1. Genomics and Evolutionary Biology
Download icon

Introduction of a male-harming mitochondrial haplotype via ‘Trojan Females’ achieves population suppression in fruit flies

  1. Jonci Nikolai Wolff Is a corresponding author
  2. Neil J Gemmell
  3. Daniel M Tompkins
  4. Damian K Dowling
  1. Monash University, Australia
  2. University of Otago, New Zealand
  3. Landcare Research, New Zealand
Research Article
Cited
0
Views
609
Comments
0
Cite as: eLife 2017;6:e23551 doi: 10.7554/eLife.23551

Abstract

Pests are a global threat to biodiversity, ecosystem function, and human health. Pest control approaches are thus numerous, but their implementation costly, damaging to non-target species, and ineffective at low population densities. The Trojan Female Technique (TFT) is a prospective self-perpetuating control technique that is species-specific and predicted to be effective at low densities. The goal of the TFT is to harness naturally occurring mutations in the mitochondrial genome that impair male fertility while having no effect on females. Here, we provide proof-of-concept for the TFT, by showing that introduction of a male fertility-impairing mtDNA haplotype into replicated populations of Drosophila melanogaster causes numerical population suppression, with the magnitude of effect positively correlated with its frequency at trial inception. Further development of the TFT could lead to establishing a control strategy that overcomes limitations of conventional approaches, with broad applicability to invertebrate and vertebrate species, to control environmental and economic pests.

https://doi.org/10.7554/eLife.23551.001

eLife digest

Insect and other animal pests pose some of the greatest challenges to biodiversity, global economies and human health. The environmental and agricultural losses caused by pests have been estimated at 120 billion US dollars a year in the US alone. Many pests also spread diseases, such as dengue fever and malaria. A variety of different strategies are used to control pests, but their effects are generally short-lived and they are often ineffective when pest numbers are low. Furthermore, many of these strategies are harmful to other wildlife, such as bees.

Most of the DNA within an animal cell is contained within a structure called the nucleus. However, some DNA is also found within other compartments called mitochondria. The Trojan Female technique has been proposed as a new strategy to control insect pests that harnesses naturally-occurring changes (known as mutations) in this mitochondrial DNA (or mtDNA for short). Introducing mutations that lower the fertility of males, but have no effects on females, into a pest population should, in theory, lead to a long-lasting decline in the size of the population, even if it is relatively small to begin with.

Wolff et al. tested this theory in fruit flies, which are often used as models of insects and other animals in research projects. Adding female fruit flies carrying a mutation in mtDNA that lowers male fertility (known as “Trojan Females”) into populations of fruit flies reduced the size of the population over several generations. The mutation was maintained in the population for at least ten generations, and no “rescue” mutations evolved in the nuclear DNA to compensate for the mtDNA mutation. This indicates that the Trojan Female technique could be effective at controlling pests, without the need for Trojan Females to be repeatedly released into the populations.

The next steps following on from this work are to test this approach in economically important pest species, and to find out whether the approach is effective in various environments outside the laboratory. If these findings do indeed translate into these pests, then the Trojan Female technique may have the potential to be used to control a wide variety of different pest species from mosquitos through to rats.

https://doi.org/10.7554/eLife.23551.002

Introduction

Pest species pose some of the greatest present-day challenges to native biota, global economies, and human health (Naranjo et al., 2015; Simberloff et al., 2013). With their emergence and spread linked to human trade, transport and the agricultural revolution, vertebrate, invertebrate and plant pest impacts have followed human global movement over past millennia like a footprint (Hulme, 2009). The environmental and agricultural losses caused by pests have been estimated at US$120 billion annually in the US alone (Pimentel et al., 2005), while many also vector diseases of concern. For example, there are around 500 million human cases of the mosquito-borne diseases malaria and dengue fever annually (World Health Organization, 2015; Bhatt et al., 2013), with other agents such as Zika virus rapidly spreading (World Health Organization, 2016). Considering the humanitarian, economic, and environmental values at stake, pest management requires large and recurrent government expenditure worldwide (Stenseth et al., 2003).

Conventional pest management programs typically rely on some form of lethal control, such as vertebrate shooting, trapping, and poisoning, or the area-wide application of pesticides to target weeds and invertebrates. The effects of these approaches are often temporary in nature and thus require regular re-application, and may have unwanted side effects on non-target species and the environment (Bergstrom et al., 2009; Tompkins and Veltman, 2006; Innes and Barker, 1999). Furthermore, vertebrate culling is generally cost- and labor-intensive and can prove ineffective at low population densities (Clout and Russell, 2006), while blanket pesticide application is often hampered by the evolution of resistance in target species (Tabashnik et al., 2008; Innes and Barker, 1999). Research efforts have thus focused on the development of novel control or eradication techniques that are target-specific, cost-effective even when applied at low population densities, and long-lasting in effect.

One promising avenue lies in the development of techniques that impair the reproductive capacity of target pest species (Cowan et al., 2002; Courchamp and Cornell, 2000). The most successful of such techniques employed to date is the Sterile Insect Technique (SIT), whereby large quantities of sterilized males are introduced into target populations, reducing the reproductive success of the females with whom they mate (Alphey et al., 2010). Although the SIT offers species-specificity, a major constraint lies in the need to continuously produce and release large numbers of sterile males for sustained population suppression, rendering eradication efforts time- and cost-intensive (Dyck et al., 2005; Alphey et al., 2010). Resource requirements could be greatly reduced if impairments to male fertility in a target population were heritable in nature. Emerging theory and experimental work suggests this is achievable via a prospective approach called the Trojan Female Technique (TFT; Gemmell et al., 2013).

The goal of the TFT is to use naturally occurring mutations in the mitochondrial DNA (mtDNA), which impair male fertility but have no effects on females, to achieve multi-generational pest population suppression. Because mtDNA is typically maternally inherited (White et al., 2008; Birky, 1978), such male-specific deleterious mtDNA mutations will to a large degree escape selection in the female germ line despite their associated high fitness cost to males, enabling their persistent inheritance across generations (Frank and Hurst, 1996; Beekman et al., 2014). In theory, ‘Trojan Females’ carrying such mutations, and their female descendants, could continuously produce males with impaired fertility across generations, achieving perpetual numerical suppression of target populations (Gemmell et al., 2013). Unlike other genetically based pest control approaches that involve transgenics, such as the Release of Insects carrying a Dominant Lethal (RIDL) (Thomas et al., 2000), and the theorized use of gene-drives (Webber et al., 2015; Taylor and Gemmell, 2016), the use of naturally occurring mutations means that TFT pest control would not necessarily require genome editing to progress. The TFT may thus offer a valuable alternative to emerging transgenic control techniques, whose potential use is currently subject to debates concerning safety and regulatory concerns (Oye et al., 2014; Lunshof, 2015).

The conceptual framework underpinning the TFT is based on a population genetic model that shows the maternal inheritance of mitochondria will facilitate the accumulation of deleterious mtDNA mutations that are male-biased in their effects (Gemmell et al., 2004; Frank and Hurst, 1996; Beekman et al., 2014). Recent empirical work in Drosophila melanogaster has substantiated this model, showing that the expression of fertility, longevity, and levels of nuclear gene expression are more sensitive to genetic variation in the mtDNA sequence in males than in females (Camus et al., 2012; Yee et al., 2013; Innocenti et al., 2011; Camus et al., 2015). Furthermore, particular mtDNA haplotypes have now been associated with sub- or complete-infertility in males, but with no apparent effects on female fertility, in Drosophila (Dowling et al., 2015; Yee et al., 2013; Patel et al., 2016; Wolff et al., 2016b), seed beetles (Dowling et al., 2007), hares (Smith et al., 2010), and humans (Ruiz-Pesini et al., 2000). Based on this theory and empiricism, Gemmell et al., 2013 explored the conditions under which TFT mutations (male-fertility impairing but female-benign) could lead to population suppression. The results were encouraging, indicating that TFT haplotypes are predicted to cause suppression across a wide range of life-histories, with such effects not only persisting across generations, but also accumulating across successive introductions (Gemmell et al., 2013).

Building on those initial models (Gemmell et al., 2013), empirical attention has focused on a particular mitochondrial haplotype sourced from a population of D. melanogaster in Brownsville (USA), which has been associated with perturbed spermatogenesis and sperm maturation (Clancy et al., 2011) and is known to confer complete male sterility when placed alongside one isogenic nuclear background (i.e. a nuclear background devoid of any allelic variation; Clancy et al., 2011), and consistent reductions in male fertility against a range of other nuclear backgrounds (Dowling et al., 2015; Yee et al., 2013; Wolff et al., 2016b). Sequence analyses revealed 10 single nucleotide polymorphisms (SNPs) that are unique to this Brownsville haplotype: one mutation located in the cytochrome b gene and nine others that reside within the A/T-rich control region (mt:Cyt-b; Wolff et al., 2016a). The SNP located in the mt:Cyt-b gene is non-synonymous, causing an amino acid change (Ala278→Thr) in complex III of the mitochondrial electron transfer chain, and it is this SNP that has previously been implicated as the putative fertility-reducing mutation (Camus et al., 2015; Clancy et al., 2011). This mt:Cyt-b SNP as the cause of the mtDNA-mediated male infertility has yet to be unambiguously confirmed; such confirmation would require further work to disassociate this mutation from the other nine SNPs that delineate the Brownsville haplotype from its counterparts, which would most likely be tractably accomplished by gene editing – an approach that remains in its infancy for the mitochondrial genome (Wisnovsky et al., 2016).

A recent study has provided further support for a key role for this SNP in fertility suppression (Camus et al., 2015). Camus et al. (2015) demonstrated that the gene in which the SNP lies (mt:Cyt-b) experiences a four-fold decrease in expression in flies carrying the Brownsville haplotype relative to flies with other haplotypes, while expression of other mtDNA protein-coding genes is unaffected. Intriguingly, this Ala278→Thr mutation in the mt:Cyt-b gene occurs naturally in a range of other species, both vertebrate and invertebrate (Clancy et al., 2011). Although the phenotypic implications of this mutation have not yet been screened outside of its putative effect in D. melanogaster, this indicates that male-fertility-reducing mtDNA haplotypes may routinely segregate in natural populations of metazoans (Frank and Hurst, 1996; Beekman et al., 2014; Gemmell et al., 2004). However, the practical utility of harnessing male-fertility-reducing haplotypes for pest control remains unclear on two fronts: first, whether the reductions in male fertility that they cause will indeed result in the numerical suppression of populations and second, whether any demonstrable suppression effects will persist across generations.

There are several mechanisms by which impaired fertility in individual males may be compensated for at both individual and population scales. First, even males with impaired fertility may provide sufficient viable sperm for the complete fertilization of the eggs of females with which they mate in a population context. Second, even if a female is viable-sperm limited when mated with an impaired male, she may still obtain sufficient viable sperm through mating with other males. Third, even if population suppression initially occurs, as yet undetected pleiotropic effects on females could select against the mutation across generations. Fourth, even if the mutation persisted, the selection pressure imposed on males could select for nuclear modifiers that compensate for the effect of the TFT mitochondrial haplotype. Evolutionary theory and empiricism suggest that fertility effects associated with male-harming mtDNA mutations will often be reduced in such a way (Yee et al., 2013; Frank and Hurst, 1996; Gemmell et al., 2004). Ultimately, when present in large and panmictic populations, TFT haplotypes may be expected to undergo changes in frequency through genetic drift and directional or balancing selection (Wolff et al., 2014; Gregorius and Ross, 1984; Clark, 1984). Furthermore, stochastic contractions and expansions in the target population may exacerbate the effects of genetic drift and facilitate the purging of introduced TFT haplotypes, even when they are not selected against (White et al., 2008; Rand et al., 2001).

