The effectiveness of a gene drive system is dependent on multiple factors beyond its selfish induction of incompatibility, including genotype frequency, outcrossing rate, and fitness in the host background. To explore these parameters, we adapted a family-based model (Figure 1D, Table 1; Wade and Beeman, 1994) with modifications to account for paternal delivery of the toxin, selfing versus outcrossing rate, and selection cost of the element.

Under a simple scenario of no fitness consequence to the host genotype (s = 0) and a completely outcrossing population (k = 1), the element spreads rapidly through the population with a maximum allele change comparable to an additive beneficial allele with a selection coefficient of 0.44 (Figure 1E), 2–4 times higher than the selection coefficient of lactase persistence in humans (Bersaglieri et al., 2004). However, gene drive is weaker than the beneficial allele at the tails of the allele frequency range: at low frequency, the rarity of the element limits how fast it spreads; at high frequency, the rarity of the vulnerable genotype slows its approach to fixation. If the element induces a carrying cost to the host genotype (e.g. s = 0.3, s = 0.6), for example, via energy expenditure or ‘leaky’ toxicity, the dynamics at the extreme allele frequencies are amplified (Figure 1F). At low frequency, the carrying cost counteracts gene drive, reducing the likelihood that the element reaches appreciable frequency by genetic drift before being lost. At high frequency, the carrying cost compounds the slowing rate of gene drive such that it reaches a stable equilibrium and does not fix.

Previous models have shown that spread of a TA element accelerates with the rate of outcrossing (Noble et al., 2021). Given a substantial carrying cost to the host genotype (s = 0.35), a TA element is likely to increase in frequency only under relatively high rates of outcrossing (Figure 1G). Under outcrossing rates (~1%) typical for C. elegans (Barrière and Félix, 2005; Frézal and Félix, 2015), the element will likely be lost from the population under all but the mildest carrying costs (0.008) (Figure 1—figure supplement 1), as increasing fitness costs require increasing outcrossing for the element to reach a stable equilibrium (Figure 1H).

Given these dynamics, we are challenged to explain how a novel TA element could rise in initial frequency in a population. One hypothesis is that TA elements in non-obligate outcrossing Caenorhabditis may have originated in an outcrossing ancestor, then persisted by other evolutionary forces such as drift or balancing selection (Noble et al., 2021; Seidel et al., 2011; Sweigart et al., 2019). Such a scenario is consistent with the recent opinion by Sweigart et al., 2019, who argue that TA elements may exist in nature with only incidental instances of ‘selfish’ activity. This shift away from the conventional framing of TA elements as consistently selfish makes sense in the context of non-obligate outcrossing populations, which permit elements to proliferate in sequestered lineages without conflict.

To investigate its potential to spread through the population without conflict, we evaluated the fitness consequences of the zeel-1;peel-1 element independent of its incompatibility cost in heterozygotes. First, we employed a previously described fitness assay (Large et al., 2016; Zhao et al., 2018) to compete N2^{zeel-1;peel-1(CB4856)}, which carries an ~140–370 kb interval spanning the zeel-1;peel-1 locus from CB4856 introgressed into N2 (Ben-David et al., 2017), against N2^marker, a modified version of N2 carrying a silent marker mutation in the dpy-10 gene. As CB4856 harbors the inactive haplotype, N2^{zeel-1;peel-1(CB4856)} lacks the toxin/antidote element, while N2^marker carries the active element native to N2. In these assays, males are not present and outcrossing is prevented, so relative fitness is estimated from true-breeding hermaphrodite genotypes. As a positive control, we used the N2^{glb-5;npr-1(CB4856)} near-isogenic lines (NILs) strain, which carries ancestral alleles of glb-5 and npr-1 that decrease fitness of animals in laboratory conditions (Zhao et al., 2018).

N2^marker outcompeted N2^{zeel-1;peel-1(CB4856)} (Figure 2A), with a relative fitness (w) of 1.18 (1.15–1.21, 95% CI). Association of the active allele with higher fitness suggests that induction of peel-1 toxicity and/or rescue by zeel-1 is not costly, that the active allele is linked to one or more mutations in the N2 background that confer an independent fitness advantage, or both. These mutations could reside within zeel-1;peel-1, within the four nearby genes within the high-diversity region, or outside the high-diversity region but within the 140–370 kb introgressed region of this strain (Figure 1B). We also measured fecundity and body size in N2 and N2^{zeel-1;peel-1(CB4856)} directly and observed similar outcomes: N2 laid 9% more embryos (p<0.001, Figure 2B) and was 9% larger 72 hr after hatching (p<0.001, Figure 2C), indicating animals grew faster, resulting in a larger body size at a similar time in development.

