Recent advances in genomics and bioinformatics are allowing science to test the boundaries of reductionism as never before: how well can biological processes observed at the level of individual molecules, genes or cells predict the behaviour of more complex systems such as whole organs, individuals, or populations? These exciting technological developments have generated renewed interest within existing, more long-established biological disciplines to seek out empirical tools for quantifying and testing the relationship between phenomena occurring at different levels of biological organization in order to generate better predictions. One such field is medical entomology. Typically research in medical entomology is conducted at two very different scales: the first being laboratory-based studies of arthropod disease vectors under controlled insectary conditions, and the second being large-scale epidemiological surveys of their abundance and distribution in nature. While the former approach is clearly advantageous for identifying potential biological mechanisms, and the latter for generating hypotheses from correlations, it is increasingly recognized that the ability to make inferences across these scales is hindered by a conceptual and methodological gap in between that limits our ability to conduct hypothesis-driven empirical research on relatively large and environmentally realistic scales 1.

Arguably for the first time in the history of their discipline, medical entomologists now find themselves in the unique position of having both the need for such experimental initiatives recognized 2, and the financial support to create them becoming available through new funding streams in global health. The need to fill the research gap between laboratory and field is also stimulated by awareness that although current approaches to important vector-borne diseases such as malaria based on artemisinin-based combination therapies and insecticide treated bed nets are proving successful 34567, their long-term effectiveness may be undermined by the emergence of drug and insecticide resistance 8910. Consequently, the need for new strategies that exploit novel aspects of vector genetics, physiology, behaviour and ecology are increasingly needed. These innovations must be drawn from an understanding of vector biology within natural transmission settings if they are to yield rapid, locally appropriate strategies for disease control.

While almost all new approaches to vector control could benefit from a closer integration of laboratory and field perspectives 11, the most prominent candidate is the development of transgenic parasite resistant and/or sterile vectors whose release into the wild could reduce disease transmission by reducing parasite and/or vector populations 12131415. In the past decade, the genetic transformation of a number of important disease-transmitting mosquito species has become possible 1617181920. Transgenes have been identified that deliver effector molecules that substantially reduce the development of rodent malaria parasites 212223 and human dengue virus within mosquitoes 24, which has fuelled optimism that mass-release of laboratory-reared genetically-modified individuals could reduce disease transmission. The greatest unknown with respect to the feasibility of this approach is whether genetically-modified mosquitoes would be able to survive and successfully compete for mates against their wild counterparts outside of the confines of the laboratory. Initial laboratory studies indicated that transgenes impose fitness costs which reduce the reproductive success of the mosquito bearer 252627. A recent study suggests this disadvantage can be reduced by use of out-crossed mosquito lines 28, although so far only under conditions where exposure to parasites is substantially greater than mosquitoes encounter in the wild. While this improvement is encouraging, it does not address the problem that all laboratory-reared mosquitoes, regardless of their genotype, may have poor competitive ability in the wild. For example, recent comparative analysis of Anopheles gambiae s.s. in captivity and in nature in southern Tanzania suggest free-living males are larger and have greater lipid reserves than those reared under apparently optimal laboratory conditions 29. Regardless of whether this reduction in energetic reserves was due to selection for smaller individuals during the colonization process and/or sub-optimal conditions of insectary environments, it suggests laboratory-reared mosquitoes could be at a sizeable disadvantage to their wild counterparts.

Additional studies have shown that the mating success of male mosquitoes depends on subtle variation in environmental conditions experienced during larval development 3031, which may not be fully captured in mass-rearing facilities. These limitations are thought to have been responsible for the failure of many genetic control trials during the 1970's and 80's, which found that laboratory-reared male mosquitoes were unable to compete in the wild 32. Clearly, to avoid repeating these failures with the new generation of transgenic mosquitoes, intermediary testing grounds between the laboratory and field within disease-endemic countries are needed.

Semi-Field Systems (SFS) have been proposed as the best mechanism for bridging this gap. A semi-field system is here defined as an enclosed environment, ideally situated within the natural ecosystem of the target disease vector and exposed to ambient environmental conditions, within which all features necessary for its lifecycle completion are present 33. In the case of mosquito vectors of human disease, this typically involves a large outdoor cage in which the movement of the disease vector of interest either in or out of the unit is restricted by netting, and within which features such as aquatic larval habitats, blood hosts for adult females, sugar sources (plants) for adults, appropriate resting sites (houses, cattle sheds, etc.) and environmental features (e.g swarm markers to stimulate mating), are present. There are no general guidelines for the appropriate size of such a unit, but ideally it should be large enough to sustain a population of similar density to that encountered in the target environment for numerous generations.

