Rapid decline in fitness of mutation accumulation lines of gonochoristic (outcrossing) Caenorhabditis nematodes

Abstract

Evolutionary theory predicts that the strength of natural selection to reduce the mutation rate should be stronger in self-fertilizing than in outcrossing taxa. However, the relative efficacy of selection on mutation rate relative to the many other factors influencing the evolution of any species is poorly understood. To address this question we allowed mutations to accumulate for ~100 generations in several sets of “mutation accumulation” (MA) lines in three species of gonochoristic (dieocious) Caenorhabditis (C. remanei, C. brenneri, C. sp5) as well as in a dioecious strain of the historically self-fertile hermaprohodite C. elegans. In every case the rate of mutational decay is substantially greater in the gonochoristic taxa than in C. elegans (~4× greater on average). Residual heterozygosity in the ancestral controls of these MA lines introduces some complications in interpreting the results, but circumstantial evidence suggests the results are not primarily due to inbreeding depression resulting from residual segregating variation. The results suggest that natural selection operates to optimize the mutation rate in Caenorhabditis and that the strength (or efficiency) of selection differs consistently on the basis of mating system, as predicted by theory. However, context-dependent environmental and/or synergistic epistasis could also explain the results.

Mutation rate varies within and among taxa, including among individuals within the same population (Sturtevant 1937; Drake et al. 1998; Lynch 2008). In principle, mutation rate may vary among groups descended from a common ancestor for three general reasons. First, the respective environments inhabited by the different groups may differ in mutagenicity (e.g., incident UV radiation), in which case the difference in mutation rates is solely due to environmental effects and does not represent evolutionary divergence, except incidentally. Second, non-adaptive processes (mutation, genetic drift, hitchhiking) may have caused the different groups to evolve different mutation rates. Third, natural selection may have favored different mutation rates in the different groups - i.e., the difference represents the adaptive outcome of divergent optimizing selection.

It is commonly asserted by evolutionary biologists that the mutation rate is subject to optimizing selection, at least in principle (e.g., Fisher 1930; Sturtevant 1937; Kimura 1967; Leigh 1973; Kondrashov 1995; Dawson 1998; Drake et al. 1998; Johnson 1999; Sniegowski et al. 2000; Andre and Godelle 2006; Baer et al. 2007). The orthodox view, initially articulated by Kimura (1967), (but see Lynch 2008) holds that direct selection on a modifier of mutation rate (almost) always favors a reduction in mutation rate so as to reduce the fitness cost of deleterious mutations, but there is also indirect selection to reduce the pleiotropic fitness costs of increasing the fidelity of genome replication. The pleiotropic “cost of fidelity” may be imposed by speed of replication and/or metabolic resources devoted to genome surveillance. The optimum (equilibrium) mutation rate occurs at the point at which opposing direct and indirect selection exactly counterbalance each other.

The extent to which observed variation among groups in mutation rate reflects underlying differences in optimizing selection is unknown, but theoretical principles provide guidance. The two clearest cases concern the role of recombination (~ sexual vs. asexual reproduction) and self-fertilization. Under the assumption that beneficial mutations are trivially rare and can be ignored (which is demonstrably untrue in asexual taxa; Denamur and Matic 2006 and references therein), the strength of direct selection on a modifier allele that reduces the genome-wide deleterious mutation rate (U) by some amount v in (diploid) sexual taxa is $v\stackrel{‒}{hs}$, where $\stackrel{‒}{hs}$ represents the average strength of selection against new deleterious mutations in heterozygotes; in asexual organisms the equivalent strength of selection is v. Comparing obligate selfing vs. obligate outcrossing organisms, the strength of selection in the obligate selfer is v/2 (Kondrashov 1995; Drake et al. 1998). If $\stackrel{‒}{hs}$ is on the order of 1–2% (Lynch et al. 1999), then the strength of selection to reduce mutation rate will be perhaps two orders of magnitude greater in asexual than in sexual taxa, and half that difference in selfers vs. outcrossers. Similar reasoning applies to other situations, e.g. sex-chromosomes vs. autosomes (McVean and Hurst 1997).

Although these theoretical predictions are clear, the relevance of the theory to organisms in nature is not. Which is to say: Do asexual and/or self-fertilizing taxa typically evolve lower mutation rates than their sexual and/or outcrossing relatives? If the answer is yes, then optimizing selection on mutation rate must be taken into account in any context in which it is relevant (e.g., genome evolution). If the answer is no, it suggests that optimizing selection is weak relative to other factors potentially underlying variation in mutation rate (e.g., genetic drift, hitchhiking with beneficial mutations), and the cause(s) of variation must be sought elsewhere.

There is considerable evidence that predominantly selfing taxa harbor less standing genetic variation than their outcrossing relatives (Hamrick and Godt 1996; Glemin et al. 2006; Ross-Ibarra et al. 2008), consistent with (among other things) selfing taxa having evolved lower mutation rates. However, N_e in a pure selfing population is expected to be half that of the same population under random mating (Pollak 1987). Moreover, populations of self-fertilizing organisms are potentially founded by a single individual propagule, with the concomitant drastic reduction in N_e. Finally, and possibly most importantly, selection at linked loci is likely to reduce N_e in selfing populations below that of equivalently-sized outcrossing populations (Hill and Robertson 1966; Gillespie 2000; Wright et al. 2008). Thus, although the preponderance of indirect empirical evidence is consistent with self-fertilizing taxa having evolved reduced mutation rates, other explanations are equally plausible and (arguably) more parsimonious and cannot be ruled out.

