Background
Any attempt to control a harmful species, whether it be a disease-causing microbe or a crop-competing weed, must consider the potential for resistance to evolve—for example, to antibiotics or herbicides. The evolution of resistance has been a recurring theme over decades of malaria control efforts, both by the
Plasmodium parasite to new drugs [
1,
2] and by the
Anopheles mosquito vector to new insecticides [
3]. Any new intervention to control or help eliminate malaria must, therefore, be assessed in terms of how sustainable impacts are likely to be in the face of potential selection for resistance.
Recent progress in molecular and population biology has raised the possibility of new interventions for vector control using genetic approaches to disrupt the survival or reproduction of the mosquitoes, or to render them unable to transmit the parasite [
4‐
7]. One possible approach is to make a synthetic driving Y chromosome by inserting, onto the Y, a gene encoding a nuclease that recognizes and cleaves a repeated sequence found only on the X chromosome, and have the appropriate control sequences such that the gene is only expressed during spermatogenesis [
8‐
10]. The idea is that expression of the gene at this time will disrupt transmission of the X chromosome, leading to a preponderance of Y-bearing sperm and male offspring, which will themselves carry the nuclease gene. As long as the nuclease gene does not affect male fitness too strongly, the modified Y chromosome is expected to increase in frequency within a population, eventually replacing the wild-type Y. As it does so, the population sex ratio will become increasingly male-biased, which will have a direct impact in reducing disease transmission (because only females bite people and transmit disease). Since females are also likely to be most responsible for the productivity of the population, a male-biased sex ratio may also lead to a reduction in the total number of mosquitoes, further reducing transmission, and if the Y drive is sufficiently strong, then spread could lead to elimination of the population [
9,
10]. Recently, there have been promising proof-of-principle demonstrations in
Anopheles gambiae that cleavage of the X chromosome during spermatogenesis can lead to male-biased sex ratios with little or no effect on male fertility, both using engineered meganucleases [
11] and with a CRISPR-based nuclease [
12].
The spread of a driving Y may be expected to select for resistant genotypes. One obvious form of resistance would be changes in the target sequence such that it is no longer cleaved by the nuclease, as has been modelled and observed in the context of homing-based gene drive constructs [
9,
10,
13‐
19]. In addition, since the spread of a driving Y will produce a male-biased sex ratio, there can be selection for autosomal suppressors that restore a 50:50 sex ratio [
20,
21]. In the context of investigating a potential population modification (as opposed to population suppression) strategy for
Aedes mosquitoes, Huang et al. [
22] show using deterministic models that release of a driving Y can result in the spread of X-linked and autosomal resistance.
Though resistance to a driving Y may evolve, it is not inevitable. For example, resistant genotypes may arise sufficiently rarely that the population is eliminated by the driving Y before resistance evolves. Or resistance may have sufficiently large pleiotropic fitness effects that prevent it from spreading. To investigate further the likelihood that either target site resistance or a trans-acting suppressor will evolve to a driving Y, a population genetic and population dynamic model is developed. Stochastic effects are incorporated by extending the time-inhomogeneous branching process method of Uecker and Hermisson [
23], which has also been used recently to analyse the evolution of resistance to homing-based gene drive elements that spread without causing population suppression [
16]. Results are checked by fully stochastic Gillespie simulations [
24]. The models identify a number of factors affecting the probability resistance evolves and rescues the population, including the mutation rate, the intrinsic rate of increase of the population, the strength of drive and the pleiotropic fitness costs of the resistant allele. In seasonally varying environments, the probability of resistance evolving is affected by the time of release of the driving Y males. Trans-acting suppressor alleles are more likely to suffer stochastic loss at low frequencies than target site resistant alleles.
Discussion
This paper has considered driving Y chromosomes that are capable of eliminating a closed, random-mating population, and modelled the factors affecting the probability that resistance evolves before that happens. This topic of ‘evolutionary rescue’ has previously been studied in the context of populations threatened by a change in the external environment [
23,
37‐
41]; the key differences here are that the risk to the population is a driving Y, with its own particular dynamics, and whose mechanism of action can itself give rise to one form of resistance. The models have identified several factors affecting the probability resistance evolves. Some are properties of the target population, in particular the size and the intrinsic rate of increase; all else being equal, higher values of both these parameters make resistance more likely. Other parameters are properties of the driving Y, and highlight two primary routes to minimizing the risks of resistance [
9]. The first strategy is to target essential sites such that resistant alleles are likely to have pleiotropic fitness costs (previously analysed in the context of homing-based gene drive construct [
9,
10,
13,
16]). This factor makes the rDNA repeat a more attractive target than some less important or nonfunctional repeat on the X chromosome; targeting functional sites within the rDNA repeat may also be better than targeting nonfunctional sites in the same repeat, though a fuller description of the mutations caused by rDNA-targeting nucleases would be helpful here. The second strategy is to ensure the mutation rate to resistance is low. In the context of a nuclease-based driving Y, targeting a sequence present in hundreds of copies will be better in this regard than targeting a single copy sequence (which may not lead to preferential inheritance of the Y in any case). Presumably
\(u\) could be further lowered by targeting a second sequence within the rDNA repeat. With no obvious limit to the number of sites that can be targeted in the rDNA, management of target site resistance would seem to be achievable.
