Experimental procedure
The experimental procedure was an adaptation of the protocol for the tunnel test described in the WHO guidelines for laboratory and field-testing of long-lasting insecticidal nets [32]. While the dimensions and the design of the tunnel followed the WHO guidelines, here, we introduced only one female at a time into the tunnel while the bait was an artificial feeding unit filled with pig blood (Abattoir Bell Schweiz AG, Basel, Switzerland) instead of a restrained guinea pig or a rabbit. To lure the host-seeking mosquito to the feeder, the experimenter (MF) sat behind the feeding unit. The mosquitoes’ flight trajectories were captured with a 3D video tracking system while mosquitoes were exposed to either an untreated or a permethrin-treated net and observed for 10 minutes, instead of the 12–15 hours as per WHO guidelines.
Net samples
We exposed mosquitoes either to a permethrin-treated net (Olyset Net, 150 denier polyethylene monofilament, colour white, 20g/kg permethrin, Sumitomo Chemical Co., Ltd, Tokyo, Japan) or to an untreated net (100 denier multi-filament polyester net, colour white).
Video tracking system
In order to record the mosquitoes’ flight trajectories, we used the software Trackit 3D Fly suite 2.0 (Scitracks GmbH, Bertschikon, Switzerland), installed on a desktop computer (Intel Core i7-8700 CPU @ 3.20 GHz, RAM 16.0 GB, Windows 10). The software was used in combination with two acA2040-90umNIR USB 3 digital cameras (Basler AG, Ahrensburg, Germany), one near-infrared (NIR) light emitting diode (LED) lamp (Raytec pulsestar illuminators PSTR-I96-HV, Raytec Ltd, Ashington, United Kingdom) controlled by a controller (Raytec pulsestar illuminators PSTR-PSU-4CHNL-HV, Ashington, UK), and three custom-made NIR LED lamps (Ingenieurbüro Warnke GmbH, Wietzendorf, Germany).
The video cameras had a resolution of 4 MP and acquired images at a rate of 50 frames per second (fps). We equipped the cameras with Fujinon DV3.4x3.8SA-SA1 lenses (Fujifilm Holdings K.K., Tokyo, Japan) and Midopt BP850 NIR band-pass filters (Midwest Optical Systems, Palatine Inc. Illinois, USA) that only pass light at a wavelength above 850 nm. Both lamp types were NIR LED floodlights, emitting light at a wavelength of 850 nm with a power of 70 W. This combination of cameras, lenses, filters and lamps allows for tracking mosquitoes in both bright light and darkness as mosquitoes are not susceptible to light in the NIR range [34].
The software of Trackit 3D determines the mosquitoes’ position every 20 ms based on stereo image pairs of dark objects against a bright background by triangulation [35] while the positional precision is improved through a Kalman filter loop [36]. We calibrated the system with the software’s built-in calibration function while we set the tracking parameters gain, threshold, exposure time and minimum object area to 0, 9, 4000 µs and 20 pixels, respectively.
In combination with the video tracking software, we used the Open Broadcaster Software version 22.0.2 (Philadelphia Works, Philadelphia, USA) to record videos of the computer screen so that we could later playback the raw images from the two cameras, validating the 3D trajectories where needed, for example, to verify through which hole the mosquitoes passed the net.
Tunnel experiments
We performed the experiments in a specialised, climate-controlled room at the Swiss Tropical and Public Health Institute (Swiss TPH) in Basel, Switzerland. The design and dimensions of the test tunnel were based on those from the WHO tunnel assay [32] with slight modifications allowing for the infrared video tracking system to follow mosquitoes inside the tunnel (Fig. 1).
