Impact of increased ventilation on indoor temperature and malaria mosquito density: an experimental study in The Gambia

In sub-Saharan Africa, cooler houses would increase the coverage of insecticide-treated bednets, the primary malaria control tool. We examined whether improved ventilation, using windows screened with netting, cools houses at night and reduces malaria mosquito house entry in The Gambia. Identical houses were constructed, with badly fitting doors the only mosquito entry points. Two men slept in each house and mosquitoes captured using light traps. First, temperature and mosquito density were compared in four houses with 0, 1, 2 and 3 screened windows. Second, carbon dioxide (CO2), a major mosquito attractant, was measured in houses with (i) no windows, (ii) screened windows and (iii) screened windows and screened doors. Computational fluid dynamic modelling captured the spatial movement of CO2. Increasing ventilation made houses cooler, more comfortable and reduced malaria mosquito house entry; with three windows reducing mosquito densities by 95% (95%CI = 90–98%). Screened windows and doors reduced the indoor temperature by 0.6°C (95%CI = 0.5–0.7°C), indoor CO2 concentrations by 31% between 21.00 and 00.00 h and malaria mosquito entry by 76% (95%CI = 69–82%). Modelling shows screening reduces CO2 plumes from houses. Under our experimental conditions, cross-ventilation not only reduced indoor temperature, but reduced the density of house-entering malaria mosquitoes, by weakening CO2 plumes emanating from houses.


Introduction
Between 2000 and 2015, massive deployment of malaria control interventions in sub-Saharan Africa reduced malaria prevalence by half and clinical malaria by 40% [1]. Of the interventions used to achieve this remarkable level of control, insecticide-treated nets (ITNs) were widely used and most effective, contributing to 68% of the reduction in malaria infection prevalence. The World Health Organization's target of universal coverage with ITNs requires the distribution of one bednet for two people at risk of malaria, repeated every three years [2]. In 2019, although 68% of households in sub-Saharan Africa had at least one net, only 46% of people reported using a net [3], mainly because it was too hot to sleep under a net [4]. Cooling a house at night, by improving ventilation, could increase net use and further reduce the malaria burden.
Anopheles gambiae s.l., the principal vectors of malaria in sub-Saharan Africa, bite predominantly indoors at night [5]. These mosquitoes locate a blood meal using a range of chemical cues generated by people, particularly carbon dioxide (CO 2 ) [6]. This gas is a major component of exhaled breath and stimulates take-off, extends flight duration [7] and is a long-range attractant for An. gambiae and other mosquitoes [6,8]. Thus to understand the role of ventilation in a building, one needs to understand how this gas accumulates in inhabited rooms and leaks out of buildings, since the concentration and shape of the odour plumes will affect how readily mosquitoes are able to locate people within a house.
One simple way to ventilate a bedroom, and cool the room at night, is to install mosquito screening on doors and windows [9]. We designed a series of experiments to explore the effect of house screening on indoor temperature and mosquito house entry and used computational fluid dynamic (CFD) modelling [10] to visualize the distribution of the CO 2 and to provide realistic approximations of CO 2 concentrations generated by people sleeping in single-roomed houses with and without mosquito screening. Our findings are relevant to those interested in designing and constructing buildings that protect people from malaria in sub-Saharan Africa.

