Experimental animals
For all laboratory experiments at Wageningen University, female Anopheles coluzzii mosquitoes were used. These came from a laboratory-reared colony that originated from Suakoko, Liberia in 1987. The colony is housed in the Laboratory of Entomology (Wageningen University & Research, The Netherlands) with a clock shifted 12 h light:12 h dark cycle, and at fixed temperature of 27 °C and relative humidity of 70%. Adults were kept in BugDorm (MegaView Science Co. Ltd., Taiwan) cages (30 × 30 × 30 cm) and fed sugar water solution with 6% glucose. Additionally, they were blood-fed daily with human blood (Sanquin, Nijmegen, The Netherlands) using a membrane feeding system (Hemotek, Discovery Workshop, UK). Female mosquitoes could lay their eggs on wet filter papers that were then moved to plastic trays filled with water. Emerging larvae were fed with Liquifry No. 1 fish food and TetraMin Baby (Tetra Ltd, UK). Finally, new pupae were placed in new BugDorm cages to emerge. Males and females were kept together so they could mate. Non-blood-fed adult females (age = 9.8 ± 1.4 days (mean ± std)) were collected between 12 and 16 h before experiments.
Semi-field experiments in Tanzania were done using 3 to 8 days old Anopheles gambiae s.s. female mosquitoes. These mosquitoes were reared under standard insectary conditions of 27 ± 5 °C (room temperature), 40–100% relative humidity and a 12L:12D cycle. Larvae were fed ad libitum on TetraMin fish flakes (Tetra Ltd., UK). Adult mosquitoes were kept in metal cages (30 × 30 × 30 cm) and fed ad libitum on a 10% glucose solution. Female mosquitoes used for the rearing were blood-fed with cow blood using a membrane feeding system (Hemotek).
Odour-baited traps
The traps used in these experiments were the BG-Suna (Biogents, Germany) and prototypes of the new M-Tego trap (PreMal b.v., The Netherlands). The BG-Suna was used in a standing orientation, thus mimicking the BG-Sentinel, as this position was found to have a better capture efficiency and attractiveness than the hanging BG-Suna [18, 28]. In both traps, an odour source containing one MB5 blend was used to simulate human skin odour [28, 29], and CO2 to simulate human breath. For the laboratory experiments, CO2 was provided using a pressurized canister, and consisted of a mixture of 5% CO2 with 95% air at a flow rate of 200 ml/min. For the semi-field tests, CO2 was produced using a mixture of 17.5 g of yeast and 500 g of molasses in 2 L of water [30, 31]. As in Cribellier et al. [18] the CO2 pipe of the BG-Suna was shortened and the top of the inlet was levelled to minimize blind spots of the cameras.
The M-Tego is a novel trap developed by the authors (see author contributions for details) at Wageningen University (The Netherlands) in collaboration with industrial designers from Delft University of Technology (The Netherlands) (see Additional file 1: Fig. S1). In this study, prototypes of this novel trap were used, from here on those prototypes will be referred to as M-Tego traps. The M-Tego trap uses a similar counter-flow principle as the BG-Suna to attract and capture mosquitoes, and both traps use the same brushless 12 V dc fan. With a diameter of 30 cm and a height of 38.8 cm, the M-Tego is smaller than the BG-Suna trap that has a diameter of 52 cm and a total height of 39 cm. Its inlet is slightly higher than that of the BG-Suna (9.5 cm vs 8.3 cm with levelled inlet) but both inlets have the same diameter of 11 cm. The M-Tego has a foldable black polyester tarpaulin bag (70 g per sq m, Gamma, The Netherlands), which makes transportation easier, as well as an HDPE insect net (Howitec, The Netherlands) on the top of the tarpaulin bag, to allow the outward circulation of the odour-saturated air. Additionally, the M-Tego uses an inlet module with an integrated catching cage that simplifies the removal of caught mosquitoes (see Additional file 1: Fig. S1). These design decisions improve user-friendliness and aim to reduce fabrication costs, which is beneficial for a vector control tool in rural Africa. To generate heat similar to that produced by a human body (37 °C), the M-Tego uses a 2-m Nichrome wire (diameter 0.5 mm) wrapped around the top of its inlet. The heater has an electric power requirement of 9.6 W (12 V and 0.8 A)). Finally, the trap can be filled with 1 L of warm water at 40 °C to increase local relative humidity and temperature. The wire is not in contact with the water and thus cannot warm it up. Instead, the water needs to be warmed up passively during the day or using exterior means.