Here, we experimentally test the capacity of the Brownsville haplotype (our candidate TFT haplotype) to suppress large and panmictic laboratory populations of D. melanogaster. Persistent numerical suppression would provide proof-of-concept for the TFT, showing that the maternal mode of mtDNA inheritance can potentially be harnessed for a eukaryotic pest control technique that overcomes several limitations of conventional approaches. Trial populations were initiated with the TFT haplotype at four different starting frequencies (0% [control], 25%, 50%, and 75%), with the expectation that the numerical suppression observed would increase with increasing TFT frequency, and these populations were then maintained for 10 generations under two environmental regimes.

Under the first regime, populations were maintained in the ecological and demographic conditions in which they are typically maintained in the laboratory, and in which egg numbers per generation are carefully regulated. Predictions from an existing simulation model of D. melanogaster populations under such a regime (see Supplementary information in [Wolff et al., 2016b]) are that the TFT haplotype utilized (documented to reduce male breeding success by 29–69%; Dowling et al., 2015; Wolff et al., 2016b) will cause mean population suppression of between 6.7–16.8%, 13.8–37.7% and 21.4–63.7% for the 25%, 50% and 75% TFT haplotype frequency, respectively. Such a magnitude of suppression, modeled under conditions of multiple mating, is predicted to translate into up to three times greater suppression in natural populations in which females re-mate at a relatively low frequency (Wolff et al., 2016b). Under the second regime, populations were maintained in conditions that allow them to experience stochastic contractions and expansions in population size that are more reflective of natural population dynamics. The purpose of this regime was to explore how such dynamics could influence the frequency of the TFT haplotype across generations, with genetic drift potentially leading to either haplotype purging (with associated loss of population suppressive effects) or fixation (with associated increase in population suppressive effects).

Together, our experiments document the first experimental test of the ability of a candidate TFT haplotype to cause and maintain population suppression. We demonstrate that the TFT haplotype caused significant numerical suppression in the laboratory Drosophila melanogaster populations relative to controls, with the magnitude of effect positively correlated with its frequency at trial inception. Furthermore, the suppressive effect persisted over the full length of the trial (10 generations), with no reduction in haplotype frequency. Our results thus provide proof-of-concept for the TFT, showing that uniparental inheritance of mtDNA could potentially be harnessed in the development of a pest control technique that would be broadly relevant across eukaryotes.

Results

Experiment 1 – population suppression under density-controlled conditions (regulated population size)

We found an interactive effect of TFT treatment and generation number on population sizes (Tables 1 and 2A), and on the frequency of the TFT mutation (Tables 1 and 2B), across the 10 generations of the experiment (Figure 1A). Replicates initiated with TFT haplotype frequencies of 0% or 25% stabilized at average population sizes of 72.99 and 72.57 individuals across the 10 generations, respectively. However, replicates initiated with TFT haplotype frequencies of 50% or 75% declined over the first six generations to average population sizes of 67.24 and 66.75 respectively, with this magnitude of suppression (8%) maintained for the remainder of the experiment.

Table 1

Mean offspring numbers, TFT haplotype frequencies, and genotyping outcomes for the two experiments (regulated conditions and fluctuating conditions).

https://doi.org/10.7554/eLife.23551.003
TFT treatment (starting frequency)0% TFT25% TFT50% TFT75% TFT
Populations [n]21212121
ExperimentRegulatedF1 Frequency0.000.26 ± 0.030.44 ± 0.030.63 ± 0.03
F5 Frequency0.000.15 ± 0.030.63 ± 0.030.71 ± 0.02
F10 Frequency0.000.17 ± 0.030.67 ± 0.030.80 ± 0.03
Loss-400
Fixation-003
Heteroplasmy-000
Mean offspring number (F10)74.15 ± 1.0072.05 ± 0.8967.43 ± 0.9966.38 ± 0.70
Population extinction1000
FluctuatingF10frequency0.000.35 ± 0.060.59 ± 0.060.75 ± 0.05
Loss-410
Fixation-144
Heteroplasmy-020
Mean offspring number (F10)80.47 ± 5.1281.19 ± 4.5978.56 ± 5.6385.86 ± 3.73
Population extinction2020
Table 2

(A) Linear mixed model showing effects of TFT treatment and generation number on mean offspring number, and (B) generalized linear mixed model of effects of TFT treatment and generation on TFT haplotype frequency of populations with regulated population size (Experiment 1). There was no evidence of overdispersion in the model of TFT haplotype frequency (dispersion parameter = 0.766), and addition of an observation-level random effect to the final model did not change this parameter (dispersion parameter = 0.767), nor the parameter estimates of the model.

https://doi.org/10.7554/eLife.23551.005
(A) Offspring number(B) TFT frequency
Fixed effectsχ2Dfpχ2Dfp
TFT treatment2.2330.5338.142<0.001
Generation14.7390.106.2120.045
TFT treatment × generation51.09270.00323.524<0.001
Random effectsSD*SD*
Biological replicate0.81--0--
Experimental population1.11--0--
Residual5.48-----
* Standard deviation
Mean offspring number and haplotype frequencies under density-controlled population conditions (Experiment 1).

(A) Mean offspring number (±SEM), and B) mean haplotype frequencies (±SEM) of experimental populations in Experiment 1 over 10 generations, under density-controlled conditions. Founding populations (F0) were established with varying proportions (0%, 25%, 50%, 75%) of the TFT haplotype. Each generation was propagated with 80 eggs. Genotyping was conducted at generations 1, 5, and 10.

https://doi.org/10.7554/eLife.23551.006

We genotyped females from each experimental population at generations 1, 5 and 10, and found that the population suppression effect across generations was associated with the frequency of the TFT haplotype (χ2=21.37, p=<0.003; Table 3, Figure 2A–C). There was no evidence of consistent TFT haplotype purging; while the 25% TFT treatment ended (at generation 10) at a mean haplotype frequency of 0.17, ending frequencies were 0.67 and 0.80 for the 50% and 75% TFT treatments respectively (Figure 1B; Table 1). Over the course of the trial, the TFT haplotype was purged from four replicates and went to fixation in three (out of total of 63 populations).

Table 3

Linear mixed model of association of TFT haplotype frequency on mean offspring number in populations with regulated population size (Experiment 1).

https://doi.org/10.7554/eLife.23551.008
Fixed effectsχ2DfP
TFT frequency21.3770.003
Random effectsSD*
Biological replicate0.31--
Experimental population0--
Generation2.77--
Residual6.40--
* Standard deviation
Mean offspring number in relation to TFT haplotype frequency under density-controlled (Experiment 1) and fluctuating population conditions (Experiment 2).

Mean offspring number (±SEM) across experimental populations (A–C) with regulated population size at generations 1, 5, and 10 in Experiment 1; and (D) with fluctuating population size at generation 10 in Experiment 2. Haplotype frequencies were determined by genotyping seven females for each experimental population at each of three generations (1, 5, and 10; n = 1323) in Experiment 1; and by genotyping nine females for each experimental population at generation 10 (n = 567) in Experiment 2.

https://doi.org/10.7554/eLife.23551.009

Experiment 2 – population suppression under stochastic dynamics (Fluctuating population size)

The less-constraining rearing conditions of this experiment led to mean changes in population size between any two generations of approximately 30% across all experimental populations (χ2=46.72, p=<0.001; Figure 3; Table 4), with four experimental populations (two in each of the 0% and 50% TFT treatments) going extinct. Under these conditions, there was no detectable effect of the TFT treatment on population size across the experiment (Table 4A). Nor did we detect an association between the TFT haplotype frequency of each experimental population at generation 10 and the final population size (Table 5, Figure 2D). However, the TFT haplotype was still stably maintained in most cases; ending frequencies were 0.35, 0.59 and 0.75 for the 25%, 50% and 75% TFT treatments, respectively (Table 1, Table 4B, Figure 3). Over the course of the trial, the TFT haplotype was lost from five replicates and went to fixation in nine (out of a total of 63 populations). In two replicates of the 50% TFT treatment, we detected 11 cases where flies carried both the TFT and wildtype haplotype (five individuals in one replicate and six in the other), indicating at least two cases of biparental mtDNA inheritance in the experiment.

Mean offspring number and haplotype frequencies under fluctuating population conditions (Experiment 2).

Mean offspring number (±SEM, on vertical axis on left-hand side), and mean haplotype frequencies (±SEM, right-hand side) at generation 10, of experimental populations. Founding populations (F0) were established with varying proportions (0%, 25%, 50%, 75%) of fruit fly pairs harboring the mt:Cyt-b TFT mutation. Each generation, each experimental population was propagated with all offspring of the previous generation.

https://doi.org/10.7554/eLife.23551.011
Table 4

(A) Linear mixed model showing effects of TFT treatment and generation on mean offspring number; and (B) generalized linear mixed model of effects of TFT treatment on TFT haplotype frequency of populations with fluctuating population size (Experiment 2). When reanalyzing mean offspring number (A), having excluded the 15 zero values in the dataset resulting from three vial extinctions (two from the TFT 0% treatment, and one from the TFT 50% treatment), the effect of TFT treatment on offspring number remained statistically non-significant (χ2=3.2, p=0.36). The binomial model of TFT frequency (b) indicated overdispersion (overdispersion parameter = 1.85), and thus an observation-level random effect was added (experimental population) to the model (overdispersion parameter of final model = 1.07).

https://doi.org/10.7554/eLife.23551.013
(A) Offspring number(B) TFT frequency
Fixed effectsχ2DfP2DfP
TFT treatment4.9830.1720.332<0.001
Generation46.729<0.001---
Random effectsSD*




Biological replicate3.1--0--
Experimental population11.2--1.32--
Residual20.85-----
* Standard deviation
Table 5

Linear model of association of TFT haplotype frequency on mean offspring number in populations with fluctuating population size (Experiment 2).

https://doi.org/10.7554/eLife.23551.014
Fixed effectsχ2DfP
TFT Frequency4.9590.839
Random effectsSD*
Biological replicate0--
Residual24.93--
* Standard deviation

Discussion

Working with laboratory populations of D. melanogaster, we have demonstrated that the compromised male fertility caused by our candidate TFT haplotype can suppress population sizes across generations. The magnitude of suppressive effect was dependent on the frequency of the TFT haplotype and the conditions under which populations were propagated. When the experiment was conducted under regulated population sizes, persistent numerical suppression was observed (Figure 1A). While all treatments reduced in population size over the first two generations, as the trial moved toward equilibrium dynamics, those seeded with at least 50% TFT haplotypes continued to decline to generation six, and then remained suppressed at sizes averaging 8% below control populations. No such effect was apparent for populations seeded with only 25% TFT haplotypes.