These results indicate that variants associated with the active zeel-1;peel-1 haplotype promote fitness in the host genotype, providing a potential mechanism for proliferation and persistence of the element in selfing lineages.

To test the fitness consequences of the peel-1 toxin directly, we used CRISPR/Cas9 to engineer a knockout of peel-1 (kah126, or peel-1(trunc)) in the N2 background. N2^{peel-1(trunc)} produces a truncated protein of 46 amino acids (relative to 174) via an early stop codon (Figure 3A). We verified loss of function by embryo killing assays: N2 crossed to CB4856 produced the expected 25% embryonic lethality from selfed F1 hermaphrodites; the N2^{peel-1(trunc)} cross produced zero dead embryos (Figure 3B). Interestingly, the peel-1(trunc) allele affected fitness proximal traits and fitness in laboratory conditions. The N2^{peel-1(trunc)} produced 6% fewer offspring (Figure 3C) and were 7% smaller 72 hr after hatching than N2 (Figure 3D). Competition experiments between N2^{peel-1(trunc)} against N2^marker also demonstrated a fitness increase associated with the active peel-1 allele (w = 1.06, 1.04–1.07, 95% CI) (Figure 3E); this fitness difference accounts for 32% of the difference arising from the N2^{zeel-1;peel-1(CB4856)} comparison. Thus, while peel-1 acts as a toxin in the context of outcrossing cross-progeny, it increases the fitness of selfing hermaphrodites in laboratory conditions. These results suggest that peel-1 is not simply a toxin gene and plays some other biologically relevant role in C. elegans. The fitness differences may be mediated via egg-laying rate. The higher total fecundity suggests that the number of self-sperm produced differs among strains, which would also affect the earliest timepoint eggs may be laid; additional experiments are needed for confirmation. These results also suggest that additional genetic variations linked to the zeel-1;peel-1 locus play a role in laboratory fitness as the peel-1 mutations did not fully phenocopy the fitness of the peel-1 NIL line.

Our work indicates that peel-1 plays an additional biological role outside of its role as a selfish element. Since the experiments on peel-1 relied on a single CRISPR/Cas9-generated strain, we were worried that background mutations could account for the differences in fitness and fitness-proximal traits of this strain. To address this, we generated six additional alleles modifying peel-1 (Figure 4). First, we created two replicate alleles (kah1000 and kah1001) that revert the original peel-1 mutant allele (kah126) back to wild-type. Second, we created three replicate alleles (kah1003-5) with an edit in the third codon to induce an early stop. Finally, we created an allele (kah1006) with a 5 bp deletion that excised the ATG start codon. Unlike the original N2^{peel-1(trunc)} strain, with a stop codon in the third exon that could potentially lead to a truncated protein product, these latter four strains are predicted to create true null alleles that should prevent the production of any peel-1 protein. We verified peel-1 activity using embryo killing assays on one strain of each allele type; as expected, self-progeny of the heterozygous offspring of the strain with the reversion allele of peel-1 showed 25% lethality, and self-progeny from crosses with the loss-of-function alleles of peel-1 showed zero lethality (Figure 4C). We tested these strains in competition experiments and observed equivalent performance among the replicate genotypes within the reversion and early stop allele classes, and further, equivalent performance among all peel-1 mutants (Figure 4D). As expected, strains carrying the reversion alleles showed no significant difference in fitness in competition with wildtype, while strains carrying the new loss-of-function alleles, like the original N2^{peel-1(trunc)} strain, were significantly outcompeted (Figure 4D). These experiments strongly support a role for peel-1 outside of its role as a selfish element. We did not test these additional strains to confirm that the egg-laying and growth rate phenotypes that we measured in the original peel-1 loss-of-function, so we cannot exclude the possibility that this phenotypic difference is due to background mutations.

This is not necessarily surprising, as the role of peel-1 in a secondary biological process was considered in its initial characterization (Seidel et al., 2011). Such a role would help the initial spread of the element during its formation, when its low frequency (where gene drive is ineffective) and its initial toxicity (before zeel-1 could evolve to counteract it) should prevent its spread. Our work supports that model, suggesting that both roles of peel-1 could co-evolve together. But then, why has not the element fixed? The zeel-1;peel-1 locus shows a signature of balancing selection, which appears widespread in C. elegans. Hyperdivergent regions, including that spanning zeel-1;peel-1, punctuate the genome; balancing selection across diverse ecological niches may explain their maintenance (Lee et al., 2021). Previously, maintenance of the zeel-1;peel-1 element was hypothesized to arise from tight linkage to a nearby polymorphism under balancing selection (Seidel et al., 2008). Our results suggest that peel-1 could be under balancing selection itself. peel-1 confers a fitness benefit within the lab environment, and it may pleiotropically influence other life history traits or affect fecundity and growth rate differently in different environments, providing alternate fitness strategies for local adaptation.

