We model gene drive releases in sixteen areas of 1° in latitude by 1° in longitude located across Western Africa (Fig. 1a), which were chosen to span variation in a set of factors that may be important to gene drive impacts. These include the prevalence of Plasmodium falciparum malaria in humans, vector abundance and species composition, and seasonality, estimated from fine-resolution geospatial layers (refs. 35,36,37 and see the Methods and Fig. S4). In each area, we model a mosquito metapopulation where the mosquito populations are linked to the locations of human settlements, which we estimated using fine-resolution settlement footprint data. We used a Voronoi tessellation around the settlement locations to define a locality covered by each population (e.g. Figure 1b and see Methods). Each locality has a specific human population density, rainfall profile, distribution of water bodies and vector species composition (see Methods). The estimated number of settlements/localities varied from 168 in the area located in Cameroon to 4039 in the Lagos region of Nigeria (Fig. 1a).Fig. 1: The areas of west Africa in which we model the impacts of gene drive releases.a The sixteen areas are coloured according to the estimated prevalence of P. falciparum malaria in humans residing in the area in 201935 (see Methods). Each area is labelled according to the countries within the area, and numbers in brackets show the estimated number of settlements in the area. Grid lines divide the region into areas of 1° in latitude by 1° in longitude. b An example of the landscape model for The Gambia/Senegal area showing the estimated settlement locations (black dots), localities (polygons) and the estimated number of mosquitoes associated with each settlement. Data show the estimated annual maximum number of mosquitoes in the year 2018.For each of three vector species groups (An. funestus, An. arabinesis and the combination of An. gambiae and An. coluzzii), we assume that the abundance of each population is regulated by the amount of larval habitat, which is estimated from geospatial data on weekly rainfall, the presence of rivers and lakes and the human population size. The carrying capacities of each species group were assumed to be proportional to their estimated relative abundances at each location36. In each area, we scaled the carrying capacities by an equal amount such that the seasonal peak number of mosquitoes in the area is consistent with that estimated by a malaria transmission dynamic model that relates mosquito abundance to the entomological inoculation rate (EIR) and the prevalence of P. falciparum malaria (e.g. Fig. 1b and see Methods). Among the sixteen study areas, we estimated that the largest vector populations are in the Côte d’Ivoire area (\(4.2\times {10}^{8}\) mosquitoes across all the species groups at the seasonal peak), and the smallest are in the Cameroon area (\(3.4\times {10}^{6}\) mosquitoes), a span of roughly two orders of magnitude.Impacts of gene drive releases on vector abundanceWe first consider a gene drive that reduces the egg laying rate of females carrying a single copy of the gene drive by 35% compared with wild females (we later investigate sensitivity to this fitness cost). We further assume that the gene drive allele is inherited by 97.5% of the offspring of male and female heterozygotes (gene drive paired with wildtype), while the remaining offspring inherit either the wildtype allele (1.25%) or a non-functional resistant allele (R2 allele; 1.25%) due to non-homologous end-joining (NHEJ) during the homing reaction38. We assume females are sterile if they do not carry at least one wildtype allele.We modelled simultaneous releases in all three vector species groups, and each simulation tracked the population dynamics for 12 years from the release date. For each area, we replicated the simulations over a broad range of mosquito movement rates and population sizes to account for uncertainty in these factors. Specifically, we varied the dispersal propensity \({\rho }_{n}\) (the probability an adult mosquito moves to a connected locality on a given day, see Methods) across five values from \({\rho }_{n}\) = 0.001 to \({\rho }_{n}\) = 0.025, and we varied the mosquito population size across five values from half to double the area-specific estimates (by adjusting carrying capacity parameters, see Methods), resulting in 25 parameter combinations of dispersal and population size. For each parameter combination, area, and species group, we simulated releases of 1000 gene drive heterozygous male mosquitoes in each of fifty randomly selected settlements. These releases were predicted to result in substantial reductions in biting females in all sixteen areas over 12 years (Fig. 2a). There were consistent differences between areas, however, with some areas showing average suppression at 12 years following releases of >95% (Cote d’Ivoire, Nigeria/Lagos, Liberia/Sierra Leone, Benin/Togo, Ghana, Northern