Modeling impact and cost‐effectiveness of driving‐Y gene drives for malaria elimination in the Democratic Republic of the Congo

Abstract Malaria elimination will be challenging in countries that currently continue to bear high malaria burden. Sex‐ratio‐distorting gene drives, such as driving‐Y, could play a role in an integrated elimination strategy if they can effectively suppress vector populations. Using a spatially explicit, agent‐based model of malaria transmission in eight provinces spanning the range of transmission intensities across the Democratic Republic of the Congo, we predict the impact and cost‐effectiveness of integrating driving‐Y gene drive mosquitoes in malaria elimination strategies that include existing interventions such as insecticide‐treated nets and case management of symptomatic malaria. Gene drive mosquitoes could eliminate malaria and were the most cost‐effective intervention overall if the drive component was highly effective with at least 95% X‐shredder efficiency at relatively low fertility cost, and associated cost of deployment below 7.17 $int per person per year. Suppression gene drive could be a cost‐effective supplemental intervention for malaria elimination, but tight constraints on drive effectiveness and cost ceilings may limit its feasibility.

malaria cases occur in sub-Saharan Africa (SSA), accounting for 93% of total malaria cases worldwide (WHO, 2019a(WHO, , 2020c. With 12% of all cases in SSA, the Democratic Republic of the Congo (DRC) is the second highest-burden country on the continent (WHO, 2019a).
Nearly all of the DRC's population lives in high malaria transmission zones (Vector Link, 2019). Consequently, the disease remains one of the country's most serious public health problems and is the number one cause of death (IHME, 2018;Ngatu et al., 2019).
Despite sustained malaria control, malaria incidence in the DRC has increased in the last few years (WHO, 2019a), and more than 40% of children who fell ill because of malaria did not receive adequate care (Unitaid, 2019;WHO, 2019a). Health system weaknesses and gaps in the coverage of core interventions caused by financial and programmatic limitations are likely responsible for this recent rise in cases (WHO Malaria Policy Advisory Committee, 2018), and elimination remains elusive. Sustained access to vector control has been a central strategy in the DRC's complex operating environment, where challenges are compounded by domestic political conflicts (Ngatu et al., 2019) and insufficient funding for malaria control (Head et al., 2017). These challenges emphasize the urgent necessity of de- Research in Non-Human Organisms, 2016). Gene drive is a novel method that involves the inheritance of specific traits from one generation to the next at rates higher than the 50% chance afforded through Mendelian inheritance in heterozygotes, and gives certain genes a substantially higher or lower probability of inheritance and thereby alters the frequency of such genes in the population. A gene that alters the fertility or survival of the target species could thereby alter the species population size, depending on the species and the drive system applied (Beaghton et al., 2017;Buchman et al., 2018;Burt & Deredec, 2018;Gantz et al., 2015;Hammond & Galizi, 2017;Marshall et al., 2011;North et al., 2019North et al., , 2020. Given rising resistance to existing insecticides and antimalarial drugs (Bhagavathula et al., 2016;Bull et al., 2019;Mnzava et al., 2015;Protopopoff et al., 2018;WHO, 2014), gene drive mosquitoes might hold great potential to accelerate and achieve lasting gains in malaria control (Committee on Gene Drive Research in Non-Human Organisms, 2016). The future utility of gene drives also depends on their economic aspects compared with existing or future alternatives (WHO/ TDR & FNIH, 2014). This study assesses the cost-effectiveness of gene drives together with conventional interventions by estimating Disability-Adjusted Life Years (DALYs), DALYs averted, and the costeffectiveness of vector control methods in the DRC.
Although gene drive has yet to pass the research and development stage, with driving-Y gene drive yet to be developed in the laboratory and lead candidates of gene drives only tested in confined cage trials (ENSSER, 2019;Simoni et al., 2020), public concern has been voiced over gene-related technologies that intend to alter the targeted species population, including previous techniques such as Wolbachia-based and sterile insect techniques. For example, concerns on previously developed genetic controls, such as a genetically modified version of Aedes aegypti for control of mosquitotransmitted arboviral diseases, have led to a debate on potential hazards including the unexpected contamination of transgenes in the environment, possible harms to the targeted species' morphology, consequences of transgenes to gene flows (Paes de Andrade et al., 2016), and whether releasing modified mosquitoes to control vector-borne diseases is suitable for a large-scale implementation (Flores & O'Neill, 2018). These questions remain for gene drives. At the same time, proof of efficacy presents a challenge, and informed decision-making on gene drive releases into the wild will require a step-wise approach for safety monitoring, additional information about potential effectiveness, and the evaluation of potential environmental risk and benefits to health in terms of disease control (Committee on Gene Drive Research in Non-Human Organisms, 2016;.
Disease modeling is a powerful tool that can complement laboratory findings and help develop control strategies involving transgenic mosquitoes. The scientific community, including the WHO and other policy groups, has increasingly recognized the importance of disease modeling in guiding the development of gene drives and genetically modified organisms (Committee on Gene Drive Research in Non-Human Organisms, 2016;James et al., 2018;.
In this work, we explore the possible outcomes of applying gene drives as an intervention for malaria control in SSA settings in combination with established control programs-including ITNs and ACT distributions-while also evaluating the economic cost of the resulting programs.
We model areas in eight provinces of the DRC by calibrating the transmission intensity of the selected areas to malaria prevalence estimates from open data sources, accounting for existing intervention coverage, and using local rainfall and temperature to drive seasonality in vector abundance. In each selected province, we determine effective release strategies of gene drive mosquitoes and define parameter regimes of a sex-ratio-distorting suppressive gene drive system, the driving-Y system, that results in the elimination of malaria. In the driving-Y system, the process of shredding the male's X chromosome results in male-biased progeny as the Y chromosome can still be carried through unaffected sperm and driven into the next generation (Hammond & Galizi, 2017). The system leads to fecundity reduction, the reduction of the potential to produce offspring, which affects the egg batch size and has implications for the success of the driving system (Bradshaw & McMahon, 2008;Moro et al., 2018). We simulate various intervention scenarios, including both conventional and gene drive approaches to vector control, identify combinations of interventions that lead to malaria elimination, and use modeled predictions of malaria burden to estimate DALYs averted and compare the cost-effectiveness of driving-Y gene drives and existing vector control interventions in the DRC.

