The introduction of blood-fed females into the greenhouse resulted in the presence of eggs in the breeding sites on day 3 (2.5 days after release), and eggs continued to be observed in the sites until day 7 (Fig. 3). Larvae (from L1 to L4 stage) were seen feeding at the water surface until day 23, when the last L4 larva pupated. The first five pupae were seen in the breeding sites in the evening of day 10, meaning that the variation in maturation time from egg to pupa was 7–20 days. In spite of having conditions with higher larval density and smaller water surface area, Gimnig et al. [53] recorded much reduced periods to pupation, from 5 to 12 days. In total, 57 pupae were counted in the breeding site 3.8 m in front of the hut and 130 in the site 1.1 m behind it, and on average they harboured only 0.08 and 0.18 larva/cm², respectively. These densities are lower than those observed in natural habitats [53], and given the fact that we observed algal growth, considered important for larval growth [53], it seems that the prolonged time to pupation in this trial may be attributed to the relatively small range over which temperatures fluctuate. Alternatively, re-introducing mosquitoes that had been maintained under laboratory conditions for several years in a more natural setting may have caused these effects, and poor adaptation to these conditions may have stunted their development.

The first adults were seen inside the hut on day 11, and continued to be present until the end of the experiment (day 27). Starting in the morning of day 22, we observed new eggs in the breeding sites and subsequent larval development.

From the above it can be deduced that specific behaviours of the adult insects occurred during certain time periods (Fig. 3). Oviposition activity took place twice during this trial, meaning that females survived until eggs were mature, that they successfully located a breeding site, accepted it for oviposition, and laid eggs. Other potential breeding sites, like the leaf axils of the banana trees in the sphere, were examined but were not found to harbour eggs, larvae or pupae. The period for reaching sexual maturity for males may be at least one day and for females up to 60 hrs [54], so mating of the F1 generation may not have taken place until dusk on day 14. In spite of regular observations during dusk, we did not see any swarming activity typically associated with mating in An. gambiae [55-57]. This is a particularly interesting point, because in contrast with other settings (e.g. [56]), where mating swarms were frequently observed we failed to do so over a 3-year period in the Mbita area as did Charlwood (pers.comm.) who observed only one anopheline swarm during 4 years in the Kilombero valley of Tanzania, raising the question as to whether swarming is an obligate component of the An. gambiae life-cycle or a facultative adaptation to local ecological and/or seasonal conditions. Alternatively, maintenance of mosquitoes in laboratory cages for several years forced this strain to become stenogamous (i.e. the ability to mate in small cages), and this adaptation may have interfered with its ability to swarm when introduced into the sphere.

As newly emerged adults rarely survive for more than 48 hrs without the availability of an energy source [58], mosquitoes must have supplemented their energy reserves with carbohydrates from the plants in the sphere (Table 1) for up to 6 days before they were allowed access to a blood source, in the form of a human volunteer sleeping in the hut. Some plants like Castorbean (Ricinus communis L.) were flowering at the time of the experiment, and may have provided nectar sources. In cage experiments (Impoinvil et al., in prep.) we have observed a mean survival time of 7.0 ± 0.2 days on this plant, which is comparable to mosquitoes given 6% glucose (8.7 ± 0.2 days). However, given the fact that An. gambiae mosquitoes emerge from the pupal stage with a deficit not only in carbohydrates, but also lipid and protein [59,60], which usually is compensated for by consuming a (small) blood meal within the first few days of adult life [61], it is likely that mortality during the 6-day post-emergence period in the current trial was too high to have a good number of the 80–90 females that emerged survive long enough to obtain their first blood meal.

Within 15 min of entering the hut at night, the volunteer noticed the sound of mosquitoes and subsequently felt mosquito bites on his exposed lower limbs. This implies that females were receptive to host cues, entered the hut, probably through the eaves [48], and then successfully located and fed on the human host. At sunrise, several engorged females were seen resting on the walls, indicating successful blood feeding and endophily (indoor resting), which is typical for this species [62]. Following maturation of eggs, the second oviposition began during the night of day 21, thus completing the life-cycle. We continued to observe both immature and adult insects (presumably mostly from the F1 generation) until day 27, when the experiment was terminated (by refraining from entering the sphere for about 1 month).