The external dimensions of the experimental tunnel were 60 cm x 25 cm x 25 cm. The four rectangular panels were made of 5 mm thick acrylic glass with the top, bottom and back panels being opaque white to produce a homogenously illuminated background against which the mosquitoes could be tracked as dark objects. One square panel consisted of a sliding acrylic glass door inserted into a slot located at one end of the tunnel (to the right in the image shown in Fig. 1). The door panel was attached to an L-shaped wooden handle allowing to open and close the tunnel manually. Attached to the door panel was a 30 cm x 30 cm x 30 cm insect rearing cage (BugDorm-1, MegaView Science Co., Ltd, Taichung, Taiwan) into which 10 mosquitoes were released. The cage had openings at the top and at the side in contact with the tunnel to allow for releasing mosquitoes into the cage and for mosquitoes flying from the cage into the tunnel, respectively. The other square panel located at the other end of the tunnel (to the left in the image shown in Fig. 1) consisted of a mesh allowing some airflow and mosquitoes to feed through. Here, we placed a Hemotek membrane blood feeding unit (Hemotek Ltd, Blackburn, UK) filled with 36°C warm, stirred pig blood and covered with parafilm was placed in contact with the mesh on a 12 cm high plastic support outside the tunnel. To track mosquitoes near the blood feeding unit, we placed a piece of white paper between the blood unit and the PSTR-I96-HV NIR lamp. The paper produces a homogenous light by which mosquitoes can be tracked as a dark object against the light background. The net sample, held in a metal frame by magnetic strips, was placed at two thirds the length of the tunnel from the door panel side, dividing the tunnel into a larger and smaller compartment (Fig. 1). The tunnel itself rested on two tables separated from each other by 55 cm.
We used 5- to 10-day-old females from the insecticide susceptible An. gambiae s.s. Kisumu-1 strain and the multi-resistant Tiassalé strain that is also resistant to pyrethroids [37]. The Kisumu-1 strain (MRA-762) was obtained from the Malaria Research and Reference Reagent Center (MR4) in 2011 and since then reared as a colony at Swiss TPH. The An. gambiae s.s. Tiassalé colony was originally obtained from the Centre Suisse de Recherches Scientifiques en Côte d’Ivoire and established at Swiss TPH in 2017. The mosquito colonies were reared in two separate rooms to avoid cross-contamination between them. We maintained the mosquito colonies at 26.7°C (ranging between 26.4°C and 27.1°C), a relative humidity of 69.9% (ranging between 55.6% and 78.8%) and a day-night-cycle of 12–12 hours.
While in the standard tunnel test, 100 female mosquitoes are introduced into the tunnel, here, we tracked only one mosquito at a time. Even though the video system may track multiple mosquitoes at a time, tracking single mosquitoes allows to unambiguously assigning all flight trajectories to the same individual as mosquitoes may fly outside the cameras’ field of view.
In preparation of the experiment, we removed the female mosquitoes from a stock cage in which females and males were allowed to mate, and mosquitoes had access to water and a 10% sucrose solution ad libitum. Prior to the tunnel assays, we starved the female mosquitoes of the sucrose solution for 8 hours.
We performed experiments with both strains under subdued light at an intensity of 50 Lux to simulate the conditions according to the mosquito species’ diurnal biting pattern. The room with the experimental setup had an average temperature of 25.9°C (ranging between 24.9°C and 27.5°C), an average relative humidity of 68.6% (ranging between 53.4% and 83.1%) and an average CO2 concentration of 564 ppm (ranging between 478 ppm and 636 ppm). Ambient temperature, humidity and CO2 were constantly monitored with Tinytag data loggers (Gemini Data Loggers Ltd, Chichester, UK).
In preparation of an experiment, we cut the net in pieces of 20 cm x 20 cm. We holed each net piece with 9 holes of 1 cm in diameter with 1 hole in the centre and 8 holes equidistant from each other and at 5 cm from the border as stated in the WHO guidelines [32].
At 16:00 hours, 30 minutes before the start of the test, we introduced 10 starved mosquitoes into the release cage for acclimation. The Hemotek feeding unit was filled with 1 ml of pig blood, covered with parafilm, connected to a Hemotek PS6 Power Unit to keep the blood at 36°C and placed in contact with the mesh at the end of the tunnel. Then, the holed net sample was inserted into the tunnel. The sequence of treated and untreated net samples was randomised using the function ‘sample’ in the R version 4.1.2 [38].