Nematodes in the genus Caenorhabditis provide an ideal opportunity to test these theoretical predictions. Self-fertilization has evolved independently in the genus at least three times, in the species Caenorhabditis elegans and C. briggsae (Kiontke et al. 2004) and in the recently discovered C. “species 11” (C. Braendle, personal communication). C. briggsae and C. elegans are both believed to be primarily (but not exclusively) self-fertilizing in nature (Cutter et al. 2006). Several lines of evidence suggest that the mutation rate in C. briggsae is greater than in C. elegans (Baer et al. 2006; Ostrow et al. 2007; Phillips et al. 2009; D. Denver et al., unpublished). Two studies have attempted to compare the genomic deleterious mutation rates (U) of these species with the gonochorisitc (= dioecious) species C. remanei indirectly using base substitution rates at four-fold degenerate sites in coding genes. One (Cutter and Payseur 2003) found no difference between the three species C. elegans, C. briggsae, and C. remanei; a much larger study (Artieri et al. 2008) inferred a modestly (~15%) but significantly lower U in C. remanei than in C. briggsae, contrary to the theoretical prediction.

We report here the results of a comparative mutation accumulation (MA) experiment in which mutations were allowed to accumulate under relaxed natural selection for 100 generations in several highly inbred populations of the dioecious taxa C. remanei, C. brenneri, and C. “species 5”, as well as in a population of C. elegans rendered functionally dioecious. We characterize two mutational parameters, the per-generation decline in mean fitness (ΔM) and the per-generation increase in genetic variance in fitness due to new mutations (V_M). These parameters are composite functions of the genomic mutation rate for the trait (U) and the average homozygous effect of a mutation on the trait (E[a]); ΔM=UE[a] and V_M=UE[a²] (Lynch and Walsh 1998, Ch. 12).

MA studies have advantages and disadvantages relative to indirect inferences from comparative genomics. Two of the most important advantages of MA studies are that generation time and population size are known with certainty; the most obvious disadvantage of MA experiments is that inferences are necessarily drawn from a very small number of genotypes (usually one, typically highly homozygous) in a very small number of environmental contexts (usually one), either of which may not accurately reflect the typical properties of the species. MA and comparative studies are complementary and thus each is more informative in light of the other.

Methods and Materials

Systematics and Natural History of Nematode Strains

We chose three gonochoristic species in the genus Caenorhabditis: C. remanei, C. brenneri, and the unnamed “C. species 5” (Supplementary Figure 1; also see Kiontke et al. 2007). Gonochoristic reproduction is thought to be the ancestral character state in the genus (Kiontke et al. 2007), so the mating system in these species presumably represents the ancestral character state. All Caenorhabditis species are believed to have a XO system of chromosomal sex determination, in which females/hermaphrodites are XX and males are XO (Pires-DaSilva 2007). We have previously characterized the mutational properties of C. briggsae and C. elegans, predominantly self-fertilizing (androdioecious) species in which the androdioecious mating system evolved independently (Baer et al. 2005; Baer et al. 2006; Ostrow et al. 2007; Joyner-Matos et al. 2009; Phillips et al. 2009; Braendle et al. 2010).

Inbred lines of the gonochoristic species were constructed by 20 generations of single-female descent, in which a single gravid adult female was taken as the parent of the next generation. This method allows the possibility of multiple male parents of the next generation, but construction of advanced generation inbred lines of gonochoristic Caenorhabditis by full-sib mating is extremely difficult and usually fails (S. Baird personal communication; P. Phillips, personal communication; CFB, unpublished results; Dolgin et al. 2007). C. remanei was represented by three inbred lines, two of which (PB4642, PB4643) were derived from the EM464 strain and one of which (PB2282) was derived from the PB228 strain. C. brenneri was represented by two inbred lines (PB2801 and PB2802) derived from the LKC28 strain. C. sp5 was represented by two inbred lines, JU727.3 and JU727.5, derived from the JU727 strain. Lines designated “PB” were inbred and kindly provided by Scott Baird; JU727 was kindly provided by Marie-Anne Félix and inbred in our lab. All ancestral stocks are descended from wild isolates that were cryopreserved after only a few generations in the lab, thus the opportunity for adaptation to the lab and/or loss of ancestral variation was minimal.

To account for possible systematic differences in mutational properties as an inherent function of mode of reproduction per se (as opposed to evolutionary history), we constructed a strain of C. elegans in which the capability for self-fertilization was blocked. fog-2 is a recessive mutation that destroys spermatogenesis in hermaphrodites, thereby rendering hermaphrodites functionally female (Schedl and Kimble 1988). We introgressed the fog-2 allele into the canonical wild-type N2 genetic background (Vassilieva and Lynch 1999; Baer et al. 2005) by backcrossing fog-2 homozygotes to N2 males for 12 generations. fog-2 homozygotes were recovered by first allowing the offspring of the 12th generation backcross to self, mating full-sibs of the F1 generation of the self-cross (expected to segregate 1:2:1 fog-2 and N2), isolating immature hermaphrodite/female F2 offspring and allowing them to self. Individuals that did not produce offspring were presumed to be fog-2 homozygotes.