These results raise two questions. First, why was the suppressive effect only 8% at the population scale, when a-priori modelling predicted suppression relative to controls of at least 21.4% in the 75% TFT haplotype frequency treatment (see Supplementary information in Wolff et al., 2016b). Second, why was there no apparent effect of the 25% TFT treatment? These two issues are likely linked, and may be due to compensation at the scale of the individual, through females obtaining more fertile sperm than would be expected on an additive basis, either wholly from the sub-fertile TFT males with whom they mate and/or through multiple mating (i.e. the TFT haplotype used may have caused less reduction in male breeding success than modelled, or females may have mated with more males than included in our modelling (see Supplementary information in Wolff et al., 2016b; Gemmell et al., 2013)). Such compensation would explain why population size was not affected in the 25% TFT treatment (complete compensation) and was less than predicted in the 50% and 75% TFT treatments (partial compensation). The reduction in haplotype frequency that occurred in the 25% TFT treatment (Figure 1B; Table 1) may also have contributed. However, irrespective of these or other mechanisms that may be responsible, a significant population suppression effect of the TFT haplotype, which persisted to the end of the trial, was observed for populations of the 50% and 75% TFT treatments.

When the experiment was conducted under conditions in which populations experienced large stochastic contractions and expansions in size, no suppression effects were detected. This was not driven by overall changes in TFT haplotype frequency over the course of the trial. While, as was expected, there were more cases of haplotype loss and fixation under this regime than under the more stable regime (totals of 14 versus 7 such events), the haplotype went to fixation more often than it was purged, and no overall decline in TFT frequency from starting conditions was observed (Figure 3). It is thus most likely that the suppression effect of around 8% observed under the more stable experimental regime, was masked by the underlying population dynamics in the more stochastic regime. Although mean TFT haplotype frequency declined slightly (from 25% to 17%) across the 10 generations of Experiment 1 in the 25% TFT treatment, the frequency of the TFT haplotype actually increased in the 50% and 75% treatments (to an average of 67% and 80%, respectively; Figure 1A–B) of this experiment. Furthermore, in Experiment 2, haplotype frequencies increased in the 25% and 50% treatments and were stably maintained in the 75% treatment at generation 10. Thus, across six treatments and two experiments, the frequency of the TFT haplotype decreased in just one treatment, and this decrease was modest. These data suggest there was no strong selective pressure against the TFT haplotype, even though it was causing population suppression. This result further supports both the theory underpinning the TFT, that male-infertility caused by a mutation in the mtDNA will generally escape selection due to its maternal inheritance (Frank and Hurst, 1996; Gemmell et al., 2004; Beekman et al., 2014), and the previous anecdotal observations of no negative pleiotropic effects on female fertility being linked to the TFT haplotype (which would also incur selection against it; Clancy et al., 2011; Dowling et al., 2015; Yee et al., 2013). Intriguingly, the frequency of the TFT haplotype increased in four of the six treatments (Table 1). While this contention requires further experimental testing, if haplotype frequency increases are occurring one potential explanation may lie in an observation from a previous study that the TFT haplotype is linked to increased pupal viability (Wolff et al., 2016b), suggesting an antagonistic pleiotropic effect (low male fertility, but high pupal viability) that is under positive selection due to its benefits to females.

As noted in the introduction, persistent maintenance of TFT haplotypes within a population is expected to result in selection on males for nuclear modifiers that compensate for the negative TFT effect, and restore male fertility (Frank and Hurst, 1996; Yee et al., 2013; Wolff et al., 2016b; Dowling et al., 2015). The capacity for an effective compensatory response will depend largely on levels of standing nuclear allelic variance already present within populations. Our previous work indicated that although the effects of the TFT haplotype on male fertility consistently conferred lower fertility in males relative to other haplotypes, it was indeed modulated by the nuclear background of different populations (Wolff et al., 2016b; Dowling et al., 2015). However, even though the experimental populations utilized in the current study were large and expected to maintain high levels of segregating nuclear allelic variance (Gardner et al., 2005; Griffin et al., 2016), there was no apparent rapid selection for fertility-restoring nuclear components over the course of our trials (which would have been expressed as restoration in population sizes over time). Thus, although nuclear mutations could arise over time in a population that compensate for the negative effects of the TFT mutation, in the population of Drosophila that we used there were no apparent segregating nuclear modifiers that had the capacity to completely restore male fertility and be rapidly selected. Extending our experiments by placing the TFT haplotype alongside additional outbred nuclear backgrounds, including the nuclear background from the population the TFT haplotype was originally sourced from, could potentially inform at what frequency (if at all) such nuclear modifier alleles may occur. In this regard, it would also be interesting to evaluate whether the use of fertility-reducing mutations that have evolved naturally are likely to be more successful in the long-term (in terms of heritability and sustained effect) to suppress population size than the use of artificial gene-drive constructs, whose introduction are predicted to almost inevitably lead to the emergence of drive-resistant alleles in most natural populations (Unckless et al., 2017; Noble et al., 2016). Notably, if the suppression effect of these mutations (be they natural or gene-drive constructs) when placed into new target pest populations is large, this will presumably act to reduce both the efficacy by which selection can target standing nuclear variation, and the likelihood of spontaneous compensatory mutations, that restore fertility in the target population.

Interestingly, we identified cases in which offspring were heteroplasmic for both the TFT and Dahomey mtDNA haplotypes in two of the replicate populations. Heteroplasmy has previously been found in Drosophila sourced from Brownsville (Kann et al., 1998). Whether the Brownsville population is predisposed to sporadic episodes of biparental inheritance is unclear, but the repeated observation of low rates of biparental inheritance of mtDNA in populations across the globe suggests that paternal leakage may be common in Drosophila (Wolff et al., 2013; Nunes et al., 2013; Dokianakis and Ladoukakis, 2014). The intra-individual co-occurrence of both TFT and wildtype haplotypes enables the possibility for recombination between divergent mtDNA molecules to create novel mitochondrial haplotypes carrying the TFT mutation (Ma and O'Farrell, 2015). Whether the fertility-suppressing effect of the candidate TFT mutation(s) would be moderated by its placement alongside a different mitochondrial genetic background, within a recombinant haplotype, is unknown and would depend on the capacity for epistasis within the mitochondrial genome to affect fitness outcomes. Furthermore, to come into effect, such a novel recombinant haplotype must then be at a selective advantage if it is to become rapidly fixed within populations. However, a more likely scenario is that rare recombinant haplotypes will be purged under drift while segregating at low population frequencies, or under purifying selection if harmful to females (Wolff et al., 2011; Bergstrom and Pritchard, 1998; Ma and O'Farrell, 2016).

While we have provided proof-of-concept for the TFT, demonstrating experimentally that its male fertility effects can achieve persistent population suppression, the question remains of its utility for field application to pest populations. Critically, while the conceptual foundation for the work was based on TFT haplotypes conferring complete male sterility (Clancy et al., 2011), subsequent work has demonstrated that reduced fertility is the more likely scenario (Dowling et al., 2015; Wolff et al., 2016b), and our current laboratory trials indicate that compensation due to females remating with wild-type males, or other processes such as mitonuclear, or gene-by-environment interactions, could result in reduced levels of population suppression than would otherwise be predicted by the underpinning theory (Gemmell et al., 2013). However, modeling predicts the magnitude of effect to be much greater in natural populations in which females are expected to re-mate at a relatively low frequency (Wolff et al., 2016b), due to lower encounter rates between individuals of each sex and lower densities of cohabitation (Gemmell et al., 2004). Previous experiments have further revealed that individuals harboring the Brownsville haplotype exhibited increased pupal viability, which is likely to aid the introgression of the TFT haplotype into target populations (Wolff et al., 2016b).

The utility of the TFT in the field will also depend on the efficacy of specific TFT haplotypes to decrease male fertility (Gemmell et al., 2013). Multiple-release strategies can be employed in order to reach TFT haplotype frequencies required to achieve population suppression of natural populations. This way, eradication strategies may still be achievable even if the desired effect in population suppression is reliant on TFT haplotype frequencies that are high. In addition, there is potential to augment the sterilizing effects of the TFT haplotype through linking it with further candidate TFT mutations. Evolutionary theory and empiricism both suggest that plant and animal mitochondrial genomes should be naturally enriched for male-harming mtDNA mutations (Innocenti et al., 2011; Gemmell et al., 2004; Frank and Hurst, 1996; Camus et al., 2012; Beekman et al., 2014; Dobler et al., 2014). Indeed, a male-sterilizing but female-benign mutation has recently been discovered in the gene encoding the cytochrome c oxidase subunit 2 (mt:COII) in D. melanogaster (Patel et al., 2016). Linking multiple TFT mutations within a single TFT haplotype, or the release of multiple TFT strains each bearing a distinct set of TFT mutations holds great promise to further the capacity of the TFT to efficiently suppress population size. The pairing of multiple TFT mutations within the one mtDNA sequence should soon be quickly achievable, given the rapid advances in genome editing technologies (Reddy et al., 2015). Thus, although the TFT does not necessarily require transgenics to progress, genome editing could enable the time- and cost-efficient placement of TFT mutations into target populations (without the need for long-running mutagenic and breeding approaches to first generate and then implant the candidate TFT mutations). Placement of single mutations into test populations would also allow the unambiguous identification of the fertility-reducing mutation(s) harbored by the Brownsville haplotype, and to confirm whether it is indeed the mt:Cyt-b mutation that causes the decrease in male fertility. If confirmed, the utility of the mt:Cyt-b mutation holds particular promise for pest control given that this candidate mutation has already been identified in a broad range of invertebrate and vertebrate species (Clancy et al., 2011).

Combined with previous studies, and a solid theoretical conceptual basis, our results lend credence to the utility of the TFT as a novel approach to pest control, deserving of continued development. Underpinning work needs to continue in the fruit fly model system to both explore the effects of linking multiple candidate TFT mutations within single mtDNA sequences (and quantifying their effects), and whether female-beneficial (but male fertility-impairing) haplotypes can be harnessed to drive the spread of TFT haplotypes through pest populations to effectively suppress population size. However, it would now also be timely to explore the capacity of TFT candidate mutations to decrease male fertility in other species, particularly real-world pest species that could be suitable targets for TFT control. If applicability and consistency of effect can be confirmed, the potential use of the TFT to suppress populations holds promise for a broad range of metazoan pests.

Materials and methods

Fly strains

The experiment harnessed a laboratory population of fruit flies that was originally collected in Dahomey (Benin, West Africa) in 1970 (Partridge and Andrews, 1985), and which has been kept at large effective population sizes since (at 25°C on a 12:12 light: dark cycle). To maintain the high levels of nuclear allelic variation segregating within the Dahomey population (Gardner et al., 2005; Griffin et al., 2016), populations have been kept in large replicate populations on a discrete-generations cycle since these were obtained from Prof Linda Partridge in 2010. This is achieved by propagating each generation with around 900 adult flies of 4 days of adult age, dispersed across three 250 ml bottles, each containing 75 ml of a potato-dextrose-agar food substrate. The flies are provided with a one to two hour ovipositioning period, after which the number of eggs per bottle is manually reduced (trimmed) to 300–350. Adult flies are then removed from the bottles and then, for the subsequent generation, emerging adult offspring that eclose from each bottle are admixed prior to their re-sorting into three separate bottles to start the propagation procedure for the following generation.

We initiated six replicates of the Dahomey population, and introgressed the TFT mtDNA haplotype (sourced from Brownsville [BRO] Texas, USA; Rand et al., 1994) into three of these replicates. The other three replicates were designated to the control, and they hosted their own coevolved mtDNA haplotype sourced from Dahomey [DAH] (Partridge and Andrews, 1985). While the BRO haplotype confers low male fertility, with no recorded negative effects on female fertility, the DAH haplotype is putatively healthy and confers normal fertility in both sexes (Camus et al., 2012; Yee et al., 2013; Wolff et al., 2016b; Dowling et al., 2015; Camus et al., 2015).