Study design
A detailed description of the experiments is provided in the electronic supplementary material. Briefly, in year 1, four square single-roomed experimental houses constructed with mud walls, metal roofs and closed eaves, each sleeping two men, were used to examine the effect of insecticide-free screened windows on indoor climate and mosquito house entry (figure 1). In all experiments, each house had narrow slits above and below two metal doors, which constituted the main entry point for mosquitoes, and represented badly fitting doors, commonly found in local villages. We carried out three experiments in year 1. Experiment 1 compared houses with badly fitting screens (control) to houses with one small-screened window, two small-screened windows, one small-screened window and one medium-screened window. Experiment 2 maintained the same control compared to houses with one, two or three large-screened windows. Experiment 3 compared unscreened houses with narrow slits above and below the doors with those with slits above or below the door. In each house, temperature and relative humidity were measured using a data logger and mosquitoes captured using a CDC light trap. Collections were made over 25 nights and typologies rotated between houses each week, so that at the end of each experiment, each typology had been tested in each house.
We found that increasing ventilation, by increasing the number and size of windows, had a small cooling effect, but, surprisingly, also markedly reduced the numbers of malaria mosquitoes entering the house. We formulated three hypotheses to explain why screened windows reduced mosquito house entry through the slits above and below the doors. These were that screening: (i) acts as a site of attraction, with mosquitoes attracted to host odours leaking from the screening, accumulating on the outside of the window screening; (ii) reduces indoor concentrations of host odours by a combination of increased ventilation and a cooler house reducing sweating and the production of host odours; and (iii) changes the shape of odour plumes emanating from a house, attracting fewer mosquitoes.
To test these hypotheses, we carried out a further series of field studies using two experimental houses in year 2, repeating the basic study design from year 1, but, in addition to routine monitoring, also measured indoor CO 2 concentrations. In experiment 4, one house with solid doors was compared with one with screened doors, both doors having narrow slits top and bottom. If hypothesis 1 was correct, more mosquitoes would be collected in the house with screened doors than solid doors, since mosquitoes accumulating outside the screened door, attracted to odours from the door, would enter the house through the door gaps in larger numbers than houses with solid doors. In experiment 5, both houses had solid doors, one had screened windows and one had no windows. This experiment allowed us to confirm the findings of experiments 1 and 2, but in addition allowed us to measure indoor CO 2 concentrations to test hypotheses two and three. Typologies were rotated between houses weekly and each experiment lasted 20 nights.

Study area
In year 1, in 2017, the study took place outside Wellingara village (N 13°33.365 0 , W 14°55.461 0 ), adjacent to a large area of irrigated rice, and, in year 2, in 2019, it took place approximately 2 km to the north, in the grounds of the Medical Research Council's field station at Wali Kunda (N 13°34.25 0 , W14°55.28 0 ). Both sites are situated on the south bank of the River Gambia, in the Central River Region. The study was done during the rainy season, from June to October, when malaria transmission is high and people are most likely to sleep under an ITN [11].

Experimental houses
In year 1, experiments were conducted from 2 July to 23 October 2017. Construction of the experimental houses has been described in detail elsewhere [12]. Briefly, four square single-roomed mudblock houses were constructed on the western edge of Wellingara village and two in Wali Kunda field station. Each house was the average size of single-roomed houses in the Upper River Region and was built along a line, 10 m apart. Each house was 4.20 m by 4.20 m in floor area, with walls 2.20 m high and a saddleshaped roof, made of corrugate sheeting, with closed eaves. Each house had two doors (1.75 m high × 0.75 m wide), built on opposite sides of the house. In all experiments (apart from experiment 3), screened and unscreened doors had narrow slits, 0.02 m high and 0.75 m wide, above and below the door to simulate badly fitted doors common in local villages (electronic supplementary material, figure S1). In experiments 1 and 2, the reference house was a metal-roofed house which had narrow slits above and below the solid metal doors and windows (figure 1). In experiment 3, the reference house had no windows, only badly fitting doors. Polyester netting (2 × 2 mm mesh) screens mounted on metal frames were used for screening windows.
In year 2, field experiments were conducted during the rainy season, from 28 August to 11 November 2019. Experiments were conducted using two experimental houses of a similar design to those used in year 1, apart from the saddle roof apex in year 2 houses faced the front and back facades, rather than the sides, i.e. the roof was rotated 90°and the roofs had been painted red in year 2, rather than bare metal in year 1. We compared mosquito entry in screened and unscreened houses. As described earlier, two men slept in separate beds, under ITNs during the night and mosquitoes collected using light traps. We measured indoor and outdoor climate as before and also measured CO 2 concentrations in the houses, to inform the modelling of CO 2 movement within and outside the houses. At the start of each experiment, each typology was randomly allocated to one house position. Each house was arranged with two single beds located parallel to one another on opposite walls to the doors. Two adult men slept on separate beds in each house under an ITN (Olyset, Sumitomo Chemical, Japan) from 21.00 h to 06.00 h. The men remained in the house in the same position throughout each experiment, so that the relative attractiveness of the house positions to mosquitoes was combined with that of the pair of sleepers. For each experiment, CDC light traps were used to collect mosquitoes indoors for five weeks, each week for five nights. Each house typology was rotated weekly between houses using a replicated Latin rectangle design.