Although the inbred lines were expected to be highly homozygous after 20 generations of (essentially) full-sib mating, whole-genome sequencing of three inbred lines of dioecious Caenorhabditis (one of which, PB2801, was included in this study) revealed extensive residual heterozygosity in certain regions of the genome (Barrière et al. 2009). For example, approximately 30% of the PB2801 (C. brenneri) genome and ~ 10% of the PB4641 genome (C. remanei, derived from the EM464 strain) is heterozygous. We genotyped the ancestral progenitor lines of the fog-2 lines at a panel of 36 dinucleotide repeat loci; we found no unambiguous evidence of residual heterozygosity in that strain. The genotyping protocol is presented in Supplementary Document 1. Residual genetic variation complicates the interpretation of the results considerably; we consider these complications at length in the Discussion.

Mutation Accumulation

In May 2006, 48 replicate MA lines consisting of a single visibly gravid female were initiated from the strains PB4642, PB4643, JU727.3, JU727.5, PB2801, and PB2802 and 96 plates were initiated from the fog-2 strain. In August 2006 we initiated 96 MA lines from the PB2282 strain. We used standard C. elegans husbandry conditions (Baer et al. 2005), consisting of NGM agar plates seeded w/ 100 μl of the OP50 strain of E. coli. Worms were maintained at 20°C. All ancestral (control) stocks were cryopreserved at −80°C using standard methods (Wood 1988) at the time MA lines were initiated.

We initially transferred a single gravid female from each MA line at four-day intervals (a “bottleneck”); the preceding two generations were maintained as backups of the leading-generation plate. If a line failed to reproduce we re-constituted the line by picking a single gravid female from the previous generation (“going to backup”); if only immature individuals were present the plate was “held over” for another four days at the same generation. However, it quickly became apparent that many lines had greater than a four-day generation time and that the failure rate of single-gravid-female reproduction was very high. After ~20 generations we changed our protocol to (1) transfer a single L4 (fourth instar juvenile) stage female and a L4 or young adult male and (2) increase the generation time to five days for all strains except fog-2. Generation time was lengthened at several points during the experiment; by generation ~80 we had gone to a seven-day interval for all strains except fog-2. We refer to the maximum possible number of generations experienced by a MA line as G_MAX and the number of bottlenecks actually experienced by a MA line as G_BOT.

Different lines experienced different numbers of bottlenecks, and different strains experienced different average numbers of bottlenecks (Supplemental Table S1). Census size fluctuated between N=2 in generations in which there is a bottleneck and N >> 2 (~ several hundred) in generations in which we had to go to backup. When census size N fluctuates, the effective population size (N_e) is equal to the harmonic mean N (Crow and Kimura 1970, p. 109), which depends strongly on the number of generations in which N = 2 and only very weakly on the particular magnitude of N>>2. We calculated an approximate N_e for each line by assuming that N=2 (i.e., full-sibs) at each bottleneck and N=200 in generations in which we went to backup (Table 2).

Table 2.

Mutational variances. Abbreviations are: N lines, number of lines included in the assay; N_e, estimated effective population size at G50 and G100; l_L,10, among-line component of variance in the ancestral control (G0); I_M,i, mutational variance after i generations of MA; I_E,i, environmental (within-line) component of variance after i generations of MA; $H_0^2$, standing broad-sense heritability at G0; $h_{m,i}^2$, mutational heritability after i generations of MA. See text for details of calculations.

Species	C. brenneri		C. elegans	C. remanei			C. species 5
Strain	PB2801	PB2802	fog-2	PB2282*	PB4642	PB4643	JU727.5
N lines	30/16/22	30/3/11	49/50/51	35/-/50*	30/9/25	35/19/30	20/9/10
IL,0 (× 103) (95% CI)	18.62 (0, 101.0)	99.59 (0, 275.3)	39.86 (11.0, 68.3)	66.65 (0, 252.1)	240.53 (37.1, 499.2)	0.95 (0, 12.5)	108.2 (0, 298.7)
IM,50 (× 103) (95% CI)	5.66 (0, 22.83)	4.43 (0, 13.13)	1.12 (0.51, 1.80)	-	0.12 (0, 1.10)	4.46 (0, 9.22)	4.29 (0, 12.81)
IM,100 (× 103) (95% CI)	6.60 (1.97, 14.96)	0.78 (0, 6.92)	0.15 (0, 0.34)	1.93 (0.76, 3.29)	2.33 (0, 5.62)	4.17 (0.88, 7.91)	7.74 (0, 19.25)
IE,0 (SEM)	0.68 (0.11)	0.76 (0.16)	0.16 (0.02)	1.58* (0.28)	0.98 (0.17)	1.01 (0.24)	1.18 (0.17)
IE,50 (SEM)	1.13 (0.46)	0.89 (0.27)	0.26 (0.04)	-	0.80 (0.30)	1.63 (0.46)	0.88 (0.16)
IE,100 (SEM)	2.02 (0.58)	7.64 (2.62)	0.31 (0.04)	1.75* (0.24)	4.21 (2.07)	2.30 (0.40)	0.73 (0.43)
$H_0^2$	0.029 (0.046)	0.144 (0.127)	0.259 (0.106)	0.045 (0.051)	0.245 (0.116)	0.001 (0.008)	0.090 (0.069)
$h_{m,50}^2$ (× 103) (SEM)	8.83 (0.15)	4.15 (4.08)	4.38 (1.41)	-	0.13 (0.36)	2.71 (0.13)	4.70 (3.43)
$h_{m,100}^2$ (× 103) (SEM)	3.26 (1.44)	9.55 E-5 2.25 E-4	0.49 (0.29)	1.12* (0.38)	0.48 (0.29)	1.87 (0.86)	26.60 (70.33)

Variation in N_e among lines and/or strains due to fluctuating population size will affect the mutation accumulation process in two ways. First, through its effect on the efficiency of selection: mutant alleles with deleterious effects s < 1/4N_e will be effectively neutral (Keightley and Caballero 1997; Kondrashov et al. 2006). Although in any MA line the substitution rate is expected to be equal to the (effectively) neutral mutation rate, the fraction of mutations that are effectively neutral will vary inversely with N_e. Second, the per-generation rate of accumulation of genetic variance (the mutational variance, V_M) depends on N_e (Lynch and Hill 1986).