Nigeria) with the remaining areas showing average suppression of >70%, with some realisations showing <50% suppression (Southern Mali and Cameroon).Fig. 2: Population suppression predicted by the mosquito metapopulation model.a The suppression of the total number of biting females in each area in the years following gene drive releases at time zero in all vector species groups. The boxes represent the interquartile ranges across 25 parameter sets that differed in both dispersal (\({\rho }_{n}\,\&\,{\rho }_{b}\)) and population size parameters (\({K}_{1,a}\,\&\,{K}_{2,a}\)) (see Methods and Supplementary Material), with 5 replicate simulations per parameter set. The white lines in the boxes show the medians, the whiskers represent 1.5 times the interquartile ranges, and the individual points are outliers. b–d. The sensitivity of these results to (b) average population size, (c) population density (the number of populations in the simulated area), and (d) the mosquito dispersal propensity. In each plot, the points show average 12 year suppression across simulations sharing the same focal parameter, and we highlight the results of five simulation areas (coloured; the grey lines/points represent the remaining eleven simulation areas). All simulations followed the default release strategy, described in the text, of 1000 males released in 50 localities in all three species groups.Sensitivity to population size, density, dispersal, and seasonalityTo understand these differences in suppression, we assessed the role of four factors on the simulation results described above: average population size, population density (the number of populations per unit area), dispersal propensity, and seasonality (defined as the average number of dry weeks per year in an area; Table S2). A multivariate regression model with these predictors explained much of the variance in suppression across all simulations (\({R}^{2}=0.79\)). There was a positive relationship between suppression and each of the first three factors, i.e. suppression is higher where mosquito populations are larger (Fig. 2b), more densely packed (Fig. 2c), and if dispersal is at the higher end of the range that we considered (Fig. 2d). Dominance analysis39 revealed that the first three factors had similar power to explain the variance in our suppression results; population size contributed 30% of the overall \({R}^{2}\) (standardised Dominance 0.30), dispersal contributed 32%, and population density contributed 29%.The role of recolonisation rateThe commonality of the above three factors—population size, population density, and dispersal propensity – is that they all affect the numbers of mosquitoes moving between populations. We contend that suppression increases with population-level mobility because higher mobility increases gene drive spread into wildtype populations, reducing the likelihood that wildtype mosquitoes recolonise habitats where local extinctions have occurred. If this explanation is correct, we can expect reduced rates of populations cycling between extinction and recolonisation as mobility increases.To test our assertion, we computed the rate at which populations cycle through states of extinction and recolonisation for each simulation. Specifically, we defined the local suppression \({s}_{i,d}\) at locality \(i\) on day \(d\) as \({s}_{i,d}=1-\frac{{n}_{i,d}}{{n}_{i,d}^{*}}\) where \({n}_{i,d}\) is the number of biting females in that population and \({n}_{i,d}^{*}\) the equivalent number from a simulation of the same scenario except without gene drive releases. We defined a recolonisation event in locality \(i\) as a transition from extinction (\({s}_{i,{d}_{1}}=1\)) to a state of recolonisation (\({s}_{i,{d}_{2}} < \,0.5\)) in the time-series \(\left({s}_{i,1},{s}_{i,2},\ldots \right)\), and defined the population cycling rate to be the average number of recolonisation events per year among the populations where the gene drive was released. Each of the three explanatory variables – population size, population density, and dispersal propensity – negatively associated with population cycling rate (Fig. S1a–c). Moreover, we found a strong negative association between population cycling rate and area-wide suppression (Fig. S1d). These results support our assertion that the three explanatory variables affect suppression via their influence on population-level mobility.Variation in spatial dynamicsThe effects of population-level mobility on extinction and recolonisation cycling, and thus suppression, resulted in markedly different spatial dynamics in the different study areas. In areas with low mobility, the wildtype allele frequently escapes the influence of the gene drive allele to recolonise habitat where the mosquito had previously become extinct. For example, in the Western Mali area where the populations are small and widely spaced, a typical simulation reveals irregular waves of extinction followed by waves of wildtype recolonisation such that