Adaptive molecular evolution of a defence gene in sexual but not functionally asexual evening primroses


Erika I. Hersch-Green, Department of Biological Sciences, Michigan Technological University, Houghton, MI 49931, USA.
Tel.: +1 906 487 3351; fax: +1 906 487 3167; e-mail:


Theory predicts that sexual reproduction provides evolutionary advantages over asexual reproduction by reducing mutational load and increasing adaptive potential. Here, we test the latter prediction in the context of plant defences against pathogens because pathogens frequently reduce plant fitness and drive the evolution of plant defences. Specifically, we ask whether sexual evening primrose plant lineages (Onagraceae) have faster rates of adaptive molecular evolution and altered gene expression of a class I chitinase, a gene implicated in defence against pathogens, than functionally asexual evening primrose lineages. We found that the ratio of amino acid to silent substitutions (Ka/Ks = 0.19 vs. 0.11 for sexual and asexual lineages, respectively), the number of sites identified to be under positive selection (four vs. zero for sexual and asexual lineages, respectively) and the expression of chitinase were all higher in sexual than in asexual lineages. Our results are congruent with the conclusion that a loss of sexual recombination and segregation in the Onagraceae negatively affects adaptive structural and potentially regulatory evolution of a plant defence protein.


Sex is the dominant mode of reproduction among euka-ryotic life, but why sex prevails over asexual reproduction is among the most important unresolved and contested questions in biology (Bell, 1982; Barton & Charlesworth, 1998; Salathéet al., 2008b; Otto, 2009; Becks & Agrawal, 2010; Lively, 2010). Despite being less common, asexual reproduction is widespread across the tree of life and is associated with many ecological and evolutionary advantages over sexual reproduction. For example, asexual individuals can avoid ecological and genetic costs of sex by reproducing faster (‘cost of males’, Maynard Smith, 1978), and they also exhibit greater relatedness to offspring (‘cost of meiosis’, Williams, 1975; Lively & Lloyd, 1990). In contrast, the prevalence of sex is attributed to evolutionary advantages associated with recombination and segregation of alleles, which are greatly suppressed or completely absent in asexual individuals. For example, evolutionary theory predicts that recombination and segregation enable more effective purging of deleterious mutations (Muller, 1964; Lynch & Gabriel, 1990) and promote beneficial (adaptive) allele combinations (Muller, 1932; Agrawal, 2006; Otto, 2009). There has been some empirical work in animal systems, showing that asexual lineages have the propensity to accumulate more deleterious mutations than related sexual lineages (Paland & Lynch, 2006; Johnson & Howard, 2007; Neiman et al., 2010), but we are unaware of any studies that examine whether asexual lineages also experience slower rates of adaptive molecular evolution (positive selection) than related sexual lineages.

Natural enemies of plants are ubiquitous in nature and can have major negative fitness consequences for host species. Under such circumstances, natural enemies are expected to impose selection on host species to evolve adaptive defence traits that help them avoid attack and/or minimize negative fitness consequences of attack (Ehrlich & Raven, 1964; Zangerl & Berenbaum, 2005; Stenberg et al., 2006). Understanding the effects of recombination and segregation on the molecular evolution of host defence genes could provide key insights into whether sexually reproducing hosts have an advantage over asexually reproducing hosts when responding to natural enemies and whether sex promotes adaptive molecular evolution. For example, if plants adapt to natural enemies by altering proteins that recognize or attack natural enemies, then sexual lineages should exhibit stronger signatures of positive selection on defence proteins than asexual lineages. Such adaptive changes might not only involve structural changes to proteins (e.g. amino acid replacements in coding regions) but might also rely on changes in regulatory regions or a combination of these factors (Doebley, 1993; Purugganan, 1998; Hoekstra & Coyne, 2007).

Here, we first show that pathogens can potentially exert selection on Onagraceae species and then we test whether a loss of sexual recombination and segregation in plants negatively impacts adaptive molecular evolution in a plant gene involved in defence against pathogens. We focus on a class I basic chitinase (chiB) because plant chitinases are well studied and have a clear function in plants. Specifically, plant chitinases are disease resistance enzymes that are present in virtually all plants, and they aid in both the recognition and attack of fungal pathogens (Collinge et al., 1993; Kasprzewska, 2003; Ferreira et al., 2007). ChiB also offers an excellent model to test whether recombination and segregation (two fundamental components of sex) promote adaptive molecular changes within a plant defence gene because previous work on plants, without regard to variation in sexual reproduction, showed that specific amino acid sites within chiB are under positive selection (Bishop et al., 2000; Tiffin, 2004). We test our hypotheses by examining species belonging to the evening primrose family (Onagraceae). This family is an ideal plant system for testing the effects of sex on plant defences because it contains many sexual and functionally asexual species, where a breakdown in sex arises because of a genetic system called permanent translocation heterozygosity (PTH). PTH species retain fertilization but experience a near-complete suppression of both recombination and segregation such that all seeds produced by a maternal plant are genetically identical (Stebbins, 1950; Cleland, 1972; Rauwolf et al., 2008, 2011). Furthermore, there have been many (> 20) independent transitions between sexual and functionally asexual (PTH) reproduction in the Onagraceae (Holsinger & Ellstrand, 1984; Johnson et al., 2009, 2011), which allows for a naturally replicated and robust test of the effects of sex on plant defence evolution.