Procedures
In both years, mosquitoes were collected from each house using a CDC light trap (Model 512, John W. Hock Co., Gainesville, USA), sampled nightly for five nights each week for five weeks. Traps were suspended from the roof with the light 1 m above the ground in the centre of the room between the two sleepers at the foot end of the bed and operated from 21.00 h to 06.00 h. Mosquitoes were collected from each house at 06.00 h and killed by freezing. Mosquitoes were identified morphologically and female An. gambiae s.l. identified to species by PCR [13,14].
Indoor temperature and relative humidity were measured every 30 min using data loggers (Tiny tag, TGU 4500, Gemini Data Loggers, Chichester, UK) suspended 1 m from the floor in the centre of the room. In year 2, CO 2 concentrations were recorded outdoors with the logger situated on a tripod at a height of 1.3 m, midway between the two experimental houses in Wali Kunda during the rainy season, from 2 July to 25 July 2017. Recordings were made on 12 nights from 21.00 to 06.59 h.

Outcomes
Primary outcomes were the mean indoor temperature, and the mean number of An. gambiae s.l./light trap/night for each house typology. Night-time temperature was analysed from 21.00 to 23.59 h, when adults in the study area go to bed and make a decision to sleep under a bednet or not [15], and from 00.00 to 06.59 h, when most are asleep.

Data analyses
Sample size calculations were based on mosquito counts since these are considerably more variable than the environmental recordings made. In year 1, the sample size was estimated via simulation based on data from a similar trial conducted in the study area, [16] where the mean number of An. gambiae s.l. collected indoors was 10.8/hut/night (standard deviation, s.d. = 8.7). The study was powered to detect an effect size of 50% at the 5% level of significance and 80% power and required 25 nights of observation. In year 2, we based our calculations on a study conducted in 2017 in nearby Wellingara village [12]. In this study, there was a mean of 6.2 An. gambiae/night (s.d. = 5.1). To detect 66% fewer mosquitoes indoors at the 5% level of significance, with 80% power, required 20 nights of observation.  The effect of house typology on indoor climate and mosquito house entry was assessed using generalized linear modelling, using a negative binomial model with a log link function for mosquito count data, while comparisons of indoor temperatures and CO 2 were made using linear regression. In addition to house typology, house position and day were included in the model as fixed effects. For analysis of temperature and CO 2 we calculated mean values for two separate periods, from 21.00 h to 23.30 h and 00.00 h to 07.00 h, for each house each night.
LadyBug software (LadyBug Products, Athol, ID, USA) was used to estimate the percentage of time occupants of various house typologies spent in the 'comfort zone' [17]. The human comfort index is widely used by those working in the built environment to assess how comfortable a building is and is based on experiments with hundreds of volunteers of different genders, ages and ethnicity wearing different amounts of clothing [18]. The index is used in the ANSI/ASHRAE Standard 55: Thermal Environmental Conditions for Human Occupancy. It is an American National Standard published by ASHRAE that establishes the ranges of indoor environmental conditions to achieve acceptable thermal comfort for occupants of buildings [19]. χ 2 -test was used to look for trends in human comfort with increasing numbers of large-screened windows and for comparisons between typologies. Analyses, apart from psychrometric analysis, were done using Stata version 16 (StataCorp, College Station, TX, USA), except for chi-square calculations which were done with Epi Info v. 3.01 (CDC, Atlanta, USA).
CFD modelling was used to simulate CO 2 concentrations using Ansys ® Fluent (v. 19). Model assumptions and set-up configurations were as follows: (i) the house was based on the structure of an experimental house and was assumed air-tight, except for    The funders had no role in study design, data collection, data analysis, data interpretation or writing of the report. All authors had full access to all data in the studies and had final responsibility for the decision to submit for publication.

