Grass-Like Plants Release General Volatile Cues Attractive for Gravid Anopheles Gambiae S.S. Malaria Vector Mosquitoes


 Background: Understanding the ecology and behaviour of disease vectors, including the olfactory cues used to orient and select hosts and egg-laying sites, are essential for the development of novel, insecticide-free control tools. Selected graminoid plants have been shown to release volatile chemicals attracting malaria vectors, however, whether the attraction is selective to individual plants or more general across genera and families is still unclear. Methods: To contribute to the current evidence, we implemented bioassays in two-port airflow olfactometers and in large field cages with four live graminoid plant species commonly found associated with malaria vector breeding sites in western Kenya, Cyperus rotundus, Cyperus exaltatus of the Cyperaceae family and Panicum repens and Cynodon dactylon of the Poaceae family. Additionally, we tested one Poaceae species, Cenchrus setaceum, not usually associated with water. The volatile compounds released in the headspace of the plants were identified using gas-chromatography/mass spectrometry.Results: All five plants attracted gravid vectors, with the odds of a mosquito orienting towards the choice-chamber with the plant in an olfactometer being 2-5 times higher than when no plant was present. This attraction was maintained when tested with free-flying mosquitoes over a longer distance in large field cages, though at lower strength, with the odds of attracting a female 1.5-2.5 times higher when live plants were present than when only water was presented in the trap. Cyperus rotundus, previously implicated in connection with an oviposition attractant, consistently elicited the strongest response from gravid vectors. Volatiles regularly detected were limonene, β-pinene, β-elemene and β-caryophyllene among other common plant compounds previously described in association with odour-orientation of gravid and unfed malaria vectors. Conclusions: The present study confirms that gravid Anopheles gambiae use chemical cues released from graminoid plants to orientate. These cues are released from a variety of graminoid plant species in both the Cyperaceae and Poaceae family. Given the general nature of these cues, it appears unlikely that they are exclusively used for the location of suitable oviposition sites. The utilisation of these chemical cues for attract-and-kill trapping strategies must be explored under natural conditions to investigate their efficiency when in competition with complex interacting natural cues.

For example, studies have shown that gravid malaria vectors are attracted to headspace volatiles released from wetland rice plants (Oryza sp. [22]) and to volatiles released from pollen of maize (Zea mays [43]) and sugar cane (Saccharum o cinarum [44]). The authors of that work suggest that mosquitoes have selectively adapted to habitats dominated by agricultural grasses of the Poaceae family which in turn would suggest that these grasses release a unique odour pro le that separates them from wild grasses.
On the other hand, the grass-like sedges in the Cyperaceae family are frequently indicators of wetlands [45,46] and have been associated with productive Anopheles breeding sites in a multitude of studies [28, 40,47]. Cedrol, a sesquiterpene alcohol, was identi ed from the headspace of aqueous infusions that were made from soil and rhizomes taken from a productive Anopheles habitat, that was densely vegetated by the sedge, Cyperus rotundus. The infusion as well as water treated with synthetic cedrol attracted Anopheles gambiae and An. arabiensis in laboratory, semi-eld and eld experiments [17,21]. Plant-based chemical compounds might either be released from roots and submerged plant parts into the water [36, 48] of the potential oviposition site or might be released into the air from the emergent parts of the plant [36,49]. Cedrol, has been identi ed directly from rhizome extracts of sedges [50] as well as from associated microbes [41].
It is against this background, that we set out to contribute to the current knowledge base by further investigating wild graminoid plant species from the Cyperaceae and Poaceae families for their potential to attract gravid malaria vectors with the volatiles they release when presented as intact plants. The four selected test plant species dominate natural aquatic habitats around the shores of Lake Victoria in western Kenya [40]. For comparison, an ornamental dry-land grass usually not associated with malaria vector breeding sites was included in the study. The overall aim of this work was to investigate if chemical cues released from graminoid plants result in species or family speci c volatile pro les and selective responses from gravid Anopheles gambiae or if the chemical cues are of more general nature.

Study site
All experiments and plant volatile collections were conducted under ambient climate conditions at the International Centre of Insect Physiology and Ecology (icipe), Thomas Odhiambo Campus (TOC), Mbita (00° 26' 06.19'' S; 34° 12' 53.13'' E; 1137 meters above sea level), western Kenya. The area is characterized by equatorial tropical climate with daily average minimum and maximum temperatures ranging from 16°C to 28°C. The chemical analyses of the volatile samples were done at laboratories at KTH Royal Institute of Technology in Stockholm, Sweden.

