, TX).