All six replicates went through the same handling procedures leading into the experiment, over successive generations, which ensured effective population sizes across all replicates were carefully controlled with the expectation that levels of segregating nuclear variance were highly similar across replicates. Specifically, to initiate replicates harboring the TFT haplotype, 45 virgin females were collected from a genetic strain in which the BRO haplotype is placed alongside an isogenic nuclear background, called w1118 (Bloomington #5905; Ryder et al., 2004). These females were then crossed to 50 males from the Dahomey lab population. In the next generation, 45 virgin daughters were collected from each strain replicate, and again backcrossed to 50 males from the Dahomey lab population. This backcrossing procedure progressed for 12 generations. To initiate replicates harbouring the DAH haplotype, the same procedure was followed, but the haplotypes were sourced directly from the Dahomey lab population, via an initial mating of 45 virgin females to 50 males collected from Dahomey. Then in the next generation, the virgin daughters of this cross were backcrossed to males of the stock Dahomey population, and this procedure repeated each generation (Figure 4).

Experimental breeding scheme.

The TFT-bearing BRO haplotype and the putatively healthy DAH haplotype were introgressed into the outbred nuclear background Dahomey in three independent replicates (i.e. BRO:Dahomey1-3; DAH:Dahomey1-3). Experimental populations were established by supplementing DAH:Dahomey test populations (DAH:Dahomey1-3) with varying contributions of fly pairs bearing the TFT haplotype (0%, 25%, 50%, 75%) sourced from the corresponding BRO:Dahomey replicate population (i.e. DAH:Dahomey1/BRO:Dahomey1; DAH:Dahomey2/BRO:Dahomey2; DAH:Dahomey3/BRO:Dahomey3). For each of the three biological replicates and for each treatment class (0%, 25%, 50%, 75%), we established seven technical replicate populations. These experimental populations were further duplicated, one cohort providing populations for the regulated population size approach (Experiment 1), and one cohort for the fluctuating population size approach (Experiment 2).

https://doi.org/10.7554/eLife.23551.015

Theoretically, each generation of introgression of the TFT haplotype into the Dahomey nuclear background increases the contribution of the Dahomey nuclear background by 50%, and thus, after 12 generations of introgression the contribution of Dahomey nuclear alleles to each TFT population replicate should have exceeded 99.98%. Thus, following the introgression procedure, we had six replicate strains, each of which contained a large representative sample of the nuclear alleles segregating within the Dahomey laboratory population; three of which however harbored the BRO haplotype harboring the candidate TFT mutations (denoted BRO/Dahomey), and the other three the DAH haplotype (denoted DAH/Dahomey).

All experimental crosses involved 4-day-old flies in 40 ml vials containing 6 ml of potato-dextrose-agar medium (at 25.0°C and a 12 hr: 12 hr light:dark cycle). Populations were maintained at a density of 80 flies per vial because at this density fly populations are sufficiently large to be stably maintained while limiting nutritional stress (e.g. food scarcity) which otherwise may impact development, fitness and behavior of affected fly populations (Santos et al., 1994). Further, sensitivity analyses within our a-priori demographic modeling, which informed our TFT treatments, showed that predictions were robust with respect to a modelled lab population size of 80 individuals (see Supplementary information in Wolff et al., 2016b).

Experimental design

To test the capacity of the candidate TFT haplotype to effect population suppression within a multi-generational framework, in each experiment we established experimental target populations (DAH/Dahomey) that were seeded with varying contributions of the TFT haplotype-bearing BRO/Dahomey individuals (a TFT treatment with four levels). All populations were established with 80 adult fruit flies at 1:1 sex ratio, and with the TFT haplotype contributing to 0% (control), 25%, 50%, and 75% of the starting population. All four levels of the TFT treatment were established for each of the three biological BRO/Dahomey replicates, with each BRO/Dahomey replicate matched to a corresponding DAH/Dahomey replicate. Within each strain replicate, each level of the TFT treatment was itself replicated seven times (i.e. three biological replicate BRO:Dahomey populations × four treatment levels × seven technical replicates = 84 experimental populations; see Figure 4).

We conducted two separate experiments, which ran concurrently. In the first experiment, we matched the conditions (in terms of density) under which fly populations are typically maintained in our laboratory, in which egg numbers per generation are carefully regulated, and thus, the population size is maintained around a constant density of 80 individuals at both juvenile and adult life stages (hereafter referred to as: Experiment 1; Regulated population size). Each experimental population was initiated using virgin flies at three days of age. The flies of each experimental population were then allowed 24 hr to mate, after which flies were transferred into vials with fresh food substrate for 4–6 hr for ovipositioning on day 4, until each vial contained in excess of 80 eggs. Immediately after ovipositioning, flies were collected and stored at −20°C, and the number of eggs per population reduced to 80 eggs by manually removing surplus eggs in each vial. To select eggs for retaining, the circular-shaped food source was divided into stripes. Eggs within each stripe were then counted starting from one side of the circle moving toward the opposite side of the circle until 80 eggs had been counted. This way, eggs were selected from the periphery through to the center of the food source regardless of egg density. Surplus eggs were discarded. We also aimed to minimize sampling effects by regulating egg density prior to the manual cull of eggs, by having females lay eggs over a short time period of only 4 to 6 hr. This ensured a sufficient number of eggs per vial, and also ensured that the majority of eggs contained in any one vial was used to give rise to the next generation. Each clutch of eggs was counted twice to minimize error in egg counts. Despite this precaution, eggs can be covered by food and escape detection, thus for 11 out of 840 vials (1.3%) offspring counts >80 were observed. The 80 eggs of each experimental population were then allowed to develop until eclosion to give rise to the next generation. Once eclosed into adults, flies of each population were transferred onto fresh food daily, until 4 days of age, when the next round of ovipositioning and egg-trimming occurred. We continued this procedure for 10 consecutive generations in total. In each generation, the total number of flies eclosing per population, from the initial pool of 80 eggs, was counted. The experiment was conducted blind to the identity of the experimental vials.

In the second experiment, all procedures were identical with the exception that the 84 experimental populations were not subjected to egg-trimming each generation (hereafter referred to as: Experiment 2; Fluctuating population size). Instead, once established, each ovipositioning period per experimental population was stopped when around 50% of the population vials were estimated to contain in excess of 80 eggs. Once this threshold was reached, flies were collected and stored at −20°C, eggs were left to develop to eclosion, and the number of eclosed flies per population in each generation was counted. This protocol diverges from the rearing conditions under which our laboratory populations are typically maintained, with the populations experiencing high levels of competition at both larval and adult stages for the 6 ml of available food, which routinely led to generations of population contractions, interspersed by generations of population expansions. This experimental design was used as an approximation for the demographic conditions natural populations may be exposed to, where single populations potentially transition through severe population bottlenecks, in which genetic drift would be expected to play a larger role in shaping frequencies of co-occurring mtDNA haplotypes relative to the regulated conditions of Experiment 1.

Genotyping

DNA from single females was extracted in 96-well format using Wizard Genomic DNA Purification Kit (Promega, Madison, WI 53711, USA) following the manufacturer’s instructions for single sample extractions, and using a third of recommended volumes to adjust for 96-well plate well volume. We invested most genotyping resources on Experiment 1, extracting DNA from seven females in each of three generations (1, 5, and 10) from each of the 63 populations that had been seeded with the TFT haplotype BRO (25%, 50%, and 75% BRO:Dahomey contribution; sample size: 63 populations × seven females × three generations = 1323 females).

For Experiment 2, we extracted DNA from nine females, all from generation 10 only, from each of the 63 populations that had been seeded with the TFT haplotype (25%, 50%, and 75% BRO:Dahomey contribution; sample size: 63 populations × nine females × one generation = 567 females). Populations established with DAH:Dahomey only (control populations) were not genotyped for either experimental cohort (their genotypes were fixed at 0% TFT haplotype). All DNA extracts were adjusted to DNA concentrations of 5 ng*µl−1 and a final volume of 50 µl per sample. Genotyping was conducted using a custom iPLEX Gold genotyping assay on the Sequenom MassARRAY Analyzer four system at Geneworks Pty Ltd, Thebarton, Australia.

Statistical analysis

The structure of the data is outlined in Figure 4. We fitted linear mixed models to phenotype data (offspring number), and generalized linear mixed models to genotype data (frequency of the TFT haplotype), using the lme4 package 1.1.12 (Bates et al., 2015) in R 3.0.3 (R Development Core Team, 2013).

In the phenotypic data analyses, the response variable (number of offspring produced) was modeled using a Gaussian distribution. Although this data is strictly count data, it conformed to a normal rather than Poisson distribution as expected of large sample sizes under the Central Limit Theorem. For example, the analysis of Experiment 1 contained only 1 zero value in the dataset. Although the analysis of Experiment 2 contained 15 zero values, removal of these values did not change the statistical inferences; and importantly, the residuals of these models were normally distributed. For the linear mixed models, fixed effects parameters were estimated using maximum likelihood estimation, and random effects were estimated using restricted maximum likelihood estimation. In the genotype data analyses, the frequency of the TFT haplotype was modeled as a binomial vector comprising the number of TFT haplotypes genotypes per vial and the number of wild-type haplotypes, using a binomial distribution and logit link.

For both analyses, we treated both the identity of the experimental population (i.e. individual vials) nested within Biological Replicate, and Biological Replicate, as random effects, and both TFT treatment (control, 25% TFT, 50% TFT, 75% TFT), and generation (F1–F10), as fixed effects. The TFT treatment control level was removed from analyses modeling the TFT haplotype frequency across the TFT treatments, since this level was invariably zero and its inclusion violated the model assumption of homogeneity of variance across classes. Generation was not added as a factor to the model of TFT haplotype frequency for Experiment 2, since genotyping was only conducted at generation 10 in this experiment. Consequently, each data-point represented that of a specific experimental population (i.e. 63 data points in the dataset, equaling 63 experimental populations), and thus, experimental population represented an observation-level effect in the models of Experiment 2. For the binomial models of TFT haplotype frequency, the blmeco package (Korner-Nievergelt et al., 2015) indicated overdispersion; ‘experimental population’ (an observation-level random effect) was thus added to the models to account for this.

We also examined the correlation between the number of offspring produced per experimental population and the TFT haplotype frequency, for both experiments. We fitted linear mixed models, with the number of offspring produced per experimental population as the response variable, and the TFT frequency of the same population as a fixed factor (of eight levels in Experiment 1, and 10 levels in Experiment 2), with the identity of the experimental population nested within the Biological Replicate, Biological Replicate, and generation number added as random effects to the model of Experiment 1, and with Biological Replicate added as a random effect for the model of Experiment 2.

Significance of fixed effects in each model was assessed using Type III sums-of-squares, χ2 tests in the car package of R (Fox and Weisberg, 2011).

