TRANSPOSABLE ELEMENTS, MUTATIONAL CORRELATIONS, AND POPULATION DIVERGENCE IN CAENORHABDITIS ELEGANS
Transposable element activity is thought to be responsible for a large portion of all mutations, but its influence on the evolution of populations has not been well studied. Using mutation accumulation experiments with the nematode Caenorhabditis elegans, we investigated the impact of transposable element activity on the production of mutational variances and covariances. The experiments involved the use of two mutator strains (RNAi-deficient mutants) that are characterized by high levels of germline transposition, as well as the Bristol N2 strain, which lacks germline transposition. We found that transposition led to an increase in mutational heritabilities, as well as to the intensification of correlation patterns observed in the absence of transposition. No mutational trade-offs were detected and mutations generally had a deleterious effect on components of fitness. We also tested whether the pattern of mutational covariation could be used to predict observed patterns of population divergence in this species. Using 15 natural populations, we found that population divergence of C. elegans in multivariate phenotypic space occurred in directions only partially concordant with mutation, and thus other evolutionary factors, such as natural selection and genetic drift, must be acting to produce divergence within this species. Our results suggest that mutations induced by mobile elements in C. elegans are similar to other spontaneous mutations with respect to their contribution to the microevolution of quantitative traits.
Mutation is fundamental to the evolutionary process because it creates genetic variation that can be acted upon by natural selection and random drift, thus leading to evolutionary change. Because organisms are integrated and complex wholes underlain by pleiotropic genetic pathways, the study of phenotypic evolution can benefit from a multivariate approach, a perspective that directly takes into account interdependence among traits. A family of models based on this approach has been developed and provides a link between microevolutionary processes and macroevolutionary patterns (Lande 1979; Riska 1989; Arnold et al. 2001). These models focus on the interaction between quantitative genetic constraints (as expressed in G, the matrix of additive genetic (co)variances between traits) and evolutionary forces, such as natural selection and random genetic drift. Most of the work in this field of research has focused on testing whether G remains constant across large numbers of generations, a critical assumption underlying the long-term prediction of evolutionary trajectories (Roff 2000; Steppan et al. 2002). However, little work has been accomplished on other important issues, such as how various factors shape the G matrix, and how well genetic constraints predict patterns of evolutionary divergence (McGuigan 2006).
Although the G matrix is central to the predictive equations of quantitative genetics, few attempts have been made to study the underlying factors that shape it (but see Jones et al. 2003). One of these factors is the mutational (co)variance matrix (which we will call M, also referred to as U by some researchers), a population characteristic embodying the per-generation effect of pleiotropic mutation on the phenotype. In turn, the total mutational (co)variation can be divided into various causal factors. One particularly likely source of mutation is transposable element (TE) activity (Finnegan et al. 1978; Green 1988), whose impact on the multivariate evolution of populations has not been well studied. Transposable elements are DNA segments that have the ability to reinsert in new genomic locations. TEs are common across all types of organisms and, in some species, account for a large fraction of all mutations. In addition to TE-mediated disruption of exons or regulatory regions, transposition could potentially lead to significant phenotypic changes if, for instance, a TE that had previously incorporated a regulatory sequence happened to insert upstream of another gene (i.e., a transduction event). This could either up- or down-regulate entire physiological pathways (Moran et al. 1999; Morgante et al. 2005) and influence multiple quantitative traits.
Another important issue pertains to the role of quantitative genetic constraints in shaping long-term evolutionary change. The “constraint hypothesis” (Sokal 1978; Björklund and Merilä 1993; Schluter 1996) is akin to a neutral hypothesis in the sense that it postulates that genetic constraints are the sole determinants of the rate and direction of evolutionary change in a lineage. It proposes that a high genetic correlation between two traits within an ancestral population forces the distribution of mean trait values of daughter populations to be similarly correlated (i.e., the D matrix, the (co)variance matrix of mean phenotypic trait values of the extant populations, should be proportional to the ancestral G matrix), whereas an absence of correlation in the ancestral population imposes no such constraint. A major challenge in conducting this type of study is that it is often impossible to have access to accurate information on the ancestral G matrix. As a surrogate for this matrix, investigators typically employ the G matrix of one of the extant populations or use phylogenetic methods to reconstruct an ancestral matrix (Bégin and Roff 2004; Merilä and Björklund 2004). Although there is no theory that explicitly links the M and D matrices, the extrapolation is straightforward because M is a major component of G. Thus, it is worthwhile to test the constraint hypothesis using the M matrix, in particular because it allows one to isolate the role of mutations, as opposed to that of standing genetic variation (G), in directing multivariate evolution.