Gravid mosquito preparation
Anopheles gambiae s.s., Mbita strain insectary-reared mosquitoes, were used for all experiments. Mosquitoes were reared under ambient conditions following the protocol described by Okal et al. [51]. Adult mosquitoes were held in 30x30x30 cm netting-covered cages at 25-28°C temperature and 68-75% relative humidity in a 12 h: 12 h light: dark photoperiod. Equal numbers of 2-3 days old adult female and male mosquitoes were transferred into a clean cage and starved for six hours starting at 13:00 h before allowed to feed on human arm at 19:00 h for 15 minutes. Blood feeding was repeated the next day at 19:00 h using the same procedure. After each blood-meal, the mosquitoes were provided with 6% glucose solution ad libitum. A wet towel was placed on top of the cages to provide water and keep humidity high. After the second blood-meal, the mosquitoes were kept for another two days with access to glucose solution. On the third day, gravid females were selected and used in bioassays.

Preparations of test substrates
Four graminoid plant species, naturally occurring frequently in malaria vector breeding sites in western Kenya [40], namely the grass-like sedges (Cyperaceae), Cyperus rotundus (nut grass), and Cyperus exaltatus (giant sedge), as well as the true grasses (Poaceae), Panicum repens (torpedo grass) and Cynodon dactylon (Bermuda grass) were collected from wetlands along the shores of Lake Victoria, around Mbita and Rusinga towns, western Kenya. The plants were uprooted carefully and the plants with soil transported to icipe-TOC for bioassays in olfactometers and large eld cages, and for volatile collections. A drought-tolerant grass, not native to wetlands and frequently used as ornamental grass in gardens, Cenchrus setaceum (purple fountain grass; Poaceae) was obtained from plant nurseries in Kisumu town and maintained at icipe-TOC. The plants were used in their non-owering stage (roots, stems and leaves only). In preparation for bioassays, the plants were washed thoroughly using lake water to remove the soil. Fresh plant samples were used for every round of bioassays. A bunch of several individual plants, weighing approximately 350 g, was used for every replicate bioassay.
Soil collected from the habitat where Cyperus rotundus was uprooted, was used for a preliminary bioassay. The soil was taken from the upper 10 cm of the habitat and plant material sieved out before use. For each replicate bioassay, 4kg of fresh soil was used.
Water was used in all bioassays (4 litres per test substrate), acknowledging that water vapour is a major oviposition attractant [20]. The water was taken from Lake Victoria and sediments allowed to settle before the clear supernatant was used for experiments.
A hay-infusion previously shown to be repellent for gravid An. gambiae [21] was prepared for the initial calibration of the olfactometer bioassays. The infusion was prepared by mixing 24 L of lake water and 90 g of hay in a bucket and kept in a dark place for three days before use for the bioassays.