the landscape is an ever-shifting mosaic of neighbourhoods in different states (Fig. S2; a simulation of gene drive dynamics in An. arabiensis in Western Mali is viewable here) [https://github.com/AceRNorth/Animations/blob/main/anim201.gif]. By contrast, in regions with larger and more numerous populations, vacated habitat tends to be rapidly recolonised by neighbouring populations that contain the gene drive such that a degree of population suppression is maintained (e.g. An. arabiensis in the Côte d’Ivoire area; Fig. S2; animation viewable here) [https://github.com/AceRNorth/Animations/blob/main/anim211.gif]. The dynamics in most of the study areas fell between these two types. A simulation in the Southern Mali area, for example, shows that local extinctions are typically recolonised by neighbouring populations containing the gene drive, yet there are also occasions where populations become free of the gene drive and the wildtype allele temporarily spreads locally (Fig. S2; animation viewable here; An arabiensis) [https://github.com/AceRNorth/Animations/blob/main/anim203.gif].Sensitivity to heterozygote gene drive fitnessPrevious studies of female fertility gene drives have highlighted the importance of the effect of the gene drive on fertility in heterozygous females13,18, which we now consider (Fig. 3). Predicted suppression is zero in the extreme case of heterozygous sterility, because such a drive allele will not increase from rare even in a local well-mixed population13,18. As the fitness of heterozygous females increases from zero, the predicted 12 year suppression gradually increases in all study areas, up to a fitness of about 0.4 (meaning that one-copy females produce 60% fewer eggs than wildtype females). Indeed, up to this fitness level, the predicted suppression in all sixteen areas is near identical. As fitness increases further, the study areas diverge in predicted suppression as discussed above. Across all areas, however, suppression tends not to increase as heterozygous fitness levels increase beyond 0.5.Fig. 3: The sensitivity of predicted population suppression to the fitness of heterozygous females.The points represent the total suppression of all three species groups 12 years after releases in each species group, averaged across 5 replicate simulations. All other model parameters are set to their default values (Table S3).To understand these fitness effects we must consider how the gene drive allele affects individual populations. If the fitness of heterozygous females is low (0–0.4), the drive allele will rarely become locally fixed. Instead, it will rise to an equilibrium frequency with both the wildtype doublesex gene and R2 alleles being present. This equilibrium results in population suppression rather than extinction, to a degree that depends on the drive allele fitness18. Such a gene drive will thus tend to spread radially from its release rather than induce extinction and recolonisation dynamics, which is why the predicted results are similar in all the areas we consider. At fitness levels greater than ~0.4, local suppression more frequently causes local extinction, thereby causing a switch to extinction and recolonisation dynamics, whose precise form will depend on the area.Impacts of gene drive releases on the malaria burdenFor each of the sixteen areas, we use our malaria transmission dynamic model to compare the impacts of gene drive releases in each of the three vector species groups on the prevalence of P. falciparum malaria in the human population. In each area, we parameterise the model to approximate historical trends in malaria prevalence in humans over the period 2000–2018 accounting for changes in the coverage of malaria control interventions throughout this time, including vector control interventions such as ITNs and IRS, and drug treatments and SMC. This model assumes that the human population is well-mixed throughout each modelled area (see Methods). We first consider the impacts of releases in a single vector species group only (Fig. 4; blue, green and red lines and markers show releases in An. funestus, An. arabiensis, and both An. gambiae and An. coluzzii respectively).Fig. 4: Impacts of gene drive releases in different vector species groups on the prevalence of P. falciparum malaria.Markers show the average annual prevalence of P. falciparum malaria in each year following gene drive releases, in either An. arabiensis (green), An. funestus (blue), both An. gambiae and An. coluzzii (red), or in all four vector species (dark green). Markers and error bars show the means and the 95% central quantiles, respectively, from 125 simulations. Pie charts show the proportion of each of the three vector species groups in each area. Columns (a–d) divide the sixteen areas into the quartiles of the average annual prevalence in the year prior to gene drive releases, where (a–d) show the first, second, third and fourth quartiles, respectively.In the majority of areas, the