Systematic identification of plausible pathways to potential harm via problem formulation for investigational releases of a population suppression gene drive to control the human malaria vector Anopheles gambiae in West Africa

The World Health Organization (WHO) estimated in 2019 there were 229 million cases of malaria globally, accounting for 409,000 deaths [1]. Persistent threats from insecticide resistance have created the need for additional, complementary strategies to control the mosquito vectors that are responsible for transmission of disease [2‐5]. Based on mathematical modelling studies, the use of gene drive in vectors such as Anopheles gambiae has been proposed as one such complementary approach to vector control [6‐12].

Recent advances in the development of synthetic gene drives involve the use of a transgene encoding CRISPR-Cas9 endonuclease that is expressed from a germline promoter, coupled with expression of a guide RNA targeting a specific DNA sequence [13‐19]. In heterozygous transgenics, the transgene is inserted at the target site of the guide RNA on one chromosome, while the target site on the homologous chromosome is wild type. In the germline of such heterozygotes, Cas9 uses the guide RNA to recognize and cleave the target sequence of the wild-type chromosome. This causes a double-stranded break in the germline at the target sequence, which is repaired using the homologous chromosome containing the transgene as a template, so that the transgene is copied onto the homologous chromosome that had previously been wild type, in a process known as ‘homing’.

Thus, homing converts a proportion of the germline in transgenics to homozygosity so that a greater proportion of gametes contain the transgene, and therefore a greater proportion of offspring will inherit the transgene than would otherwise be the case. Homing, therefore, allows the transgene to spread through a population with super-Mendelian inheritance. This means that, unlike some other genetically modified insect vector control approaches, gene drive transgenes are intended to increase in frequency and persist in the target organism (TO) to achieve lasting impact on disease transmission [20].

‘Population suppression gene drive’ is a potential way to exploit such a homing transgene in order to decrease the density of mosquito vector populations and thus reduce malaria transmission [5, 12, 21]. One method involves exploitation of a haplo-sufficient female fertility gene, mutations of which can cause female sterility when homozygous, but not when heterozygous [6, 12, 16, 17, 21]. The CRISPR-Cas9 system is encoded by a transgene that uses a guide RNA to target such a female fertility gene allowing the transgene to spread at super-Mendelian levels through the population via homing, disrupting the female fertility gene as it does so. As the heterozygous transgenics are fertile, their frequency steadily increases in the population after introduction of the transgene. As heterozygous transgenic females and males begin increasingly to encounter and mate with each another, the proportion of the female population that is homozygous for the transgene increases. However, as they are sterile, the number of viable eggs produced decreases until eventually the population is suppressed [17].

The Target Malaria not-for-profit research consortium has been engaged in the co-development of gene drive as a malaria vector control tool in Africa [22]. Before a population suppression gene drive system undergoes regulatory review for field release, plausible risks to human health and the environment must first be identified and assessed [3, 4, 23‐35]. Indeed, James et al. [30] proposed that “the safety standard for moving an investigational gene drive product from physical confinement to field testing should be a well-reasoned justification that it will do no more harm to human health than wild-type mosquitoes of the same genetic background and no more harm to the ecosystem than other conventional vector control interventions.”

Problem formulation is a rigorous scientific analysis that defines the overall parameters for an environmental risk assessment (ERA) and facilitates the systematic identification of potential harms or hazards, as well as their routes of exposure, whilst being transparent about the assumptions that have been made during the process (see Table 1 for Glossary of Terms) [33‐39]. The WHO has developed a guidance framework for testing genetically modified mosquitoes to ensure that organisms are effective and competitive and that risks are reduced to acceptable levels [40], so that gene drive organisms intended for release should have already been tested for deleterious phenotypes or unacceptable, unintended effects during product development. Thus, ERAs can focus on the introduced genetic construct and the intended outcome. ERAs should specifically address protection goals that are identified from policy, legislative, regulatory and community requirements from the region where the intervention is being considered. Referring to such protection goals, a wide range of potential harms is initially considered in a highly iterative, systematic approach involving a diverse range of expert input. An ERA of the release of a genetically modified organism needs to consider both direct effects on individual organisms that the transgenic itself generates, such as via predation, competition, habitat alteration, hybridization and introduction of new parasites and diseases, and indirect effects such as those on individual organisms in the wider environment without immediate contact with the transgenic (Table 1) [26]. Based on the biological information available on both the transgenic and its parental species, as well as the wider environment into which the transgene will be released, the plausibility of each pathway to potential harm to the identified protection goals is examined by establishing the causal chain of events, sometimes condensed to “pathway to harm” [41]. To determine the cogency of each pathway, risk hypotheses are constructed that can be used to interrogate individual steps in the pathway. Next, an analysis plan is developed that includes both defined measurement endpoints to test risk hypotheses and other potential sources of evidence aimed at reducing identified areas of uncertainty surrounding pathways to each potential harm. Such evidence may be sought from pre-existing literature, modelling, new experimental investigations, previous experiences [35], or any combination thereof. The analysis plan will make important contributions to subsequent steps of hazard, exposure and risk characterization in the ERA. This also allows the identification of hypotheses that can be tested most reliably and efficiently. The potential for harms to occur can be considered negligible where a risk hypothesis can be accepted on the basis of unequivocal evidence and minimal remaining uncertainties surrounding the pathway. Where this is not possible, a number of risk hypotheses must be tested using a ‘weight of evidence’ approach that draws upon several sources of evidence for assessment of the pathway [35, 42].

