Drive-RIDL performance in a mosquito model

We initially used our discrete-generation model to compare Drive-RIDL to fsRIDL and SIT, but population elimination likely occurred too readily in this model compared to natural settings (see Fig. S5 and accompanying text). To assess our drives in a more realistic setting, we used a mosquito model with specific life stages and competition at the juvenile/larvae stage. We also adapted this model to have three different density response curves, the “concave” Beverton-Holt curve, a linear curve, and a “convex” curve (Fig. S4). In all cases, higher drive conversion efficiency allowed for far smaller release sizes, and indeed, the reduction in the release size was directly proportional to the drive conversion efficiency. For fsRIDL (on the figure with drive conversion of 0), the concave curve in the mosquito model needed a relatively low minimum release size (0.85) to achieve population elimination because competition (Fig. 2). For the linear density response curve, the drop ratio was far greater (3 for fsRIDL), with a similar increase for convex curve (5.2 for fsRIDL). A similar pattern was seen for SIT (Fig. S6). Population elimination in these other curves was far less effective because higher suppression power was needed to initially reduce the population and further increase the actual ratio of release males to native-born males.

Fig. 2

Effect of density dependence in the mosquito model. Males homozygous for Drive-RIDL were released into a population every week based the drop ratio, which specifies the relative number released each generation (3.17 weeks) compared to the male population at equilibrium. The density dependence of the model was varied (see Fig. S4), with a fixed low-density growth rate of 6. Gray indicates failure to eliminate the population after 100 generations (317 weeks). Each spot shows an average of 20 simulations

We also assessed the effect of varying the low-density growth rate parameter on our suppression drive constructs, using a linear density response curve (Fig. S7). We found that the necessary drop ratio was proportional to the low-density growth rate, as expected. Considering the major potential seasonal and ecological [24] variation in this parameter, it is thus a critical consideration for developing a successful release program.

Long-term persistence of Drive-RIDL

To demonstrate the self-limiting nature of Drive-RIDL, we tracked a short-term release that took place over 10 weeks (Fig. 3). Though drive efficiency went up rapidly (with a corresponding population reduction) at a high drop ratio of 3, regardless of drive conversion efficiency, this was halted shortly after the drive release. At this point, drive frequency increased only for a short period when drive conversion was possible, after which it rapidly declined, and the population was rapidly restored to equilibrium. One exception was the ideal drive with 100% drive conversion and no fitness costs, each of which would be difficult to achieve in practice. This drive is able to persist in the population without any decline in frequency while also keeping the population low. However, it cannot increase in frequency on its own, exception for stochastic fluctuations.

Fig. 3

Self-limiting nature of Drive-RIDL. Male mosquitoes homozygous for Drive-RIDL were released into a population every week between weeks 10 and 20 with a drop ratio of 3, which specifies the relative number released each generation compared to the male population at equilibrium. The low-density growth rate was 6, and the drive conversion efficiency was varied as shown. The graphs show the drive frequency among juvenile mosquitoes as well as the adult female population size. Population elimination occurred in half the simulations with 100% drive efficiency. Each line is an average of 20 simulations

We also examined confinement of the drive between two demes with unidirectional migration from the target deme to a nontarget deme. Even with high levels of migration, Drive-RIDL could not reach a high level in the nontarget deme before the target population was eliminated (Fig. S8). The population size in the nontarget deme was not substantially affected, declining modestly and quickly recovering.

Drive variants and fitness costs

While most gene drives do not have substantial unintended fitness costs aside from possible cargo genes or somatic targeting of essential genes in suppression drives, some fitness costs are occasionally observed. Using a linear density response and a low-density growth rate of 6, we thus investigated the effect of fitness costs of Drive-RIDL, fsRIDL, and SIT. We found that unless males had very low fitness, release sizes would not be substantially affected (Fig. 4). Note that this refers to genetic fitness costs affecting all drive carriers, not to any additional fitness costs of reared insects compared to wild-borne insects.

Fig. 4

Effect of fitness costs in the mosquito model on different suppression systems. Males of the indicated types (Drive-RIDL males for both the complete drive and split drive had a 50% drive conversion efficiency, and Drive-RIDL TARE males had a 100% germline cut rate) were released into a population every week based the drop ratio, which specifies the relative number released each generation (3.17 weeks) compared to the male population at equilibrium. The fitness of each construct was varied (the Cas9 allele for the split drive had no fitness costs), with a fixed low-density growth rate of 6 and a linear density response curve. Grey indicates failure to eliminate the population after 100 generations (317 weeks). Each spot shows an average of 20 simulations

Confinement of Drive-RIDL could potentially be lost if the fsRIDL element was mutated. Thus, we investigated a split drive, where Cas9 is placed on a separate locus that is not copied by the drive, forming an intrinsically self-limiting system even without fsRIDL [25]. This method suffered only a small efficiency loss compared to the complete Drive-RIDL system (Fig. 4).