Two-port air ow olfactometer bioassays
Four two-port olfactometers were constructed from galvanised iron sheets ( Fig. 1) to test the odourorientation of gravid An. gambiae s.s. in response to test substrates. The olfactometers were placed in a netting-screened makeshift shed where experiments were run overnight under ambient conditions. The olfactometers had two large substrate holding chambers (1 x 0.9 x 1 m), two trapping chambers made of polyvinyl chloride (PVC) pipes (30 cm long and 10 cm diameter), a fan and mosquito release chamber (0.5x0.2x0.3 m). The size of substrate holding chambers was su cient to carry whole live plants. Mosquitoes were introduced into the release chamber through an opening at the bottom. The electricitypowered fan drew air from the two substrate holding chambers through the holding chamber to the outside. Funnels inserted into the trapping chamber prevented mosquitoes from returning to the release chamber.
Test substrates were placed in both holding chambers. The fan was switched on ve minutes before releasing 100 gravid An. gambiae to the choice chamber at 18:00 h. The choice made by mosquitoes was recorded the following morning at 8:00 h by counting the number of mosquitoes trapped in each trapping chamber. The positions of the two test substrates were randomly rotated between chambers and olfactometers.
All choice experiments are listed in Table 1. Prior to testing intact plants, the olfactometers were calibrated by evaluating their accuracy of generating valid and reproducible results and to gauge the response rate that can be expected under standard test conditions. This was done by providing (1) two equal-choices in both chambers, (both containing water and both being empty) and (2) by providing two different choices with predictable outcome (water vs. empty; hay-infusion vs. water).
After calibration, a series of choice tests were done with intact plant materials (Table 1). Each comparison was replicated over 16 nights using a new batch of mosquitoes and fresh test substrates for every replicate. The replicate was discarded and repeated when mortality was ≥ 20% in the release/choice chamber or when less than 50% of the released mosquitoes responded (meaning majority remained in the central release chamber for the night). Table 1 Summary of behavioural bioassays with gravid An. gambiae in two-port air ow olfactometers and in large eld cages in relation to research questions. Large eld-cage experiments with free-ying mosquitoes Test treatments that elicited a positive response in olfactometer bioassays were then further evaluated with free-ying gravid An. gambiae in large eld cages (11.8 m long × 6.8 m wide × 2.4 m high; Fig. 2A) under ambient environmental conditions to mimic a more natural setting and test for longer-range attraction. The test substrates were placed inside BG-Sentinel traps (Biogents AG, Regensburg, Germany) and these traps were buried into the ground so that only the netting top of the trap and collection funnel containing the fan, was visible [51]. A black plastic bucket, 34 cm high and 30 cm in diameter, was inserted in each trap to hold the test substrates (Fig. 2B). Two traps with either equal or different test substrates included were set up per eld cage ( Table 1). The two traps were placed 4 m apart and 1.4 m away from the nearest wall. Mosquitoes were released from the opposite side, 9 metres away from the traps (Fig. 2C). The two test substrates were allocated to the location randomly and the position of the two traps were exchanged between the two shorter walls of the cage in consecutive nights. Every experimental night, 200 gravid An. gambiae s.s. were released in the eld cage at 18.00 h. The next morning at 08.00 h the traps were collected, and the number of mosquitoes recaptured in the traps' catch bags counted. Every experiment was repeated over 16 nights.

Sample size considerations for bioassays
The sample size for replication was estimated using the formula developed by Hayes and Bennett [53] for comparing proportions of clustered data. For equal choices an equal proportion responding to either choice was assumed for the reference (p1 = 0.5). Based on previous work [52], we aimed to be able to detect an increase in attraction by 16% (p2 = 0.66). Assuming a coe cient of variation (k) of 0.25 based on preliminary nightly test runs, and assuming at least 50 responding mosquitoes per night (n in each group), 16 replicates would be required for both treatment arms (p1 -equal choices; p2 -two choices) to detect the effect with 80% of power at a 5% signi cant level.

Bioassay data analysis
Choices experiments using olfactometers and BG-sentinel traps were analysed with generalized linear models with quasibinomial distributions tted to cater for overdispersion. The proportions of gravid females responding to the 'test' (as opposed to the 'control') in two choices experiments with two different choices were compared to the proportion of gravid mosquitoes responding to the 'test' in the experiments where 'test' and 'control' treatments were the same (lake water vs. lake water) [54]. The experiment was included as the xed factor and the 'equal choice' experiment was used as a reference to estimate the odds ratios (OR) and their 95% con dence intervals (CI). All reported mean proportions and their 95% con dence intervals (CIs) were estimated based on the model by transforming the log odds (logit) of the outcome to the odds scale and from the odds scale to the probability scale. R statistical software version 4.0.3 was used for the analyses [55].

Sampling of headspace from intact plants
Volatile chemicals released from test plants were trapped from intact live plants using dynamic headspace sampling. For this, several non-owing plants (approximately 350 g) were placed with some soil in a bucket with water, similar to the experimental conditions. The sampling was done for 48 hours under ambient conditions in the eld cage (Fig. 3). The aerial parts of the intact plants were enclosed into heat-resistant roasting bags (Sainsbury's Supermarkets Ltd, London EC1N 2HT) which were kept in an oven at 200°C for two hours prior to use. Porapak Q (50 mg, 50/80 mesh; Supelco) sorbent material was packed in a glass liner with glass wool on both ends to hold the sorbent in place. The Porapak Q traps were washed using 4 ml of hexane and kept in an oven for 2 hours at 50°C before use. Headspace collection was done by pumping 500 ml/min charcoal-ltered air into the bags through the inlet port and drawing the air out at a rate of 300 ml/min through the outlet port [56]. Headspace collections were done on two different dates, sampling four replicates of every plant species per date (total 8 headspace samples per plant species). Collections were also done from three replicates of empty cooking bags to account for the background chemicals concurrently for the two dates. After sampling, the traps were sealed with polytetra uoroethylene (PTFE) tape and kept in a freezer at -71°C. The lters were shipped to KTH Royal Institute of Technology, Stockholm, Sweden, where they were rst eluted using 3 ml hexane to decrease the likelihood of chemicals remaining in the trap and then concentrated to 250 µl using a desiccator connected to a duo rotary vane pump before chemical analysis.
Chemical analysis based on gas chromatography coupled with mass spectrometry

