Introduction

Almost a half-century after the concept of a molecular clock was introduced (Zuckerkandl and Pauling 1965), the data that have been amassed by DNA sequencing at an ever increasing pace leave no doubt that DNA substitution rates differ strongly not only among different genomic regions, but also between different organisms or lineages: molecular clocks rarely tick at a constant tempo across much more than individual clades or even branches on phylogenetic trees (e.g., Aris-Brosou 2007). While substitution rate contrasts among different genomic regions are often straightforward to explain by differential functional constraints, a variety of competing, though not necessarily mutually exclusive, hypotheses have been put forward to explain rate differences between organisms. Among these are the metabolic-rate hypothesis (Martin and Palumbi 1993), and the DNA-repair-efficiency hypothesis (Britten 1986), which are hard to test without extraordinary experimental effort. In contrast, a speciation-rate hypothesis (e.g., Barraclough and Savolainen 1996, 2001) that links accelerated rates of molecular evolution with speciation, can be tested based merely on a properly sampled phylogenetic DNA dataset. Unfortunately, the reported correlation between lineage diversification rate and molecular substitution rate has recurrently been due to a methodological artifact called “node density effect” (NDE) rather than to the supposed underlying biological phenomenon (see Hugall and Lee 2007). Likewise, the correlation of phenotypic traits with molecular rates can be tested based on phylogenetic trees and sequence data. One such factor that has most commonly been found to correlate with substitution rate at selectively neutral DNA in plants is the generation time (e.g., Wu and Li 1985; Bousquet et al. 1992; Gaut et al. 1992; Whittle and Johnston 2003; Smith 2008), although this has also recently been found in animals (Tsantes and Steiper 2009).

Generation time is the time necessary to produce a gamete from a gamete. If inheritable mutations occur in organisms exclusively during recombination, then organisms that produce new generations in shorter time should have more mutations per unit time (Page and Holmes 1998). Annuals are believed to have shorter effective generation times than perennials, although this neglects the effect of seed banks. Even though the effect of seeds older than 2 years may be negligible (Kalisz and McPeek 1992), this may be enough to blur the distinction from that of short-lived perennials. The generation-time hypothesis furthermore assumes that mutations are either mostly accumulated during meiosis or that generation time is correlated with cell divisions per generation (Eyre-Walker and Gaut 1997). These assumptions have been doubted by Whittle and Johnston (2003) due to the lack of significant correlation in their study comparing substitution rates in several pairs of taxa differing in generation time. It should therefore be considered that faster substitution rates in annuals could also be due to other factors correlated with this life history.

The correlation of annual life history and selfing breeding system has already been noticed by Stebbins (1957) and predicted theoretically (Morgan et al. 1997; Pannell and Barrett 1998; Aarssen 2000). Thus, higher substitution rates in annuals when compared to perennials could be a secondary effect of higher substitution rates in selfers, which are predominantly annuals, misleading us about the cause of higher substitution rates of annuals among angiosperms as a whole (Bousquet et al. 1992; Laroche and Bousquet 1999), in specific groups (Ainouche and Bayer 1999; Andreasen and Baldwin 2001; Archibald et al. 2005), or sister species (Morgan-Richards and Wolff 1999), in which the generation-time hypothesis was used as an explanation but breeding system was not investigated.

Two factors have been invoked to imply a higher substitution rate in selfers when compared to outcrossers. First, larger effective population size in outcrossing organisms is supposed to homogenize the gene pool, at least in the absence of directional selection, and thus decrease the probability of fixation for new substitutions (Morgan-Richards and Wolff 1999). Thus, outcrossers may have higher within-population diversity when compared to selfers (Hamrick and Godt 1990, 1996; Glémin et al. 2006), but older mutations are also retained longer. In contrast, small effective population size in selfers reduces the efficiency of selection and allows the accumulation and fixation of weakly deleterious mutations, although the effect may be different for other selective regimes and dominance levels (Glémin 2007). The retention of more weakly deleterious mutations combined with faster fixation of mutations could then lead to a higher substitution rate. Second, with little effective recombination, i.e., recombination between divergent alleles, loci that increase the mutation rate are preferred, because they can hitchhike with a fitness-enhancing locus (Sniegowski et al. 2000), which would lead to higher mutation rates in selfers when compared to outcrossers. Theoretical considerations demonstrated that selfing plants but not outcrossing plants may become fixed for modifier loci increasing mutation rate (Holsinger and Feldman 1983). Although these two factors are theoretically sound, empirical evidence is lacking for some of their components. Nevertheless, analysis of the breeding system is warranted in analyses that explore sources of divergent substitution rates among DNA sequences.