Based on evolutionary theory, we predicted that the repeated losses of recombination and segregation in the Onagraceae cause decreased rates of adaptive molecular evolution in chiB. To test this prediction, we applied maximum-likelihood comparative phylogenetic approaches to infer rates of structural molecular evolution of the chiB enzyme across 16 sexual and 16 asexual evening primrose (Onagraceae) plant lineages. We further compared chiB gene expression across a subset of these lineages to determine whether a loss of sex is also associated with changes in regulatory evolution.

Materials and methods

Pathogen severity and plant fitness

We examined plant fitness and the severity of damage by a generalist foliar pathogen, powdery mildew (Erysiphe sp.), on different genotypes of Oenothera biennis (a functionally asexual PTH species) to determine whether pathogens could select on genetic variation in resistance. We measured the severity of damage on 12 different genotypes of O. biennis planted at a single field site (University of Toronto’s Koffler Scientific Reserve) 50 km north of Toronto, Canada, a relatively high-latitude site for this and other Oenothera species. These plots were established for a separate study, and detailed methods of the experimental design are provided in this previous study (Johnson et al., 2006). Pathogen severity was not previously reported but was measured as percentage leaf area covered with Erysiphe spp. on the upper surface of the first 15 leaves, counting from the base of the plant, where area covered was scored between 0 and 4 (0 = no mildew, 1 = 1–25%, 2 = 25.1–50% covered, 3 = 50.1–75% and 4 = 75.1–100%). Plant fitness was measured as lifetime total number of fruits produced.

Plant and genetic material

Total genomic DNA was extracted from 30 Oenothera and 2 Gayophytum spp. (representing 16 PTH and 16 sexual species; Fig. 1) following standard CTAB protocols. For RNA extraction and cDNA synthesis, we first grew replicate plants of 25 Oenothera spp. (Fig. 1) at the same time and under controlled environmental conditions in a randomized design. Then, for all plants, we harvested leaf tissue from the tips of rapidly expanding true leaves of seedlings when they reached the four-leaf stage by immediately placing tissue in liquid nitrogen. Total RNA was extracted using Qiagen’s RNeasy Plant Mini Kit (Qiagen, Hilden, Germany) with minor modifications to remove interfering mucilage and polyphenols (Greiner, 2008). RNA concentration and purity were determined spectrophotometrically (Nanodrop 2000c, Thermo Scientific, Wilmington, DE, USA), and 150 ng of total RNA from each sample was reverse-transcribed using M-MLV reverse transcriptase (Invitrogen, Karlsruhe, Germany) and an Oligo(dT)24 primer (GeneRacer Kit; Invitrogen). Residual RNA complements were removed from the synthesized cDNA by incubation with RNase H (Invitrogen, Carlsbad, CA, USA). RNA extraction and cDNA synthesis were performed for all samples simultaneously.

Figure 1.

 Evolutionary history of sexual and functionally asexual permanent translocation heterozygote (PTH) reproduction in the Onagraceae, adapted from the study of Johnson et al. (2009). Transitions between sexual (blue/dark grey in print edition) and PTH (red/mid-grey in print edition) reproduction are shown, and Bayesian posterior probabilities are shown above branches and are indicated when > 50%; *posterior probability > 95%. Rates of molecular evolution (ω = Ka/Ks) are displayed below each branch and were estimated in CODEML from a model that allowed each branch to evolve independently (lnL = −3172.75, d.f. = 63). Squares show species used in gene expression analyses.