Two-port air ow olfactometer bioassays
The preliminary calibration experiments helped gauge the performance of the bioassay design and apparatus. During the majority of the preliminary experimental runs, around 50% of the released mosquitoes responded, whilst the others remained in the release chamber. This proportion could not be increased even when the experimental set up was modi ed. Hence, for all following experiments, it was de ned that for a viable outcome the response rate must be 50% or above. When two equal choices of water were provided in the chambers, the released gravid mosquitoes distributed equally between the two chambers as expected (Table 2). When both chambers were empty, mosquitoes still responded, likely ying upwind in search of cues, and again distributed equally between the two chambers. The response rate, however, was overall slightly lower (46%) than when water was provided. When a choice between water in one chamber and no substrate in the other chamber was provided, > 80% of the responding females chose water. This con rmed that water vapour acts as an attractant for gravid mosquitoes. Moreover, it was con rmed that fermented three day-old hay infusion repels gravid An. gambiae. Out of all responding females, > 70% oriented away from the infusion and towards the chamber with water.

CI-con dence interval
After con rming the consistent performance of the bioassay, three sets of experiments were implemented. Equal choice experiment where the mosquitoes were provided with lake water in both chambers randomly allocated as 'test' and 'control', were run in parallel for all three sets of experiments. Expectedly, these reference test resulted in an approximate 1:1 distribution of gravid females (Fig. 4). Any preference for a speci c test substrate in choice tests was expected to lead to a signi cant deviation from this balanced distribution.
Previous work [17] implicated soil from the Cyperus rotundus collection site as attractive oviposition substrate for gravid An. gambiae s.s.. Consequently, we evaluated in a rst step, whether wet soil from the location might be equally or more attractive in olfactometer bioassays than the live Cyperus rotundus plants in the same wet soil. However, the odds of a gravid female selecting the test chamber with the plants was nearly three-fold higher than in the reference experiment (OR 2.93; Fig. 4). Removing the soil completely from the bioassay increased the odds further when compared to the reference (OR 4.95). Consequently, another four graminoid plants were tested and all of them released volatile chemicals attractive to gravid An. gambiae females (Fig. 4) (Fig. 4).
Large eld-cage experiments with free-ying mosquitoes Bioassays with free-ying gravid mosquitoes con rmed olfactometer results with higher proportions of the released gravid females trapped with BG-Sentinel traps containing live plants than with traps that contained water only (Fig. 5). The odds of a female being captured in the test traps in the two-choice experiments were 1.5-2.5 times higher than in the reference experiment. Differences in the effect size of attraction between the plant species were not very pronounced under these more natural, long-range conditions, though Cyperus rotundus volatiles did slightly outcompete volatiles from P. repens in a similar way as in the olfactometer bioassays (Fig. 5).
Volatile organic compounds identi ed from the graminoid test plants  Table S1. The qualitative analysis shows that almost half of the detected compounds were sesquiterpenes. The second most common chemical class was monoterpenes, followed by a number of cyclic and straight compounds such as cyclic ketones, aliphatic esters and aromatic compounds. Table 3 shows compounds that have been detected in any one headspace sample of a plant species. There was a slight overlap in the pro les of monoterpenes and sesquiterpenes which were identi ed from different plant species (Table 3)