This definition of a SFS differs from others that apply 'semi-field' to studies that actually involve observation of vectors in a non-contained setting or habitat, where only one part of its life cycle is present 34. A major goal of SFS is to establish multiple generations of a vector population within a contained setting, without outside intervention 35 in addition to facilitating short-term behavioural or ecological studies based on a single cohort. The main advantage of this approach is that because the abundance and composition of vectors within the SFS can be known, and if desired experimentally manipulated (either at the time of introduction, or through removal of some target individuals), much more precise estimates of the value and variability of demographic and life-history parameters can be obtained than would be from the field. Additionally, they allow researchers to conduct high throughput assays of control tools and ecological phenomena year round without risk of exposure to infection, as all mosquitoes used within the SFS will be free of parasites.

The concept of simulating the natural environment within contained settings in order to experimentally test ecological hypotheses does not originate in medical entomology. This approach has a long history in aquatic ecology, where hundreds of studies have successfully employed pond meso- and microcosms to examine the impact of biotic and abiotic factors on population and community dynamics 36. Furthermore, neither is this approach new within medical entomology. Almost 70 years ago, Hackett and Bates 37 commented on this need for ecological experimentation within natural disease transmission settings: "The study of behavior under natural, semi-natural and laboratory conditions necessitates locating the laboratory at the source of material. Self evident as this may seem, there are very few laboratories of this kind functioning at present in malarial regions". Since that time, only a handful of attempts have been made to create large-scale research facilities within semi-natural conditions in disease endemic settings, with the majority being initiated only in the last decade (Table 1). Early work in Albania and India used outdoor cages (< 75 m²) to conduct basic ecological observation of anopheline species 3738. Thirty years later this approach was revived for comparative evaluation of different genetically-based population suppression methods for the Indian vectors Aedes aegypti and Culex fatigans 353940 but was discontinued after the abandonment of the Sterile Male Release programme that motivated this research 41. Within the last decade, several research programmes in Africa, Asia, Europe and Australia have revitalized SFS for examination of mosquito vector ecology and control (Table 1). This approach has been used particularly productively in western Kenya 33, where SFS studies of the malaria vector An. gambiae s.s. within 85 m² modified greenhouses have yielded valuable insights into basic ecology and vector-parasite interactions 424344 and novel control and monitoring methods 45464748. Here the establishment of what is currently the largest SFS in the world for the purpose of experimental study of the ecology and control of African anopheline malaria vectors is described. This facility was built over a two-year period at the Ifakara Health Institute (2004–2006) and is the site of several new studies on vector behaviour, ecology and control.

In just under two years a 625 m² state-of-the-art SFS for large-scale experimentation on anopheline mosquito ecology and control was established within a remote area of southern Tanzania where malaria transmission intensities are amongst the highest ever recorded 4950515268. This unique facility is more than 4 times larger than any SFS previously or currently in existence, and has capacity for a wide variety of research activities including mass-rearing of African malaria vectors under natural conditions, high throughput evaluation of novel control and trapping techniques, short-term assays of host-seeking behaviour and olfaction, and long-term experimental investigation of anopheline population dynamics and gene flow within a contained environment that simulates a local village domestic compound. This was accomplished through a multidisciplinary collaboration between entomologists, senior public health scientists of the IHI, architects, engineers, site managers and a dedicated team of labourers who built this structure largely in the absence of electricity or any other mechanized construction aids.