Problem formulation has previously been used to identify potential harms or hazards and routes of exposure for the control of insect pests [43] and potential releases of generic gene drive systems for malaria vector control in Africa [44‐46]. Romeis et al. [43] reported on pathways to harm for population suppression gene drive in the agricultural pest Drosophila suzukii, identifying (i) indirect food-web effects; (ii) toxicity from the transgene; and, (iii) gene transfer to other species as potential harms. David et al. [44] identified ecological harms as a first step towards an ERA, distinguishing between transient and steady state harms but without specific reference to protection goals or pathways. Roberts et al. [45] derived protection goals on human health, animal health, biodiversity and water quality for broad classes of gene drive, identifying numerous high-level potential harms belonging to each of these categories, although they were not mapped onto individual pathways to harm, so that risk hypotheses and analysis plans were not developed or reported. Based on a series of four workshops in different African countries, Teem et al. [46] reported a problem formulation exercise based on four different types of gene drive to control malaria, identifying many protection goals similar to those previously reported [45], along with six high-level, consensus pathways to harm. However, the authors recognized that the approach was not designed to constitute a comprehensive first step in a specific ERA exercise and developed only limited analysis plans for each pathway [46].

Here, problem formulation was used to systematically and comprehensively map 46 plausible pathways to potential harm from the simulated investigational release in West Africa of a population suppression gene drive that would employ the CRISPR-Cas9 system to target the doublesex female fertility gene (dsxF^CRISPRh) [17, 19]. For each of these pathways, risk hypotheses and an analysis plan were developed [47], which will be used to inform subsequent steps of hazard and exposure characterization in an ERA for population suppression gene drive in West Africa.

The aim of this problem formulation exercise was to generate a comprehensive, if not exhaustive, set of plausible pathways to potential harm to previously defined protection goals from simulated field releases of the dsxF^CRISPRh transgene in West Africa. Although based on the gene drive strain containing the dsxF^CRISPRh transgene as described in Kyrou et al. [17], this work should also broadly be applicable to other similar CRISPR/Cas9-based population suppression gene drive strains targeting doublesex, for example those exploiting multiplexed guide RNAs (see Table 2). However, some of the pathways presented here may need to be re-examined on a case-by-case basis, depending on the strain under investigation and release protocol to be used. Indeed, new or altered pathways may also be identified as the scientific evidence base expands. These pathways are, therefore, likely to be revised and updated periodically, including with feedback from stakeholders and the wider scientific community.

Although the logical, stepwise process allows for the systematic identification of plausible pathways to potential harm, the strict linear format does not allow for branching pathways, parallel pathways or feedback loops. This may sometimes force the ‘shoehorning’ of parallel processes into a series of sequential events. The approach does, however, offer a transparent communication tool for the unbiased identification and description of pathways, potential harms and analysis plans.

Another difficulty encountered in this exercise was in defining the criteria for plausibility; pathways were considered plausible based on scientific and technical knowledge, but also based on expert judgement, introducing an element of subjectivity to the exercise [41]. This situation is not unique to problem formulation, however, as expert elicitation in quantitative risk assessment, for example, also involves expert judgement [137].

The precise classification of TOs and NTOs may also evolve over time. Recently, membership of the An. gambiae complex has expanded with the addition of Anopheles fontenillei from Central Africa [52]. Moreover, some species of the complex, are currently considered to be only minor vectors of malaria but are still TOs, so that one expected consequence of the intervention could be suppression of species that do not efficiently transmit disease. However, it may still be worthwhile to reduce the density of such vectors, notwithstanding them being inefficient at transmitting disease at present [136].

It could also be argued that many of the pathways identified here, although plausible, would not, in any event, be expected to occur because the potential harm would be identified during standard product development and result in the strain being eliminated from further advancement towards release. There are already regulatory precedents for the development and successful field release of transgenic mosquito strains [20, 138] so that increased vector competence, decreased insecticide sensitivity, broadened tolerances for abiotic conditions, or toxicity and allergenicity of transgenic products are all recognized as having the potential to contribute to harms to health and environmental protection goals. In that respect, many of the pathways identified here are not uniquely drawn from problem formulation. However, this problem formulation approach did help to unpick many of the more complex environmental and ecological pathways to potential harm that need to be further investigated in subsequent steps of ERA for population suppression gene drive. This exercise also allowed for the systematic definition of analysis plans and made the process of identifying and describing potential harms and their plausible pathways demonstrable, transparent and systematic, allowing for any potential for erroneous assumptions or unconscious bias to be challenged.