greatest reductions in prevalence are found when releases occur in the An. gambiae/An. coluzzii group (Fig. 4; Senegal/Guinea Bissau, Ghana, Nigeria (Lagos), Cameroon, Sierra Leone/Liberia, Guinea, Togo/Benin, Liberia, Côte d’Ivoire). These areas have a high relative abundance of An. gambiae and An. coluzzii (Fig. 4). Moreover, these two vector species have higher rates of blood feeding on humans compared to An. arabiensis, which takes more blood meals from non-human hosts (see the parameter Q0 in Table S4). Targeting these species is thus more effective in reducing transmission compared to targeting An. arabiensis. When releases occur in An. gambiae/coluzzii, the five areas with the greatest relative reductions in prevalence were 68% (95% central quantile: 37–86%) in Senegal/Guinea Bissau, 65% (CI: 51–82%) in Nigeria (Lagos), 64% (CI: 44–91%) in Ghana, 40% (CI: 29–51%) in Sierra Leone/Liberia, and 38% (CI: 28–49%) in Guinea. Larger relative reductions in prevalence occurred in areas where the malaria prevalence prior to the releases was relatively low, such as in Senegal/Guinea Bissau, Nigeria (Lagos) and Ghana because reductions in transmission have larger impacts on prevalence at lower initial prevalence levels (as discussed further below). Releases in An. arabiensis produced strong reductions in prevalence in Niger/Nigeria and northern Nigeria (Fig. 4a), which are northern areas where An. arabiensis is the predominant vector. In these two areas, pre-release malaria prevalence was low, and releasing in An. arabiensis was able to halt malaria transmission. Releasing in An. funestus was the best strategy in Western Mali, Southern Mali, Western Burkina Faso and Benin/Burkina Faso (Fig. 4).We next compare releases in a single vector species group with releases in all three vector species groups simultaneously. The latter strategy results in much greater reductions in prevalence that exceed the additive effects of releasing in each of the three vector species groups individually. After >5 years from the release date, prevalence remains below 5% in all areas, except for Liberia and Cameroon, where prevalence remains at 5–10% (Fig. 4; dark green triangles). Strong reductions in prevalence occur when releases are made in all vector species because the relationship between malaria transmission and prevalence is non-linear, with prevalence falling away rapidly once malaria transmission drops to low levels (Fig. S3).Impacts of gene drive releases in different species combinationsWe now compare a range of release strategies that target different combinations of vector species to investigate the reductions in the disease burden that could be achieved, as measured by the numbers of clinical malaria cases. Here, we also consider alternative scenarios where an additional non-target vector species is present, to represent the likely possibility that other vector species may contribute to transmission. We model the impacts of releases targeting An. gambiae and An. coluzzii only (Fig. 5, bars labelled GC), An. gambiae, An. coluzzii and An. arabiensis (Fig. 5, bars labelled GCA), and all three vector species groups (Fig. 5, bars labelled GCAF). In the case where releases occur in all three vector species groups, we consider additional alternative scenarios in which a non-target vector species is present in the area with a relative abundance of 10% (GCAF*10) and 20% (GCAF*20) of the combined abundance of the three vector species groups targeted by gene drive releases. We assume that this non-target species is identical to An. arabiensis in terms of all demographic and behavioural parameters (see Methods). We measure impacts by the cumulative number of clinical malaria cases averted in children aged 0–5 years over the 12 year period following releases. We also calculate the cumulative number of cases occurring over the period and show the proportional as well as absolute reductions in cumulative cases in each area achieved by each release strategy (Fig. 5).Fig. 5: Impacts of gene drive releases on malaria cases when implemented with and without vaccination and new types of ITNs.Releases in three different species combinations are modelled, denoted GC, GCA and GCAF, where for the third strategy the asterisks denote the presence of other non-target vector species at different relative abundances (see the main text). a Red bars show the cumulative number of cases occurring in children aged 0–5 years over the 12 year period following the implementation of gene drive interventions in each area. Hashed bars show the cumulative number of cases averted by gene drive releases compared to the counterfactual scenario where no gene drive release occurred. b For the eight areas with the highest pre-release malaria prevalence, red bars show the cumulative number of cases occurring when gene drive releases are combined with RTS,S vaccination and switching to pyrethroid-PBO ITNs. Orange patterned bars show the cumulative number of cases averted