Mosquito colony
All mosquitoes used in this study were colony-reared An. gambiae (Kisumu strain). Eggs to establish the colony were initially obtained from the Malaria Alert Centre, Blantyre, Malawi, and the colony was maintained in the laboratory facilities at Majete Wildlife Park in Chikwawa District, Malawi. The colony rearing facility was not climate controlled, and the temperature and relative humidity in this facility ranged from 24 °C to 36 °C and from 62% to 85%, respectively. Mosquitoes were allowed to blood-feed twice per week on a human arm, and eggs were distributed over larval rearing trays (46 × 30 × 9 cm) filled with water from a well near the laboratory facility or from a tap at the nearby Kapichira Power Station. Each tray held 300–400 larvae, fed on ground pellets of Marltons koi and pond fish food (Marltons Pet Care Pty Ltd, Westmead, South Africa). Pupae were collected daily and placed in cages for emergence to adults. All cages with adult mosquitoes were provided a 10% sucrose solution via a piece of soaked cotton wool. Cages with experimental mosquitoes were not provided with a blood meal prior to the experiments.
Experimental set-up
Experiments were performed in a semi-field screened enclosure measuring 12.0 × 12.0 × 2.1–4.0 m (length, width, height) at the Majete Wildlife Park in Chikwawa District, Malawi. The walls of the screened enclosure were made from fiber glass, mosquito-proof screening (Phifer Inc, Tuscaloosa, AL, USA), and the roof was a waterproof tarpaulin. Within this enclosure, we built an experimental house measuring 5.0 × 3.0 × 2.2–2.7 m (length × width × height) (Fig. 1). The walls of the experimental house were constructed from locally produced bricks and plastered with cement, and the roof was made of corrugated iron sheets, including a 20-cm overhang. The front wall of the experimental house was fitted with a door (inner dimensions 197 × 60 cm) in the middle, two windows (30 × 30 cm) and four removable eave frames (inner dimensions 90 × 10 cm per frame). The back wall of the house was also fitted with two windows and four eave frames, but no door. The wooden door, window frames and eave frames were painted black with water-based chalkboard paint.
The windows and eave frames could be left completely open, fitted with insect screens or closed completely with wooden shutters. The screens were made of charcoal-colored fiber glass (Wire Weaving Co. Dinxperlo, The Netherlands), and the shutters were made of plywood painted black with water-based chalkboard paint. Using this system, we were able to systematically investigate the effect of window and eave closure and screening on mosquito house-entry behavior. The door remained closed overnight for all experimental treatments.
Two beds were positioned inside the house, one along each outside wall, and each bed was covered with an untreated bednet. During experimental nights, one adult man slept in each bed, under the bednets, to act as a bait for mosquitoes. Three pairs of adult men volunteered to sleep in the house for 10 experimental nights each. Written informed consent was obtained from the volunteer sleepers. The College of Medicine Research and Ethics Committee (COMREC) in Malawi approved the study (Proposal Number P.02/19/2598). A US Centers for Disease Prevention and Control (CDC) light trap (John W. Hock Ltd, Gainesville FL, USA) was placed near each bed to collect a sample of the mosquitoes that entered the house [28, 29].
Camera and real-time mosquito tracking set-up
A multi-camera videography system was used to track the three-dimensional (3D) flight kinematics of An. gambiae mosquitoes around the experimental house. The videography system consisted of four synchronized machine-vision cameras (Basler acA2040-90umNIR, USB 3.0; Basler AG, Ahrensburg, Germany), equipped with 16-mm f1.4 wide-angle lenses (Kowa LM16HC; Kowa Optical Products Co., Ltd., Nagoya, Japan), with lens aperture set at f2.8. The cameras were operating at 50 frames per second (fps), with a 1-ms exposure time. To improve the light sensitivity of the cameras, pixels within each 2 × 2-megapixel camera image were binned 2 × 2. Binning combines the charge from adjacent pixels (in this case, 2 × 2-pixel bins), resulting in increased light sensitivity but a reduced spatial resolution (in this case, reduced to 1 × 1 megapixel). Image capture on the cameras was synchronized by means of an external trigger pulse, generated by an Arduino Uno microcontroller board (Arduino, Monza, Lombardy, Italy) (https://github.com/strawlab/triggerbox.git).
To protect the cameras and lenses from water, heat and dust, each set was placed in a camera housing (Transpac THP 4000; Basler AG, Ahrensburg, Germany). These camera housings were mounted onto an aluminum frame (MayTec Aluminium Systemtechnik GmbH, Olching, Germany) that was fixed to the concrete floor on which the house was built (Fig. 1b). The cameras were placed at an approximate distance of 2.5 m from the front wall of the house, at heights between 0.8 and 1.3 m. As a result, the camera system imaged the front, right side of the experimental house, including half the door and one window. The cameras were oriented slightly upwards to film the volume below the roof near the eave area. The dimensions of the area in front of the house where mosquitoes could be tracked were approximately 2.5 × 1.0 × 1.5 m (Fig. 1d).