Here we report the results of experiments and observations designed to explore the impact of TE activity on the M matrix and on phenotypic divergence in the nematode Caenorhabditis elegans. The first part of this investigation involved three mutation accumulation (MA) experiments conducted with the Bristol N2 strain (which served as a control, because it lacks germline transposition; Eide and Anderson 1985), along with two transpositionally active mutator strains, mut-7 and mut-14 (Ketting et al. 1999; Tijsterman et al. 2002). These mutator strains undergo on average two new TE insertions per generation (a minimum estimate), and have higher deleterious mutation rates than N2, as assayed using phenotypic studies (Bégin and Schoen 2006). The mutator strains used here were produce by EMS mutagenesis in the N2 background, and are characterized by a deficiency in the RNA interference (RNAi) pathway, a lesion that is associated with highly increased germline activity of TEs (Ketting et al. 1999; Tijsterman et al. 2002). RNAi has been hypothesized to have evolved as a genomic equivalent to the immune system in the sense that it may help prevent the disruption of the genome by selfish genetic elements (Plasterk 2002). In the second part of the work reported here, we studied phenotypic divergence in fifteen natural populations of C. elegans distributed across several continents. We tested the constraint hypothesis and examined the relationship between patterns of phenotypic divergence and the M matrix, as estimated with and without germline TE activity.
Materials and Methods
STUDY ORGANISM AND MAINTENANCE
All nematode strains used in this study, including natural populations (Table 1), were obtained from the Caenorhabditis Genetics Center (Minneapolis, MN). Nematodes were maintained on agar plates using a slightly modified standard protocol (Stiernagle 1999)—we used HB101 E. coli as a food source, kept worms at 20°C, added 1 mL/L of 200 mg/mL streptomycin sulfate to both LB medium and NGM, used 20 g/L of agar in the NGM, and used 10 g/L of NaCL in the LB media.
Table 1. Name, sampling location, and mean trait values (SE) for all laboratory strains and natural population of Caenorhabditis elegans. Two values are given for N2, mut-7, and mut-14: before (g0) and after (MA) mutation accumulation
| AB3 || Adelaide, Australia || 270.5 (11.9) || 16.3 (1.5) || 1937.1 (34.8) || 127.6 (2.3) |
| CB4855||Palo Alto, CA||243.9 (15.0)||15.9 (1.1)||1965.0 (41.5)||117.9 (2.8)|
| CB4856 || Hawaii, U.S.A. || 240.5 (13.8) || 20.2 (1.4) || 1932.2 (27.2) || 123.1 (1.9) |
| CB4857||Claremont, CA||252.2 (14.7)||13.1 (1.0)||1912.6 (34.6)||125.6 (3.1)|
| CB4932 || Taunton, England || 274.5 (8.0) || 24.3 (1.2) || 2074.8 (20.3) || 143.5 (2.2) |
| DH424||El Prieto Canyon, CA||239.8 (6.8)||22.5 (1.4)||2173.5 (29.3)||123.7 (1.7)|
| JU258 || Ribeiro Frio, Madeira, Portugal || 210.5 (9.5) || 23.5 (0.8) || 1887.6 (32.5) || 122.4 (2.0) |
| JU262||Le Blanc, France||267.7 (13.4)||13.6 (1.1)||1885.0 (30.4)||122.1 (2.6)|
| KR314 || Vancouver, Canada || 281.1 (11.2) || 19.4 (1.1) || 2130.1 (22.6) || 144.7 (2.8) |
| MY1||Lingen, Germany||290.9 (12.3)||16.5 (0.9)||2068.9 (39.7)||151.9 (5.2)|
| MY16 || Mecklenbeck, Germany || 262.6 (6.7) || 15.1 (1.0) || 1933.8 (18.8) || 129.5 (2.3) |
| N2||Bristol, England||285.8 (12.4)||21.7 (1.2)||2173.4 (24.2)||128.2 (2.1)|
| PX174 || Lincoln City, OR || 272.4 (11.8) || 17.3 (1.2) || 2013.6 (27.8) || 124.9 (1.7) |
| RC301||Freiburg, Germany||260.4 (6.0)||15.3 (0.5)||2234.4 (29.1)||144.7 (2.7)|
| TR403 || Madison, WI || 174.0 (5.3) || 20.5 (1.1) || 1970.5 (23.4) || 107.3 (1.4) |
| N2 g0 || 290.2 (10.0) || 20.8 (0.7) || 2107.0 (22.7) || 133.9 (2.5) |
| N2 MA|| ||273.6 (5.2)||19.0 (0.4)||2251.4 (15.4)||140.0 (1.2)|
| mut-7 (NL917) g0 || 101.5 (7.6) || 20.0 (0.8) || 1801.4 (36.2) || 110.1 (2.8) |