The genus Veronica (Plantaginaceae) is suited well for a study trying to differentiate between effects due to annual life history or selfing because both the factors are not as tightly linked in the genus as in other groups and both traits have evolved multiple times (Albach et al. 2004a; Albach and Greilhuber 2004, see below). Furthermore, Veronica has been the study group in which another hypothesis regarding the evolution of annuals has been refuted. Annuals are generally considered to have a smaller genome than perennials (e.g., Torrell and Vallès 2001), although this is a misinterpretation of Bennett (1972). An investigation of genome size in Veronica, however, has shown that a small genome size is correlated with a selfing breeding system (Albach and Greilhuber 2004). It therefore seemed rewarding to investigate the correlation of substitution rate and annual life history in the same study group. The disadvantage of the group is that relative rates cannot be converted to absolute rates, because no reliable dating is possible for the group. There are neither appropriate fossils available nor are there species known from oceanic islands with reliable date of origin. Although divergence estimates for deeper splits in asterids exist, extrapolating from these would greatly inflate error. However, these problems may be common to most genera that could serve as study group to address reasons for strong rate heterogeneity in taxa with variable life history.

This article presents a detailed analysis of substitution rates in Veronica. We use likelihood ratio tests to establish whether annual lineages display significantly higher substitution rates in two DNA markers than their most closely related perennial lineages. Similarly, we contrast selfing with outcrossing lineages, and examine whether it is generation time or the breeding system that is more strongly coupled with higher substitution rates. For this purpose, we analyze the plastid trnLF region (trnL spacer, trnL 3′exon, trnL-F spacer) as well as the nuclear ribosomal ITS region for 131 members of the tribe Veroniceae. The combined trnLF region and the ITS region are widely used to infer intrageneric relationships and have proven useful in Veroniceae for this purpose (Albach et al. 2004a, b, 2005; Albach and Chase 2004). By using data from two different genomes we control for genome-specific rates as an explanation for rate differences. Plastid genome data is expected to differ in rates from nuclear genome data, because the more or less strictly maternal inheritance reduces the effective population size for this genome. Furthermore, the general absence of recombination in the plastid genome in nature uncouples it from generation-time effects, which are linked to the frequency of meiosis.

Materials and Methods

DNA Extraction and Sequencing

For our purpose increased taxon sampling with 13 new sequences (5 trnLF-, 8 ITS-sequences) when compared to previous analyses was necessary to better represent lineages variable for life history and/or breeding system. Total genomic DNA was extracted from herbarium material and silica-gel-dried leaf samples according to the 2x CTAB procedure of Doyle and Doyle (1987) and then washed twice with 70% ethanol. DNA pellets were dried and resuspended in TE buffer. The trnLF region (trnL intron, trnL 3′exon, and trnL-F spacer) was amplified with primers c and f of Taberlet et al. (1991), while the ITS region was amplified using primers 17SE (Sun et al. 1994) and ITS4 (White et al. 1991).

PCR products were run on a 1.0% TBE-agarose gel, cut out of the gel, and cleaned using QIAquick™ PCR purification and gel extraction kit (Qiagen GmbH, Hilden, Germany) following the manufacturer’s protocols. Sequencing reactions (10 μl) were carried out using 1 μl of the Taq DyeDeoxy Terminator Cycle Sequencing mix (Applied Biosystems Inc.). Reactions were run out on a Prism 377 automated sequencer (Applied Biosystems Inc.), and both strands were sequenced. Sequences were assembled and edited using Sequence Navigator™ (Applied Biosystems Inc.). Assembled sequences were manually aligned prior to analysis using Quickalign (Müller and Müller 2003) following the guidelines of Kelchner (2002). Aligned sequence matrices are available from the corresponding author by request. Information on sequences used in this study including GenBank accession numbers and voucher information for new sequences is provided in Table S1 (supplementary material).

Phylogenetic Analyses

Parsimony

Maximum parsimony analyses used the parsimony ratchet (Nixon 1999). PRAP (Müller 2004) was used to output command files to be executed in PAUP (Swofford 2002), using 10 random addition cycles with 200 ratchet iterations each. During bootstrapping, searches for the optimal trees were far less intense, with only one tree swapped upon (TBR) during each replicate (Müller 2005). To minimize standard errors of bootstrap proportions, 10,000 replicates were run.