The filming volume was illuminated with eight near-infrared light-emitting-diode (NIR-LED) lights (two ABUS TVAC71000-60° lights and six ABUS TVAC71070-95° lights; ABUS, Volmarstein, Germany). The NIR-LED lights were mounted on a frame placed on the concrete slab directly below the area of interest (Fig. 1b), and the lights were directed upwards and arranged to uniformly light the filming volume near the eave and window, aiming for optimal contrast between the illuminated mosquitoes and the dark background of the house.
We used an automated tracking software [30] to track in real-time the positions of multiple mosquitoes flying in the four camera views, and from these we reconstructed the 3D flight tracks. The tracking software ran on a single laptop (Lenovo ThinkPad P51; Lenovo, Beijing, China) with an Intel Xeon E3-v6 processor and Ubuntu Linux operating system, which performed the real-time image analysis and object tracking for all four cameras, as well as the 3D flight track reconstruction. Based on pilot recordings, sensor gain was set to 1.0 for all cameras, and the maximum number of simultaneously tracked mosquitoes was set to 10. Tracks were reconstructed only when the mosquito was visible in at least two of the four camera views. A dynamic background model was used with update intervals for each 100 frames and a 1% weight factor to compensate for slow changes in illumination conditions.
Cameras were calibrated with the multi camera self-calibration routine [31] by tracking a single moving LED light with each of the four cameras (Cree SunBright 535 nm Green LED; CreeLED Inc, Durham, NC, USA). This calibration was aligned to world reference points based on landmarks on the experimental house. The resulting coordinate system in the world reference frame was defined as X, Y, Z, with the X-axis oriented perpendicular to the house front wall, the Y-axis oriented parallel to the house front wall along the ground and the Z-axis oriented vertically. We defined values within this coordinate system as {x, y, z}, with the origin {x, y, z} = {0,0,0} located against the house front wall (x = 0), on the ground in front of the house (z = 0) and (y = 0) at the right side of the door frame as observed from the cameras.
The calibration procedure was repeated every experimental day to correct any inadvertent change in camera position. A correction for lens distortions was generated for each camera at the start of the experiment, using a 6 × 10 checkerboard pattern with 90-mm squares. Distortion parameters were computed using openCV procedures (https://docs.opencv.org). Tracking results were corrected for lens distortions.
Videography experiments were performed from 20:00 to 04:00 h. If volunteers briefly left the experimental house during the night, a 5-min buffer period was marked prior to leaving and post re-entering the experimental house. Tracking data within those time slots were removed from further analyses.
Eave and window modifications
We evaluated five experimental house modifications (Fig. 2). For our control condition, both the eaves and windows were fully open (eaves open – windows open [EO-WO]). We used two treatments to test the effect of window modifications on mosquito house entry behavior. In the first treatment, we screened the windows and left the eaves open (EO-WS), and in the second treatment we closed the windows while leaving the eaves open (EO-WC). To test the effect of eave modifications on mosquito house entry behavior, we used treatments in which we screened or closed the eaves while, in each case, screening the windows (ES-WS and EC-WS, respectively).
Experimental procedure
Before each experiment, the house was prepared by closing, screening or leaving open the eaves and windows, as randomly assigned for each replicate night of the study (Fig. 2). Each treatment was in place for 6 replicate nights (see experimental treatment schedule in Additional file 1: Table S1).
On the day of each experimental replicate, 500 female mosquitoes (5–8 days old and not previously blood-fed), were selected before 12:00 h and set aside in the insectary in a release bucket (diameter 12.5 cm, height 12.5 cm), covered with a mesh and provided with water-soaked cotton wool. Two volunteers slept inside the house under untreated bednets, starting at 19:30 h. The volunteers’ heads were positioned towards the front (door) side of the house, and each pair of volunteers shifted beds (to the left or right side of the house) after each replicate. At 19:30 h, the two CDC light traps at the end of each bed were turned on, with their lights switched off, and the bucket with mosquitoes was placed in the screened enclosure, 5.8 m in front of the experimental house.
At 20:00 h, mosquitoes were released from the bucket by lifting the mesh using a fishing line operated from outside the screened enclosure. Mosquito flight was tracked until 04:00 h, after which the CDC light traps were turned off, and the volunteers could leave the house. Any temporary absence of volunteers during the recording period was recorded in a logbook. A Prokopack aspirator (John W. Hock Company) was used to collect mosquitoes from inside the experimental house at 04:00 h. Together with these Prokopack catches, CDC light trap catches were briefly frozen, and the collected mosquitoes were then counted. Mosquitoes remaining in the release bucket were also counted, and the number of responding mosquitoes for each replicate night was defined by subtracting the number remaining in the release bucket from the initial 500 mosquitoes. Remaining mosquitoes found inside the screened enclosure later that day were removed with the Prokopack and discarded after freezing. Experimental replicates were carried out no more frequently than every other day to ensure proper preparation and to allow any uncaught mosquitoes remaining in the screened enclosure to die before the next experimental replicate.