| mut-7 (NL917) MA|| ||85.0 (5.1)||18.4 (0.6)||1764.5 (28.2)||106.7 (2.1)|
| mut-14 (NL1838) g0 || 99.6 (5.7) || 15.7 (0.5) || 1840.0 (34.8) || 134.9 (3.7) |
| mut-14 (NL1838) MA|| ||74.4 (4.7)||14.2 (0.4)||1796.7 (36.3)||131.5 (2.9)|
MUTATION ACCUMULATION EXPERIMENTS
Each of the three MA experiments (N2, mut-7[NL917], and mut-14[NL1838]) began with a single hermaphrodite nematode, used as the initial progenitor of 20 generations of single worm descent through self-fertilization. This process should have led to the three resulting stocks of worms being nearly completely homozygous prior to the start of the MA experiments. For each MA experiment (i.e., one conducted with each C. elegans strain described above), 75 replicate lines were initiated, each from a single homozygous hermaphrodite, and these were independently propagated for 40 generations by transferring a randomly selected larva each generation (for more details see Vassilieva and Lynch 1999). Two generations of back-up worms were kept at all times and used to restart a line if it failed, which would result in extinction of that line only when all remaining back-up individuals failed to reproduce. All surviving lines where cryopreserved at generation 0, 20, and 40 (Stiernagle 1999). Mutation accumulation experiments allow mild effect mutations to become fixed in lines via random drift, at a rate approximately equal to that of their production, whereas lethal and sublethal mutations are not maintained under these conditions (Keighley and Caballero 1997).
Worms were thawed prior to fitness assays and allowed to freely proliferate for one to three generations. Large numbers of eggs where then bleached, to remove E. coli cells, and larvae were synchronized (Stiernagle 1999) by transferring to a plate with E. coli. Four fitness components were measured on adult hermaphrodites: fecundity, longevity, body length, and body width. Fecundity was estimated by manually counting daily egg production and then summing over lifetime. This trait is likely overestimated because some nonviable eggs (i.e., eggs incapable of hatching) may be included in the counts. Longevity was scored on the worms used for fecundity. This trait represents the number of days that a worm lived, and was measured by checking daily for the lack of body and pharynx movement. Body length and width were scored on different worms than fecundity and longevity, but all worms came from the same plate and MA line. Four-day-old worms were sacrificed and immersed for at least a week in a fixative (3 mL of lactic acid, 1 mL of acetic acid, 5 mL of glycerol, 60 mL of 100% ethanol, and 31 mL of water). Measurements were performed on worms placed in multiwell microslides filled with fixative. The software programs QCapture version 2.68.2 (QImaging, Burnaby, BC, Canada) and Image J version 1.32 (National Institute of Health, Bethesda, MD) were used for image capture and measurement, respectively. Length and width data are reported as numbers of pixels—1 mm equals 1575.2 pixels. Additional details of how fitness measurement data were collected are available elsewhere (Bégin and Schoen 2006).
For the mutation accumulation experiments, the number of lines assayed was 63, 62, and 46 for N2, mut-7, and mut-14, respectively. On average, approximately 10 replicate worms per line were scored for fecundity and longevity, and six for body length and width. Assays were performed at generation 40 for N2 and at generation 20 for the mutator strains. For the test of the constraint hypothesis, fitness assays on naturally occurring strain isolates were performed as above, except that these strains were not divided into replicate lines. Sample sizes were between 19 and 34 individuals per strain, depending on the trait.