Experimental set-ups
Three experiments were performed. First, in dual choice testing in the laboratory, two traps were placed next to each other in a flight tent. Mosquitoes were released in the tent where they were free to fly around the two traps. Using this set-up, trap capture percentages were compared to each other. Secondly, semi-field experiments were performed inside three large screen houses in Tanzania. In each screen house, one of the tested traps was placed next to a replicate of a rural African house. The numbers of released mosquitoes that were captured by each trap were then compared. Third, to study the flight behaviour of mosquitoes around the M-Tego, with or without additional host cues, their flight trajectories were tracked in the laboratory in the vicinity of the trap using machine vision techniques.
Dual-choice experiments
The goals of the dual-choice tests were, first, to benchmark the capture performance of the M-Tego by comparing it to the well-established BG-Suna and, secondly, to quantify the effect of adding short-range host cues on the capture performance of the M-Tego. Five trap conditions were tested versus the same BG-Suna: the BG-Suna #2 (control), the M-Tego without additional cues, the M-Tego with heat, the M-Tego with warm water, and the M-Tego with heat and warm water (Fig. 1a).
The dual-choice tests were performed in a netting cage of 2.9 × 2.5 × 2.5 m (Howitech, The Netherlands, see Additional file 1: Fig. S2) inside a climate-controlled room (ambient temperature = 26.1 ± 0.9 °C (mean ± std), and relative humidity = 72.9% ± 3.9%). On each side of the cage, a trap could be placed above the centre of a 1 × 1 m horizontal white ground plate. These two plates were placed in front of two 1 × 2 m vertical white plates and next to each other, separated by another 1 × 2 m vertical plate. All traps were placed in the cage such that the top of the trap inlet was at a height of 54.5 cm, in order to be consistent with our previous study [18]. During the experiments, the room was illuminated only by a single nightlight (0.4 W), placed above the centre of the cage.
Each trap was equipped with a MB5 odour source (OS1 or OS2) and placed on the left or right side of the cage. The position of the traps (left or right) and the odour source used (OS1 or OS2) were chosen following a quasi-randomized planning where all combinations of conditions were tested at least 7 times. See Additional file 2: Database S1 for a summary of all test conditions.
Before each experiment, the traps and the experimental set-up were cleaned using a 15% ethanol solution. All handling of the materials and mosquitoes was done wearing nitrile gloves to minimize the risk of skin odour contamination. After setting up the traps, 50 mosquitoes were released from a holding container on the opposite side inside the netting cage by pulling a string outside of the cage. Then, the experimenter left the room. Mosquitoes could then choose to fly around their preferred trap. After 20 min, the experimenter re-entered the room, closed the traps, killed the remaining mosquitoes in the cage using an electric mosquito zapper and cleaned the cage with a vacuum cleaner. Each trap capture bag was then placed in a freezer to kill captured mosquitoes, which were manually counted later. Relative humidity and temperature inside the room were recorded before and after each trial using a weather station (TFA Dostmann/Wertheim, Kat. Nr. 30.5015). Four such dual-choice trials were done during each experimental morning, which coincided with the dark period of night-active mosquitoes.
Semi-field testing
To verify the results of the dual-choice laboratory tests, the capture percentage of the BG-Suna and the M-Tego with or without heat and warm water in semi-field experiments were compared. These experiments were performed at the Ifakara Health Institute (Ifakara, Tanzania) during the first week of November 2018. For the experiments, three screen houses of 10 × 10 m each were used with a slightly scaled-down house inside (see Additional file 1: Fig. S3). These houses were built from local materials such as bricks, corrugated sheet metal, straw or mud. Three traps were tested each experimental night, a BG-Suna, a M-Tego without additional host cues and an M-Tego with heat and warm water (Fig. 1b). The tested traps were placed outside the houses, with their inlet pipe at a height of 65 cm. The fact that this height is higher than the one in the laboratory experiments (65 vs. 54.5 cm) might have resulted in a small overall difference in capture performances [15]. However, it is unlikely that it would have affected the comparative results between traps. The M-Tego traps were hung from the house, whilst the BG-Suna was placed in a metal wire frame on the ground. The traps were cleaned before use with 70% ethanol and handled with gloves afterwards. Each trap was powered using a 12 V car battery and contained an MB5 odour source. CO2 was produced using 5.5 L plastic containers with a yeast and molasses mixture, which was placed next to each trap, and replaced daily. Inside each house, a set-up with CO2, an MB5 blend and a fan (same as used in the traps) was placed below a bed net to simulate human presence. The fan was positioned such that it produced an airflow that directed the odour and CO2 towards the nearest window. Screen houses were cleaned before and after the experiments. Only natural (moon) light was illuminating the screen house during the night.