Molecular sequence data

We used a partial Oenothera chiB sequence (Mráček et al., 2006, GenBank accession FG122686.1) and the Arabidopsis class I orthologous chitinase sequence (Samac et al., 1990, GenBank accession M38240) to design primers to amplify a 756-bp coding region corresponding to class I basic chitinase (chiB) from our samples (Table S1). PCR products were cloned into a pGEM®-T Easy vector (Promega, Madison, WI, USA) and transformed into Escherichia coli grown on LB with ampicillin to facilitate blue-white screening of successfully transformed colonies. Cloned products were PCR-amplified with M13 forward and reverse primers and sequenced in both directions. Initial sequencing of eight clones from each of a phylogenetically diverse set of eight species showed that chiB is likely present as a single copy gene within our species (no individual showed > 2 alleles/locus). We therefore amplified and sequenced the forward and reverse sequences of four colonies per species for all 32 species. All sequencing was performed using an ABI 3730xl capillary sequencer following the ABI Big Dye terminator sequence protocol (ABI Biosystems, Foster City, CA, USA). Raw sequence data were edited using sequencher 4.10.1 software (Gene Codes Corporation, Ann Arbor, MI, USA), and consensus sequence data were generated for each species. Consensus sequences of the coding region for all 32 species (756–777 bp) were aligned with clustalw followed by manual manipulations in BioEdit (Hall, 1999); the nucleotide alignment used for subsequent analysis is provided (Table S2), and sequences are available as GenBank accessions and in Dryad (doi:10.5061/dryad.sk94j). Lastly, to compare our results for chiB with a gene not involved in defence, we also examined molecular evolution of cytosolic phosphoglucose isomerase (PgiC), a gene that is known to play a highly conserved role in plant metabolism of glucose into pyruvate during ATP synthesis. The sequence data used for this analysis were taken directly from the study of Johnson et al. (2009), where 459 bp of coding sequence was available for 10 sexual and 10 PTH species.

RT-qPCR expression data

A single primer pair was designed from a highly conserved region of chiB to amplify the gene across all species (Table S1). After screening several potential reference control genes, we selected elongation factor 1a (EF1α) as our control gene because analysis showed that it gave the most consistent amplification and expression results for the species considered in this study (Supporting information). For each species and gene, we assayed expression from two to four biological replicates (individual plants) and two technical replicates per biological replicate. We also included ‘no-template’ controls to detect contamination. All RT-QPCRs were performed in a Stratagene Mx 3000P (Stratagene, La Jolla, CA, USA) as a 25-μL reaction, which contained 2× Brilliant II SYBR Green QPCR mastermix (with low ROX; Stratagene), 0.2 μm of each forward and reverse primers and 20 ng RNA equivalent of cDNA. PCR cycling conditions followed normal protocols (Table S1).

Statistical analysis

Selection by pathogens

To determine whether pathogens affect plant fitness and could act as selective agents, we first analysed pathogen severity and plant fitness data separately to obtain mean values for each genotype using generalized linear models with type III error of sas (Proc GLM; SAS Institute, Cary, NC, USA) as described in the study of Johnson et al. (2006). We then correlated mean genotype pathogen damage and relative plant fitness values to calculate a standardized selection gradient on resistance to the pathogen (Lande & Arnold, 1983).

Molecular sequence evolution

We tested for selection on chiB using maximum-likelihood methods implemented in CODEML of the paml package (Yang, 2007). The input tree for all analyses was an unrooted tree generated using protein coding and noncoding sequences from five genes (Johnson et al., 2009) and does not include sequence information from chiB analysed here. We chose this phylogeny because relationships between taxa were well supported (e.g. high Bayesian posterior probabilities) and the mode of reproduction (sexual or PTH) could be reliably mapped onto the phylogeny. This ancestral state reconstruction showed at least 12 independent transitions between sex and PTH, as published in the study of Johnson et al. (2009) and shown in Fig. 1. Comparisons of the ratio of nonsynonymous (Ka) to synonymous (Ks) substitution rates (ω = Ka/Ks) among lineages and codon sites were used to infer the type of selection acting on chiB, where ω < 1 corresponds to purifying selection, ω = 1 corresponds to neutral evolution and ω > 1 corresponds to positive selection (Li et al., 1985; Nei & Gojobori, 1986; Yang & Bielawski, 2000). We used three different types of models to address the questions of our study: ‘branch models’, ‘site models’ and ‘branch-site models’, and the goodness of fit of different nested models was assessed using likelihood ratio tests.

Branch models test whether different lineages are evolving at different rates by allowing for variable ω among branches in the tree, but invariable ω among different amino acid sites (Yang, 1998; Yang & Nielsen, 1998). We used a one-ratio model as the null model to determine the average overall selection pressure acting upon chiB (H0, one estimate of ω), and then we compared this model with two alternative models (tests 1 and 2), both of which allowed for variable ω between PTH (ωPTH) and sexual (ωSEX) lineages. The alternative model in test 1 was a two-ratio model (H1, two estimates of ω: ωPTH and ωSEX), in which the character states (PTH or SEX) of internal branches of the phylogeny were estimated as shown in Fig. 1. To determine whether the estimated character state of the internal branches influenced overall estimates of ω between PTH and sexual lineages, we also compared H0 (test 2) with a three-ratio model (H2, three estimates of ω: ωPTH, ωSEX and ωinternal). Although branch models test whether ω varies among lineages, it is not clear whether any observed differences are due to difference in Ka, Ks or both. To address this limitation, we obtained estimates of nonsynonymous (Ka) and synonymous (Ks) substitutions for each lineage using a model in CODEML that allowed each branch to evolve independently, and then we examined the effects of sex on these rates with analysis of variance (anova) tests.