Effect of resistance alleles on Drive-RIDL suppression power

All CRISPR-based gene drives are potentially vulnerable to resistance allele formation, and Drive-RIDL is no exception. Resistance alleles would allow females to remain viable. However, unlike homing suppression drive, where a single functional resistance allele (referring to alleles that preserve the function of the drive’s target gene) could prevent suppression, resistance alleles for Drive-RIDL would merely reduce its power. These could be “functional” if there is a target gene or could refer to any resistance allele if there is no target gene and all resistance alleles have the same fitness effects. If the entire population was composed of such alleles, then Drive-RIDL would act identically to fsRIDL, still retaining moderate suppressive power. To investigate the effect of resistance, we fixed the drive conversion efficiency to 50% and implemented a germline resistance allele formation rate, allowing some of the wild-type alleles to be converted to resistance alleles (Fig. 5). We found that when the germline resistance allele formation rate was lower than around 10−4, the necessary drop ratio to suppress the population was largely unchanged from a scenario without resistance alleles (which required a male drop ratio of 1.5 to achieve suppression). When resistance allele formation was maximum (0.5—so no remaining wild-type alleles after drive effects in male drive heterozygotes), Drive-RIDL was substantially impaired, and the required release size was 2.5. This was still a small improvement over fsRIDL (which required a drop ratio of 3 under similar conditions).

Fig. 5

Effect of resistance alleles in the mosquito model. Males homozygous for Drive-RIDL were released into a population every week based the drop ratio, which specifies the relative number released each generation compared to the male population at equilibrium. With a low-density growth rate of 6 and a drive conversion efficiency of 50%, the resistance allele formation rate was allowed to vary. The left panel assumes all resistance alleles are functional (note the logarithmic scale), while the center and right panel assume nonfunctional resistance alleles. Grey indicates failure to eliminate the population after 100 generations (317 weeks). Each spot shows an average of 20 simulations

Homing modification drives have been developed that target an essential gene and provide rescue, thus allowing nonfunctional resistance alleles to be removed [15, 16]. The use of multiplexed gRNAs and conserved target sites could allow nearly all resistance alleles to be nonfunctional [26]. Homing suppression drives lacking a rescue element, such as those targeting female fertility genes, could have similar advantages. However, targeting such a gene with Drive-RIDL would complicate rearing and prevent the release of all homozygous males unless efficiency was 100%.

We thus modeled two rescue strategies where all resistance alleles are nonfunctional, with Drive-RIDL systems that target haplolethal genes or haplosufficient but essential genes (Fig. 5). We found that nonfunctional resistance was actually substantially beneficial to the haplolethal targeting drive. This is because any individuals inheriting a resistance allele are immediately nonviable. Such alleles thus contribute to population suppression, even if not as much as drive alleles. The Drive-RIDL allele with a haplosufficient target saw a similar effect, but greatly reduced in magnitude because only resistance allele homozygotes are nonviable. Indeed, a small resistance allele formation rate actually very marginally hampered this drive because such resistance alleles can prevent drive conversion in male drive carriers. These results suggest that Drive-RIDL using Toxin-Antidote Recessive Embryo (TARE) as the driving element in lieu of a homing drive could also improve suppressive power. Modeling indicated a substantial improvement compared to fsRIDL if the total germline cut rate was 100%, though not quite as much as a homing drive (Fig. 4).

Demonstration of Drive-RIDL in flies

To demonstrate the concept of Drive-RIDL in a model organism, we modified our haplolethal homing rescue drive to contain a tTAV element with a tetO/hsp70 promoter (Fig. 6A). This allows the tTAV (which binds to tetO to increase expression) to have runaway lethal expression in the absence of tetracycline, but only in females due to the presence of a female-specific intron.