References

  1. 1
  2. 2
  3. 3
  4. 4
    Germline bottlenecks and the evolutionary maintenance of mitochondrial genomes
    1. CT Bergstrom
    2. J Pritchard
    (1998)
    Genetics 149:2135–2146.
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
    Natural selection with nuclear and cytoplasmic transmission. I. A deterministic model
    1. AG Clark
    (1984)
    Genetics 107:679–701.
  12. 12
    The eradication of mammals from New Zealand islands
    1. M Clout
    2. J Russell
    (2006)
    In: M Clout, F Koike, M Kawamichi, M De Poorter, K Iwatsuki, editors. Assessment and Control of Biological Invasion Risks. Kyoto, Japan: Shoukadoh Book Sellers. pp. 127–141.
  13. 13
  14. 14
    Field Applications Of Fertility Control For Wildlife Management
    1. P Cowan
    2. R Pech
    3. P Curtis
    (2002)
    In: A. R Pickard, W. V Holt, J. C Rodger, D. E Wildt, editors. Reproductive Science and Integrated Conservation. Cambridge, UK: Cambridge University Press. pp. 305–318.
    https://doi.org/10.1017/CBO9780511615016.024
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
    An R Companion to Applied Regression (2nd edn)
    1. J Fox
    2. S Weisberg
    (2011)
    London, U.K: Sage Publishing.
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
    Selection with gene-cytoplasm interactions. I. maintenance of cytoplasm polymorphisms
    1. HR Gregorius
    2. MD Ross
    (1984)
    Genetics 107:165–178.
  26. 26
  27. 27
  28. 28
    Ecological consequences of toxin use for mammalian pest control in New Zealand—An overview
    1. J Innes
    2. G Barker
    (1999)
    New Zealand Journal of Ecology 23:111–127.
  29. 29
  30. 30
  31. 31
    Bayesian Data Analysis in Ecology Using Linear Models with R, BUGS. and Stan
    1. F Korner-Nievergelt
    2. T Roth
    3. S von Felten
    4. J Guélat
    5. B Almasi
    6. P Korner-Nievergelt
    (2015)
    Statistics Academic Press.
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
    R: A Language and Environment for Statistical Computing
    1. R Development Core Team
    (2013)
    R Foundation for Statistical Computing Austria, Vienna.
  43. 43
    Sexually antagonistic cytonuclear fitness interactions in Drosophila Melanogaster
    1. DM Rand
    2. AG Clark
    3. LM Kann
    (2001)
    Genetics 159:173–187.
  44. 44
    Neutral and non-neutral evolution of Drosophila mitochondrial DNA
    1. DM Rand
    2. M Dorfsman
    3. LM Kann
    (1994)
    Genetics 138:741–756.
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65
    World Malaria Report
    1. World Health Organization
    (2015)
    Geneva, Switzerland: World Health Organization.
  66. 66
    Zika Situation Report
    1. World Health Organization
    (2016)
    Geneva, Switzerland: World Health Organization.
  67. 67

Decision letter

  1. Marcel Dicke
    Reviewing Editor; Wageningen University, Netherlands

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Introduction of a male-harming mitochondrial mutation via Trojan Females achieves population suppression in fruit flies" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Diethard Tautz as the Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Liliana Milani (Reviewer #1) and Fabrizio Ghiselli (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Your manuscript describes the effects of a male-fertility impairing mtDNA haplotype on haplotype frequency change and population size variation in population cage experiments in the fruit fly Drosophila melanogaster. The experiments are well carried out and described. The three reviewers have provided constructive comments that critically evaluate your manuscript. The experiments seem to provide first support for the Trojan Female Technique. However, important issues have been identified by the reviewers that need to be addressed to fully prove this.

Important issues in this respect are:

1) Selection of 80 eggs and selection criteria of this upper limit

2) The limited effect (and replicability) and its implications for application

3) Identity of the mutation that caused the effect

Addressing these issues will be important in the light of a decision on your manuscript.

In addition, the reviewers have provided various additional valuable comments that deserved being addressed in a revision. Their comments are provided below.

Reviewer #1:

The work by Jonci N. Wolff, Neil J. Gemmell, Daniel M. Tompkins, Damian K Dowling, entitled "Introduction of a male-harming mitochondrial mutation via Trojan Females achieves population suppression in fruit flies", is the experimental evidence that the Trojan Female Technique (TFT) can be useful in population control as predicted by previous modelling (Gemmel. et al. 2013). Not only the TF mitochondrial mutation reduces male fertility, but there is no apparent pressure against the TFT haplotype considered (sometimes even showing increased frequency) that also appears to guarantee a persistent numerical suppression of target population.

The present work can be intended as a proof-of-concept for the use in TFT of the candidate mutation (mt:Cyt-b), that caused lower fertility for males (Dowling et al. 2015), in both competitive and noncompetitive mating contexts, across all nuclear backgrounds screened, with different genotype-by-environment interactions (Wolff et al. 2016).

I think the work is clear in describing the state of the art, the performed analyses, and the questions the new data are addressing. For these reasons, I recommend its publication.

Please clarify in the text what is meant for isogenic nuclear background.

Introduction, eighth paragraph: "Large and panmictic laboratory populations of D. melanogaster "

Discussion, fifth paragraph: "experimental populations utilized in the current study were panmictic”.

How can some sort of preferential mating be excluded?

Subsection “Experimental design”, second paragraph: I see from excel files that sometimes the offspring is of more than 80 individuals also in Experiment 1 (regulated population size). I imagine that the number of 80 eggs intended to be maintained had a certain error (and that it was not an error in the count of the offspring, that I imagine was of simpler calculation).

How were the eggs to be eliminated chosen? I am not an expert in Drosophila breeding, and I would like to know the best "random" way to reduce their number. From their different position in the vial? Were those closer to each other pruned? May the position of an egg be related in some way to its successful hatching, so a character to be considered?

Discussion, fifth paragraph: "[…] even though the experimental populations utilized in the current study were panmictic and characterized by high levels of segregating nuclear allelic variance".

How did the authors estimate the nuclear allelic variance of the used populations? I found no reference to this in Materials and methods when the introgression procedure was performed to obtain the 6 replicate strains.

What is the output of the genotyping analysis? Did you obtain both data on mitochondrial haplotype and nuclear alleles?

Reviewer #2:

The manuscript provides proof-of-concept for the Trojan Female Technique (TFT), a prospective population suppression technique based on naturally-occurring mitochondrial mutations that affect male fertility. TFT is a promising control strategy that could be applied to control pests.

The Introduction is clear, very well-written, and engaging. The experimental design is robust, and it required an impressive amount of work. Data collection and statistical analysis are appropriate. The authors discussed the results thoroughly.

Overall I really enjoyed reading the manuscript and in my opinion it represents an important study, very valuable for both applied (pest control) and basic research (mito-nuclear interaction and coevolution, effect of mitochondrial mutations).

I just would like to make a comment.

On the one hand, the authors observed that TFT-bearing BRO haplotype does not decrease in frequency through generations. This means that there is no direct evidence for selection against the TF haplotype. On the other hand, they also observed that the population suppression was below the expected levels in Experiment 1, and absent in Experiment 2.

I think that an explanation for this might be a sort of buffering effect consisting of "wild-type" haplotyes (DAH), complementing the suboptimal functionality of BRO haplotypes. Mitochondria carry multiple copies of mtDNA, and in order to have a phenotypic effect, a deleterious haplotype must be present in the mtDNA population of a mitochondrion at a frequency sufficient to counteract the complementing effect of functional haplotypes. The same rationale can be applied considering that there are multiple mitochondria in a cell, including of course germ cells, which are the main focus here (my collaborators and I discussed this topic extensively in [1] and [2]). Under this light, if a mitochondrial mutation is not severe enough to be purged by natural selection, it can persist in the population at a frequency that is determined by the severity itself, especially in this case where the mutation is deleterious only for males, but mtDNA has (almost always) matrilinear inheritance. That said, selection is still acting on male fertility, because each generation only the viable sperm will produce progeny, and because of this even if the suboptimal haplotype does not decrease in frequency, the observed population suppression is lower than expected (or absent).

An objection to this point could be about the TF haplotypes being segregated from "wild-type" haplotypes, but I think there are multiple ways by which an introgression can take place. Bottom line of this comment is: in my opinion, in the process of assessing the penetrance of a mitochondrial mutation, it is necessary to take into account the population dynamics also at mtDNA level (organelle, cell). I might be missing something in my considerations, so I would love to know alternative/complementary points of view.

In the future, it might be interesting to sequence samples from the populations included this study at different generations, to assess: 1) if compensatory mutations are arising in genes encoding subunits of the respiratory chain; 2) the frequency of TF and "wild-type" haplotypes, and if compensatory mutations are present also at the mtDNA level (e.g.: a mutation in cytb that compensates Ala278→Thr). Point 1) could be achieved by targeted deep sequencing of subunits of the electron transport chain (+ ATP syntase), and point 2) by deep sequencing of mtDNA.

Important note: I am including the following references only for the sake of scientific debate. I am not suggesting the authors to include such references in the Manuscript.

References:

1] Ghiselli, F. et al. Structure, transcription, and variability of metazoan mitochondrial genome: perspectives from an unusual mitochondrial inheritance system. Genome Biol. Evol. 5, 1535-1554 (2013).

2] Milani, L. & Ghiselli, F. Mitochondrial activity in gametes and transmission of viable mtDNA. Biol. Direct 10, 22 (2015).

Reviewer #3:

This is a well written manuscript that describes two experiments that test the effects of a male-fertility impairing mtDNA haplotype on haplotype frequency change and population size variation in population cage experiments in Drosophila melanogaster. The genotypes used have been previously described in other publications from the same group and that of David Clancy's group, namely a male-specific fertility defect in fruit flies harboring the Brownsville mtDNA haplotype. The main outcomes of this study show that under controlled densities, the proportion of Brownsville mtDNA haplotype is associated with a lowering of overall number of fruit flies (decreasing from 80) and this effect is numerically larger when the starting frequency of the impaired haplotype is higher. Overall, the suppression effect was a numerical reduction by ~8%. In contrast, in population cages that were allowed to fluctuate in size (around 80 flies), simulating a more natural population contraction-expansion, there was no suppression effect detected. The effects on haplotype frequency change were minimal throughout generational time, and the authors identified that heteroplasmy had established in some replicate cages.

I think the experiment was well conducted and the results well described but there are a few points that I feel require discussion or correction, outlined below.

Overall, the Introduction and Materials and methods are well described. One minor point – the reference to the fertility phenotype in the fifth paragraph of the Introduction refers to Yee et al. 2013 (male fertility), not Innocenti et al. 2011 (gene expression, which does not include any male fertility phenotypes).

One main point I have an issue with is the assertion that the defect is caused by the point mutation in the mtDNA CytB gene, which is repeatedly stated. While previous studies have suggested this association, there is no a priori reason to assume this mutation is any more important than the number of other mutations private to the Brownsville haplotype. For example, a study of mtDNA sequence variation from the same group (Wolff et al. 2015) provides sequence data across the whole mtDNA molecule and the regulatory D-loop alone harbors seven mutations that are private to Brownsville (D-loop alignment positions 303, 1466, 1469, 3244, 3250, 3251, 3709). Since these mutations are linked to the putative CytB mutation, there is insufficient evidence to suggest this mutation is the smoking gun. I think the use of safer description is warranted throughout and I suggest that the authors use the terms 'associated' and 'Brownsville haplotype' rather than the putative Ala278->Thr mutation.