Site models account for different functional or structural constraints experienced by individual amino acid sites within a protein by allowing for variable ω among sites, but invariable ω among branches (Nielsen & Yang, 1998; Yang et al., 2000). These models estimate the strength of selection and the proportion of amino acid sites that experience purifying (ω0), neutral (ω1) and/or positive (ω2) selection. Following Yang et al. (2000), we made several comparisons to test for rate heterogeneity among amino acid sites. First, we compared the one-ratio model (H0) with a discrete model (M3, three estimates of ω), which allows for purifying, neutral and positive selection to act on sites (test 3). Next, because we detected variation in selection among sites, we made two additional comparisons. First, we compared the nearly neutral model M1a (two estimates of ω for purifying and neutral selections) with the positive selection model M2a (three estimates of ω for purifying, neutral and positive selections), which also allows for positive selection (test 4). Second, we compared the M7 model that assumes a β distribution of class sites that range from purifying selection to neutral evolution (0 < ω < 1) with the continuous M8 model, which has an additional class of sites with ω > 1 that are under positive selection (test 5).

We used branch-site models to determine whether certain sites are under positive selection in only PTH or sexual lineages, because these models simultaneously account for heterogeneity in selection between different lineages and across amino sites (Yang & Nielsen, 2002; Zhang et al., 2005). In all branch-site models, we used a phylogeny with the topology and mode of reproduction (PTH or sexual) mapped, as published in the study of Johnson et al. (2009) and shown in Fig. 1, because likelihood ratio tests between various models showed that this character state topology best fits the data (Table 1; tests 1 and 2). We used the branch-site model A (Zhang et al., 2005) as an alternative hypothesis with both PTH and sexual lineages set as foreground branches because only the foreground branches are allowed to experience positive selection in CODEML. We tested these alternative hypotheses against two null hypotheses: the nearly neutral model (M1a) and the same branch-site model A, but with ω2 fixed at 1 (Zhang et al., 2005; tests 6–9). We also compared the new clade model C, which estimates ω2 separately for different lineages without any constraints (e.g. ω2 can be any value equal to or greater than zero), with the nearly neutral model (Yang, 2007; test 10), to determine whether any sites that are under positive selection in sexual lineages are also under positive selection in PTH lineages and vice versa. To assess whether any patterns of selection observed for chiB are specific to chiB or to genes showing positive selection, we repeated similar analyses in CODEML for a ‘control’ gene, PgiC, which a previous study had generated coding sequence data (459 bp) for 10 of the sexual and 10 PTH species included here (Johnson et al., 2009). To determine whether sites that we identified to be under positive selection in the above analyses were also implicated to be under positive selection in different plant species with divergent histories, we aligned the chiB sequence from O. grandiflora (sexual species) to homologous sequences published for Boechera and Zea (Bishop et al., 2000; Tiffin, 2004) to compare sites under positive selection.

Table 1.   Likelihood values and parameter estimates from PAML analyses of chiB across 32 Onagraceae species. The phylogeny and mapped character states used for these analyses are reported in the study of Johnson et al. (2009) and do not use sequence information from chiB. Results from comparative analyses that use a phylogeny derived from chiB sequence information are reported in Supporting information and Table S4.
ModelfplnLEstimates of parameters†Positively selected sites‡
  1. fp, number of free parameters; lnL, log-likelihood value for each model.

  2. †Parameters estimated in models: In branch-specific models, ωPTH and ωSEX are the Ka/Ks ratios for permanent translocation heterozygote (PTH) and sexual lineages, respectively; In site-specific and branch-site models, p0, p1 and p2 equal the proportion of sites with Ka/Ks ratios less than one (ω0), roughly equal to one (ω1) and greater than one (ω2), and the ω values reflect the estimates of Ka/Ks for purifying (ω0), neutral (ω1) and positive (ω2) selection, respectively. In branch models M7 and M8, site classes are drawn from a beta distribution with β(p,q), where p and q are the shape parameters. Values in parentheses are not directly estimated but are determined by the relationship that p0 + p1 p2 = 1 or they are fixed in the model.