This study required the work of multiple laboratory assistants. Fecundity was statistically corrected for multiple observers by multiplying each observer's counts by a factor determined through a test count of N2 controls. Longevity was not corrected because preliminary tests showed negligible variation between observers in their ability to determine whether a worm was alive or not. Size measurements were taken by a single person throughout the experiment. Efforts were made to randomize the measurement procedure with respect to lines, strains, and generation (0 or MA), but all of the individuals of a line were always assayed on the same day and stacked together in the growth chamber. Replicate lines of the various strains and natural populations were assayed continuously over several months in a semirandom order. Statistical analyses did not show substantial temporal trends, and so data were not corrected in this regard, to simplify the final multivariate analysis.
ESTIMATION OF QUANTITATIVE GENETIC PARAMETERS
The M matrix is composed of the mutational variances of traits (Vm) and mutational covariances between traits (Covm), which represent the per-generation increase in genetic (co)variance arising through mutation. The M matrices were estimated separately for mut-7, mut-14, and N2 from the fitness assay data obtained following the MA experiments. Because fecundity and longevity were measured on the same worms, and length and width were measured together on other worms of the same MA line, we randomly assigned the two sets of values to individual worms within lines. This procedure is not expected to bias the among-line variance (Estes et al. 2005). We used the program H2boot (subroutine for recombinant inbred lines, Phillips 1998a) to estimate M matrices and mutational heritabilities (h2m). The procedure involved estimating the among-line variance component from a one-way analysis of variance (ANOVA) (Vm= 1/2 among-line variance component), coupled with bootstrap resampling (5000 iterations at the line level). The estimate for a matrix element is the average of all bootstrap estimates, whereas the standard error is the standard deviation of the bootstrap estimates. This procedure assumes that genetic variation at generation 0 was nil, which was not always the case (results not shown). The proper procedure in this situation would be to subtract the initial genetic variation from the MA estimates, but the former was noisy and made the results meaningless. In principle, however, there is no reason to believe that genetic variation was indeed present at generation 0, and the significant among-line variance was probably due to common environmental effects. This effect may inflate the estimation of mutational (co)variances but, as it is reasonable to assume that common environmental effects do not change drastically between strains and between the beginning and end of the MA procedure, this confounding factor should have minimal effects on cross-strain comparisons of M matrices.
Comparison of M matrices across strains was performed using two methods: the Flury hierarchy and the jackknife–MANOVA approach. The Flury hierarchy (Cowley and Atchley 1992; Phillips and Arnold 1999) is based on the principle that two matrices share some part of their eigenstructure (i.e., principal component structure), and uses maximum likelihood to determine which hierarchically nested model best describes the structural differences. The models tested were (1) “unrelated structure”: the matrices have no eigenvectors in common; (2) “partial common principal components”: the matrices share some of their eigenvectors; (3) “common principal components”: the matrices share all of their eigenstructure, but not their eigenvalues; (4) “proportionality”: matrices share their eigenstructure, and their eigenvalues differ by a constant; and (5) “equality”: matrices share the whole eigenstructure. For each model, the Flury hierarchy calculated a log-likelihood statistic to quantify the fit of that model to the observed matrices. A likelihood ratio was then calculated for each model against the model of “unrelated structure” (jump up procedure, Phillips and Arnold 1999). To avoid the assumption of multivariate normality in hypothesis testing, and because the degrees of freedom are poorly specified under the null hypothesis, randomization was used to determine the probability that a model fits the data significantly better than the “unrelated structure” model. In the present analysis, 5000 randomized datasets were created, each iteration randomly assigning whole lines to strains. The best fitting model was determined as the model immediately under the first significant probability, going from the bottom (“unrelated structure” model) to the top (“equality” model) of the hierarchy (Phillips and Arnold 1999). This analysis was performed using the program CPCrand (Phillips 1998b). Note that because this program does not implement the estimation of M matrices, we instead compared the matrices corresponding to the total mutational output after MA. The results are expected to be very similar because the elements of the two types of matrices are exactly proportional, the factor of proportionality being the number of generations of mutation accumulation.
Another way to implement the Flury hierarchy is to use line means to estimate a product-of-moment M matrix (Phillips and Arnold 1999). This approach is expected to be biased (Roff and Preziozi 1994), but produces (co)variances that are statistically well behaved and for which different matrices can be directly tested using parametric methods. The Flury hierarchy was implemented using the CPC program (Phillips 1998c). The procedure is as described above, except that the likelihood ratio was directly tested against a χ2 distribution.