| MATERIAL S AND ME THODS
The simulations in this study use Epidemiological MODeling software (EMOD) v2.18 (IDM, 2019), an agent-based, discrete-time, Monte Carlo simulator of malaria transmission with a vector life cycle (Eckhoff, 2011) and within-host parasite and immune dynamics (Eckhoff, 2012(Eckhoff, , 2013. The modeling framework combines an epidemiological model of Plasmodium falciparum transmission between individual human agents and cohorts of mosquito agents distinguished by life stage, feeding and oviposition stage, age, and genotype. The vector lifecycle consists of four stages: egg, larvae, immature adults, and host-seeking adults, with temperaturedependent larval development, immature maturation, and sporogony. Mosquito abundance is driven by the availability of larval habitat, and mosquito mortality is also affected by temperature and humidity. In humans, the model includes asexual parasite and gametocyte densities, human immunity, effects of antimalarial drugs, and symptomatic aspects of malaria, all of which have been previously calibrated to field data Gerardin, Ouedraogo, et al., 2015;Selvaraj et al., 2018). The model dynamically simulates vector-human and human-vector transmissions during blood meals. Driving-Y is one of several gene drive strategies that can be simulated within EMOD (Selvaraj et al., 2020).
We selected eight provinces in the DRC for simulations ( Figure 1) in both nonspatial and spatial simulation frameworks. The selection was based on malaria parasite prevalence data from the DRC- populated. We applied site-specific environmental covariates based on node geolocation including climate (rainfall, temperature, humidity) and seasonality averaged from monthly vectorial capacity.
Since An. gambiae mosquitoes, the only modeled mosquito species in this study, breed primarily in temporary puddles replenished by rainfall and drained through evaporation and infiltration (Koenraadt et al., 2004), the simulations used climate data to model the availability of larval habitat, which drove the number of vectors throughout the year and thus biting intensity and transmission (Eckhoff, 2011). Weather stations and readings by the National Oceanic and Atmospheric Administration (NOAA) Global Surface Summary of the Day were used for generating temperature and dewpoint anomalies.
Baseline monthly averages were generated using WorldClim 1.4 raster files in a grid format, 2.5 arc minutes, and 30 arc seconds from WorldClim 1.4 (Hijmans et al., 2005). Rainfall files were generated by downscaling RFE 2.0 Rainfall Estimates from NOAA's Climate Prediction Center (NOAA, 2006). In both nonspatial and spatial simulation frameworks, seasonality was enforced in the models.
To calibrate seasonality of larval habitat abundance, we simulated  (Bhatt et al., 2015). The annual means of estimated parasite rate in children between the ages of two and ten (PfPR 2-10 ) from the year 2000 to 2015 were retrieved from MAP rasters (Bhatt et al., 2015) for all simulation nodes. For Haut Katanga, where parasite prevalence was lower, the larval habitat multiplier was calibrated so that the average modeled parasite prevalence for years [2013][2014][2015] was close to the MAP estimates for the same period. We set each node's population to 1000 individuals and set birth and mortality rates to 36.3 per 1000 people per year. The human population size used is large enough to sustain low-transmission malaria but not unrealistically large for rural areas. The simulation was run for 50 years to initialize population immunity.
In the final 10 years of the 50-year initialization period, the fol- Outcomes were evaluated at 5, 10, and 15 years.
For ITNs used in the model, the initial strength of the blocking effect on indoor mosquito feeding on an individual with an ITN was 0.9, and the blocking decayed at an exponential rate with a mean of 730 days. The blocking effect captures the physical barrier of a bednet that prevents a mosquito from making contact with a human.
The initial strength of the killing effect was 0.6 and decayed at an exponential rate with a mean of 1460 days. Killing and blocking parameters were obtained from calibration to clinical trial data (Eckhoff, 2013). The model assumed an individual who received an ITN had a 0.65 probability of using it on any given night, and ITNs were redistributed every 3 years. For ACT, the parameters and values used in the model followed .
The model focuses on final mosquito offspring under the gene drive intervention, and females that mate with a male carrying driving-Y will have as offspring wild-type females and males carrying the driving-Y. The fraction of offspring that are driving-Y males is then 0.5+0.5*(X-shredder efficiency), and the fraction of offspring that are females is 0.5-0.5*(X-shredder efficiency). Only females that mate with a driving-Y male have their fertility reduced, and the total egg batch size is reduced by the fecundity reduction for each female that mates with a modified male (Eckhoff et al., 2017).
We selected a gene drive release size and schedule by simulating highly efficient drives in the nonspatial framework (  per node in the nonspatial framework, we varied the X-shredder efficiency from 0.5 to 1.0 and fecundity reduction from 0 to 0.5, simulated 10 stochastic realizations per X-shredder efficiency and fecundity reduction parameter combination, and evaluated whether malaria was eliminated. A simulation was defined as reaching malaria elimination when all-age parasite prevalence in the model is not detectable, that is, dropped to zero and remained zero until the end of the 15-year simulation timeframe. We selected parameter sets of X-shredder efficiencies (0.9, 0.95, 1.0) and fecundity reductions (0.05, 0.1, 0.15) to generate 9 combinations of driving-Y parameters that could eliminate malaria in the nonspatial framework (Supporting information 2).
The driving-Y gene drive release strategy and parameters were then applied to the spatial simulation framework, where we iden- In the spatial simulation framework, vectors could move between adjacent nodes less than 10 km apart with a rate inversely proportional to their distance. Adjacent nodes included diagonally adjacent nodes. The vector migration rate ranged from 0.09 to 0.12 (9%-12% of female mosquitoes migrate out of a grid cell on any day).
We did not include human migration in the simulations and found no difference in parasite reduction outcomes upon including human migration ( Figure S6.2), likely because the simulated vector migration rates are high enough to stably suppress the vector population.