  3. ‡Positively selected sites, where *P > 95%; **P > 99%, following Bayes empirical Bayes analysis (M2a, M8, Model A_alt) or naïve empirical Bayes (M3). Final determination of significance followed likelihood ratio tests (Table 2).

H0: one-ratio1−3213.62ω = 0.17None
Branch-specific models
 H1: two-ratio2−3211.34ωPTH = 0.11, ωSEX = 0.20NA
 H2: three-ratio3−3211.06ωPTH = 0.10, ωSEX = 0.22, ωinternal = 0.18NA
Site-specific models
 M3: discrete5−3091.25p0 = 0.73, p1 = 0.22 (p2 = 0.05)
  ω0 = 0.02, ω1 = 0.40, ω2 = 3.12
13*, 32**, 37**, 126**, 145**, 154**, 207**, 208**
 M1a: nearly neutral2−3108.97p0 = 0.87 (p1 = 0.13), ω0 = 0.04NA
 M2a: positive selection4−3093.65p0 = 0.88, p1 = 0.09 (p2 = 0.03)
  ω0 = 0.05,ω2 = 3.96
32**, 207**
 M7: β2−3112.47p = 0.09, q = 0.44NA
 M8: β + ω > 14−3091.05p = 0.19, q = 1.44
  p0 = 0.96 (p2 = 0.04)
  ω2 = 3.30
32**, 37*, 126*, 145*, 154*, 207**, 208*
Branch-site models
 Model A_null (PTH)3−3107.69p0 = 0.86, p1 = 0.12 (p2 = 0.02)
  ω0 = 0.04 (ω1 = ω2 = 1)
 Model A_alt (PTH)4−3106.77p0 = 0.87, p1 = 0.12 (p2 = 0.01)
  ω0 = 0.04, ω2 = 2.89
 Model A_null (SEX)3−3106.15p0 = 0.86, p1 = 0.06 (p2 = 0.08)
  ω0 = 0.04 (ω1 = ω2 = 1)
 Model A_alt (SEX)4−3082.82p0 = 0.88, p1 = 0.09 (p2 = 0.03)
  ω0 = 0.05, ω2 = 5.65
32**, 126*, 154*, 207**
 Clade Model C5−3082.48p0 = 0.88, p1 = 0.09 (p2 = 0.03)
  ω0 = 0.05, ω2PTH = 0.05, ω2SEX = 5.69 (ω1 = 1)

Lastly, we performed several additional analyses to further determine whether (i) the topology and character state assignment of our phylogeny, (ii) variation in the rate of nonsilent substitutions (Ks) or (iii) recombination influenced the results of CODEML (Supporting information). These additional analyses are not as comprehensive as the analyses discussed above, but they have the advantage of allowing us to examine potential confounding effects that could arise from an incorrect phylogeny or from variation in the rates of nonsilent substitutions (Ks) or recombination; CODEML does not account for the latter. The results of these additional analyses were largely consistent with our analyses discussed above, indicating that our results are robust to assumptions made about the specific phylogeny and assumptions made by CODEML (see Supporting information). Therefore, we report the full results from these additional analyses online.

ChiB expression

Expression was quantified as the number of cycles needed for the amplification to reach a threshold fixed in the exponential phase of the PCR (CT value). For each species and each biological replicate, chiB expression was calculated as the average CT value of chiB relative to the average CT value of EF1α. Because preliminary analyses of the expression data (following methods of Blomberg et al., 2003; Ives et al., 2007) found that chiB trait evolution was independent of the species phylogeny (results available upon request), we compared expression levels between species with a standard one-way anova.

Results and discussion

Selection by a pathogen on a PTH Oenothera

Both the severity of pathogen damage by Erysiphe sp. and plant fitness varied among the 12 O. biennis genotypes planted in the field. Plant fitness was significantly negatively genetically correlated with pathogen damage (Fig. 2), and in comparison with other studies of natural selection (Kingsolver et al., 2001; Hersch & Phillips, 2004), the strength of genotypic selection imposed by pathogens was moderate to strong (S′ = ß′ = −0.11).

Figure 2.

 Plant relative fitness as a function of mean foliar damage by powdery mildew (Erysiphe sp.) for 12 different genotypes of Oenothera biennis, a functionally asexual permanent translocation heterozygosity (PTH) species (R2 = 0.45; < 0.02).