The second experiment, in which we released 500 virgin females together with 1500 males demonstrated that mating does occur in a relatively small, semi-field system. After the third night that a volunteer had slept in the hut, we observed eggs in the breeding sites. The production of offspring, though, was low, and we only collected 40 pupae by the end of the trial period. This may have been caused by heavy rainfall during three consecutive nights (day 2–4), which may have affected the survival of the adults and/or larvae or washed away the larvae from the breeding sites due to overflow. Since we observed few mosquitoes, we decided to conduct a human landing catch during two nights inside the hut, starting two nights after sleeping in the hut had ended. Apparently no host-seeking females were present, as no mosquitoes were collected. Nevertheless, the life-cycle was completed, as manifested by the harvested pupae, which were removed from the breeding sites to prevent emergence of the F1 generation which would have compromised interpretation of survival of the F0 generation.

The third experiment started by introducing 500 eggs into each of the breeding sites, whilst 1000 eggs (from the same original batch) were reared under laboratory conditions. In the laboratory, larvae developed at the same rate and most reached maturity by day 10, when the first pupae were observed (Fig. 4). This was similar for the breeding sites in the sphere, except that development was highly asynchronous, i.e. some larvae pupated by day 10, whereas others took up to day 24 before pupation. These times to pupation are similar to those observed in the first trial, but are again in contrast with other studies [53]. On day 7 we counted all larvae and observed 887 in the insectary, as opposed to 652 in the sphere. Overall, the laboratory batch yielded 804 pupae, versus 495 from the breeding sites. On the basis of these data, the average daily survival was 0.90 for the laboratory larvae, and 0.83 for the larvae in the sphere. With nearly half the larvae surviving to the adult stage, these results contrast sharply with much higher mortalities (up to 90%) observed in the Kisumu area [63,64] and São Tome (Charlwood, pers. comm.). The sex ratio of emerging adults in the laboratory was 2:3, which translated into a female population in the sphere of 297, on the assumption that no insects died during emergence. On day 28, after the release of eggs in the breeding sites, we observed that eggs had been laid by the F1 generation, but with only 6 and 3 eggs in the two breeding sites respectively.

Our results have shown that by starting either at the post-blood feeding, pre-mating, or egg stage, a new generation of insects can be reared under these semi-field conditions, and that all life-history behaviours were successfully completed to a lesser or greater extent. This therefore represents the first and promising step towards continuous maintenance of parasite-free An. gambiae populations under semi-natural conditions that can be experimentally manipulated in studies of malaria vector ecology and transmission control. Russell and Rao [38] used a large outdoor cage, based on a design by Hackett and Bates [37], to study swarming and oviposition behaviour of An. culicifacies Giles in India, and showed that such systems can be used to unravel aspects of the behavioural ecology of anophelines. Our study shows that such systems can now also be developed for studies on African malaria vectors in order to start filling the gaps in our understanding of the behavioural ecology of these insects [65].

This system has obvious advantages over natural outdoor conditions. First and foremost, it provides a suitable intermediate between laboratory-based studies addressing mosquito behaviour and ecology, and the field situation. Too often, conclusions are drawn from results obtained under laboratory conditions that necessitate speculation as to what may or may not happen in the field. Fixed climatic conditions, cage-experiments, olfactometers and windtunnels, in which the mosquito strains used have been laboratory-reared for sometimes decades, may readily distort behavioural and ecological phenomena. Here we have shown that, beyond introducing F1 generation malaria-free mosquitoes from wild populations, it may be possible to rear vectors in situ within a semi-natural system that may minimize such artefacts. Conversely, the advantage of using insectary-reared mosquitoes is in the level of control that may be exerted that would not otherwise be possible: Experiments can be conducted all-year-round, with fixed numbers of insects, of known age and physiological status, in a malaria parasite-free environment under ambient climatic conditions. This enables more direct inferences to be drawn from data analysis as compared to longitudinal field studies, because of constant conditions and simplified experimental design.