Likelihood

For maximum likelihood analyses, a fast likelihood ratchet approach (Morrison 2007) as newly implemented in PRAP (Müller 2004) v.2.0 was employed (http://systevol.nees.uni-bonn.de/software/PRAP2). PRAP generated command files that were handed over to PAUP (Swofford 2002). Finding the ML tree ideally includes (i) rapidly getting a starting tree not too far from the optimal score, (ii) move rapidly to a [near-] optimal tree island, (iii) getting the best tree within the island (Morrison 2007). Step (i) was achieved by calculating a BioNJ tree using LogDet distances, followed by one round of NNI and then one round of SPR branch swapping, optimizing the substitution model parameters between these steps. Similar to the ratchet originally described for parsimony, step (ii) included alternating between branch swapping on the original matrix and branch swapping on a matrix with 25% of characters upweighted. Differences to Nixon’s strategy for parsimony include that SPR branch swapping was used, only ten iterations were performed, and during the weighted analyses, only one tree was saved. ApproxLim was set to 2 and not adjusted. In particular for datasets with low levels of phylogenetic signal, this strategy was found to be more successful (Morrison 2007) than the strategies implemented in GARLI (Zwickl 2006) or RaxML (Stamatakis et al. 2005). Using JModelTest (Posada 2008) and the AIC, the best model was found to be GTR+Γ+I. To assess confidence in clades, bootstrapping was performed by executing PRAP-generated command files in PAUP. Using optimized parameter values and the maximum likelihood topology from the likelihood ratchet search, SPR branch swapping was performed for each bootstrapped data matrix, and the proportion of iterations in which a given clade was recovered was mapped onto the maximum likelihood tree using a strongly modified version of TreeGraph (Müller and Müller 2004; Stöver and Müller in preparation; http://systevol.nees.uni-bonn.de/software/TreeGraph2).

Bayesian Analysis

Bayesian analyses were conducted in MrBayes v3.1.2 (Ronquist and Huelsenbeck 2003) using the GTR + Γ model. Default priors were used, i.e., flat dirichlets (1.0,…,1.0) for state frequencies and instantaneous substitution rates, a uniform prior (0.0,50.0) for the shape parameter of the gamma distribution, a uniform prior (0.0,1.0) for the proportion of invariable sites, a uniform topological prior, an exponential prior Exp(10.0) for branch lengths. Four categories were used to approximate the gamma distribution. Two runs with 2-million generations each were run, and four chains were run in parallel for each run, with the temperature set to 0.2. The chains were sampled every 100th generation. To check for convergence of the independent runs under a given model, it was made sure that (i) plots of both runs indicate that stationary phase was reached in both runs, (ii) potential scale reduction factor approached 1 for all parameters, (iii) no supported conflicting nodes were found among the consensus trees generated from each run.

Relative Rate Tests

The comparisons made and hypotheses tested are shown in Table S2 (supplementary material). Specifically, it was tested whether species that are annuals and selfers [first row in table] have lower rates than species that are annuals and outcrossers [second column in table]; then whether species that are annuals and selfers [first row in table] have lower rates than species that are perennials and selfers [third column in table], and so on. Comparisons were made for ITS and trnLF separately. Perl scripts were used to extract triplets of sequences from the original alignments (ITS and trnLF), thereby removing positions consisting of gaps only. The triplets corresponded to the two species to be compared and an outgroup species as close as possible. Using this kind of three-way relative rate tests effectively ruled out the risk of biases caused by the node density effect (Fitch and Beintema 1990; Hugall and Lee 2007). We experimented with varying outgroup species to ensure that our conclusions were not sensitive to the choice of the precise outgroup taxon used, which was not the case. Not in all cases did we compare taxa that are immediate sisters in the phylogenetic trees (Figs. 1, 2). We tried to compare whenever possible taxa of the same ploidy level. Thus, we compared V. lycica (diploid) with V. trichadena (diploid) rather than V. cymbalaria (tetraploid). In some cases, we compared taxa of different ploidy level, because of the lack of other appropriate comparisons [V. agrestis (tetraploid) vs. V. filiformis (diploid); V. peregrina (hexaploid) vs. V. beccabunga (diploid)] and in some cases the ploidy level of the taxon compared is unknown (V. czerniakowskiana, V. donii, V. gentianoides). In the comparison of V. amoena and V. polita, V. polita was chosen as the only diploid and annual taxa of V. sect. Agrestes in our dataset. In the comparison of V. acinifolia and V. serpyllifolia, V. serpyllifolia is the only known diploid of V. sect. Serpyllifolia and V. acinifolia was chosen as representative of V. sect. Acinifolia based on branch length in a conservative manner. The same reasoning applies to the choice of V. vindobonensis over other taxa of V. sect. Chamaedrys and in the comparisons of V. amoena vs. V. czerniakowskiana and V. allionii vs. V. alpina. Ideally, confounding factors such as ploidy level would be excluded, but reducing the dataset to sister species of the same ploidy level would compromise the number of comparisons and, thus, the power to detect differences. This shortcoming is not unique to Veronica but likely present in all flowering plant genera.