| RE SULTS
This study uses mathematical modeling to explore the potential role powerful X-shredder efficiency at little to no cost of fertility (Galizi et al., 2014;Simoni et al., 2020). Across settings, gene drive was most successful at reducing malaria prevalence when the X-shredder efficiency ranged from 0.95 to 1.0, and fecundity reduction ranged from 0 to 0.15. A similar sensitivity of mosquito population suppression to X-shredder efficiency could be observed in previous studies using simpler models that studied mosquito population dynamics in a homogeneous and constant environment (Deredec et al., 2008(Deredec et al., , 2011) and an extended model that applied regional heterogeneity to model malaria mosquitoes at a national scale (North et al., 2019).
In our study, the release of drives reduced parasite prevalence, adult vectors, and wild-type vector fraction with a similar trend across all sites under a wide range of transmission intensity and seasonality   (Table 3).
The expansion paths of all sites show the order in which interventions would be selected at different levels of resources available based on the ICER (Figure 4, Table 4). The ICER indicates additional costs required to avert each additional DALY by moving from the lower-cost to the higher-cost intervention (Tan-Torres Edejer et al., 2003). It is calculated using average yearly costs and yearly F I G U R E 3 Dependence of final parasite prevalence on X-shredder efficiency and fecundity reduction for the eight modeled sites in the nonspatial framework. Simulation outputs are means of 10 stochastic realizations per parameter combination and are observed 15 years after the release of 300 gene drive mosquitoes. See Supporting information 3 for upper and lower 95% predicted interval bounds and for outcomes with 100 or 200 mosquitoes released effectiveness (Table 3) Once the mosquito population collapses, gene drives become more cost-effective over a medium timeframe.