Adaptive evolution within sexual and asexual Onagraceae

When we averaged over all sites and lineages, we detected purifying selection acting on chiB (H0: ω = 0.17 < 1, Table 1). However, when we allowed functionally asexual and sexual lineages to evolve separately, we found clear differences in molecular evolution of chiB between asexual PTH and sexual lineages. Specifically, the average rate of nonsynonymous to synonymous substitutions (ω = Ka/Ks) was significantly lower for PTH lineages (ωPTH = 0.11) than for sexual lineages (ωSEX = 0.19; test 1, = 0.033; Tables 1 and 2). Higher values of ω, even when less than one, can arise from positive selection acting on a small number of specific amino acid sites. Consistent with this interpretation, we detected significant positive selection acting on four sites in sexual lineages (ω2 = 5.65, < 0.0001 for tests 8 and 9; Tables 1 and 2), whereas zero sites exhibited significant positive selection in PTH lineages (ω2 = 2.89, > 0.1 for tests 6 and 7; Tables 1 and 2). Furthermore, the sites identified to be under positive selection in sexual lineages appear to be under purifying selection in asexual lineages (ω2PTH = 0.05 and ω2SEX = 5.69, test 10). Among the four positively selected sites in sexual lineages, one was in the active-site cleft region of the protein, which is thought to be important in pathogen recognition (Bishop et al., 2000). These results are congruent with the prediction that a loss of recombination and segregation severely constrains adaptive molecular evolution of a plant gene involved in defence against pathogens.

Table 2.   Likelihood ratio tests between different CODEML models in paml (Table 1).
TestCompared models (alt vs. null)d.f.−2ΔlnLP value
  1. Test 1: tests for differences in ω between sexual and permanent translocation heterozygote (PTH) lineages, where character state of internal branches is estimated and shown in Fig. 1.

  2. Test 2: tests for differences in ω between sexual and PTH lineages (tips), where the character state and ω of the internal branches are treated as an unknown and separate character state as the tips.

  3. Test 3: tests whether there is rate heterogeneity among amino acid sites.

  4. Test 4: tests whether rate heterogeneity among amino acid sites includes positive selection at some sites.

  5. Test 5: tests whether positive selection, not purifying selection, occurs at some sites.

  6. Test 6: tests whether positive selection occurs in PTH lineages; null model fixes ω2 = 1.

  7. Test 7: tests whether positive selection occurs in PTH lineages; null model does not include ω2.

  8. Test 8: tests whether positive selection occurs in sexual lineages; null model fixes ω2 = 1.

  9. Test 9: tests whether positive selection occurs in sexual lineages; null model does not include ω2.

  10. Test 10: tests whether rate heterogeneity among amino acid sites, which includes positive selection at some sites, differs between lineages.

 1H1 vs. H014.560.033
 2H2 vs. H035.120.077
 3M3 vs. H04244.74< 0.0001
 4M2a vs. M1a230.65< 0.0001
 5M8 vs. M7242.84< 0.0001
 6ModelA_ null vs. ModelA_alt (PTH)11.840.1750
 7M1a vs. ModelA_alt (PTH)24.400.1108
 8ModelA_ null vs. ModelA_alt (Sex)146.66< 0.0001
 9M1a vs. ModelA_alt (Sex)252.30< 0.0001
10Clade Model C vs. M2a352.98< 0.0001

Differences in ω between asexual and sexual lineages may not be entirely dependent on differences in the frequency of positive selection on amino acid substitutions (Rausher et al., 2008). For instance, higher overall ω in sexual lineages could also arise from differences in synonymous substitutions (Ks) with or without changes in nonsynonymous substitutions (Ka). We tested this possibility by examining variation in Ks and Ka among species. We found no evidence that Ks differed between PTH and sexual lineages (F1,30 = 0.113, = 0.739; Table S3). Furthermore, additional phylogenetic analyses that explicitly incorporate variation of Ks into estimates of ω identified similar patterns of selection acting on chiB as our analyses in CODEML, which does not allow Ks to vary (Supporting information). In contrast, we found that Ka marginally differed between sexual and PTH lineages (F1,30 = 3.102, = 0.088, Table S3) where Ka was three times higher in sexual species (0.006 ± 0.002) than in asexual PTH species (0.002 ± 0.002). Variation in Ka among lineages could be due to both the accumulation of beneficial (positively selected) and deleterious mutations (Kimura, 1977; Rausher et al., 2008). The presence of four positively selected amino acid sites and the fact that deleterious mutations are predicted and observed to accumulate disproportionately within asexual lineages (e.g. Muller, 1964; Paland & Lynch, 2006; Neiman et al., 2010), not within sexual lineages, both suggest that the latter explanation is less plausible than differences in positive selection. Cumulatively, our results are consistent with differences in positive selection between functionally asexual PTH and sexual lineages.