Data analysis
The real-time tracking algorithm used a Kalman predictor to reconstruct 3D flight paths from stereoscopic videography data [30], and thus the output data consisted of Kalman-filtered flight paths defined by location, flight velocity and the Kalman covariance error e(t). In post-processing, we filtered the resulting database of flight tracks in two steps. First, to remove potential extrapolation errors from the Kalman predictor, we deleted the end of tracks if either the estimated flight speed exceeded 1.5 m/s or the Kalman covariance error was > 0.01. Second, we then discarded all tracks that were shorter than 10 cm or less than 0.2 s (10 video frames at 50 fps). These settings were based on a sensitivity analysis and the assumption that flying Anopheles mosquitoes have a maximum flight speed of < 1.5 m/s. The resulting flight paths consisted of the temporal dynamics of the 3D location {x(t), y(t), z(t)} and velocity {u(t), v(t), w(t)} of each flying mosquito; these were used for our subsequent analyses.
We used all combined flight tracks per treatment to calculate average mosquito density distributions and flight velocity distributions around the house. For this, we divided the filming volume into 40 × 40 × 40 voxels (spatial bins), resulting in an approximate voxel size of 5 cm in the X- and Z-direction, and 7.5 cm in the Y-direction. In each voxel we estimated the mosquito density as the relative proportion of time mosquitoes spent in that voxel, defined as T* = Ti/Ttotal, where Ti is the time spent in voxel i, and Ttotal is the total flight time. We visualized these density distributions as heat maps projected on three two-dimensional (2D) planes (X–Y, X–Z and Y–Z). We determined the flight velocity vector in each voxel as the mean flight velocity of all mosquitoes that passed through that voxel. We visualized the velocity distributions using streamline plots derived from these velocity fields, projected on the same set of 2D planes as for the density distributions (X–Y, X–Z and Y–Z).
For measuring and comparing flight activities near the eave and window area, we defined volumes-of-interest around the eave and window (Fig. 1e, f, respectively). These volumes had the same rectangular or square shape as the eave or window, respectively, but extended 10 cm on each side (in the Y- and Z-direction). The volumes started at the wall and extended 30 cm outward in the direction perpendicular to the wall (in the X-direction). We then identified all flight tracks that intersected these volumes around the eave and window. Based on these, we quantified flight activity around the window and eave using the time that mosquitoes spent in the corresponding volumes. We determined this time spent in each volume by summing all durations that flight tracks remained in the defined volume; this was done for each experimental night and for an array of time bins with a temporal resolution of 10 min.
Next, we used the flight tracks around the window and eave to study when and how mosquitoes visited the window and eave, and when and how they arrived, departed, remained in and returned to these volumes. ‘Arrivals’ were defined as flight tracks that started at least 10 cm outside the volume-of-interest and ended within the volume. ‘Departures’ started within the volume-of-interest and ended at least 10 cm outside the volume. ‘Visitors’ started outside the volume-of-interest, entered the volume, left the volume and finally ended outside the volume. ‘Returnees’ started inside the volume-of-interest, left the volume, re-entered the volume and finally ended inside the volume. ‘Remainers’ started and ended inside the volume-of-interest, without moving outside the volume. It should be noted that if a flight track ended within the window or eave volume, the mosquito might have entered the house or might have landed on the house, because the tracking algorithm only tracked mosquitoes flying outside the house.
Based on these data, we determined the number of mosquitoes that showed each type of flight behavior (visiting, arriving, departing, remaining and returning). We then used the flight kinematics data to determine the behavior-specific flight dynamics around the eave and window. Specifically, we constructed streamline plots, both per treatment and across all 30 replicates, for all mosquitoes that arrived at the volumes around the eave and window. To focus on the approach kinematics only, we removed the parts of the tracks after arrival.
We used analysis of variance (ANOVA) to test for differences among treatments in various flight kinematics and house entry parameters. The dependent parameters were the number of responding mosquitoes, the percentage of responding mosquitoes collected inside the experimental house, flight track duration (time spent) and the number of flight tracks. We used Tukey’s HSD for pairwise comparisons when the ANOVA test showed a significant difference between treatments. We also used ANOVA to test for differences in house entry rates among the three pairs of volunteer sleepers. We defined P < 0.05 as significant, 0.05 ≤ P < 0.10 as marginal, and P ≥ 0.10 as non-significant.
![[UPDATE] Romero Games Employee Says ‘Whole Studio’ Subject to Layoffs After Microsoft Pulls Funding for New Shooter from Doom Co-Creator John Romero](https://afnnews.qaasid.com/wp-content/uploads/2025/07/1751614671_gettyimages-2148301815-1751541798386.jpg)