The second method used was the jackknife–MANOVA (Roff 2002). This approach uses the jackknife procedure (Manly 1997, pp. 24–33) to produce a distribution of pseudovalues of matrix elements for each strain separately. A pseudovalue was calculated by estimating a matrix element after deleting all individuals of one line. The number of pseudovalues calculated for a strain is equal to the number of lines. For a given line that has been removed, the pseudovalues corresponding to each matrix elements (10 elements in this case: four variances and six covariances) can be arranged in a row, which then constitutes the pseudovalue of the whole matrix. Two or more matrices can then be compared by using the pseudovalues as data in a multivariate analysis of variance (MANOVA).
THE CONSTRAINT HYPOTHESIS
To test the hypothesis that population divergence occurred in the multivariate direction predicted by the quantitative genetic constraints induced by mutation, we estimated the D matrix and compared it with each of the three M matrices (N2, mut-7, and mut-14). The D matrix was computed by using, as data points, the mean phenotypic trait values of each of the 15 natural populations (Table 1), and directly estimating the (co)variances of the four traits. Phylogenetic correction appeared to have little influence in the estimation of the components of D (see the Appendix), and was thus not used. The constraint hypothesis was tested by comparing the M and D matrices using the Flury hierarchy (CPC program, Phillips 1998c) and by calculating the angle between the corresponding principal components of the matrices. These were obtained by principal component analysis of each matrix separately (using correlations, not covariances, because the difference in scale between traits was too large). The angle θ in radians between principal component 1 of each matrix and also the angle between principal component 2 of each matrix were each estimated as
where T refers to matrix transposition, PC to principal component (eigenvector), and M and D to matrices of the same name. The angle was transformed to degrees for simplicity of interpretation. Only the first two principal components were used in the comparison because, together, they represent at least 75% of the total variation. The range of angles obtained by deleting one natural population at a time (therefore producing a total of 15 angles) was reported to illustrate data dispersion.
Mean trait values are presented for all strains and natural populations (Table 1, for a full analysis see Bégin and Schoen 2006). The mutator strains, mut-7 and mut-14, suffered significant fitness decline following 20 generations of mutation accumulation: trait means declined by 0.10% to 1.26% per generation (Bégin and Schoen 2006). Forty generations of mutation accumulation in the N2 strain produced milder results: fecundity and longevity declined by 0.14% and 0.21% per generation, respectively, whereas length and width increased by 0.17% and 0.11% per generation, respectively (Bégin and Schoen 2006).
Mutational variances, covariances, correlations, and heritabilities for N2, mut-7, and mut-14 are shown in Table 2. All mutational heritabilities except longevity in mut-14 were significantly different from 0, and all were higher in the mutators than in N2 by a factor of 2 to 8. In N2, a single mutational covariance, the one between length and width, differed significantly from 0 whereas, in both mut-7 and mut-14, three mutational covariances differed from 0. The small number of statistically significant covariances may in part be due to large standard errors around the quantitative genetic estimates. All of the correlation estimates that were significant were positive in sign (Table 2). The point estimates of correlations were more highly positive in the mutator strains than in N2 in all cases except for the correlation between fecundity and width. In the two mutator strains, the body size traits were very highly correlated, fecundity and length were moderately correlated, and longevity was almost never correlated with other traits.
Table 2. M matrices for three strains of Caenorhabditis elegans. Mutational covariances are above the diagonal, variances are on the diagonal, correlations are below the diagonal, and heritabilities are in the last column. Standard errors are given in parentheses. See Materials and Methods for units of the original measurements.