It would be advantageous to know how much autosomal sequence variation there was in the Dahomey background used in this study (how outbred the lab-maintained stock really is, especially given the egg selection routinely used). Previous studies have shown near isogenic lines modify the effects of the Brownsville mtDNA, yet there is no assessment here; the Dahomey nuclear background also tends to exaggerate the Brownsville mtDNA haplotype effects. The magnitude of suppression effects is likely sensitive to nuclear background and only one was assessed here, therefore the generality of the finding is unknown. After all, Brownsville, TX, has a viable fruit fly population and this could form the basis of some discussion because it is the most 'natural' experiment. The Discussion, fifth paragraph, is unsubstantiated. Do you have any estimates of autosomal genetic diversity in these lines? If so, I would suggest including them, or at least discuss what may be expected in nuclear backgrounds that are not so sensitive to mtDNA effects, and which are likely to be experienced in a natural setting.

Is a -8% and stable population suppression biologically meaningful in the context of pest control, especially when it is only observed in a tightly regulated laboratory population? The Discussion mentions why the results may differ from the predicted values (and were somewhat less than expected) but the usefulness of the technique is somewhat limited (based on these findings). I think this could benefit from more discussion and especially for the role of egg-to-adult survival in the initial generations, which is known to be above average in Brownsville mtDNA lines.

Is the experiment #1 population size of 80 arbitrary, or was this based on the previous simulations? At such small population sizes, the opportunity for drift is very high and coupled with low and uneven sampling for the haplotype frequency estimates across both experiments, could this influence the results? Although there was vial replication, could these estimates not suffer from the same systematic bias? On the same note, the removal of flies from egg laying in experiment #2 (when approximately 50% of the vials produced 80 eggs) is still quite artificial and not what would be experienced in nature. This deserves some discussion since even a mild deviation from a strict egg laying dynamic can nullify the suppression effects found in experiment #1. It is perhaps unfortunate the reciprocal 100% Brownsville mtDNA treatment was not used as a control. While it would not be necessary or appropriate to use pest control in a population of zero flies, it would give a good estimate of the expected population size variation in the experimental design used here, which differs from previous studies using the Brownsville mtDNA haplotype. If these data are available, I would encourage the authors to include them to aid results interpretation.

The discussion of using genotypes with relatively high fitness females (Discussion, last paragraph) with low fitness in the corresponding males seems counterproductive since the object of pest control here is surely to reduce the number of insect vectors (not increase them!)

Perhaps wrap up with a more general commentary of the problems of evolution in gene drives as a pest control and how this technique can be tailored to avoid the same pitfalls – just a minor thought, but I feel this would compliment the nuclear compensation argument.

https://doi.org/10.7554/eLife.23551.016

Author response

Your manuscript describes the effects of a male-fertility impairing mtDNA haplotype on haplotype frequency change and population size variation in population cage experiments in the fruit fly Drosophila melanogaster. The experiments are well carried out and described. The three reviewers have provided constructive comments that critically evaluate your manuscript. The experiments seem to provide first support for the Trojan Female Technique. However, important issues have been identified by the reviewers that need to be addressed to fully prove this.

Important issues in this respect are:

1) Selection of 80 eggs and selection criteria of this upper limit

Although, to the best of our knowledge, hatching success is not influenced by an egg’s position on the food source, we tried to keep egg selection as random as possible while remaining experimentally tractable. To select eggs for retaining, the circular-shaped food source was divided into stripes. Eggs within each stripe were then counted starting from one side of the circle moving towards the opposite side of the circle until 80 eggs had been counted. Surplus eggs were discarded. This way, eggs were selected from the periphery through to the center of the food source. We also aimed to minimize sampling effects by regulating egg density prior to the manual cull of eggs, by having females lay eggs over a short time period of only four to six hours. This ensured a sufficient number of eggs per vial, and also ensured that the majority of eggs contained in any one vial was used to give rise to the next generation. We have added further explanation of our process to Materials and methods (subsection “Experimental design”, second paragraph).

The upper limit of 80 eggs/larvae/individuals per vial was chosen because at this density fly populations are sufficiently large to be stably maintained while limiting nutritional stress (e.g. food scarcity) which otherwise may impact development, fitness and behavior. Accordingly, in the first experiment, all populations were started with 80 eggs in each generation. We now realize that we have failed to state this clearly, and have amended the manuscript to provide clarification in the Materials and methods section (subsection “Fly strains”, last paragraph). Further, sensitivity analyses within our a-priori demographic modelling, which informed our TFT treatments, showed that predictions were robust with respect to modelled lab population size of 80 individuals (now added as Supplementary Material).

2) The limited effect (and replicability) and its implications for application

In experimental categories where population size was affected by TFT treatment, effects were highly reproducible and were detected across all three strain replicates. In regard to the apparent lack of replicability across the two experiments (i.e. the suppression effect observed under regulated population dynamics not being observed under more stochastic dynamics), the lack of observable effect does not necessarily equate to a lack of phenotypic effect. Rather, the most probable explanation is that the TFT-mediated effects under the more stochastic dynamics in which large fluctuations in population size occur across single generations (populations experienced contractions and expansions amounting to on average 30% of their population size, per generation). This interpretation of the data is supported by the genotyping data, which showed an increase in TFT haplotype frequencies in Experiment 2 (more stochastic dynamics) for both the 25% and 50% treatments, and stable maintenance for the 75% treatment. We have amended the Discussion so that this observation is clearly articulated (Discussion, third paragraph).

The primary goal of the research presented was to provide proof-of-concept for the feasibility of using the TFT to suppress populations. This work thus presents an important step towards the implementation of the technique. As discussed in the manuscript, we acknowledge further development is required for its final and successful implementation in the field. While the 8% suppression effect we have observed in our experiments may appear modest, demographic modelling predicts that the same TFT effect will cause much greater suppression in natural populations where mating rates are likely to be much lower (details of the model used are now included with this paper, rather than referring to the Supplementary Material of a previous publication). The suggestion by the reviewer of inclusion of a discussion of the above-average pupal viability associated with the Brownsville haplotype is an excellent idea, and we have now incorporated this aspect in the Discussion (sixth paragraph). Ultimately, higher efficiencies of the TFT will likely be achievable via placement of several fertility-reducing mutations within a single TFT haplotype, or via the release of multiple TFT strains harbouring distinct TFT haplotypes. Since the writing of this manuscript another TFT mutation has been discovered independently by Patel and colleagues (Patelet al. 2016. A mitochondrial DNA hypomorph of cytochrome oxidase specifically impairs male fertility in Drosophila melanogaster. eLife, 5, e16923). This finding further supports the notion that animal mitochondrial genomes will be naturally enriched for male-harming mtDNA mutations, which can be harnessed to further increase the efficiency of the TFT. We have thus incorporated these new findings and elaborated on how these mutations can be harnessed to further develop the TFT (Discussion, seventh paragraph).

3) Identity of the mutation that caused the effect

Addressing these issues will be important in the light of a decision on your manuscript.

We agree with the reviewer that there is no definitive evidence that the mt:Cyt-b causes the male-sterility, and that additional mutations unique to the Brownsville haplotype located within the AT-rich region, may be responsible. The mt:Cyt-b has previously (and here) been highlighted, because it is the only mutation that causes a non-synonymous change. Following a classical genetics model, it is thus the only mutation with immediate consequences at the protein level in a functionally important enzyme complex. However, we acknowledge that this view is incomplete and perhaps too simplistic, and have incorporated the reviewer’s concern and recommendation into the manuscript. We have thus changed the wording throughout the manuscript, now referring to the TFT haplotype instead of the mt:Cyt-b mutation. We have further provided additional information in regard to the mutation harbored by the TFT haplotype, and in the Discussion we have further outlined that future experiments employing genome-editing technologies will be required to unambiguously single out the fertility-reducing mutation(s) harbored within the Brownsville haplotype (Introduction, sixth paragraph; Discussion, seventh paragraph).

In addition, the reviewers have provided various additional valuable comments that deserved being addressed in a revision. Their comments are provided below.

Reviewer #1:

[…] Please clarify in the text what is meant for isogenic nuclear background.

We have added a definition for isogenic nuclear background: “a nuclear background devoid of any allelic variation”.

Introduction, eighth paragraph: "Large and panmictic laboratory populations of D. melanogaster"

Discussion, fifth paragraph: "… experimental populations utilized in the current study were panmictic".

How can some sort of preferential mating be excluded?

We agree that there can still be mate preferences within our populations. But in this context, we use the word panmictic to indicate there are no genetic or behavioural restrictions between members of the Dahomey laboratory population (i.e. each fly can freely mate with other flies in this population – all individuals of the opposite sex in the population are potential partners, with no recorded reproductive isolation mechanisms causing divergence of the population). We are happy to remove the adjective panmictic to describe the population should the reviewer wish us to.

Subsection “Experimental design”, second paragraph: I see from excel files that sometimes the offspring is of more than 80 individuals also in Experiment 1 (regulated population size). I imagine that the number of 80 eggs intended to be maintained had a certain error (and that it was not an error in the count of the offspring, that I imagine was of simpler calculation).

We fully agree that a count of above 80 is most likely caused by human error during the egg-counting process. We tried to minimize error by counting each batch of eggs twice. However, eggs can be covered by food and escape detection. Values of 80 and above were recounted several times to exclude error in the offspring count. We have added an explanatory section to Materials and methods for clarification, and have also added information how often a count of 80 was observed (subsection “Experimental design”, second paragraph).

How were the eggs to be eliminated chosen? I am not an expert in Drosophila breeding, and I would like to know the best "random" way to reduce their number. From their different position in the vial? Were those closer to each other pruned? May the position of an egg be related in some way to its successful hatching, so a character to be considered?

We kept this process as random as possible, and experimentally feasible. To achieve this, we removed the circular-shaped food source from the lid (2 ml of food < 4mm thickness that sits in the lid of a cylindrical specimen container of 3 cm diameter), and divided this into stripes starting from one side of the circle moving towards the opposite side of the circle. Eggs were then manually counted, starting from one side of the food source, and moving towards the opposite side, until 80 eggs were counted. Surplus eggs were discarded. This way, eggs were chosen from all areas of the food source (e.g., from the periphery to the center) regardless of egg density. Furthermore, we aimed to minimize sampling effects by regulating egg density even prior to the manual cull of eggs, by having females lay eggs over a short time period of 4 to 6 hours). This ensured a sufficient number of eggs per vial, and also ensured that the majority of eggs contained in any one vial was used to give rise to the next generation, regardless of their position in the vial. To the best of our knowledge, hatching success is not influenced by an egg’s position on the food source. We have added further explanation to Materials and methods (subsection “Experimental design”, second paragraph).

Discussion, fifth paragraph: "[…] even though the experimental populations utilized in the current study were panmictic and characterized by high levels of segregating nuclear allelic variance".

How did the authors estimate the nuclear allelic variance of the used populations? I found no reference to this in Materials and methods when the introgression procedure was performed to obtain the 6 replicate strains.

What is the output of the genotyping analysis? Did you obtain both data on mitochondrial haplotype and nuclear alleles?

We have not directly measured the nuclear allelic variation maintained at specific genetic loci in our experimental populations. Rather, this inference is based on the results of quantitative genetic studies that routinely use the D. melanogaster Dahomey population used here, and have shown high amounts of additive genetic variance underpinning the expression of numerous life-history traits. See, for example:

Gardner, M.P., Fowler, K., Barton, N.H. & Partridge, L. (2005). Genetic Variation for Total Fitness in Drosophila melanogaster. Genetics, 169, 1553-1571.