A nonmutually exclusive alternative explanation is that differences in the ecology between sexual and asexual species, aside from differences in adaptive constraint, contribute to the observed differences in protein evolution. For example, PTH species tend to occur at slightly higher latitudes than sexual species (Johnson et al., 2010), and it has been argued that biotic selection pressures might be weakest at high latitudes (Schemske et al., 2009; but see Moles et al., 2011). Such a scenario could cause selection by pathogens to be stronger on sexual than on asexual populations. Although we acknowledge that ecological factors can influence molecular evolution, there are several reasons why this ecological mechanism is unlikely to fully explain our results. First, closely related sexual and PTH species often co-occur in the same habitat and location (Johnson et al., 2010), where they are likely to be exposed to similar selection pressures by pathogens. Moreover, a complex assemblage of pathogens, including powdery mildews (Erysiphe spp.), leaf-spot fungus (Septoria oenotherae) and rusts (Puccinia spp.; Farlow & Seymour, 1888; Klaphaak & Bartlett, 1922; Dickey & Levy, 1979), attacks both sexual and PTH taxa throughout their ranges, including at high latitudes. Finally, we found that even at relatively high-latitude sites, PTH populations can experience strong selection by pathogens. It is not possible to test whether all ecological factors are roughly equal between lineages, but there is no a priori reason to suggest that such differences are consistent and explain our findings of positive selection within sexual but not within asexual lineages. For example, Onagraceae includes annual, biennial and perennial species, and there is no association between these different life history traits and rates of synonymous or nonsynonymous substitutions within chiB in our data set (Table S3). We therefore conclude that the most parsimonious answer to explain our results are that differences in recombination and segregation directly influenced the ability of populations to respond to selection on chiB.

Comparisons of the effects of sexual reproduction on genes expected to be under purifying selection

For genes under strong purifying selection, the opposite pattern of selection is expected where asexual lineages accumulate more deleterious mutations than sexual lineages, indicated by higher rates of nonsynonymous substitutions. The influence of sexual reproduction on the molecular evolution of the nonrecombining mitochondrial genome has been examined in three recent studies involving Daphnia and snails (Paland & Lynch, 2006; Johnson & Howard, 2007; Neiman et al., 2010). These studies all reported that a loss of sex was associated with stronger purifying selection on proteins in sexual as opposed to asexual lineages (ωSEX < ωASEX < 1), consistent with relaxed purifying selection and an accumulation of deleterious mutations in asexual lineages. This result is expected because mitochondria are essential for cellular respiration and the function of mitochondrial genes is highly conserved among eukaryotes. Therefore, virtually all nonsynonymous mutations are likely to be deleterious, and these mutations are predicted to be more efficiently purged by sexual than by asexual reproduction.

Although defence against pathogens can also be critical for a plant’s survival, coevolution between plants and their parasites should cause selection on nuclear defence genes to be more dynamic than selection on mitochondrial genes. Moreover, nonsynonymous mutations of defence genes may be either beneficial or deleterious, and theory predicts that a loss of recombination and segregation (asexuality) negatively affects the fixation of beneficial mutations and thus adaptive molecular evolution of plant defence genes (Howard & Lively, 1994; Salathéet al., 2008a). To assess whether asexual PTH Onagraceae lineages are also more likely than sexual Onagraceae lineages to accumulate deleterious mutations in genes under strong purifying selection and/or that are not directly related to defence, we ran similar analyses in CODEML for PgiC (data from Johnson et al., 2010; Table S4). Overall, we found that ω = 0.17 (identical to chiB), but there was no evidence for positive selection and ω did not significantly differ between lineages (Table S4). Furthermore, we have also conducted similar analyses for four genes in the flavonoid biosynthetic pathway, which overall are under much stronger purifying selection (0.04 ≤ ω ≤ 0.09; E. I. Hersch-Green, H. Myburg & M. T. J. Johnson, unpublished data). Interestingly, for three of the genes, we found that ω was higher in the PTH than in the sexual lineages, a pattern that would be expected if deleterious mutations accumulate disproportionately in functionally asexual PTH lineages (E. I. Hersch-Green, H. Myburg & M. T. J. Johnson, unpublished data). Our results combined with previous studies provide empirical support for theoretical predictions that recombination and segregation have a potential adaptive benefit for organisms by both increasing the fixation of beneficial mutations and eliminating deleterious mutations.