|N2||Fecundity||19.8 (3.8)|| 0.21 (0.19)|| 6.5 (11.1)|| 2.3 (2.0)||0.005 (0.001)|
| Longevity || 0.20 (0.19) || 0.05 (0.02) || −0.86 (0.73) || −0.04 (0.05) || 0.001 (0.0004) |
| ||Length|| 0.11 (0.20)||−0.31 (0.27)|| 154.9 (34.9)|| 3.5 (1.5)||0.015 (0.003)|
| Width || 0.55 (0.46) || −0.19 (0.25) || 0.30 (0.11) || 0.88 (0.18) || 0.008 (0.002) |
| mut-7 ||Fecundity||38.4 (6.0)|| 2.1 (0.5)|| 97.4 (28.5)|| 4.6 (3.0)||0.018 (0.003)|
| Longevity || 0.55 (0.12) || 0.38 (0.09) || 5.8 (3.7) || 0.43 (0.28) || 0.006 (0.001) |
| ||Length|| 0.48 (0.11)|| 0.28 (0.17)||1077.3 (192.3)|| 73.0 (13.3)||0.072 (0.015)|
| Width || 0.29 (0.18) || 0.28 (0.18) || 0.89 (0.03) || 6.3 (1.0) || 0.059 (0.012) |
| mut-14 ||Fecundity||18.7 (4.2)|| 0.48 (0.24)|| 89.7 (29.8)|| 4.8 (3.2)||0.010 (0.003)|
| Longevity || 0.50 (0.29) || 0.054 (0.027) || 4.2 (2.7) || 0.43 (0.25) || 0.002 (0.001) |
| ||Length|| 0.57 (0.14)|| 0.54 (0.39)||1323.6 (338.1)|| 87.1 (29.8)||0.062 (0.019)|
| Width || 0.37 (0.24) || 0.70 (0.50) || 0.81 (0.08) || 8.6 (2.8) || 0.052 (0.019) |
Pairwise comparisons of M matrices across strains revealed that the two mutator stains have mutational properties that are more similar to each other than they are to those of N2 (Table 3). This result is consistent across three different matrix comparison methods. Inspection of the matrices revealed that this pattern is driven in part by the two body size traits.
Table 3. Pairwise comparisons of M matrices across strains of Caenorhabditis elegans. Results are reported for three methods (see text). Model refers to the principal component model that best describes the difference between two matrices; unrelated means that the two matrices are different; PCPC1 means that the matrices share only one eigenvector; CPC means that the matrices share all of their eigenvectors, but not their eigenvalues; and equality means that the two matrices share their complete eigenstructure. F is the F statistic of the MANOVA and P is the P-value of the corresponding test. See Materials and Methods for more details on these methods.
|N2 vs. mut-7||PCPC1||0.02||unrelated||<0.0001||5.5||<0.001|
| N2 vs. mut-14 || CPC || 0.03 || unrelated || <0.0001 || 3.0 || 0.002 |
| mut-7 vs. mut-14||equality||0.07||CPC||<0.0001||2|| 0.04|
The D matrix is shown in Table 4. Half of the correlations were significant, and all of those were positive in sign and moderate. We compared M to D to test the constraint hypothesis. When using the M matrix of N2, the angle between principal components 1 of the M and D matrices was 13.0°, and ranged from 5.5° to 25.7° (this range was produced by a method analogous to a jackknife procedure, see Materials and Methods). The angle between principal components 2 was 103.8°, and ranged from 75.1° to 110.5°. In other words, the multivariate direction that explained the most variation is similar in the two matrices, but the second most important direction is widely different (principal component 2 of one matrix did not correspond to principal components 3 or 4 of the other matrix, results not shown). When the M matrix of mut-7 was used, the angle between principal components 1 was 26.3°, with a range of 13.3° to 39.2°, and the angle between principal components 2 was 102.4°, with a range of 77.6° to 107.0°. When mut-14 was used, the angle between principal components 1 was 31.9°, with a range of 18.7° to 44.7°, and the angle between principal components 2 was 140.1°, with a range of 29.7° to 142.2.
Table 4. D matrix representing the multivariate divergence of 15 natural strains of Caenorhabditis elegans. Covariances are above the diagonal, variances are on the diagonal, and correlations are below the diagonal. Standard errors are given in parentheses.
|Fecundity||875.3 (432.0)||−28.2 (23.4)|| 1158.0 (637.9)||247.2 (126.1)|
| Longevity || −0.26 (0.24) || 12.2 (3.0) || 113.8 (106.5) || −1.4 (9.4) |
|Length|| 0.36 (0.18)|| 0.30 (0.28)||12371.0 (3253.4)||698.1 (269.6)|
| Width || 0.71 (0.13) || −0.04 (0.24) || 0.55 (0.15) || 132.3 (45.8) |
The general result of conservation of the first principal component is corroborated by a comparison of the M and D matrices through the Flury hierarchy, as implemented by the CPC program. The comparisons yielded the PCPC1 model in all three cases (P= 0.04 for N2, P= 0.005 for mut-7, and P= 0.01 for mut-14), which means that the matrices shared only their first principal component out of the four. The first principal component was moderately correlated with fecundity, length and width in all matrices, but not always with longevity (Table 5).