| DISCUSS ION
This study uses mathematical modeling to describe the potential role of sex-ratio-distorting gene drive mosquitoes in malaria control across the transmission spectrum in the DRC, an area where achieving effective control has historically been challenging. Our results suggest that population suppression through gene drives could be an effective strategy for malaria elimination in the DRC, either as a single intervention or in combination with other interventions. To the best of our knowledge, this is the first study that models the epidemiological impact and cost-effectiveness of gene drive mosquitoes for malaria elimination. Previous studies involving gene drives for malaria control are limited in scope to laboratory experiments (Akbari et al., 2014;Curtis, 1968;Galizi et al., 2016;Pike et al., 2017), and the development and parameterization of mathematical models Heffel & Finnigan, 2019;Noble et al., 2019;North et al., 2019). By extending previous modeling work (Eckhoff et al., 2017) to approximately estimate the costeffectiveness of gene drive in SSA settings, our work helps fill a gap in evidence about the programmatic implementation of gene drives TA B L E 2 Minimum intervention or combination that can achieve malaria elimination in each target location within 15 years after adding driving-Y mosquitoes into the simulated scenarios

Province The minimal intervention(s) that could achieve malaria elimination
Nord Kivu Elimination is possible with interventions at pre-existing levels.   Abbreviations: ACT, case management rate with artemisinin-based combination treatment (Artemether + Lumefantrine); ITNs, insecticide-treated nets; NA, not applicable, gene drives were not applied in the scenarios because malaria elimination was achieved with the indicated intervention combination without gene drives.

TA B L E 3
Average yearly cost per one million population and mean parasite prevalence reduction for interventions and combinations of interventions applied in the study for scenarios that could result in malaria elimination. 95% confidence intervals are presented in (lower, upper). (See Table S7 and Supporting information 7 for complete estimates of all scenarios.) in the context of limited resources. The study not only estimates the feasibility of gene drives in realistic malaria elimination scenarios but also evaluates the cost of gene drives in comparison to other currently available interventions. This work helps gauge the probabilities of success and possible outcomes of gene drives that are strictly laboratory-contained or in the transition from the laboratory-based research to future field-based research. Introducing modified organisms into the environment can be invasive, and preventive measures should be in place to provide timely mitigation in case of spillovers and countermeasures to halt an ongoing gene drive when necessary.