Before each trial, each trap was equipped with one of three MB5 odour sources (OS3, OS4 or OS5), and placed inside one of the three screen houses. The odour source and the screen house used for each trap were changed following a quasi-randomized planning (see Additional file 2: Database S2). The M-Tego with short-range host cues was then filled with 0.7 L of water from a water bottle that stood in direct sunlight during the day (temperature = 39.7 ± 0.5 °C, n = 2). At the start of the experiment, a release pot containing 200 females An. gambiae s.s. was placed in the corner of each screen house and the mosquitoes were released manually, at approximately 18:00 h. The experiment ended around 06:20 the following morning. Captured mosquitoes were killed by moving the capture bag or pot out of the trap and placing them in direct sunlight for a day. The desiccated mosquitoes were manually counted.
The mean run-time of the experiments was 12 h and 23 min (± 22 min). At the start of the experiment, the ambient air temperature was 32.4 ± 1.4 °C and ambient relative humidity of the air was 38.1 ± 4.7%. The next morning, the ambient air temperature was 22.9 ± 0.8 °C and ambient relative humidity was 78.0 ± 6.0%.
Mosquito flight tracking experiments
To study the flight behaviour of mosquitoes around the M-Tego with or without additional host cues, mosquitoes were tracked around the traps in the same netted cage as used for the dual-choice tests (ambient temperature = 24.9 ± 0.7 °C and relative humidity = 73.7 ± 3%). The experimental procedure was identical to the dual-choice experiments, except for the following differences. A single M-Tego was placed on the right side of the dual-choice setup and a single MB5 odour source was used inside the trap (OS1). Four trap conditions were tested, the M-Tego without additional cues, the M-Tego with heat, the M-Tego with warm water, and the M-Tego with heat and warm water. Each experimental morning, all conditions were tested using as quasi-randomized planning (see Additional file 2: Database S1).
Three infrared-enhanced cameras (Basler acA2040-90umNI) with Kowa 12.5 mm lenses (LM12HC f1.4) were used for the tracking (Fig. 2), which synchronously recorded images at temporal resolution of 90 frames per second and a spatial resolution of 1024 × 1024 pixels. Cameras were synchronized using pulses from an Arduino Uno board. Because mosquitoes cannot see infrared light [32], the tracking set-up was illuminated using two infrared light-emitting-diode (LED) lamps (Bosch Aegis SuperLed, 850 nm, 10° beam pattern – SLED10-8BD). Lens distortions were corrected using pictures of a chequerboard pattern. Calibration was done daily using a single LED manually waved inside the filmed volume to find DLT (direct linear transformation) coefficients [33]. Alignment was done with a calibration device with four LEDs that were consecutively blinking at various known three-dimensional positions. The real-time three-dimensional tracking software Flydra (version 0.20.19) was used to track the three-dimensional positions, velocities and accelerations of flying mosquitoes within a three-dimensional space of approximately 1x1x1 m around the trap [34]. For each experiment, mosquito flight trajectories were recorded for 20 min, but the first 3 min were used for tracker initialization. The remaining 17 min were used for the analysis.
Analysis of three-dimensional flight tracks
All the flight dynamics analyses were done using Matlab 2018b (MathWorks). Mosquitoes could enter and exit the filmed three-dimensional space several times per trial, therefore individuals could not be identified. Filtering of tracked points was based on the covariance matrices estimated by the extended Kalman filter used for three-dimensional reconstruction by Flydra. Three-dimensional points with too high estimated standard deviation of either their position or speed were considered as outliers and deleted. When two or fewer consecutive video frames had missing values in the three-dimensional tracks, they were linearly interpolated. If more than two consecutive frames had missing values, the trajectory was divided in two separate tracks. Finally, tracks shorter than 10 video frames in length were deleted. For all frames of each computed three-dimensional track, linear and angular flight speeds as well as linear accelerations were computed (as in [18]).