In addition of releasing one single mosquito at the time, we also adapted the standard protocol in that we left the mosquito for only 10 minutes in the assay instead of 12–15 hours. It allowed for running 10 replicates per day and preliminary experiments showed the time to be sufficient for a mosquito to reach the other end of the tunnel under normal circumstances.
At 16:30 hours, we started the video tracking. Then, the experimenter pulled the handle to release the mosquitoes into the tunnel and blew twice air from his mouth into the cage to provide CO2 and odour stimuli, triggering the mosquito to initiate host seeking [39]. Once a mosquito entered the tunnel, the release cage was closed again to keep the remaining mosquitoes in the cage. At the end of the 10-minute exposure, the video tracking was stopped and the mosquito was removed from the tunnel. We repeated the procedure until 5 mosquitoes were exposed to the same net piece. Then the net was removed from the tunnel and replaced by the other net while the same procedure with the remaining 5 mosquitoes was repeated. Thus, from the batch with 10 mosquitoes, 5 were exposed to the permethrin-treated and 5 to the untreated net. Before changing nets, and regardless which net was used first, we washed the tunnel with neutral soap to remove insecticide residues from the tunnel.
Data analysis
The readouts of the tracking system were Cartesian x-, y- and z-coordinates with a time stamp at a temporal resolution of 50 Hz. We stored the data as CSV files on the hard drive of the desktop computer while the positional data were already filtered for outliers and missing observations. We imported the data from the CSV files into R version 4.1.3 [38] for data analysis. For creating the plots we used either the ‘rgl’ [40] or ‘ggplot’ [41] packages.
As the data points correspond to a discrete time series with one measurement every 20 ms, we applied the finite helix fit (FHF) approach of Crenshaw et al. [42] which has previously been applied to analyse flight trajectories of tachinid flies [43]. The FHF reconstructs a 3D trajectory by decomposing the trajectory into its translational velocity V, with speed s, curvature κ and torsion τ as a function of time.
For the analysis of the spatial distribution of mosquito activity at the net and frame interface, missing positions of mosquitoes were first filled with the last tracked position using the ‘na.locf’ function from the ‘zoo’ package [44]. We then virtually divided the net surface and the frame area around the net into a grid with 5 mm x 5 mm raster cells (i.e. pixels). From those pixels, we created heat maps for visualisation and data analysis purposes by counting the time spent by each mosquito within each pixel. We considered mosquitoes being in contact with the net if they were within 3 mm from the virtual position of the net. Using the R package ‘spatstat’ [45], we also measured the clustering of the mosquitoes’ net contacts by transforming our data frames into point patterns and performing a nearest neighbour measure Clark-Evans test [46]. The Clark and Evans aggregation index R is a crude measure of clustering or ordering of a point pattern. It is the ratio of the observed mean nearest neighbour distance in the pattern to that expected for a Poisson point process of the same intensity. A value R > 1 suggests ordering, while R < 1 suggests clustering.
To assess differences in mortality, blood feeding and net passage rates as well as the mosquitoes’ preference to land at the upper or lower and left or right half of the net, we estimated the odds ratios and their 95% confidence intervals (95% CI) between the permethrin-treated and untreated net for each mosquito strain. The estimates were based on generalised linear models with a binomial error distribution and a logit link function using the R function ‘glm’ in the ‘stats’ package [38]. To assess the differences in velocity, curvature and torsion as well as the time spent in contact with the net and frame, and the average height at the net and frame we fitted linear models using the function ‘lm’ from the R ‘stats’ package [44]. We used the “offset” function in the model to compare the average surface coverage of net and frame as a function of the average time spent at the net. Cochran-Mantel-Haenszel and Fisher tests were used to assess overall changes in hole counts between the permethrin-treated and untreated nets.
We set the level of statistical significance at α = 0.05.