Previous work has suggested that TA elements may shape evolution by promoting selfing to escape the cost of selfish gene drive (Noble et al., 2021). Here we provide a mechanism for their spread and maintenance that helps to explain their prevalence in selfing Caenorhabditis (Ben-David et al., 2021; Noble et al., 2021; Sweigart et al., 2019). Moreover, our observation of a toxin directly affecting biological traits mirrors work in transposable elements, which are also selfish elements that can be domesticated for phenotypic benefit to the organism (Werren, 2011). This previously undescribed, non-toxin related role of a TA element is expected to shape evolutionary trajectories of both the element and the organism.

In the future, it will be interesting to parse the mechanism by which peel-1 affects fitness in the adult hermaphrodite. peel-1 transcripts are restricted to sperm (Seidel et al., 2011), but these results suggest that it can affect adult phenotypes like fecundity and growth rate. One possibility is that the PEEL-1 protein persists to adulthood and affects cellular function despite its presence at very low levels. Alternatively, PEEL-1 may induce long-lasting effects in adulthood via early developmental processes; notably, the toxic effect of PEEL-1 arises late in embryogenesis (Seidel et al., 2011). As the toxicity of PEEL-1 depends on the sex of the sperm donor, mediated in part by dosage (Seidel et al., 2011), PEEL-1 may function as a mechanism to communicate parentage to the offspring. Additionally, it is possible that environmental conditions of the parent regulate PEEL-1 levels, which could also be communicated to the offspring.

We have brought genomic editing and experimental evolution resources to bear on the study of a toxin-antidote element, addressing long-standing questions about their origin and maintenance in populations. We discovered that peel-1 plays a role in individual fitness outside of its role as a toxin, affecting growth, fecundity, and fitness of non-hybrid genotypes, supporting recent arguments that non-selfish activity in inbred lineages may explain the prevalence of TA elements in non-obligate outcrossers (Noble et al., 2021; Sweigart et al., 2019). This work adds to the complicated nature of ‘selfish’ gene elements, similar to work in bacteria that has shown that TA elements can provide fitness benefits such as improved antibiotic resistance (Bogati et al., 2022). We hypothesize that other TA elements identified in Caenorhabditis species will also play roles outside of outcrossing, explaining how they can be retained in non-outcrossing populations.

Strains were cultivated on agar plates seeded with Escherichia coli strain OP50 at 20°C (Brenner, 1974). The following strains were used in the study.

CRISPR/Cas9 was used following a previously published co-conversion method to edit the target gene and dpy-10 gene at the same time (Arribere et al., 2014). Generated strains are outcrossed to N2 more than three times before used for assay. Information on the N2 genome came from https://wormbase.org// and CeNDR (Cook et al., 2017). The following primers/sequences were used to create the CRISPR/Cas9 strains:

All code to control population dynamics parameters and then plot the trajectories were stored at https://github.com/lijiang-long/TA_modeling (copy archived at Long, 2023). To calculate the allele frequency change at different frequencies of zeel-1;peel-1, the population is initiated with Hardy–Weinberg equilibrium such that the frequency of homozygous zeel-1;peel-1 is the square of its allele frequency, and so on and so forth. The frequency of each genotype is updated each generation using the family-based toxin-antidote evolution dynamics in Table 1. This population is allowed to evolve five generations to deviate from Hardy–Weinberg equilibrium and reach the evolution trajectory of zeel-1;peel-1. The population evolves another generation, and the allele frequency change in this generation is used for plotting. To generate the heatmap where the frequency of zeel-1;peel-1 after 1000 generations is plotted against varying outcrossing rate and fitness cost, the population is initiated with half zeel-1;peel-1 allele. The genotype frequency is calculated assuming Hardy–Weinberg equilibrium. The population then evolves 1000 generations following Table 1. The final allele frequency of zeel-1;peel-1 is then plotted on the heatmap.