WHO's Estimates from model's outputs
Multiple safeguards will be needed in parallel (WHO, 2020a). In addition to the technical perspective provided in this study, further work is necessary, including on the ethical perspective, that is, standard research ethics, procedural ethics, and participatory management of the technology (Thompson, 2018), as a key component to implement this technology in wild mosquito populations (Wedell et al., 2019). We found that the success of driving-Y gene drives in all areas regardless of vector density highly depends on the ability of gene    (Galizi et al., 2014;Simoni et al., 2020), driving-Y gene drives have yet to be developed in the laboratory. Fecundity does not appear to be a major detriment to gene drives that do not directly target female fecundity (Kyrou et al., 2018). The adoption of a driving-Y strategy could be very challenging because it may be difficult to achieve a perfect X-shredder efficiency at every development stage and during implementation while overcoming the challenge of meiotic sex chromosome inactivation (Thompson, 2018). Moreover, possible resistant mutants could convert wild-type genes and spread resistance (Beaghton et al., 2017;Bull et al., 2019;Champer et al., 2021), and cleavage resistant alleles have already been observed in An. gambiae (Galizi et al., 2014).
Some sex-ratio-distorting drives may not be comparable to Y drive or X shredders, for example, if the female carries a drive that inactivates the reproduction of her male progeny. However, the difference in mechanisms by which sex ratio distortion is achieved may only lead to differences in how quickly a drive establishes itself rather than downstream outcomes regarding elimination, which is the focus of this study. The models predict that reaching malaria elimination does not always require bringing the number of mosquitoes down to zero and that there is continued mosquito biting after releasing gene drive mosquitoes. Our finding that the drive must be highly effective while mild fecundity costs are well-tolerated is likely generalizable to other suppression drives as a whole, although further explorations in settings of stronger seasonality are needed. Our economic findings on the cost-effectiveness of drives are likely to be order-of-magnitude similar for any highly effective suppression drive.
The success of suppressive gene drives such as driving-Y depends on mosquito population size and allowing enough time for the drives to propagate in the mosquito population. Understanding interactions between existing vector control methods such as ITNs and IRS that temporarily reduce the mosquito population (Alphey et al., 2010;WHO, 2015a) and gene drives will be necessary given that vector control typically reduces mosquito populations. While this work focuses on the impact of vector abundance, seasonality, and conventional vector control on gene drive outcomes, spatial connectivity of mosquito populations and terrain heterogeneity will also have important implications for gene drives success. Accurate capture of local variation in mosquito population connectivity requires data on mosquito swarms and habitats at high resolution, mosquito movement patterns, and mosquito species introgression.
Some of these quantities are measurable and known but most are unavailable for the DRC. A sensitivity analysis with vector migration rate reduced down to 3 orders of magnitude did not observe substantial change in elimination outcomes (Supporting information 6, Figure S6.1), although reduced migration led to later elimination.
Our models predicted that high coverage with ITNs and high access to treatment with ACTs could eliminate malaria in lower-transmission settings, but achieving such high coverages of existing measures is not only extremely difficult but also comes with high implementation and logistical costs (Shretta et al., 2017;Zelman et al., 2014). It may take much more investment in logistics and systems to achieve 95% coverage of both ITNs and ACT than WHO's estimates applied in the study (Table 1) . Even if theoretically achievable, it is highly improbable to sustain necessary coverage levels in the complex operational environment of high disease burden countries like the DRC (Carrel et al., 2015;WHO, 2005 (Sturrock et al., 2015). Future work that includes importation of vectors and infections is necessary to address the feasibility of sustained elimination, and to specify release schedules that are operationally practical, technically necessary for intended deployment areas, and appropriate for the local seasonality.
Because of their self-propagating and self-sustaining properties (Hammond & Galizi, 2017), gene drives would likely result in better cost-effectiveness once implemented compared to other genetically engineered mosquitoes previously developed (e.g., sterile insect techniques). Nonetheless, the payoffs are only observed once malaria elimination is reached-in most cases, after 5 years postrelease in the settings considered in this study. This waiting period can be critical, given many life losses in the interim in the DRC's context.
Our results highlight the importance of efficient gene drives over simply increasing the number of gene drive mosquitoes released.
This aligns with the result of a recent study using the simplest model of a population with one life stage and density-dependent mortality that the diffusion rate of Y drive males depends on the strength of drive (Beaghton et al., 2016).
We based our cost-effectiveness analysis on the unit costs of OX513A (Alphey et al., 2018) and Wolbachia-infected mosquitoes per person (Meghani & Boëte, 2018), since cost data of genetic control methods are limited. We performed a systematic scoping review As we demonstrated in the ICER analysis, the cost-effectiveness is cost-sensitive. The gene drive approach in malaria elimination is also effectiveness-sensitive and becomes less cost-effective compared to other strategies once its cost increases or effectiveness decreases or both.
Gene drive technology is at an early stage of development and concerns over ethics, safety, and governance, as well as questions on affordability and cost-effectiveness, must be addressed before implementation (WHO, 2020a). For high malaria burden countries such as the DRC, collaborating on testing, implementing, and regulating new technologies like gene drives poses challenges not only from within the country but also with other countries where different systems of governance can further complicate the collaborations (Dambach et al., 2014). Presently, many countries including the DRC have insufficient resources to individually follow recommendations such as extensive risk assessment and safety testing, and close monitoring after mosquito releases (WHO, 2020a), making it a challenge to enforce legislation required under the Cartagena Protocol (Kingiri & Hall, 2012 (Guerra et al., 2014) and no importation of infections and wild-type mosquitoes in the models, the study offers an evaluation framework. The framework can be generalized to look at other gene drive approaches to effectively plan gene drive strategies in malaria control, especially in other high burden countries where parasite transmission intensity varies.

ACK N OWLED G M ENTS
The authors thank the Institute for Disease Modeling and, in particular, Benoit Raybaud and Edward Wenger for their generous technical support and resource sharing. We are immensely grateful to Philip Welkhoff and David O'Brochta for helpful discussions and comments on earlier versions of the manuscript.

CO N FLI C T O F I NTE R E S T
The authors state that there are no competing interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available in GitHub repositories as follows.