Evolution of chitinase among plant lineages

Molecular evolution of chitinases has been studied in other plant lineages but without regard to variation in sexual reproduction. When we ignore differences in reproductive mode (asexual vs. sexual), we found that between two (M2a), seven (M8) or eight (M3) amino acid sites were evolving under significant positive selection in the Onagraceae [ω2 (M3) = 3.12, ω2 (M2a) = 3.96, ω2 (M8) = 3.30; tests 2–4, < 0.0001; Tables 1 and 2]. To determine how the pattern of positive selection that we observed compared with other plant lineages, we aligned a 230-AA (690 bp) region of our Onagraceae chiB sequence to orthologous sequences in dicot (Boechera chiB, from Bishop et al., 2000) and monocot lineages (Zea chiI, from Tiffin, 2004; Fig. 2). Our alignment started after the protein’s hinge region because there were few similarities across genera in the hinge region and N-terminal domain, making it difficult to align across genera. Within this 230-AA region, approximately 3% (7/230, Fig. 2) of the amino acid sites within Onagraceae were under strong positive selection [ω2 (M3) = 3.12 or ω2 (M8) = 3.30, Table 1], as compared to approximately 4% (9/230, Fig. 3) in Boechera2 = 6.8, Bishop et al., 2000) and approximately 8% (19/230, Fig. 3) in Zea2 = 1.2, Tiffin, 2004). All three genera exhibited signatures of positive selection in the active-site cleft region of the protein, where two of the seven positively selected sites in Onagraceae were in the active-site cleft region, compared to six of the nine sites in Boechera and four of the 19 sites in Zea. Differences in the occurrence, strength and location of positively selected sites across plant genera and lineages suggest that adaptive responses might be lineage-specific, depending on differences in mode of reproduction, demographic history, pathogen pressure, pathogen virulence and other extrinsic and intrinsic factors (Bishop et al., 2000; Stotz et al., 2000; de Meaux & Mitchell-Olds, 2003; Tiffin, 2004; Siol et al., 2010).

Figure 3.

 Aligned orthologous amino acid sequences of mature chitinase proteins from Boechera (chiB, GenBank AAF69783), Zea (chiI, GenBank AY532743) and Oenothera (chiB, Dryad: doi:10.5061/dryad.sk94j). Amino acids under positive selection identified in Boechera (Bishop et al., 2000) and Zea (Tiffin, 2004) are shaded light grey, amino acids under positive selection identified in sexual Oenothera lineages (branch-site models, Table 1) are shaded darker grey, and three additional sites identified as being positively selected under models that allow evolution to vary among amino acid sites but not among lineages (site-specific models M3 and M8, Table 1) are outlined. Active-site residues are underlined.

Effects of recombination and segregation on chitinase expression

We found that a loss of recombination and segregation was also associated with significant differences in chiB expression (F1,165 = 18.14, < 0.0001). For example, constitutive chiB expression (CT value of chiB/CT value of EF1α) was 8% higher in sexual lineages (1.53 ± 0.02) than in asexual PTH lineages (1.41 ± 0.02). This suggests that sexual lineages maintain slightly higher basal levels of chitinase than asexual lineages, which could aid in the initial defence against pathogens. Presently, we do not know whether the observed differences in gene expression arise from adaptive changes in regulatory regions (Garfield & Wray, 2010) or are due to an overall accumulation of deleterious mutations (higher genetic load) within these regions. The relative role of structural vs. regulatory changes in adaptive evolution is vigorously being debated in evolutionary and developmental biology (Doebley & Lukens, 1998; Hoekstra & Coyne, 2007; Wagner & Lynch, 2008; Wray, 2008; Martin et al., 2010), and this system represents an exciting avenue for future research to address this problem.


Interactions between plants and their natural enemies are ubiquitous in nature, and evolutionary dynamics between them are thought to be fundamental to plant evolution and the origin and maintenance of biodiversity (Ehrlich & Raven, 1964; Agrawal, 2011). Here, we show that a breakdown in recombination and segregation (sex) is negatively correlated with adaptive structural molecular evolution and could potentially affect regulatory evolution of a plant protein involved in defence against pathogens. Specifically, we show that sexual lineages have a higher incidence of positive amino substitutions and elevated constitutive expression of a chitinase pathogen resistance protein (chiB). Our results suggest that sexual reproduction (recombination and/or segregation) influences the evolution of a defence gene and more specifically that a loss of sex in plants might constrain adaptive evolutionary responses to pathogens.


We thank M. Rausher and S. Smith for helpful comments on this research. S. Greiner and J. Muerer shared sequences and provided advice on molecular work. The Ornamental Plant Germplasm Center, A.A. Agrawal, S. Barrett, G. Hoggard, R. Hoggard, R. Raguso and P. Van Zandt provided seeds. H. Sederoff kindly shared equipment. This work was supported by Michigan Technological University, NC State University, University of Toronto, and grants from NSF (DEB-0919869) and NSERC to MTJJ.

Data deposited at Dryad: doi: 10.5061/dryad.sk94j