MA lines from all strains except PB2282 and JU727.3 were cryopreserved at G_MAX ~52 and ~100. MA lines from PB2282 were cryopreserved at generation 83 only. The rate of extinction in the JU727.3 lines was so great that we abandoned that set of lines before the generation 52 freeze. Lines were considered extinct when they had failed for three consecutive MA generations (i.e., there was no viable backup stock).

Fitness Assay

We assayed fitness of the ancestral controls and MA lines at generations 52 and 100 concurrently for all sets of MA lines except PB2282, which was assayed at generation 83 only. Lines that were present at G_MAX=52 (G52) but went extinct prior to G_MAX=100 (G100) were included in the assay. Certain lines that were present at G52 remained in the MA experiment at G100 but were not successfully cryopreserved at G52. Fitness assays were conducted in three blocks. MA lines were randomly assigned to the blocks; each line × generation combination of a MA line (e.g., line 1 at G52) was present in only one block.

The fitness assay procedure was similar to our canonical assay (Baer et al. 2005), with one major difference (see below). At the beginning of the block, randomly chosen MA lines from each line × generation combination were thawed. A sample of each ancestral control population was thawed and individual worms were chosen to begin replicate lines. We initiated ten replicate ancestral control lines for PB4642, PB4643, JU727.3 (blocks 1 and 2 only), PB2801, PB2802; 15 (block 1) or ten (blocks 2 and 3) lines for PB2282; and 20 (block 1) or 15 (blocks 2 and 3) lines for fog-2. The different numbers of control lines between strains and blocks reflects the different sample sizes of MA lines in the different groups. Repeated attempts to propagate JU727.3 controls during the third assay block failed.

The fitness assay procedure differs from previous assays in that we did not carry lines through two generations to remove parental and grandparental effects; we chose to directly assay the initial worm because the mortality rate was much greater than in prior assays with self-fertilizing worms. For each MA or ancestral control line, five replicates were started by placing a single randomly chosen female worm in the fourth larval stage (L4 stage) and 3–4 young adult males onto a seeded NGM plate. These plates were designated the Reproduction 1 (R1) day plates. Plates were assigned a random number and were handled only in random numerical order throughout the assay. The next day, the parental (R1) worms were removed and placed on fresh seeded plates (R2 plates). The plate from which the parental worms were removed was incubated overnight at 20°C to allow eggs to hatch and then stored at 4°C to prevent further development. The next day, parental worms were removed from the R2 plates and placed on fresh (R3) plates. R2 plates were incubated overnight at 20°C and then stored at 4°C. The R3 plates were incubated for 48 hours at 20°C (to allow one day for egg laying and one day for hatching) and then stored at 4°C; in most cases, reproduction after the third day of the reproductive period was negligible. When the female could not be found, the plate was “held over” to the next reproductive day. On rare occasions, the female was found dead (did not respond to repeated tactile stimulation); these plates were incubated overnight at 20°C and then stored at 4°C. Upon completion of the assay, plates were stained with 0.075% toluidine blue and worms were counted under a dissecting microscope at 20× magnification.

Data Analysis

We refer to the total lifetime reproduction of a set of worms (female and males on a plate) as “Total Fitness” (W); when worms that failed to reproduce are excluded we refer to total lifetime reproduction as “productivity”. W thus includes the effects of survivorship as well as fecundity.

i) Per-generation change in the mean

The data are not normally distributed and could not be transformed into normality, so the usual statistical tests of different group means (F-test, REML) are inappropriate. To test the hypothesis that the % per-generation rate of change the trait mean (ΔM) differs between groups we used a variation of the bootstrap protocol of Baer et al. (2005). For each strain, data were resampled with replacement at the level of line (i.e., all replicates within a line were included) for control and MA lines separately. We then fit the general linear model y = block + generation + line + replicate(line) using REML as implemented in the MIXED procedure in SAS v. 9.2; block and generation were considered fixed effects with the other effects random. Among-line and among-replicate (= error) components of variance were estimated separately for each generation (SAS option GROUP=). Trait means were estimated for each generation by least squares (SAS option LSMEANS). We then calculated the % change per generation as $\frac{{\stackrel{‒}{W}}_{MA}-{\stackrel{‒}{W}}_0}{t{\stackrel{‒}{W}}_0}$ where ${\stackrel{‒}{W}}_{MA}$ and ${\stackrel{‒}{W}}_0$ are the least-squares estimates of the MA and ancestral control means, respectively and t is the number of generations of MA (G_MAX). This resampling procedure was repeated 1000 times; empirical 95% confidence limits were established from the upper and lower 2.5% of the distribution. Groups for which the 95% confidence intervals do not overlap are considered to be significantly different at the 5% level. Although this procedure does not account for within-line variance, we have found that it is invariably conservative when compared with formal statistical tests when the latter are appropriate (Joyner-Matos et al. 2009; CFB unpublished results).