Fig. 6

Schematic of Drive-RIDL and homing drive performance. A The Drive-RIDL allele is transformed into a previous split homing rescue drive, rendering the DsRed marker inactive. The Drive-RIDL allele has an EGFP gene expressed in the eyes, and a tTAV gene that can activate itself in the absence of tetracycline. The drive retains a recoded rescue copy of its haplolethal target and two gRNAs. B The drive showed high inheritance from male and female heterozygote parents (with one paternal copy of Cas9) in the presence of tetracycline. Each dot represents progeny from a single drive individual. Black bars represent the average and standard error

To assess drive performance, the Drive-RIDL line expressing EGFP was crossed to a nanos-Cas9 line (Cas9 with the nanos promoter, 5′ UTR, and 3′ UTR) expressing DsRed in the presence of tetracycline to generate drive offspring with both DsRed and EGFP fluorescence. These drive offspring were further crossed to a w1118 line, and their progeny were screened for fluorescence. The percentage of EGFP flies represented the drive inheritance rate, which was 80% for female heterozygotes and 83% for male heterozygotes (Fig. 6B, Data Set S1). Though moderately high, this was somewhat lower than the 90% of the haplolethal homing drive without tTAV [15], perhaps due to the larger size of the Drive-RIDL construct.

To assess the ability of the Drive-RIDL allele to induce female nonviability, several crosses were conducted, all in a Cas9 homozygous background. First, control crosses with only Cas9 revealed a high egg-to-adult viability of 93% with tetracycline and 85% without tetracycline (Fig. S9, Data Set S2). Tetracycline is not likely to have a large effect on viability, so this difference was perhaps due to batch effects of the flies or from the food. To assess Drive-RIDL heterozygote viability, most representative of field conditions, Drive-RIDL homozygous males were crossed to Cas9 females. All offspring were thus heterozygous. Viability was 95% with tetracycline and 56% without tetracycline (Fig. 7A–B, Data Set S2). Without tetracycline, 76% of progeny were males, indicating that females had 31% relative viability. This high viability may be due to the genomic position of the construct, or it could be because the tTAV gene relies on the P10 terminator that is oriented for the EGFP gene, possibly leading to reduced tTAV mRNA stability. We also performed a cross of Drive-RIDL heterozygous males and Cas9 females in the absence of tetracycline (Fig. 7C, Data Set S2). Here, the drive inheritance remained at 83% and the female drive carrier relative viability at 33%, indicating that drive and female lethality could occur together. Unusually, non-drive progeny had a high female bias (Data Set S2), though the reasons for this were unclear. Finally, we assessed drive homozygotes as would be found a rearing facility. Viability was lower, even in the tetracycline vials, likely due to poor food batches (Fig. 7D, Data Set S2). However, in the absence of tetracycline, no females were viable, indicating an efficient construct (Fig. 7E, Data Set S2).

Fig. 7

Female lethal effect of the Drive-RIDL allele. Egg viability was recorded in vials in various crosses, all of which were with a Cas9 homozygous background. Adult progeny were also phenotyped for sex. Crosses were between drive homozygotes males and wild-type females A with and B without tetracycline, C between drive heterozygous males and wild-type type females without tetracyclines, and between drive homozygous males and females D with and E without tetracycline. Each dot represents progeny from a single drive individual. The green dot represents the mean for all individuals, and black bars represent the average and standard error

Because some female heterozygotes survived, we were curious to check their relative fitness compared to non-drive females. However, these females appeared fertile and did not have significantly different fecundity than Cas9 females (Data Set S3). It is possible that they may suffer other fitness costs such as reduced longevity, but these were not assessed. For Drive-RIDL, the high female viability would be close, but not quite enough for the drive to persist in a population (as opposed to being self-limiting due to female-lethal effects), based on drive performance and estimated embryo resistance allele formation rate in females [15], but only if the drive was a “complete” drive rather than a split drive, which would always be self-limiting.

Because incomplete lethality could potentially affect release candidates, we modeled this with a low-density growth rate of 6, linear density response, 50% drive conversion efficiency, and 50% resistance allele formation rate. Drive-RIDL homozygous females still had zero viability. For both haplolethal and haplosufficient targets, high female survival actually resulted in more efficient population elimination because the drive gained some character of a self-sustaining homing suppression drive (Fig. S10). This could potentially allow Drive-RIDL strategies to succeed even in species where a nearly perfect RIDL element could not be constructed. However, for the haplosufficient target, female survival over 30% prevented successful population suppression with any release size because the resistance alleles inhibited drive conversion, essential to produce nonviable female homozygotes. With lower resistance allele formation rates, the haplolethal target drive would also suffer from similar effects if drive conversion remained low. Depending on the drive conversion rate and fitness costs, sufficiently high female heterozygote viability could also prevent the Drive-RIDL system from being self-limiting.