Experimental activities have only recently been initiated within the SFS, and the ultimate value of this facility as a research tool will be realized as the studies now underway reach conclusion. Preliminary results from short-term behavioural assays of An. gambiae host-seeking behaviour using odour-baited traps and live animal baits suggest that realistic and repeatable results can be obtained within the SFS in a relatively short period of time (I. Lyimo & S. Moore, pers. comm.). The longer-term task of establishing a self-replicating, free-living population of An. arabiensis within simulated village conditions is currently underway. Although too early to forecast the outcome of this objective, the fact that an amenable spectrum of climatic conditions can be generated within the SFS is encouraging. Although daily temperatures within exposed areas of the SFS routinely exceeded the optimum temperature of An. gambiae [26.5 C under insectary conditions, 69] for some periods of the day, they were not significantly higher than those of the natural environment immediately outside of the SFS. Had the aim been to create an insectary facility for efficient mass production of An. gambiae, the regular daily periods of excessive ambient temperatures within the SFS (>35°C) would be a cause for concern. However the goal was instead to simulate the ambient climatic conditions within the Kilombero region and this was accomplished. Furthermore, the features that were built within the SFS provided microclimatic refuges in which mean temperature and variability was substantially reduced and stayed within the acceptable limits for adult survival and reproduction. For example, the mean temperature within the mud house inside the SFS was 3.5°C lower than exposed areas of the SFS. This observation matches reports from South Africa of air temperature inside mud and thatch houses being 3–6°C cooler than ambient conditions 70. Temperatures within the mud house in our SFS were significantly higher than those reported in a similar structure within the SFS at Mbita, western Kenya 33, which may be more reflective of the different climatic conditions between study sites than structural differences in SFS design. Water temperatures within the artificial larval habitats in our SFS were higher than the reported optimal value gauged from insectary studies [24–26 C, 71] and those reported for the Mbita SFS 33, but remained within the natural range observed in An. gambiae s.l. larval habitats in east Africa 7273. Given that free-living An. gambiae s.s. were able to complete their life cycle within the slightly cooler and much smaller confines of the Mbita SFS 33 there is optimism that the same can be achieved in the IHI SFS with An. arabiensis, a species known to have greater tolerance of hot and arid environments 6674.

While early observations are promising, much still remains to be known about how representative conditions inside the SFS will be of mosquito ecology in the wild. Open questions include whether a self-replicating population can be maintained over numerous generations on this spatial scale, what carrying capacity this population will reach under ambient climatic and host (bovine) conditions, whether additional climatic refugia or controls will be needed, and if existing plant and nectar sources within the SFS will be sufficient to maintain the adult male population. Importantly, the identification of limitations in the ability of our SFS to replicate natural mosquito dynamics as experimental work progresses will in itself provide valuable knowledge of the crucial determinants of anopheline population growth and stability that would not be possible under natural field conditions.

As research begins at the SFS in Ifakara and similar facilities around the world (Table 1), it is useful to consider the major challenges to the successful use of this research tool. These challenges are varied and range from the purely scientific to those of logistics and ethics. Five key areas merit discussion. The first is the possibility that although biological inferences made from SFS may be much more realistic than those from cage studies, they may still misrepresent some areas of mosquito ecology and population processes in nature. For example, although full life cycle completion of An. gambiae s.s. was achieved within the SFS in western Kenya, it was noted that the artificial breeding sites within it gave rise to considerably fewer larvae than expected 33. Whether this reduced efficiency was due to a problem with the environmental conditions inside the SFS, or maladaptation of the laboratory population used in these experiments to ambient conditions is unknown, but suggests there could be unique constraints or bottlenecks acting on population growth within these systems. Conversely, absence of the full range of environmental risks within the SFS such as stochasticity in host encounter rates, insecticide treated bed nets, predation by small vertebrates, pathogens and extreme environmental conditions such as flooding may result in an overestimation of life-history and demographic rates. For example, Knols et al estimated the daily survival of An. gambiae s.s within their SFS to be 90% which is higher than reported in many field studies 75. In order to reduce the risk of accidental parasite introduction, non-amplifying animal hosts will be used as the main source of blood in many SFS. While numerous disease vectors include the blood of non-human animals in their diet, many of the species that are most problematic exhibit a pronounced preference for humans 76. Several studies have shown that the fitness haematophagous insects derive from blood varies with host species 777879808182. It is also known that selection for divergent preference for human or cow hosts in An. gambiae mosquitoes can be generated in as little as 5–6 generations of selection 83. Thus constantly exposing vector populations within SFS to non-human hosts could result in the generation of individuals with different phenotypes, genotypes and population dynamics than those who feed on and transmit disease to humans. Continued monitoring and comparison of SFS results to those observed in the field will be useful to identify which, if any, of these issues pose serious obstacles to interpretation and reinforce the point that SFS studies are intended to complement but not replace field studies.

A second scientific concern is that vector populations established within SFS will likely be considerably smaller than those in the wild and thus experience inbreeding and a resultant reduction in genetic diversity which could impede fitness. It is well known that genetic diversity within insect vectors can be considerably reduced during laboratory colonization 848586. Free-living populations established within SFS may be considerably larger than typical laboratory colonies and thus avoid a similar intensity of inbreeding, however it is unlikely they will escape some bottle-necking and an associated loss of diversity from founder populations. Within only a few generations of laboratory colonization, mosquitoes can develop significant behavioural divergence from wild populations which restricts mating between them 87. This phenomenon may also occur within SFS, although perhaps at a slower rate than in small laboratory cages. As genetic and phenotypic divergence between contained SFS and wild populations may be unavoidable, the need for repeated comparative sampling of individuals in both settings is advocated to track if and how genetic diversity is reduced in captivity, and provide guidelines for how frequently captive populations should be enriched by fresh genetic material to maintain representative levels of diversity.