Griffin, R.M., Schielzeth, H. & Friberg, U. (2016). Autosomal and X-Linked Additive Genetic Variation for Lifespan and Aging: Comparisons Within and Between the Sexes in Drosophila melanogaster. G3: Genes|Genomes|Genetics, 6, 3903-3911.

Since receiving the Dahomey population in our lab in 2010, we have kept it at very large effective population sizes (~900 adult flies per generation), to ensure that there is no bottleneck in allelic variation. We can thus be confident the experimental populations we have created harbor high levels of segregating nuclear variance. It was not our intention to suggest that we characterized the nuclear genetic variation in this experiment but we now see how current wording could have led to misunderstandings. We have thus amended that section for clarification, and have added relevant references (Discussion, fourth paragraph; subsection “Fly strains”, first paragraph).

Reviewer #2:

[…] I just would like to make a comment.

On the one hand, the authors observed that TFT-bearing BRO haplotype does not decrease in frequency through generations. This means that there is no direct evidence for selection against the TF haplotype. On the other hand, they also observed that the population suppression was below the expected levels in Experiment 1, and absent in Experiment 2.

I think that an explanation for this might be a sort of buffering effect consisting of "wild-type" haplotyes (DAH), complementing the suboptimal functionality of BRO haplotypes. Mitochondria carry multiple copies of mtDNA, and in order to have a phenotypic effect, a deleterious haplotype must be present in the mtDNA population of a mitochondrion at a frequency sufficient to counteract the complementing effect of functional haplotypes. The same rationale can be applied considering that there are multiple mitochondria in a cell, including of course germ cells, which are the main focus here (my collaborators and I discussed this topic extensively in [1] and [2]). Under this light, if a mitochondrial mutation is not severe enough to be purged by natural selection, it can persist in the population at a frequency that is determined by the severity itself, especially in this case where the mutation is deleterious only for males, but mtDNA has (almost always) matrilinear inheritance. That said, selection is still acting on male fertility, because each generation only the viable sperm will produce progeny, and because of this even if the suboptimal haplotype does not decrease in frequency, the observed population suppression is lower than expected (or absent).

An objection to this point could be about the TF haplotypes being segregated from "wild-type" haplotypes, but I think there are multiple ways by which an introgression can take place. Bottom line of this comment is: in my opinion, in the process of assessing the penetrance of a mitochondrial mutation, it is necessary to take into account the population dynamics also at mtDNA level (organelle, cell). I might be missing something in my considerations, so I would love to know alternative/complementary points of view.

In the future, it might be interesting to sequence samples from the populations included this study at different generations, to assess: 1) if compensatory mutations are arising in genes encoding subunits of the respiratory chain; 2) the frequency of TF and "wild-type" haplotypes, and if compensatory mutations are present also at the mtDNA level (e.g.: a mutation in cytb that compensates Ala278→Thr). Point 1) could be achieved by targeted deep sequencing of subunits of the electron transport chain (+ ATP syntase), and point 2) by deep sequencing of mtDNA.

Important note: I am including the following references only for the sake of scientific debate. I am not suggesting the authors to include such references in the Manuscript.

References:

1] Ghiselli, F. et al. Structure, transcription, and variability of metazoan mitochondrial genome: perspectives from an unusual mitochondrial inheritance system. Genome Biol. Evol. 5, 1535-1554 (2013).

2] Milani, L. & Ghiselli, F. Mitochondrial activity in gametes and transmission of viable mtDNA. Biol. Direct 10, 22 (2015).

We thank the reviewer for valuable comments, and welcome the reviewer’s thoughts on our results.Ultimately, we believe the modest levels of TFT-mediated population suppression in Experiment 1 were due to most females mating multiple times, with the negative effects of mating to a TFT male buffered by many females mating with both TFT and wild type males; thus leading to a compensation of reproductive success in the females. In Experiment 2, we observed no detectable population suppression at all, as the reviewer points out. We believe that the sheer magnitude of changes in population size between single generations (via population contraction and expansion, on average 30% per population per generation) in this second experiment were likely to have masked the more modest phenotypic effects (average of 8%) associated with the TFT mutation. We were able to show that despite strong population expansions and contractions, the introduced TFT haplotype was nonetheless stably maintained in most cases.

The reviewer suggests that these results could plausibly be explained by compensatory effects associated with each of the TFT mtDNA and wild type DNA segregating in heteroplasmy in different frequencies at the level of the organelle, cell and the individual; and/or the occurrence of compensatory modifier alleles either in the nuclear DNA sequence or the mtDNA. We agree with the reviewer that this explanation is hypothetically plausible, but believe that in our case, it is an unlikely scenario. When it comes to mitochondrial population dynamics, generally a mutant mtDNA molecule must reach between 60 and 80% frequency within a given tissue, relative to the wild type molecule, to exert its phenotypic effect, and this can lead to selection favouring shifts in the frequencies of mutant to wild type mtDNA molecules segregating within the cells of a single individual. As such, the reviewer suggests that future analyses of population dynamics of mtDNA variants would benefit from investigation across multiple levels – from organelle, to cellular, to individual. We agree that future analyses would benefit from deep-sequencing approaches that might be able to fully resolve the population dynamics of mtDNA evolution across organelle, cellular and individual levels. However, in our study, we detected occurrences of heteroplasmy in just a few cases, thus indicating that if heteroplasmy did commonly exist across our experimental flies, then it occurred at levels that were too low to be detected by conventional genotyping, and thus at levels that are unlikely to have yielded significant evolutionary implications (or implications for our experiment).

What about the interpretation that the modest effects of the TFT seen in Experiment 1, and the lack of effects in Experiment 2, could be caused by adaptive compensatory modifiers that have evolved either in the nuclear or mtDNA sequences? This interpretation has merit but, it is unlikely to explain our observed patterns (a suppression effect in Experiment 1, but not Experiment 2), given that the conditions under which Experiment 2 were run were much more restrictive than Experiment 1, in terms of the capacity for adaptive compensatory variants to evolve. That is, the population fluctuations that occurred throughout Experiment 2 would greatly magnify the influence of genetic drift in shaping allele frequencies across generations, relative to the conditions imposed in Experiment 1. Thus, compensatory adaptive variants should have been much less likely to evolve in Experiment 2 than Experiment 1. Although the overall TFT-mediated suppression effect is arguably modest in Experiment 1, we observed no clear signal of a compensatory response across 10 generations of the experiment (the 50 and 75% TFT treatments led to gradual reductions in population size with advancing generations, with no subsequent convergence, relative to the control treatment). Under a compensatory coevolution scenario, we would predict initial divergence between the control treatment (0% TFT) and the treatments with high starting frequencies of TFT variants, followed by a convergence, which would be indicative of adaptive compensation evolving as the experiment progressed across generations. In Experiment 2, we saw no such divergence in population sizes between the TFT and control treatment in the early generations of the experiment, suggesting there would never have been strong selection for a compensatory response.

We have currently not included this response in our revised manuscript as we believe its addition would distract from the discussion of our main finding. However, we would be happy to include our response if the editor feels that this addition would improve the manuscript.

Reviewer #3:

[…] I think the experiment was well conducted and the results well described but there are a few points that I feel require discussion or correction, outlined below.

Overall, the Introduction and Materials and methods are well described. One minor point – the reference to the fertility phenotype in the fifth paragraph of the Introduction refers to Yee et al. 2013 (male fertility), not Innocenti et al. 2011 (gene expression, which does not include any male fertility phenotypes).

We have corrected the reference from Innocenti et al. 2011 to Yee et al. 2013.

One main point I have an issue with is the assertion that the defect is caused by the point mutation in the mtDNA CytB gene, which is repeatedly stated. While previous studies have suggested this association, there is no a priori reason to assume this mutation is any more important than the number of other mutations private to the Brownsville haplotype. For example, a study of mtDNA sequence variation from the same group (Wolff et al. 2015) provides sequence data across the whole mtDNA molecule and the regulatory D-loop alone harbors seven mutations that are private to Brownsville (D-loop alignment positions 303, 1466, 1469, 3244, 3250, 3251, 3709). Since these mutations are linked to the putative CytB mutation, there is insufficient evidence to suggest this mutation is the smoking gun. I think the use of safer description is warranted throughout and I suggest that the authors use the terms 'associated' and 'Brownsville haplotype' rather than the putative Ala278->Thr mutation.

We agree with the reviewer that there is no definite evidence that the mt:Cyt-b causes male-sterility, and that additional mutations unique to the Brownsville haplotype located within the AT-rich region may cause the observed phenotype. The mt:Cyt-b has previously (and here) been highlighted as the putative candidate mutation because it is the only mutation that causes a non-synonymous change. Thus, following a classical genetics model, this mutation is the only mutation with immediate consequences at the protein level in a functionally important enzyme complex. However, we acknowledge that this view is incomplete and perhaps too simplistic, and have thus incorporated the reviewer’s concern/recommendation into the manuscript. We have changed the wording throughout the manuscript (including the title), now referring to the TFT haplotype instead of the mt:Cyt-b mutation. We have further provided additional information in regard to the mutation harbored by the TFT haplotype and, in the Discussion, we have further outlined that future experiments employing genome-editing technologies will be required to unambiguously single out the fertility-reducing mutation(s) harbored within the Brownsville haplotype (Introduction, sixth paragraph; Discussion, seventh paragraph).

It would be advantageous to know how much autosomal sequence variation there was in the Dahomey background used in this study (how outbred the lab-maintained stock really is, especially given the egg selection routinely used). Previous studies have shown near isogenic lines modify the effects of the Brownsville mtDNA, yet there is no assessment here; the Dahomey nuclear background also tends to exaggerate the Brownsville mtDNA haplotype effects. The magnitude of suppression effects is likely sensitive to nuclear background and only one was assessed here, therefore the generality of the finding is unknown. After all, Brownsville, TX, has a viable fruit fly population and this could form the basis of some discussion because it is the most 'natural' experiment. The Discussion, fifth paragraph, is unsubstantiated. Do you have any estimates of autosomal genetic diversity in these lines? If so, I would suggest including them, or at least discuss what may be expected in nuclear backgrounds that are not so sensitive to mtDNA effects, and which are likely to be experienced in a natural setting.

We agree that the magnitude of population suppression is sensitive to the nuclear background. We have previously examined the effect of the Brownsville haplotype on male fertility across several nuclear backgrounds and across different thermal environments. In previous experiments, we have shown that the Brownsville haplotype confers sub-fertility across a range of isogenic nuclear backgrounds; but also a range of “outbred” backgrounds in which the Brownsville haplotype had been placed inside of many flies, each fly of which contained a unique, but “population-representative” genotype that had been captured from one of three different lab-maintained populations (see Dowling et al. 2015, Evolutionary Applications). In these cases (where the Brownsville mtDNA was tested in an “outbred” background), we have verified that its fertility-impairing effect is general across 100s of different nuclear backgrounds. We also agree with the reviewer’s assessment that the Dahomey background tends to exaggerate the effect of the Brownsville haplotype relative to some other sample nuclear backgrounds (i.e. Coffs Harbour, LHM), but not relative to the Brownsville nuclear background or a nuclear background from Puerto Montt; also, note that we maintain backgrounds in our lab in which the Brownsville haplotype confers complete male-sterility.