To systematically analyse the flight behaviour of mosquitoes around the M-Tego trap, the visualization technique developed by [18] was used: based on the verified assumption that the three-dimensional flight behaviour is on average axially symmetric around the trap axis of symmetry, the three-dimensional flight movements can be projected onto a two-dimensional sub-space [18]. For this, the three-dimensional space was divided into multiple three-dimensional rings centred around the M-Tego axis of symmetry (Fig. 2a, c). Within each ring, all relevant flight dynamics metrics (such as mean flight speed) were computed. Results were visualized by projecting the metric value in each ring onto a two-dimensional parametric space with the radial and vertical positions as coordinates (Fig. 2d). All rings had a constant volume to allow metric comparison. In addition, all metric results were visualized using two-dimensional heat maps from a top-down view and from a side view close to the background walls (see Additional file 1: Figs. S5, S7, S9 and S12). For these the three-dimensional space around the trap was divided into vertical and horizontal columns, respectively.
For each tested condition, the following metrics were computed in all cells (rings or columns) around the trap: the positional likelihood of mosquitoes in each cell (Figs. 3a, d, 4a), the average time spent in a cell, the average flight velocity, flight speed, upward acceleration, angular speed, and capture probability.
The capture probability of a mosquito flying in a specific cell is defined as the number of tracks in that cell that ended by a capture divided by the total number of tracks detected in the cell. The positional likelihood of mosquitoes in a cell is defined as the normalized probability of a mosquito to fly in a cell (i.e., a horizontal ring projected in the two-dimensional parametric space). This metric allows the identification of regions of high flight activity. The positional likelihood \(P_{i}\) of mosquitoes to fly in cell i, was computed as \(P_{i} = \frac{{n_{i} }}{N} \times I\), where \(N\) is the sum of the length of all tracks in the filmed three-dimensional space (i.e. the total number of frames), \(n_{i}\) is the sum of the length of the parts of these tracks with mosquitoes flying in cell i, and \(I\) is the total number of cells within the three-dimensional space. In this way, if flight behaviour was random and the number of tracks was high, the positional likelihood heat-maps would be uniform with a value of 1. The average time spent by mosquitoes in a cell is defined as the total time spent by all mosquitoes in a cell divided by the number of tracks detected in that cell. Because the positional likelihood in a cell is equivalent to the normalized total time spent by all mosquitoes in a cell, the average time spend is a good metric to weight the measure of positional likelihood. For details about the other metrics, see [18].
To test how close-range cues affected the flight dynamics of mosquitoes, the difference in each metric between treatments were computed by subtracting the results distribution around the M-Tego without additional cues from the treatments with heat and/or warm water (Fig. 4).
Statistical analysis
For all statistical tests, generalized linear models (GLM) were used. For each test, the minimal model was determined by successively removing the least significant predictors from the model until all remaining predictors were significant (p value < 0.05). We then chose the GLMs with the lowest AIC (Akaike Information Criterion) values (see Additional file 2: Table S1).
For the dual choice experiments, a GLM with a binomial distribution, logit link function and estimated dispersion [35] was used. The response variable of the GLM was the capture percentage, defined as the ratio between number of captures by one trap and the number of captures by both traps. The predictors for the full model were: trap (BG-Suna or M-Tego), with or without heat, with or without warm water, mean humidity, mean temperature. Location (left or right side) and the odour source used in the trap (OS1 or OS2) were also included. When determining the minimal model, the predictors of interest “trap”, “heat” and “warm water” in the model were always kept.
For the semi-field experiment, a GLM with a binomial distribution, logit link function and estimated dispersion was also used. The response variable of the GLM was the capture percentage, defined as the ratio between number of captures of one trap and the total number of captures of all traps across all three screen houses. The full model predictors were trap (BG-Suna or M-Tego), with/without heat and warm water, mean humidity, mean temperature, day of the experiment, house number (H1, H2 or H3), and the odour source used in the trap (OS3, OS4 or OS5).
To compare the flight behaviour of the mosquitoes around the M-Tego trap with or without host cues, three doughnut-shaped volumes were defined around the top rims of the trap with various radii of their tube (r = \(\left[ {\frac{1}{2}, \frac{3}{4}, 1} \right] \times \left( {r_{bag} - r_{inlet} } \right)\), with \(r_{bag}\) and \(r_{inlet}\) being, respectively, the radius of the tarpaulin bag and radius of the inlet of the trap). The positional likelihood of mosquitoes to be tracked inside these volumes as well as how long on average they stayed inside for each trial was then computed (Fig. 6). Because mosquitoes were not tracked during the dual-choice and semi-field experiments, these two metrics could not be used as covariate in the previously described models. GLMs with a gamma distribution, negative inverse link function and estimated dispersion were used to model how these two metrics varied as a function of the presence of heat and warm water. Full model predictors were with heat (yes or no), with warm water (yes or no), mean humidity, mean temperature, volume of the doughnut and age in days of the adult female mosquitoes.