We have recently evaluated the efficacy of several plants traditionally used by the Luo community as repellents in a similar semi-field set-up, and simple logistic regression, on data collected during four nights per plant, yielded significant results. Within a year of nearly continuous experimentation, the repellency of 8 plant species and 3 combinations thereof was evaluated through thermal expulsion or direct burning [41], and 9 species and 2 combinations thereof were tested in potted form [43]. Such studies would have taken several years under field conditions, and would be limited to times when mosquito densities would have been sufficiently high to permit meaningful experiments. Another recent study [44] focused on the survival of An. gambiae maintained on a variety of diets (blood or sugar alone, or a combination of both) and revealed similar results to those obtained under laboratory conditions [66]. Within one year of starting studies on the behaviour of mosquitoes around bednets in a semi-field setup, we transformed a regular conical bednet into a trap that may catch up to 70% of the females released [42]. Recent field evaluation of this trap shows it to be a promising replacement for the human landing catch (Mathenge et al., in preparation.). With the trap now being considered for commercial manufacturing (it took a mere two years to reach this stage), this would have been impossible without the availability of a semi-field set-up in which continuous experimentation ensured rapid progress towards product development.

Even though our system resembles more closely the field situation, it remains to be ascertained to what extent. Our current study was mainly qualitative in design and focused on life-cycle completion. Various observations were made that have been reported before from field studies. For instance, observation of eggs in the breeding sites in the morning of day 22 during the first trial implies that these originated from females that fed once on day 19, as those that fed on day 17 should have laid before. However, it is likely that these females fed twice, on day 17 and day 19, and should be classified as pre-gravids [61,67], before fully maturing a batch of eggs. Furthermore, in the absence of a human host, mosquitoes survived up to six days after emergence, confirming field results that feeding on plant sugars does occur and may be an important feature of the life-cycle in this species [68] (Foster and Knols, unpublished data). Obviously, there are limitations associated with these studies. Some phenomena, like dispersal, cannot be studied. There may be other, yet unknown, factors that affect the behavioural ecology of the insects in such systems. Or, as Bates [69] wrote, following his outdoor cage studies in Albania: "One still cannot be sure that the reactions of the mosquito are 'natural' because there is always the barrier of wire liable to be encountered on extended flights; and when the flight of a mosquito has been interrupted by this wire barrier its further activity may be definitely unnatural". Rightly so, and even though we did not observe any obvious distorted behaviours, it is imperative that findings from semi-field studies be verified under natural outdoor conditions.

Additional studies in which the release and performance of field-collected, blood-fed mosquitoes in the sphere is compared with that of laboratory specimens in terms of egg-recovery, developmental periods and important behavioural characteristics (like swarming) will provide further insight to what extent such systems mimic the natural Anopheles environment and colony adaptation impairs natural behaviours.

Nevertheless, since Bates' days, advances in science merit a renewed impetus towards semi-field studies in contained near-natural environments, particularly with respect to transgenic mosquitoes. Fitness evaluations of engineered strains of vectors are mandatory for transformation technology to become an established disease control tool in Africa. Perhaps this alone, is ample justification for more intensive sphere studies, hopefully not only in Kenya, but also in other African countries likely to be involved in this endeavour. Studies on gene flow, mating behaviour and reproductive fitness, combined with studies on the effects of laboratory maintenance on the genetic make-up of transformed strains to be released, can be conducted in semi-field systems [28,36]. Such systems, particularly when used to study genetically-engineered mosquitoes will require more advanced containment levels than the system described here. Guidelines for facility location, physical and biological containment, safety practices and calamity control need to be developed and adapted from existing arthropod containment guidelines [36,70].

There are several good reasons to further such studies in disease-endemic settings. Under such conditions it will be possible to transform offspring from wild mosquitoes, conduct experiments under local ambient climatic conditions and evaluate transgene spread and fixation in offspring from field-collected gravid females that emerge in a semi-field setup. Last, but not least, it will enable scientists from developing countries to become more directly involved in evaluating the potential use and application of transgenic mosquitoes for future malarial disease control.