In the current study we introgressed the Brownsville mtDNA haplotype into replicated outbred populations of the Dahomey laboratory population. As we have outlined above, in our response to reviewer 1, we did not directly measure the nuclear allelic variation maintained at specific genetic loci in our experimental populations. Rather, our inferences are based on the results of numerous quantitative genetic studies that routinely use the Dahomey population used here, and which have shown high amounts of additive genetic variance underpinning the expression of numerous life-history traits (e.g. Gardner et al. 2005 Genetics 169, 1553-1571; Griffin et al. 2016 G3 6, 3903-3911). Since receiving the Dahomey population in our lab in 2010, we have kept it at very large effective population sizes (about 900 adult flies per generation), to ensure that there is no bottleneck in allelic variation. When creating the six replicate Dahomey populations used in the current study (three of which harboured the BRO haplotype, and three of which harboured the DAH haplotype), we were careful to ensure that we did not put our populations through a bottleneck that would sizeably reduce levels of genetic variation. To this end, during the introgression procedure, we backcrossed 45 females of each replicate population to 50 males of the Dahomey laboratory population each generation. These numbers per population replicate are high when benchmarked against most studies utilizing experimental evolution approaches (e.g. artificial experiments in which an investigator seeks to maintain high levels of segregating genetic variance across numerous replicate population, while subjecting them to one of two divergent selection regimes; see Innocenti et al. 2014 BMC Evolutionary Biology 14:239; Bolstad et al. 2015 PNAS, 112-13284-13289. Thus, our approach is gold-standard from an experimental evolution perspective, and utilizes a well-studied laboratory population of Drosophila that has previously provided numerous insights into the adaptive capacity of populations. We are confident the experimental populations we have created each harbour high levels of segregating nuclear variance that are all highly representative of levels in the stock Dahomey population we continue to maintain in our laboratory.

Finally, it was not our intention to suggest that we characterized the nuclear genetic variation maintained in the Dahomey population in this experiment, but we see how the wording in the previous version of the manuscript could have led to misunderstandings. We have incorporated the additional information regarding the Dahomey nuclear background (Discussion, fourth paragraph; Subsection “Fly strains”, first paragraph), and suggestions regarding the incorporation of additional backgrounds, into the discussion (Discussion, fourth paragraph).

Is a -8% and stable population suppression biologically meaningful in the context of pest control, especially when it is only observed in a tightly regulated laboratory population? The Discussion mentions why the results may differ from the predicted values (and were somewhat less than expected) but the usefulness of the technique is somewhat limited (based on these findings). I think this could benefit from more discussion and especially for the role of egg-to-adult survival in the initial generations, which is known to be above average in Brownsville mtDNA lines.

The primary goal of the research presented here was to provide proof-of-concept for the feasibility of using the TFT to suppress population size. We achieve 8% suppression under the conditions of Experiment 1. Nonetheless, as noted by the reviewer, this effect was lower than what we had been expecting. However, our a-priori demographic modelling predicts that the detected effects in our study are likely to underrepresent the likely efficacy of the TFT in wild populations where mating rates are likely to be much lower. The inclusion of the above-average pupal viability for the Brownsville haplotype is an excellent suggestion for further discussion, and we have now incorporated this into the manuscript (Discussion, sixth paragraph).

We also note that higher efficiencies in future development of the TFT are likely to be achievable via the combination of several fertility-reducing mutations within a single TFT haplotype, or via the release of multiple TFT strains harbouring distinct TFT haplotypes. Since the writing of this manuscript, another TFT mutation has been discovered independently by Patel and colleagues (Patelet al. 2016. A mitochondrial DNA hypomorph of cytochrome oxidase specifically impairs male fertility in Drosophila melanogaster. eLife, 5, e16923). This finding provides support for the notion that animal mitochondrial genomes should be naturally enriched for male-harming mtDNA mutations, and which can be harnessed to further increase the efficiency of the TFT. We have thus incorporated these new findings and elaborated on how these mutations can be harnessed to further develop the TFT. We have also incorporated the reviewer’s suggestion relating to multiple-release strategies (Discussion, seventh paragraph).

Is the experiment #1 population size of 80 arbitrary, or was this based on the previous simulations? At such small population sizes, the opportunity for drift is very high and coupled with low and uneven sampling for the haplotype frequency estimates across both experiments, could this influence the results? Although there was vial replication, could these estimates not suffer from the same systematic bias? On the same note, the removal of flies from egg laying in experiment #2 (when approximately 50% of the vials produced 80 eggs) is still quite artificial and not what would be experienced in nature. This deserves some discussion since even a mild deviation from a strict egg laying dynamic can nullify the suppression effects found in experiment #1. It is perhaps unfortunate the reciprocal 100% Brownsville mtDNA treatment was not used as a control. While it would not be necessary or appropriate to use pest control in a population of zero flies, it would give a good estimate of the expected population size variation in the experimental design used here, which differs from previous studies using the Brownsville mtDNA haplotype. If these data are available, I would encourage the authors to include them to aid results interpretation.

The upper limit of 80 eggs/larvae/individuals per vial was chosen because at this density fly populations are sufficiently large to be stably maintained while limiting nutritional stress (e.g. food scarcity) which otherwise may impact development, fitness and behavior. Accordingly, in the first experiment, all populations were started with 80 eggs in each generation.

We now realize that we have failed to state this clearly, and have amended the manuscript to provide clarification in the Materials and methods section (subsection “Fly strains”, last paragraph).

Our a-priori demographic modelling, which informed our TFT treatments, was parameterised based on a population size of 140. However, sensitivity analyses showed that predictions were robust with respect to modelled lab population size of 80 individuals. We have added this additional information as supplementary material. As we have argued in our response to the previous comment, sample sizes of ~ 40 breeding pairs represent the gold standard in experimental evolution studies of invertebrates. Since egg-to-adult viability rates are high in D. melanogaster, we expect that the effective population sizes arising from the 80 eggs per generation in Experiment 1 were high, and thus the efficacy of natural selection would have greatly outweighed the effects of drift. The fact that our high TFT treatments steadily diverged from the other treatments, and that this was maintained across the 10 generations of the trial, suggests that genetic drift was not the predominant driver of haplotype frequencies per vial. Critical to this discussion, however, is the fact that we had replicated each of our treatments across three strains, and within each strain we had 7 experimental populations. Thus, each treatment was effectively replicated 21 times. Because the effects we detected were consistent (i.e. the effects detected statistically were strong, and the patterns were upheld across vials), this confirms that the haplotype frequencies across vials were not determined by drift (since this would have led to random haplotype frequencies across vials across the treatments – detectable via large vial-specific or replicate-specific effects, and no TFT-mediated effect). We can confidently conclude that the results of Experiment 1 are not affected by genetic drift.

We also agree that stopping ovipositioning in Experiment 2 when around 50% of the population vials contained in excess of 80 eggs may be a simplistic attempt to imitate the complex population dynamics of natural populations. However, this experimental design was used as an approximation for the demographic conditions natural populations may be exposed to in terms of contractions and expansions of population size, where populations potentially transition through severe population bottlenecks which may affect the frequencies of co-occurring mtDNA haplotypes via drift.. We have highlighted our intention and associated limitations in the manuscript (subsection “Experimental design”, last paragraph). Thus, in contrast to Experiment 1, haplotype frequencies in experimental populations of Experiment 2 were much more likely to be affected by drift. Remarkably, however, while we did not detect suppression of population sizes at the higher TFT treatments in this second experiment, the frequencies of the TFT haplotypes were nonetheless stably maintained in the 50 and 75% treatments; thus suggesting that even under these conditions, drift was not the primary determinant of the competing mtDNA haplotypes.

While we agree with the reviewer that adding a 100% TFT treatment would be interesting, we were under heavy logistical constraints since we were already maintaining 164 experimental populations, and were working at the upper level of what was possible in the lab. We thought it much more valuable to have the 0% control, with which we could compare the effects of the other TFT-treatments to. The 100% treatment would have been informative, since it would have led to the greatest selection for compensatory adaptations (given that all males would have been carrying the male fertility-reducing haplotype). These data are not available however. We have added a section to highlight that inclusion of this approach may aid the interpretation in future experiments (Discussion, fourth paragraph).

The discussion of using genotypes with relatively high fitness females (Discussion, last paragraph) with low fitness in the corresponding males seems counterproductive since the object of pest control here is surely to reduce the number of insect vectors (not increase them!)

We see the reviewer’s point, and have changed the wording in this section to clarify that population suppression is the ultimate goal (Discussion, last paragraph). However, a mutation that confers high female fertility, but has devastating effects on male fertility, is potentially the perfect TFT mutation because the high female fertility will act to “drive” the TFT mutation into the pest population. Once the TFT mutation is at high frequencies, most males would have reduced fertility, and populations would crash.

Perhaps wrap up with a more general commentary of the problems of evolution in gene drives as a pest control and how this technique can be tailored to avoid the same pitfalls – just a minor thought, but I feel this would compliment the nuclear compensation argument.

We agree, and have added a section to the Discussion, highlighting/discussing the problem of the emergence of resistance alleles in response to both the TFT and gene-drive systems (Discussion, fourth paragraph).

https://doi.org/10.7554/eLife.23551.017

Article and author information

Author details

  1. Jonci Nikolai Wolff

    1. School of Biological Sciences, Monash University, Victoria, Australia
    Contribution
    JNW, Data curation, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    For correspondence
    1. jonci.wolff@gmail.com
    Competing interests
    The authors declare that no competing interests exist.
    ORCID icon 0000-0002-8809-5010
  2. Neil J Gemmell

    1. Department of Anatomy, University of Otago, Dunedin, New Zealand
    Contribution
    NJG, Conceptualization, Funding acquisition, Writing—original draft, Writing—review and editing
    Competing interests
    The authors declare that no competing interests exist.
    ORCID icon 0000-0003-0671-3637
  3. Daniel M Tompkins

    1. Landcare Research, Dunedin, New Zealand
    Contribution
    DMT, Conceptualization, Formal analysis, Funding acquisition, Writing—original draft, Writing—review and editing
    Competing interests
    The authors declare that no competing interests exist.
  4. Damian K Dowling

    1. School of Biological Sciences, Monash University, Victoria, Australia
    Contribution
    DKD, Conceptualization, Formal analysis, Funding acquisition, Writing—original draft, Writing—review and editing
    Competing interests
    The authors declare that no competing interests exist.
    ORCID icon 0000-0003-2209-3458

Funding

Ministry of Business, Innovation and Employment (New Zealand, Smart Ideas Grant)

  • Neil J Gemmell
  • Daniel M Tompkins
  • Damian K Dowling

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Belinda Williams and Winston Yee for their help with fly husbandry, members of the Tompkins, Gemmell, and Dowling groups for helpful comments on the manuscript, and the three reviewers for their constructive comments and suggestions. This work was funded by a Smart Ideas grant from the New Zealand Ministry of Business, Innovation and Employment (MBIE).

Reviewing Editor

  1. Marcel Dicke, Reviewing Editor, Wageningen University, Netherlands

Publication history

  1. Received: November 22, 2016
  2. Accepted: April 27, 2017
  3. Accepted Manuscript published: May 3, 2017 (version 1)
  4. Version of Record published: May 23, 2017 (version 2)

Copyright

© 2017, Wolff et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 609
    Page views
  • 188
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: PubMed Central, Scopus, Crossref.

Comments

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Developmental Biology and Stem Cells
    Cyrille Ramond et al.
    Research Article