ii) Mutational variance

Under an additive neutral model, the variance among MA lines (V_L) is given by the equation

$$V_L=2V_M\left[t-2N_e(1-\text{exp}(-t∕2N_e))\right]+2V_0\left[1-\text{exp}(-t∕2N_e)\right]$$ (1)

(Lynch and Hill 1986; Mackay et al. 1992), where t = the number of generations of MA, V_M is the mutational variance and V₀ is the among-line variance present in the ancestor. Although the assumption of additivity is probably violated in this case (see below), that assumption is not expected to be seriously misleading (see Lynch and Walsh 1998, p. 330 for justification of that claim). V₀ was estimated from the among-line variance of the ancestral control. The hypothesis that V_M differs between strains was tested using the same bootstrap protocol as for the change in the mean. Variance components were estimated by REML using the same general linear model presented above. V_M was calculated from equation (1). N_e was calculated separately for each MA line as described above; the mean N_e of each strain was used in equation (1) to calculate V_M. 95% confidence limits on V_M were determined in the same way as described in the previous section. All variances are scaled by the square of the group mean (I, “opportunity for selection”; Crow 1958), which is the most appropriate way to scale variances of traits expected to be under directional selection (Houle 1992). We also report mutational heritabilities $\left(h_m^2\right)$, the mutational variance scaled by the environmental (error) variance.

Results

The most obvious difference between the results of this study and the results of similar studies with self-compatible Caenorhabditis (Keightley and Caballero 1997; Vassilieva et al. 2000; Baer et al. 2005; Baer et al. 2006; Begin and Schoen 2006, 2007) is that the gonochoristic (outcrossing) taxa experienced vastly greater mortality (and thus line extinction) and quickly evolved substantially longer generation times than did the self-compatible taxa. Because at least one (PB2801, C. brenneri) and presumably all of the starting stocks were substantially heterozygous to varying degrees (Barrière et al. 2009), the higher mortality cannot be unambiguously attributed to newly accumulated mutations (but see below). In contrast, mortality and line extinction in the fog-2 strain of the historically self-fertile C. elegans did not differ from previous MA experiments.

Change in the mean (ΔM)

Averaged over all strains, the historically outcrossing groups declined about fourfold faster than did the fog-2 strain of C. elegans (Figure 1; Table 1) at both G52 and at G100 (G83 for the PB2282 strain of C. remanei). At G52, fog-2 declined significantly more slowly than all strains except PB2801 (C. brenneri); at G100, fog-2 declined significantly more slowly than all strains except JU727.5 (C. sp. 5). In all strains except PB2801, almost all of the total decline in fitness was between generation 0 and 52, with very little additional decline between generations 52 and 100.

Figure 1.

- ΔM (% per-generation decay) for Total Fitness (W). The ancestral progenitor is assigned a value of 0. Error bars delimit 95% confidence intervals. To prevent overlap, data points are offset horizontally in intervals of two generations centered around G0, G50, and G100. Diamonds are C. brenneri, triangles C. elegans, circles C. remanei, squares C. sp.5.

Table 1.

Per-generation change in mean fitness $\left(\stackrel{‒}{W}\right)$. Abbreviations are: N lines, number of lines included in the assay, numbers represent G0/G52/G100; N_e, mean effective population size (see Methods for details of calculations); ${\stackrel{‒}{W}}_0$, mean fitness of G0 lines; ${\stackrel{‒}{W}}_{52}$, mean fitness of G52 lines; ${\stackrel{‒}{W}}_{100}$; mean fitness of g100 lines; ΔM_0,52 % change per-generation in $\stackrel{‒}{W}$ between G0 and G52; ${\stackrel{‒}{W}}_{100}$, % change per-generation in $\stackrel{‒}{W}$ between G0 and G100; ΔM_0,100, % change per-generation in $\stackrel{‒}{W}$ between G0 and G100; ΔM_52,100, % change per-generation between G52 and G100.

Species	C. brenneri		C. elegans	C. remanei			C. species 5
Strain	PB2801	PB2802	fog-2	PB2282	PB4642	PB4643	JU727.5
N lines	30/16/22	30/3/11	49/52/51	35/-/52*	30/9/25	35/19/30	20/9/10
Ne (G52, G100)	3.5, 5.3	7.3, 10.9	2.3, 2.3	-, 11.2	5.0, 6.5	2.8, 3.9	6.1, 6.9
${\stackrel{‒}{W}}_0$ (SEM)	112.4 (8.5)	51.9 (5.0)	288.3 (10.1)	16.5 (1.5)	28.8 (3.8)	65.5 (4.0)	60.2 (7.8)
${\stackrel{‒}{W}}_{52}$ (SEM)	57.6 (11.7)	12.9 (6.6)	239.1 (13.0)	-	5.7 (1.5)	24.0 (4.6)	15.0 (4.0)
${\stackrel{‒}{W}}_{100}$* (SEM)	8.6 (2.7)	11.2 (4.7)	235.6 (10.5)	5.8 (0.6)*	6.4 (1.7)	10.6 (1.9)	24.0 (10.1)
ΔM0,52 (× 103)	−9.70	−14.95	−3.39	-	−16.01	−12.64	−14.91
(95% CI)	(−13.34, −4.85)	(−20.00, −10.26)	(−5.37, −1.29)		(−18.07, −13.52)	(−15.24, −9.76)	(−17.81, −11.51)
ΔM0,100 (× 103) (95% CI)	−9.23 (−9.65, −8.71)	−7.84 (−9.36, −5.89)	−1.82 (−2.68, −0.88)	−7.77* (−8.75, −6.43)	−7.72 (−8.75, −6.15)	−8.38 (−8.97, −7.76)	−5.91 (−8.87, −1.83)
ΔM52,100 (× 103) (95% CI)	−16.86 (−18.82, −14.02)	−12.53 (−30.46, 32.97)	−0.23 (−2.89, 2.89)	-	−5.77 (−9.07, 41.14)	−10.80 (−15.13, − 4.40)	14.40 (−11.47, 51.69)