by applying both RTS,S vaccination and pyrethroid-PBO ITNs, and hashed bars show the additional number of cases averted when gene drive releases are implemented in combination with RTS,S vaccination and pyrethroid-PBO ITNs. Data are shown per 100,000 children.Gene drive releases in the An. gambiae/An. coluzzii species group (GC) reduced the average number of cumulative clinical cases across the sixteen areas by as much as 70% (19–100%) in Senegal/Guinea Bissau. In many areas, releasing gene drives in An. arabiensis, An. gambiae and An. coluzzii (GCA), had a modest additional benefit compared to releasing in An. gambiae and An. coluzzii only (GC) (Fig. 5a; Côte d’Ivoire, Liberia, Togo/Benin, Guinea, Cameroon, Western Burkina Faso, Nigeria (Lagos), Ghana). In these eight areas, the greatest additional benefit of releasing in An. arabiensis occurs in the Western Burkina Faso area where the average reduction in clinical cases increases to 37% (14–66%) from 19% (0.05–41%). In areas with lower pre-release transmission levels (Fig. 5a, top panel), there is often a much greater additional benefit to releasing in An. arabiensis. For example, in the Niger/Nigeria area, gene drive releases in all An. gambiae complex species (GCA) averted 37,728 (5630, 75,552) clinical cases per 100,000 children over the 12-year period, which is a 97% reduction in average cases compared to a scenario where no releases occurred. This release strategy also achieved large reductions in cases in northern Nigeria, Western and Southern Mali, Senegal/Guinea Bissau, and The Gambia/Senegal (Fig. 5a). This is due to the predominance of An. arabiensis in these areas and their relatively low pre-release malaria prevalence (Fig. S3).In most areas, releasing in all three vector species groups (GCAF) gives much greater benefit compared to releasing only in species from the An. gambiae complex (GCA), even when a non-target vector species is present (Fig. 5a). This strategy achieved large reductions in absolute numbers of clinical cases in the highest transmission settings, such as Côte d’Ivoire, Liberia, Togo/Benin and Benin/Burkina Faso (Fig. 5a, bottom panel). For example, in Côte d’Ivoire, 421,153 (344,678, 481,496) clinical cases per 100,000 children were averted over the 12-year period by releasing in all three vector species groups where a non-target vector species had a relative abundance of 20% (GCAF*20). This represents a 69% reduction in average cumulative cases compared to a scenario where no releases occurred. The corresponding reduction in average cumulative cases in Côte d’Ivoire achieved by this strategy when no non-target vector species were present (GCAF) was 88% (Fig. 5b; GCAF).Impacts of gene drives combined with new interventionsNew interventions applied concurrently with gene drive releases have the potential to reduce the malaria burden to lower levels, which will be especially critical when gene drives are not able to target all vector species. We estimate the reductions in clinical malaria cases that could be achieved by combining gene drives with two other more recent interventions, RTS,S vaccines and switching from pyrethroid-only to pyrethroid-PBO ITNs, where we assume that switching between ITN types does not change the ITN coverage across the human population in each area (see Methods). We show the impacts of combining gene drive releases with these other interventions for the top 50% of areas with the highest pre-release malaria prevalence (Fig. 5b), because the impacts of combining interventions are greatest in high transmission settings.In most of the high transmission areas, when the vaccination and pyrethroid-PBO net interventions are in place gene drive releases produced substantial added benefit in averting clinical cases (Fig. 5b). Gene drive releases in An. gambiae and An. coluzzii (GC) increase the average number of cases averted by at least 60% (in western Burkina Faso) to as much as 170% (in Liberia) across the eight areas, relative to when vaccines and pyrethroid-PBO net interventions are applied without gene drive releases (Fig. 5b, comparing orange patterned bars versus white hashed bars). Additional releases in An. arabiensis (GCA) increase the average number of cases averted by at least 116% (in western Burkina Faso) to as much as 226% (in Guinea) across areas. Releasing in all three vector species groups (GCAF) increases the average number of cases averted by at least 210% (in Togo/Benin) to as much as 397% (in Côte d’Ivoire), or by at least 180% (in Togo/Benin) to as much as 332% (in Guinea) when a non-target vector species has a relative abundance of 20% (GCAF*20). The number of cases averted by combining interventions is, however, less than the sum of the cases averted by each intervention in isolation. This is because an intervention averts more cases (in absolute terms) when case numbers are higher, therefore additional interventions lower the number of cases that are averted by each intervention.