Experimental houses
In year 1, there were roughly equal numbers of An. arabiensis (50.7%) and An. coluzzii (49.3%), while in year 2, An. coluzzii (69.0%) was more common than An. arabiensis (30.5%; electronic supplementary material, table S1). There were no differences in how each species responded to the different housing typologies. In experiment 1, installing small-screened windows did not cool houses and were associated with an increased mean temperature of 0.2°C, both before and after midnight, compared to the reference house with badly fitting small windows (table 1, p < 0.001). By contrast, in experiment 2, installing two large-screened windows lowered indoor temperature by 0.4 to 0.5°C, and three large-screened windows by 0.6°C ( p < 0.001) compared with the reference house with badly fitting large windows. The psychrometric analysis of large-screened windows in experiment 2 shows that from 21.00 to 23.59 h, human comfort increased with the increasing number of screened windows ( p = 0.009), while from 00.00 to 05.59 h human comfort decreased with increasing numbers of windows since it became too cold ( p = 0.0007, figure 2). Human comfort, however, was statistically different from the reference house only when there were two or more large-screened windows on opposite walls (21.00-23.59 h, χ 2 = 3.9, p = 0.048; 00.00-05.59 h, p = 0.009).
As the number of small-or large-screened windows increased, the number of An. gambiae s.l., and other mosquito species (electronic supplementary material, table S1), collected indoors decreased (table 2). A summary figure combining the results from experiment 1 and 2 shows the percentage of An. gambiae s.l. collected indoors declines with increasing number and area of the screened window, with two or three windows providing best protection (figure 3).
In experiment 3, reducing the number of gaps around the door from two to one or installing two small-screened windows into the house did not reduce the numbers of An. gambiae s.l. collected indoors compared with the reference house (table 2). Numbers of Mansonia spp. and Culex spp., however, were lower in houses where there was no gap at the bottom of the door, compared to the reference house with gaps both at the top and bottom of the door (electronic supplementary material, table S1).
In experiment 4, there were 76% fewer An. gambiae (95% CI 69-82%, p < 0.001) in houses with screened doors and screened windows than those with solid doors and screened windows. Temperatures in houses with screened doors were 0.6°C less than houses with screened doors before midnight (95% CI 0.4-0.7°C, p < 0.001) and after midnight (95% CI 0.5-0.7°C, p < 0.001). Indoor concentrations of CO 2 rose above background levels shortly after two men entered each house at 21.00 h, to a maximum roughly 1 h later, before declining gradually through the night, before a small rise around 05.00 h, one hour before the men left the houses (figure 4). Indoor CO 2 concentrations were 152 ppm (95% CI 109-195 ppm, p < 0.001) lower in screened door houses than those with solid doors from 19.00 to 23.59 h ( p ≤ 0.001) and 120 ppm lower (95% CI 81-159 ppm, p ≤ 0.001) from 00.00 to 05.59 h.

Computational fluid dynamic modelling
The exchange of CO 2 between the indoor and outdoor environment of a house is shown through sections made through the doors and windows of the experimental houses royalsocietypublishing.org/journal/rsif J. R. Soc. Interface 18: 20201030 wind, unscreened houses produce horizontal jets of CO 2 from gaps below and above the doors reaching more than 4 m from a house. When screened windows are installed, little CO 2 is streamed from the doors, with most leaking from the windows, rising vertically into space. In houses with screened doors and windows, less CO 2 accumulates indoors and low levels are projected from the doors and windows, rising upwards. On windy nights, screening reduces the concentration of CO 2 indoors ( figure 7). On the windward side of the house, little or no CO 2 escapes from all house typologies. In marked contrast, on the leeward side, high concentrations of CO 2 leak from above and below the door of unscreened houses. Screening results in markedly less CO 2 leaking from the screened doors and windows, presenting a diffuse source of the gas.