^*represents G83 for PB2282 lines.

The fog-2 strain used here is a 12-generation introgression of the fog-2 allele into our canonical N2 (wild-type C. elegans) genetic background. In previous experiments, mutations have been allowed accumulate by maintaining MA lines as single hermaphrodites. In this experiment, C. elegans was maintained as single mating pairs (or potentially half-sib families). Averaged over the G52 and G100, the fog-2 strain of C. elegans declined at a rate of ~0.2–0.3% per generation, slightly (but not significantly) greater than previous estimates (see Discussion).

Mutational Variance (V_M)

Changes in variances are inherently more difficult to measure than changes in means (Shabalina et al. 1997), and the small sample size in several of the strains (Table 1) exacerbates the problem, particularly at G52. Nevertheless, the overall trend observed in ΔM is largely recapitulated in V_M (reported as I_M in Table 2): with a few exceptions, the mutational variance is qualitatively lower in fog-2 than in the gonochoristic strains.

Ideally, the among-line variance in the ancestral (G0) genotype should be 0 if the stock is homozygous, and any residual among-line variance is due to residual environmental effects (e.g., maternal effects). Residual genetic variation is expected to have two effects: first, it will inflate the among-line variance in the G0 stock above 0 and second, it will contribute to the within-line variance, which in the absence of residual genetic variation is the environmental component of variance. We know that PB2801 (C. brenneri) is heterozygous over a large fraction of its genome (Barrière et al. 2009), and we strongly suspect that the other gonochoristic stocks also harbor residual heterozygosity, with the possible exception of PB4643 (C. remanei). Interestingly, the among-line variance (I_L,0) in the G0 ancestors is significantly different from zero only in fog-2 and PB4642 (C. remanei). The lack of significant G0 among-line variance in the other gonochoristic strains is not obviously due to differences in power, because sample sizes at G0 were similar in all gonochoristic strains except for JU727.5 (C. sp.5). The broad-sense heritability $\left(H_0^2\right)$ is qualitatively greater in fog-2 and PB4642 than in the other stocks.

There is no a priori reason to expect the within-line (environmental) component of variance (I_E,i in Table 2) to differ between generations of MA. However, we have previously found in MA experiments with self-fertilizing Caenorhabditis that the within-line variance (scaled by the mean) almost invariably increases with increasing generation of MA (Baer 2008; Braendle et al. 2010). With one exception (JU727.5, C. sp.5), we observe the same trend in the present experiment. I_E,i is qualitatively smaller in fog-2 than in the gonochoristic strains at every generation, but the % per-generation rate of increase in I_E in fog-2 (~1%/gen at G100) is quite similar to the median of the gonochoristic strains (~1.6%/gen), of which only PB801 (C. brenneri) is substantially different (~9%/gen).

Discussion

The immediately obvious result is that gonochoristic (outcrossing) taxa invariably exhibit a much more rapid decay in fitness with mutation accumulation than do related self-compatible taxa (> 4×, averaged over strains and generations). This result is exactly as predicted by theory, which says that the strength of natural selection to reduce the mutation rate will be much greater in self-fertilizing taxa than in obligate outcrossing taxa (Kondrashov 1995). Unfortunately, the presence of residual segregating genetic variation in at least some and probably all but one of the ancestral progenitors of the MA lines compromises that conclusion. The more rapid decay in fitness with MA of the outcrossing lines is exactly what would be expected from inbreeding depression due to the fixation of segregating deleterious recessive (or overdominant) alleles. The possibility of a role for inbreeding depression is reinforced by the observed non-linear relationship between ΔM and generation of MA, such that fitness drops off rapidly between G0 and G52 but much less rapidly between G52 and G100 in all strains except PB2801 (C. brenneri). That pattern is what would be predicted if the initial decline was due largely to fixation of deleterious recessive alleles, followed by a much slower decline due to the accumulation of new mutations. A previous experiment with C. remanei revealed strong inbreeding depression that had not abated after 13 generations of full-sib mating (Dolgin et al. 2007). Estimates of synonymous substitution rates in the genus Caenorhabditis do not suggest a consistent difference in base substitution rate between selfing and gonochoristic taxa (Cutter 2008).

Obviously, the results of this experiment must be interpreted with considerable caution. However, there are several lines of circumstantial evidence that are that are not easily reconciled with the difference between gonochoristic and selfing taxa being due primarily to inbreeding depression in the outcrossers. First, the among-line variance in the G0 progenitor of fog-2 was not substantially lower than in the gonochoristic strains, with the exception of PB4642 (C. remanei). Of 36 microsatellite loci we genotyped in the fog-2 ancestor, all but two were unambiguously homozygous and none was unambiguously ancestrally heterozygous; the ambiguity was a result either of microsatellite “stutter” or of new mutations that occurred in during population expansion and were segregating at low frequency in the genotyping population. Most of the among-line variance in the G0 fog-2 is presumably due to environmental variance, which in turn suggests that most of the among-line variance in G0 in the other strains is also due to environmental effects.