The modelling has also revealed two other factors that can reduce the probability of resistance evolving. Increasing the transmission rate of the driving Y (
m) reduces the opportunity for resistant mutations to arise. Note, though, in spatial models without resistance, suppression may be maximised at an intermediate optimum
m [
13,
42]. Releasing the construct at the right time of year can also reduce the probability of resistance evolving. In this model, this occurred just before the peak of abundance and also just as it was entering the trough, but other models of population fluctuation and other release rates ought to be examined. Note too that if releases are made in one location with the intention that the construct spread to other locations, one will have little control over when those migration events occur.
The evolution of trans-acting suppressors was also briefly considered. It is difficult to predict the most likely molecular mechanisms for such suppressors, and in the absence of a clear expectation, a simple model is considered in which mutations do not pre-exist in population before release, but can arise in any individual, not just in progeny of driving Y males. Pre-existing mutations could be included in the analysis using the approach of Hermisson and Pennings [
43]; see also [
16]. It is also possible to imagine ways that a suppressor mutation might arise from the nuclease gene, which would only be possible in the progeny of driving Y males—for example, a duplicated or retro-transposed copy of the nuclease gene that interferes with the original gene at the RNA level (e.g. RNAi), or at the protein level (e.g., competitive binding of a non-active enzyme).
Several of the modelling assumptions used here are worth highlighting. First, it is assumed that resistance is all-or-nothing. This is a reasonable first step, but partial target site resistance may occur if the target site is present in hundreds of copies, as in the rDNA repeat. Indeed, the action of the nuclease would likely lead to many different alleles being produced, with different degrees of resistance and different pleiotropic fitness effects. The overall probability of resistance evolving would then be a sum across possible alleles of P
1 calculated for each allele, where P
1 for the ith allele would depend on the mutation rate u
i
and fitness effects w
i
for that allele. Ideally, for all possible mutations, either the fitness, the mutation rate, or the degree of resistance provided is sufficiently low as to have a low probability of rescuing the population. Unfortunately, there is likely to be a limit to how thoroughly one will be able to test for resistance and suppressors in the lab, before release in the field. If high fitness resistance is seen in the lab, it will likely arise in the field, but the failure to see it in the lab will not guarantee it will not arise in the field.
It is also assumed that the driving Y has no fitness effects except the sex ratio distortion, so its frequency increases monotonically in the deterministic model as long as there are any susceptible genotypes remaining in the population (i.e., there is no complex dynamics, such as cycling [
22,
44]). In the context of evolution of resistance, the main impact of a driving Y that has a cost on survival or mating success would likely be to slow down the spread of the driving Y and therefore slightly increase the probability of resistance evolving. A further assumption is that the daughters of driving Y males, which could harbour resistant mutations, have normal fitness. Galizi et al. [
11] found that the ~5% of females produced had low fitness, presumably due to disruption of the rDNA. If the same were true of resistant types (e.g., they were also missing many rDNA repeats), then the effective transmission of the Y chromosome could be closer to one, reducing the probability of resistance evolving. It is further assumed that each offspring is derived from an independent mating, rather than, as usual with
Anopheles gambiae, females mating only once in their life; incorporating this effect into the model would increase the amount of stochasticity, and reduce the probability of resistance evolving.
Finally, another assumption is that the population is closed and random-mating, whereas real
An. gambiae populations exist over a landscape. Previous modelling has used a range of approaches to investigate the spatial spread of a driving Y [
13,
25,
42], and these models should be extended to investigate the evolution of resistance. Some insight into the likely dynamics can be gained even from the current model. For example, consider a landscape of, say, 10,000 patches, each of which individually is a randomly mating population. If a driving Y is released into all of them, and
\(P_{1 } = 0.001\) for each patch, then it would be expected to spread and eliminate 9990 patches, and for resistance to evolve in ten of them; depending on how mosquitoes move on the landscape, those resistant types could eventually spread out from those patches and recolonize the landscape. The duration of protection offered by the driving Y will then vary from patch to patch, being shortest in those patches where the resistance first evolves, and longest in the last patch to be recolonized. If the patches vary, the modelling suggests resistance is more likely to arise in patches with a higher density of mosquitoes and, separately, that have a higher
R
m
; these could act as source populations for the others. An interesting precedent in this regard is given by the evolution of insecticide resistance, in which the same nucleotide change has arisen and established multiple times in different genetic backgrounds [
45], which implies that the (continental scale) population size of
An. gambiae is larger than the inverse of the nucleotide mutation rate [
46]. Note that on a landscape, it may be difficult for multiple resistant types to establish. In this example, if there is simultaneous release of a second construct that is sufficiently different from the first that resistance to one would not provide resistance to the other, then resistance to the second may also arise in ten patches, but if they are a different ten patches, then there may be no opportunity for the multiply-resistant type to evolve, and the population could be eliminated across the landscape. But even in the absence of resistance, there can be complex landscape dynamics between local elimination and recolonization [
13,
42] which warrant further analysis.
It will be worthwhile investigating the interactions between genetic technologies and other interventions more broadly. There is an automatic synergy between different interventions that reduce population size in terms of reducing the probability of resistance evolving: anything that reduces population size, like bed nets or IRS, will reduce the probability that resistance evolves to a driving Y. Likewise, release of a driving Y will likely reduce the probability that resistance evolves to new insecticides, or even to new anti-malarial drugs.