Table 5. Principal component analysis of the M and D matrices. Only the first two principal components (PC) of each matrix are described, along with the percentage of the total variance that they explain.
|% variance|| 42.60%|| 32.36%||60.21%|| 26.13%||68.97%|| 16.45%|| 52.33%|| 31.59%|
| Fecundity || 0.527 || 0.54 || 0.463 || 0.489 || 0.428 || −0.859 || 0.584 || 0.317 |
|Longevity||−0.234|| 0.72|| 0.403|| 0.622|| 0.497|| 0.112||−0.02||−0.836|
| Length || 0.47 || −0.418 || 0.577 || −0.379 || 0.535 || 0.94 || 0.507 || −0.447 |
|Width|| 0.668|| 0.119|| 0.539||−0.48|| 0.533|| 0.491|| 0.634|| 0.039|
The mutation accumulation (MA) experiment performed with the N2 strain of C. elegans was successful in producing mutational variance in all traits (see also Bégin and Schoen 2006), although the range of mutational heritabilities obtained for the four traits is slightly higher than that seen in other studies of spontaneous mutation in N2 (Keightley and Caballero 1997; Vassilieva et al. 2000; Azevedo et al. 2002; Ajie et al. 2005). In our study of N2, only one mutational covariance, the one between length and width, was significant and positive, as logically expected from two body size traits. There is no evidence for mutational trade-offs and our results rather suggest that mutations are mostly deleterious. The fact that body size went up during MA (Bégin and Schoen 2006), which is contrary to what was found elsewhere (Azevedo et al. 2002; Ajie et al. 2005), may, on the surface, appear to reflect that mutations are not deleterious, but it is also possible that stabilizing selection in natural populations maintains size at an optimum, so that any deviation from that optimum causes a decrease in fitness. Other mutation studies using N2 have reached conclusions similar to ours with these or other traits—they have found moderately positive and significant correlation estimates (Keightley et al. 2000; Azevedo et al. 2002; Estes et al. 2005).
The MA experiments performed on the two mutator strains, mut-7 and mut-14, showed stronger mutational effects. Mutational correlations were generally more positive and significant in the mutators compared with N2, but still revealed an overall pattern similar to that of N2—no mutational trade-offs were observed, and mutation caused a generally deleterious effect on traits. We conclude that germline TE activity, which occurs in the mutators, but not in N2, causes an increase in the mutation rate (Bégin and Schoen 2006) as well as an intensification of the covariation pattern, compared to that of other types of spontaneous mutations (as in the N2 strain). It is, however, not clear why similar mutational patterns were produced in the two different mutator strains. There is no a priori reason to believe that random insertions of TEs will produce the same pattern of multivariate mutations in two different strains, unless insertional hot spots existed for a substantial portion of all TEs. Alternatively, this similarity could just be a reflection of the high numbers of generally deleterious mutations seen in both of the strains (Bégin and Schoen 2006), causing traits to be correlated approximately in the same way. Regardless of the underlying reason, germline transposition does have an impact on the rate and pattern of mutation, and may therefore influence population evolution.