Second, with two exceptions, the within-line variance (I_E,i in Table 2) consistently increases with MA. If the observed rapid initial decline in fitness followed by relative stasis is primarily due to inbreeding depression resulting from ancestrally segregating deleterious alleles that fixed between G0 and G52, the expectation is that the segregating alleles will initially contribute to the within-line variance, but once a line is fixed at all loci the within-line variance will be strictly due to environmental effects, in which case the within-line variance should decrease with increasing MA. Only JU727.5 (C. sp. 5) exhibited a consistent decrease in within-line variance, and the number of lines was small. PB4642 (C. remanei) showed a modest decrease in within-line variance from G0 to G52, followed by a large increase from G52 to G100, but the number of lines assayed at G52 was small (9) relative to G100 (25). However, PB4642 also was the only strain other than fog-2 with significant among-line variance at G0, so there is some reason to believe that inbreeding depression may be important in PB4642.

Third, if there was significant purging of deleterious alleles, one would expect that the failure of individual replicates to reproduce, and thus the necessity of “going to backup”, would decline over time as deleterious alleles were purged from the line. If so, the ratio of the actual number of bottlenecks experienced between two time points relative to the maximum possible number of bottlenecks, G_BOT/G_MAX, would be lower between G0 and G52 than between G52 and G100. To the contrary, G_BOT/G_MAX either remained the same or decreased in every case (Supplementary Table 1).

Fourth, and perhaps most convincingly, there is no evidence for inbreeding depression in PB4643: there is no among-line variance at G0 and the within-line variance consistently increases with MA, and the sample size is quite large. At both G52 and G100, ΔM is consistently about 4× greater in PB4643 than in fog-2 (Table 1), almost identical to the overall average.

Finally, the simple fact that 20 generations of isofemale descent did not remove almost all of the segregating variation clearly shows (1) there was a great deal of genetic variation present in the ancestor and (2) suggests at least the possibility that a lot of new mutations occurred during the 20 generations of inbreeding. Both of these are consistent with a high mutation rate in the gonochoristic taxa. Moreover, given that a substantially deleterious allele is still segregating after 20 isofemale generations, it is not obvious why to expect that allele to fix in another 20 (or 50, or 100) isofemale generations.

In what follows we tentatively assume that the differences between our fog-2 strain of C. elegans and the gonochoristic taxa are due to some difference in mutational properties and not due to inbreeding depression, although it is important to recognize that the two factors are not mutually exclusive and both probably contribute to the observed results. In this study, Total Fitness in C. elegans declined at about 0.2–0.3% per generation. Averaging over several previous assays, this N2 strain declines in fitness at about 0.1–0.15% per generation when maintained as single hermaphrodites (Baer et al. 2005; Baer et al. 2006). Results from several other MA studies in which N2 was maintained as single hermaphrodites are extremely similar, with lifetime reproductive output consistently declining at about 0.1% per generation (Keightley and Caballero 1997; Vassilieva et al. 2000; Begin and Schoen 2006). A different wild-type strain of C. elegans, PB306, declines in fitness when maintained as single hermaphrodites at a rate indistinguishable from N2, between 0.1–0.15% per generation (Baer et al. 2005; Baer et al. 2006). Although ΔM of fog-2 in the present experiment does not differ significantly from previous estimates of ΔM from hermaphrodites, the two-fold difference of the point estimates does suggest that C. elegans may decline somewhat more rapidly when mating is male-female rather than hermaphroditic. Taking the difference at face value, we can imagine two potential (not mutually exclusive) reasons. First, because males produce sperm throughout their lifetime whereas hermaphrodites undergo spermatogenesis prior to oogenesis (Schedl 1997), it is at least possible that the average male sperm has undergone more rounds of genome replication than the average hermaphrodite sperm. Second, Total Fitness has a contribution from the male (e.g., mating success) and fitness is multiplicative across components, leading to a potentially greater decay in fitness in male-female reproduction than in hermaphrodites.

We have previously shown that both ΔM and V_M are significantly greater (~ 2×) in two strains (HK104, PB800) of the congeneric self-fertile hermaphroditic species C. briggsae than in C. elegans when mutations are accumulated by single hermaphrodite descent (Baer et al. 2005; Baer et al. 2006; Ostrow et al. 2007). The mutation rate at dinucleotide repeat (microsatellite) loci is also about 2× greater in those strains of C. briggsae than C. elegans (Phillips et al. 2009), which suggests that the different ΔM and V_M may be due to different overall mutation rates rather than different distributions of mutational effects.