Should self-replicating vectors be successfully established in SFS, a logistical obstacle to the estimation of precise demographic rates from them will be the problem of disentangling overlapping generations. While much more precise estimates of mosquito population size will be possible within the contained environment of an SFS than in nature, it will remain difficult to accurately monitor individual-level activities such as mating behaviour and resource acquisition, and its resultant impact on fitness. The development of novel marking schemes using stable isotopes 8889 or distinct genetic traits may permit more precise monitoring of the behaviour and reproductive success of specific subsets of individuals, or individuals themselves.

For greatest public health relevance, SFS should be situated within or as near as possible to natural disease transmission environments as possible. Placing a large contained population of competent disease vectors within an appropriate transmission setting will always raise biosecurity concerns. Any breach of containment could result in increasing the disease transmission within the local area, and the accidental introduction of parasites into contained populations from asymptomatic carriers working within the facility could also generate the potential for infection. Awareness and discussion of how to prevent those risks are absent from early accounts of SFS use, but are justifiably coming to the forefront as plans for large-scale studies with genetically-modified disease vectors come under development. Recently an international committee of scientists formalized guidelines on recommended biosecurity measures and precautions for contained SFS trials with genetically-modified mosquitoes 55. The publication of these guidelines represents a significant step forward in thinking regarding the ethical responsibility for good practice within these facilities.

A final, crucial issue for the expansion of SFS-based research programmes throughout the world is the need to engage and promote awareness within the communities that host these facilities. The communities surrounding SFS research facilities should be the primary beneficiaries of research conducted within them, and their particular needs as end-users must be kept in mind when using these facilities to trial new vector control strategies. While researchers working in SFS may have this goal clearly in mind, it will be of little value unless clearly and regularly communicated to local communities in an open and discursive manner. Understandably local residents may be apprehensive about the placement of an SFS containing live insect vectors near their home, and misinformation about the purpose of this work and risks associated with it could cause considerable friction. This lesson was painfully learnt by scientists working at the Indian Council for Medical Research in the 1970's who had their research unit on Aedes mosquitoes shut down when journalists falsely alledged that the actual purpose of the work was biological warfare and human population control 41. That such a debacle could occur in one of the most successful disease vector control research programmes of all time [with 104 papers published in 6 years and numerous insights into population suppression gained, 41] is a sobering thought for all those involved in this new generation of SFS research. Community awareness activities have begun in Ifakara, and must be sustained and scaled up if both local community members and researchers working at the SFS are to obtain maximum benefits from this research facility.

Engagement must extend beyond local communities to include scientists and students working within the disease endemic countries that host SFS. These facilities can provide substantial indirect benefits by acting as state-of-the-art training tools for young vector biologists in which methodological skills can be honed, and independent research hypotheses experimentally tested in a disease-free setting. Currently, at the IHI, there are three east African postgraduate students pursuing their PhD studies on research based within the SFS and plans to recruit several more underway. Thus this research tool will contribute to the IHI's goal of substantially increasing Ph.D-level capacity in malaria vector research within Tanzania and east Africa. Much of the recent motivation for initiating SFS programmes has been driven by laboratory-based research on genetic modification of disease vectors that has occurred almost exclusively in developing countries. For both the transgenic approach and other emerging vector control strategies to fulfill their potential, it is absolutely imperative that endemic country scientists are actively involved in driving SFS-based research and taking forward innovative techniques developed within it.

The authors declare that they have no competing interests.

HMF was the primary project coordinator of SFS establishment at the IHI and drafted the manuscript. KRN designed and set up the areas of the SFS where a free-living An. arabiensis population will be established in simulated village conditions, and assisted with collection of temperature data. TW was the lead architect and DK the head of the maintenance unit that carried out construction. SJM oversaw parts of the construction and set up the olfaction study chamber. IL designed and set up the experimental hut and insectary area of the SFS. TR oversaw parts of the construction and helped supervise research activities within it. HU and HM provided institutional support and guidance in logistics, ethics, and community sensitization. GFK provided support with project planning and coordination. BGJ initiated this project, obtained financial support for it, and provided scientific and logistical guidance.