Although TEs have an impact on mutational patterns, the extent to which mildly deleterious mutations, TE-mediated or not, affect long-term evolution is not clear. Theory proposes that the G matrix is constantly inflated by pleiotropic mutations (M), but that the magnitude and pattern of the mutational variances and covariances are eroded and reshaped by stabilizing and correlational selection (Cheverud 1984; Bürger 2000) until the equilibrium G matrix is attained. Most deleterious mutations may thus be transient (Houle et al. 1996). One way to test for the importance of new mutations in shaping the evolutionary trajectory of a lineage is to test the constraint hypothesis. Although genetic variation across natural populations of C. elegans is known to be very low (Jovelin et al. 2003; Sivasundar and Hey 2003), significant but modest divergence in phenotypic traits was detected in this study. Comparing population divergence (D) to mutational covariances (M) revealed that only the first principal component of the two matrices is similar, although not equal. This indicates that the divergence of natural C. elegans populations has proceeded, more or less, along the genetic line of least resistance (gmax, Schluter 1996). The second principal component of the M matrix, however, differs widely, which indicates that the corresponding constraint was overridden during population divergence. A principal component interpretation reveals that C. elegans populations diverged mostly along an axis that combines in a positive manner both size and fecundity. Because that axis is also characteristic of the M matrix (the importance of longevity differs between strains), random genetic drift is all that is necessary to explain divergence (Phillips et al. 2001), although natural selection differentially favouring the large and fecund individuals across populations could also be compatible with the pattern. Population divergence along the second most important axis, which mainly polarizes long- and short-lived worms, must have required a strong push by natural selection, because the genetic constraints have been overridden. This interpretation suggests that only some portion of the mutational output has been useful during the long-term evolution of the populations compared here.
An important caveat is that we assumed that the M matrices characterized here are adequate surrogates for the ancestral C. elegansM matrix, an assumption that we cannot test. A reassuring result, however, is that the three different M matrices provide similar results when compared to D, even though one includes germline transposition whereas the others do not. This may be explained by the fact that whereas some strains in nature, such as N2, repress germline transposition, others such as the Bergerac strain, do not (Eide and Anderson 1985). Moreover, other natural strains such as TR403 and DH424 have large numbers of copies of specific TEs (Egilmez et al. 1995), which may imply that transposition has occurred in the past in those strains or their ancestors. Alternatively, the similar results for N2, mut-7, and mut-14 may indicate that, although TEs influence M matrices on the short term, most of the mutations they cause are deleterious and do not persist in the population, in which case TEs would not ultimately influence the microevolution of quantitative traits in natural populations anymore than the average spontaneous mutation. This does not preclude the fact that, sometimes, genomes recruit some conveniently inserted TE copies to produce adaptations (Kidwell and Lisch 1997; Pardue and DeBaryshe 2003).
Associate Editor: C. Goodnight
This work was supported by the Canadian Institutes of Health Research, the Natural Sciences and Engineering Research Council of Canada, and le Fonds Québécois de la Recherche sur la Nature et les Technologies. Nematode strains were provided by the Caenorhabditis Genetics Center, which is funded by the National Institutes of Health National Center for Research Resources. We thank T. Bureau and J. Dent for technical advice and discussions, and C. Belair, K. Brown, É. Godin, M. Kooistra, A.-M. L'Heureux, K. Martin, A. Pastor, D. Raab, F. Saadé, I. Shahin, J. Sharma, and V. Tzaneva for laboratory assistance.
According to phylogenetic theory (Felsenstein 1985; Garland et al. 1992), estimating trait correlations across taxa requires the use of a phylogenetic tree with proper branch lengths. This information, however, is not available for all of the populations of C. elegans that we used in this study (together, the phylogenies published by Denver et al.  and Cutter  include only nine of our 15 populations). Not correcting for phylogeny can lead to increased type I error rates, but not to substantial bias in the point estimates of the correlations (Rohlf 2006). To empirically test whether our lack of phylogenetic information was problematic, we generated 1000 random phylogenetic trees using the program COMPARE (Martins 2004; standard branching process with default settings), and estimated correlations between all pairs of traits for each iteration using the Independent Contrasts option. The mean of such a distribution of correlation coefficients can be used as a reasonable estimate of the evolutionary correlation of interest (Martins 1996). As expected, point estimates of correlation coefficients were not affected substantially by phylogeny, although standard errors were much smaller when no phylogenetic correction was applied (Table A1). For simplicity, and because standard errors were not used in the test of the constraint hypothesis in this study, we did not proceed to conduct any phylogenetic correction.
Table A1. Correlation coefficients (standard errors) with and without phylogenetic correction.
|fecundity–longevity||−0.26 (0.24)||−0.27 (0.40)|
| fecundity–length || 0.36 (0.18) || 0.29 (0.38) |
|fecundity–width|| 0.71 (0.13)|| 0.72 (0.20)|
| longevity–length || 0.30 (0.28) || 0.27 (0.40) |
|longevity–width||−0.04 (0.24)||−0.05 (0.41)|
| length–width || 0.55 (0.15) || 0.49 (0.35) |