Taken together, the available evidence suggests that, on average, the gonochoristic Caenorhabditis accumulate mutational damage at about twice the rate as C. briggsae, which in turn accumulates mutational damage at about twice the rate as C. elegans. Interestingly, the relative positions of C. briggsae and C. elegans in the phylogeny of Caenorhabditis are at least qualitatively consistent with the observed differences among taxa in the rate of accumulation of mutational damage. Gonochoristic reproduction is parsimoniously interpreted as ancestral in the genus (Kiontke et al. 2004), and self-compatible hermaphroditism has apparently evolved at least three times in the genus; in C. elegans and C. briggsae and in the recently discovered C. sp 11 (C. Braendle, personal communication). C. briggsae is positioned at the tip of a clade containing several gonochoristic taxa, whereas C. elegans is positioned much more basally in the genus (Supplementary Figure 1; Kiontke et al. 2007). The relative phylogenetic positions of the two species are consistent with the possibility that self-compatible hermaphroditism evolved more recently in C. briggsae, although there is evidence that self-fertilization evolved relatively recently in both C. elegans (Cutter et al. 2008) and C. briggsae (Cutter et al. 2010). Although this possibility is obviously highly speculative, the different mutational properties of the two species are exactly what would be predicted if natural selection is acting to reduce the mutation rate in the selfing taxa and has simply had more time to operate in C. elegans than in C. briggsae.

The available phylogenetic evidence is not consistent with a higher long-term base-substitution rate in gonochoristic than in selfing Caenorhabditis (Cutter 2008), although there are many reasons why base substitutions may not reflect the true mutational process (Denver et al. 2009). However, other classes of mutations (e.g., indels, transposable elements) may differ between groups.

A very plausible alternative explanation for the observed differences in the rates of accumulation of mutational damage between C. elegans, C. briggsae, and the gonochoristic taxa is that the mutation rates do not differ between taxa but the effects of mutations are context-dependent, such that deleterious effects are larger in sub-optimal environments. All comparative fitness assays of new mutations to date have been conducted under laboratory conditions optimized for the N2 strain of C. elegans (with one exception; see Baer et al. 2006). If deleterious effects are larger in “stressful” (i.e., low-fitness) environments, the observed differences between groups are exactly what would be predicted given the histories of the strains involved (i.e., fog-2 is derived from N2), Although the evidence for context-dependent mutational effects in C. elegans and C. briggsae is equivocal (Vassilieva et al. 2000; Baer et al. 2006), there is evidence from other taxa that the effects of deleterious mutations are often magnified under stressful conditions (Szafraniec et al. 2001; Houle and Nuzhdin 2004; Xu 2004).

A variation of context-dependence is epistasis. Total fitness was qualitatively greater in the fog-2 progenitor than the other strains, in some cases much greater. Moreover, several of the inbred strains in this experiment are known to have substantially lower fitness than their outbred progenitors (Barrière et al. 2009). If deleterious mutations interact synergistically, then strains that begin MA with a greater mutational load will experience a faster rate of decay due to the proportionally greater effects of new mutations. However, synergistic epistasis also predicts an accelerating decay of fitness with MA, which is the opposite of what is observed. We believe the most likely explanation is that in many of the strains (except fog-2), fitness declined to the point where any further reduction in fitness was essentially lethal, and by maintaining lines by going to backup we artificially maintained fitness at some lower bound. The cessation of mutation accumulation past a certain point was also observed by Mackay et al. (1995) with mutations affecting bristle number in Drosophila melanogaster, which they attributed to epistasis.

A different alternative possibility is that the mutation rates themselves are context-dependent. Specifically, several lines of evidence from various taxa suggest that individuals in suboptimal conditions (and therefore in poor condition) have higher mutation rates than individuals of the same genotype under optimal conditions (Hall 1992; Foster 1993; Goho and Bell 2000; Rosenberg and Hastings 2004; Agrawal and Wang 2008). Since MA with Caenorhabditis has always been undertaken in conditions that were potentially sub-optimal for taxa other than the N2 strain of C. elegans, if the mutation rate itself is condition-dependent, mutation rates may in fact differ between strains, but as a result of differing environmental conditions rather than adaptive differences due to mating system (or anything else).

To our knowledge, there has been only one other study in which the mutational properties of related selfing and outcrossing taxa have been directly compared. Schoen (2005) performed MA with related selfing and outcrossing species of plants in the genus Amsinckia. The focus of that study was on estimation of the genomic mutation rate for fitness (U) which did not differ significantly between the selfer A. gloriosa and the outcrosser A. douglasiana. Estimates of U are notoriously noisy, however (Keightley and Eyre-Walker 1999; Begin and Schoen 2006), and both ΔM and V_M for a fitness-related trait were substantially greater in the outcrossing A. douglasiana than in the selfer A. gloriosa (see Table 1 of Schoen 2005), consistent with both theoretical prediction and the results of this study.

Conclusions

Historically outbreeding, gonochoristic strains of three species of Caenorhabditis declined in fitness with MA about four-fold faster than the historically predominantly self-fertilizing C. elegans. This result is consistent with the theoretical prediction that the strength of natural selection to reduce mutation rate should be substantially stronger in self-fertilizing taxa than in outcrossing taxa. Residual genetic variation in the progenitors of some (probably all) of the gonochoristic strains leaves open the possibility that the difference in the rate of mutational decay could be at least partially due to inbreeding depression in the outcrossing taxa. The difference in the rates of decay could also result from context-specific environmental and/or epistatic effects. Large-scale molecular genotyping can potentially resolve these questions in this experiment, but more such comparative studies in disparate taxa will be required to establish the generality of these results.

Figure 2.

Among-line component of variance in Total Fitness (V_L). Error bars delimit 95% confidence intervals. To prevent overlap, data points are offset horizontally in intervals of two generations centered around G0, G50, and G100. Diamonds are C. brenneri, triangles C. elegans, circles C. remanei, squares C. sp.5.