Discussion
Our study con rms and expands the evidence that odour cues released from graminoid plants play a role in the orientation of gravid An. gambiae females. Volatiles released from these plants add signi cant attraction to water vapour alone. Generally, all graminoid plant species tested, including the dry-land ornamental grass, Cenchrus setaceum, usually not associated with mosquito breeding sites, signi cantly attracted gravid females and behavioural differences in response to different test plants were not very pronounced especially under the more natural, longer-range trapping conditions.
Whilst the behavioural response of gravid An. gambiae mosquitoes appeared to be slightly stronger in reaction to the sedge, Cyperus rotundus, than to most other test plants, we were not able to exactly establish any unique differences in the chemical pro les that might explain this. This is likely, in part, due to the chemical sampling method. To the best of our knowledge, our bioassays are the rst to use live plants rather than eluted headspace extracts for testing for attractiveness to gravid malaria vectors. Our In our study, and across published work, we see very little variation in the strengths of the behavioural response of gravid mosquitoes to varied graminoid plant species, despite the fact that volatile pro les appear variable. The behavioural response of gravid An. gambiae induced by the wild graminoid plants in our bioassays was in the same ranges as those reported previously for An. arabiensis and An. coluzzi in response to low release rates of headspace extracts from rice plants [22] and from the tropical African wetland grasses (Poaceae) Echinocloa. pyramidalis, Echinocloa stagnina and Typha latifolia [42]. It was also in a similar range as observed for the attraction of unfed females to plant-based volatiles [58,60,70]. A limitation of our study was our inability to access equipment for electro-antennography to determine exactly which volatile chemicals released from the test plants were detected by the gravid female's antenna. However, when comparing the volatile chemicals identi ed in our study with those published for rice plants and pollen from sugar cane and maize in the context of oviposition [22,43,44], as well as with those published for a range of plants preferentially visited by malaria vectors for sugar feeding [32,[58][59][60], it becomes apparent that there is signi cant overlap in the chemical compositions.
It might be useful to explore their potential to manipulate odour-orientation of Anopheles mosquitoes in follow-up studies, since they have been implicated as semiochemicals for other insect species [75][76][77][78][79]. For example, 1,1-dimethyl-3-methylene-2-vinylcyclohexane was attractive to the beech leaf-mining weevil [76], guaiene has been suggested to play a role in the attraction of the litchi stem-end borer [80] and βelemene has been implied to contribute to attraction of the gravid tobacco moths [77]  , all of which present strong long-range cues likely used by gravid mosquitoes to evaluate the location and quality of potential oviposition sites [2]. In this context, it remains therefore unclear, if attractive, yet common, plant-based semiochemicals in odour-baited traps will be able to compete in an attract-and kill approach, with the complex interaction of cues provided by natural aquatic habitats. To date over 100 semiochemicals have been identi ed for mosquitoes of all physiological stages, yet synthetic odour-baited traps hardly play any role in contemporary surveillance and control of malaria vector mosquitoes [4]. Synthetic odour-baits mimicking human body odour have shown to perform poorly in attracting host-seeking Anopheles mosquitoes when presented in close vicinity to natural human blood hosts [86] and eld evaluations of the oviposition attractant cedrol, showed that visual cues provided by an open water surface were essential in combination with the chemical cue to attract wild oviposition-site searching females [17]. In order to develop vector control interventions that manipulate the odourorientation of malaria vectors in their natural environment, less emphasis might be placed in future on detecting more semiochemicals but more emphasis on how to formulate and present these chemicals in combination with other essential cues used by mosquitoes, to improve the e cacy of such interventions [4].

Conclusions
Our results suggest that plant volatiles provide a more general cue for gravid malaria vectors rather than vectors being highly adapted and evolved in context to speci c plant species and environments. All the graminoid test plants were very common, occurring in high abundance in grasslands and wetlands in sub-Saharan Africa and beyond [87][88][89][90][91]. Our results also challenge a previous suggestion [42] that volatile chemicals released from the grass family, Poaceae, are in general more attractive to gravid Anopheles mosquitoes than those released from the sedge family, Cyperaceae. The variations in chemical pro les and behavioural responses have been shown to be subtle across all studies. Productive breeding sites have been associated with species from both plant families in a number of eld surveys [24,26,29,40]. In nature, plant-based chemical cues interact with many other biotic and abiotic environmental cues to help gravid malaria vectors to orient and select suitable egg-laying sites, including non-plant-based chemicals [17,18,41,92,93], light and re ection [82], contrast [94], structure including plant height [29], conspeci c immature stages [16, 95,96], and other macroinvertebrates [14,97]. These complex interactions will need to be taken into consideration when designing 'attract-and-kill' strategies targeting gravid vectors with odour-baited traps.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. Table3.docx Graphicalabstract.tif