Conclusion
Well-ventilated houses are not only cooler at night and more comfortable to live in, they also have markedly fewer malaria mosquitoes than poorly ventilated houses. Installing two or more large-screened windows, each 0.5 m 2 in area, lowered the indoor temperature at night by 0.2-0.6°C, depending on outdoor wind speed. Adding two screened doors to a house with screened windows further reduced the temperature. Cross-ventilation of the room, where windows are positioned on opposite walls, without any internal obstruction, is essential to encourage airflow and reduce the indoor temperature. Increased ventilation makes it more comfortable for the occupants before midnight, making it more likely they will use an ITN. By contrast, after midnight, houses with screened windows or doors cool down further and become uncomfortably cool. Being too cool at night, however, is easily resolved by sleeping under a sheet, while being too hot, is impossible to resolve without fans, air conditioning or sleeping under a wet towel or sheet. The cooling effect of screening is also supported by our previous findings in The Gambia and elsewhere [9]. Nonetheless, the cooling effect of cross-ventilation is limited in the hot humid tropics, especially where using a bednet will reduce airflow by an average of 64% [20]. Surprisingly, we discovered that increasing both the number and size of screened windows in a Gambian house reduced house entry of An. gambiae s.l. mosquitoes. With royalsocietypublishing.org/journal/rsif J. R. Soc. Interface 18: 20201030 three screened windows and a total surface area of 1.5 m 2 , there was a 95% reduction in the number of An. gambiae s.l. found indoors. This level of protection against malaria vectors is equivalent to that seen with some ITNs in the same area [21], but in our case, no insecticides were used on the screening or doors. The only source of insecticide was on the ITNs. Similar findings were observed with other taxa of mosquitoes, suggesting that the addition of screening can reduce the house entry of many species of African mosquitoes. While it is well known that a screened house provides a physical barrier against mosquitoes, this study is the first to show screening reduces mosquito entry in houses which are not perfectly closed and are therefore more representative of typical village houses. We explored three hypotheses to explain the indirect protective effect of screening. First, our findings suggest that mosquitoes are not attracted in large numbers to external faces of screening, since houses with slits above and below screened doors had markedly fewer, not more, mosquitoes than those with solid doors and slits. Second, and more plausibly, the mechanism of protection results from screening reducing the concentration of CO 2 dispersing from a house. CO 2 concentrations were approximately 80 ppm less in houses with screened windows compared to houses without windows and adding two screened doors to a house with screened windows resulted in a further 120-152 ppm reduction. Houses with screened doors will leak CO 2 more rapidly than the door gaps common in most rural houses. The reductions in CO 2 levels are of biological significance since female mosquitoes can detect differences in CO 2 concentration as small as 40 ppm [22]. Interestingly, in poorly ventilated houses, CO 2 concentration increased with increasing indoor temperature. This is probably related to higher rates of metabolism associated with sweating and keeping the body cool. Our third hypothesis about the structure of the odour plume emanating from the house is also likely to be important. Computer simulations show that jets of CO 2 project from unscreened houses from above and below the doors, both accessible entry points for malaria mosquitoes. In marked contrast, in screened houses, the odour plume is  ). Note that CO 2 is released from narrow gaps above and below all doors and windows, and through screening. In (a)ii, (b)ii and (c)ii, the section through the house includes a sleeping person generating CO 2 from the head. Note that the legend limits were chosen to enable comparisons between the different typologies and that the maximum value of 557 ppm of CO 2 in the legend was exceeded in simulations (a), with maximum values of over 1000 ppm; typically around the mouths of the sleepers.
royalsocietypublishing.org/journal/rsif J. R. Soc. Interface 18: 20201030 weaker and rises upwards. Thus blood-seeking mosquitoes are likely to detect an unscreened house from greater distances than a screened house, potentially explaining why more mosquitoes find their way into an unscreened house compared to a screened one. In nature, the picture is more complicated, since a blood-seeking mosquito finds itself in an environment where there is a continuous plume of CO 2 from a house, when there is no wind, and turbulent pulses of CO 2 , when the wind blows and eddies around the house. Anopheles gambiae is, however, able to orientate readily along such turbulent odour plumes, moving towards the odour source.
There are four main limitations to this study. First, the experiment was carried out without considering the variability in 'real world' human behaviour. The sleepers went to bed earlier than many adults, the doors remained closed during the night, unlike those in the villages which are frequently opened and closed until midnight [11], and there was no wood smoke or bright lights indoors. Second, while our focus is on CO 2 as a long-distance attractant, there are other volatiles produced by humans that are also strongly attractive to An. gambiae s.l. [23,24]. We do not know how these short-distance attractants affect mosquito movement in our study. Thirdly, there were only two men in each house, fewer individuals than most village houses. Increasing the number of people sleeping indoors is likely to increase CO 2 concentrations, and hence indoor mosquito collections [25]. Fourth, our study used single-roomed houses, and the effects reported here are likely to vary according to different house typologies including those with multiple rooms, ceilings, porosity of structure, and presence and proximity of adjacent buildings.
Findings from the present study suggest that screening is not only an effective barrier against mosquitoes, it also increases ventilation, cools the house and reduces CO 2 concentration indoors. In houses with screening, the CO 2 -odour plumes released from the house are likely to attract mosquitoes from shorter distances, reducing the number that enter the house, compared to unscreened houses with badly fitting doors. The World Health Organization recommends house improvements, such as installing window screens, for reducing malaria [26,27]. There has never been a better time for developing new ways of screening houses and facilitating the scale-up of this intervention, since it coincides with a period when Africa's housing stock is modernizing [28], and more housing is urgently needed to accommodate the more than one billion new Africans that will need homes in rural areas by 2050 [29]. Building healthy homes that reduce the threat from malaria and other mosquito-borne diseases, that prevent over-crowding, increase airflow, reduce indoor air pollution, and keep the occupants clean, secure and comfortable during the day and night is an imperative and a basic human right. Data accessibility. All data used in this paper are available as electronic supplementary material.