Competition experiments followed previous work (Zhao et al., 2018). Pairwise competition assays in Figures 2 and 3 were done in parallel with the same start date. The competition assays in Figure 4 were performed parallel with a different start date. All pairwise competition assays were performed on 9 cm NGM plates, seeded with OP50 bacteria, and stored at 4°C until 24 hr before use. At the beginning of the experiment, 10 L4 worms of each strain were transferred onto the same plate. This plate was then incubated at 20°C for 5 d. To propagate the next generation, a 1 cm agar chunk was transferred to a new 9 cm NGM plate. The old plate was then washed with 1 ml of M9 buffer to collect worms and stored at –80°C. Subsequently, this transfer and collection procedure was held every 3 d for a total of seven transfers. The genomic DNA from the first, third, fifth, and seventh transfer was isolated using Zymo 96-well DNA isolation kit (cat# D4071). Isolated genomic DNA was fragmented using EcoRI-HF by incubation at 37°C for 4 hr and purified using a Zymo 96-well DNA purification kit (cat# D4024). After purification, DNA concentrations were measured using Qubit DNA HS assay and adjusted to 1 ng/μl. To quantify the relative proportion of the two strains, a previously designed TaqMan probe was used targeting the dpy-10 gene. After this, the DNA and TaqMan probe were mixed with the digital plate PCR (ddPCR) mix and processed through standard ddPCR procedures. The fractions of each strain were quantified using the BioRad QX200 machine with standard absolute quantification protocol. To estimate relative fitness, a linear regression model was applied to the DNA proportion data using the following equation with the assumption of one generation per transfer:log(p(a)0p(a)t−p(a)01−p(a)0) = (log(WaaWAA))t

where p(a) represents the relative allele proportion calculated from the ddPCR fluorescence signal and W_aa and W_AA represent the estimated fitnesses of the competing genotypes.

On day, a 10 cm NGM plate with plenty of gravid adults was bleached following standard protocol for each strain. Embryos were deposited to a 10 cm plates and incubated a 20°C. On day 2, 4–5 L3, young L4s hermaphrodites, and 8–10 CB4856 young L4 males were transferred to a mating plate for each of strain. Plates were incubated a 20°C. On day 5, adult hermaphrodites were singled on 6 cm plates (four plates per strain) and incubated at 20°C. on day 7, Plates were checked for males (F1) to determine if cross was successful. 20–30 F1 L4s from successfully crossed F0 herms were transferred to a 6 cm plates. Plates were incubated at 20°C. On day 8, four adult egg-laying adults (F1) were transferred to a 6 cm ‘assay’ plate for each replicate (six replicates per strain). Adults were on plates for 4 hr at room temperature. Adults were removed and plates incubated at 20°C for 18 hr. On days 9 and 12, dead embryos (F2) and adult worms were counted on days 9 and 12, respectively, and embryonic lethality was calculated for each replicate.

Fecundity assays were performed at 20°C using 3 cm NGM plate seeded with 50 μl of OP50 bacteria with OD₆₀₀ of 2.0. The plates were allowed to dry overnight and stored at 4°C until 24 hr before use. At the beginning of the assay, six fourth larval stage (L4) worms were transferred to each assay plate. The worms were allowed to grow and lay eggs for the first 24 hr after the assay began before being transferred to a new plate. This process was repeated every 12 hr thereafter until animals ceased laying eggs. The number of eggs laid was counted using a standard dissecting microscope. This process is repeated every 12 hr thereafter until 100 hr or there is no egg on the new plate. The average fecundity was calculated by summing over all time points and dividing by the total number of worms in a single assay plate. While the data was initially collected every 12 hr, only the total fecundity was recorded for each assay.

Growth rate assays were performed on standard NGM plates seeded with OP50 bacteria as previously described (Large et al., 2016). At the beginning of the assay, 10–20 adult worms were transferred onto an assay plate to lay eggs. After 2 hr, they were transferred off of the plate, leaving ~80 eggs per plate. The plates were incubated for 72 hr at 20°C. At this point, the assay plate was mounted onto a video tracking camera and recorded for 1 min. The video clip was analyzed using a customized MATLAB script that tracks each animal and calculates the average size of each worm. The average size from each plate was then normalized by the average size of three N2 plates.

All hypothesis tests were performed using non-parametric analyses. One-sample comparisons to a null hypothesis value were assessed by the Wilcoxon test, and differences between pairs of samples were assessed by the Mann–Whitney U test. For experiments with multiple comparisons, p-values were adjusted using the Benjamini and Hochberg method. For the fitness competitions testing replicate peel-1 alleles against the wild-type control, we first evaluated each independently derived allele separately; as expected, the reversion alleles (kah1000, kah1001) showed no significant fitness differences while the early stop mutant alleles (kah1003, kah1004, kah1005) each showed a fitness disadvantage (p<0.05 following correction for multiple tests). We also observed no significant differences among independently derived replicates within allele class, so we pooled replicate outcomes for further analyses (as reported in Figure 4D). For these tests, multiple comparison groups included the competition outcome tests and tests of replicate genotypes within allele class.