Fig. 1
figure 1figure 1

Consensus tree of the phylogenetic analyses of trnLF. Numbers above the branches indicate support in the MP (left) and ML bootstrap analyses, respectively; those below the branches are posterior clade probabilities from the Bayesian analysis. Species names of annuals are printed in bold and those of selfers are capitalized. Asterisks indicate species for which the breeding system was scored as unknown. Legends to the right indicate subgeneric names

Full size image
Fig. 2
figure 2figure 2

Consensus tree of the phylogenetic analyses of ITS. Numbers above the branches indicate support in the MP (left) and ML bootstrap analyses, respectively; those below the branches are posterior clade probabilities from the Bayesian analysis

Full size image

Likelihood ratio tests (LRT) were performed with the help of HYPHY (Kosakovsky Pond et al. 2005). A series of analyses was coded in the program’s high-order programming language Hyphy batch language (HBL). Following AIC-based model selection in HYPHY, the nucleotide substitution model employed in all tests was the GTR + Γ model. For each of these tests [Table S2 (supplementary material)], the null hypothesis was that rates of branches leading to species of type A [rows] are lower (or equal) across all three-taxon trees than those of the sister branches leading to species of type B [columns] [Table S2 (supplementary material)]. For the alternative hypothesis rates were free to vary along the branches. Rates were constrained accordingly in HBL by constraining global rate variables to be in the interval [0,1] and then constraining rates of branches leading to species of type A to equal this rate variable times the rate of the sister branches leading to species of type B. Joint likelihood functions were built for all trees that contained species pairs relevant to the comparison at hand, and these functions were optimized, first without constraints and then under constraints. To illustrate this with an example, consider the comparison of annual selfers versus annual outcrossers (Table S2, top left cell). For this, three taxon triplets could be considered and were defined as top1 through top3 in HBL: Tree top1 = (polita, amoena, rubrifolia); Tree top2 = (trichadena, lycica, panormitana); Tree top3 = (triphyllos, donii, aznavourii). After constructing filtered datasets (containing sequences of the relevant taxa only) filteredData1 through filteredData3, a joint likelihood function was built using the according HBL command: LikelihoodFunction lf_top1top3 = (filteredData1, top1, filteredData2, top2, filteredData3, top3). This unconstrained likelihood function was optimized by HYPHY and the results (including the log likelihood found) stored in an array ‘res_Alt’: Optimize (res_Alt, lf_top1top3). Then, constraints were formulated via constraining global variables r_1 through r_3 to be in the interval [0,1] and then constraining rates (‘mu’) of branches leading to annual selfers to equal this rate variable times the rate of the sister branches leading to species of annual outcrossers: global r_1 = 1; r_1: <1; r_1: >0; top1.polita.mu:=r_1*top1.amoena.mu; global r_2 = 1; r_2: <1; r_2: >0; top2.trichadena.mu:=r_2*top2.lycica.mu; global r_3 = 1; r_3: <1; r_3: >0; top3.triphyllos.mu:=r_3*top3.donii.mu. Again, the likelihood function was optimized, now under these constraints: Optimize(res_Null, lf_top1top3);

The likelihood ratio test statistic was calculated as the log likelihood found under constraints minus the log likelihood found without constraints. For nested hypotheses, two times this difference is approximately distributed as Chi square. However, even for nested hypotheses the Chi-square distribution is only the asymptotic distribution for the LRT statistic, and because we employed non-nested hypotheses and inequality constraints, we used parametric bootstrapping with 500 replicates to estimate the test statistic distribution and calculate type-I-errors-based thereon. To obtain the null distribution, 500 data matrices were simulated using the trees and parameter values from the joint likelihood function after optimization under constraints. On each of these simulated datasets, calculation of the LRT was done the same way as for the original data matrix.

As additional tests, all annuals were compared to all perennials, regardless of breeding system, and all selfers were compared to all outcrossers, regardless of life history. Finally, we used a LRT to directly assess which of the factors (life history or breeding system) may be more strongly associated with rate heterogeneity. We estimated the significance of the difference between likelihood scores under two different optimization schemes as follows. The first likelihood score was calculated when rates on all branches leading to annuals were constrained to be lower (or equal) than those on sister branches leading to perennials. The second was based on constraining rates on all branches leading to selfers to be lower (or equal) than those on sister branches leading to outcrossers. For this purpose, a joint likelihood function was built for all taxon triplet trees and optimized initially without constraint, then under the first constraint (as stated above), and finally under the second constraint mentioned above.

Results

Relationships among the major branches of Veronica and Veroniceae (Figs. 1, 2) are congruent with earlier results (Albach and Chase 2001, 2004; Albach et al. 2004a, b, 2005; Albach and Meudt 2009) despite the increased taxon sampling in the present analysis. Veroniceae, Veronica, and all subgenera of Veronica are monophyletic with the exception of subgenus Stenocarpon, which is only monophyletic in the ITS-analysis and in some most parsimonious trees and in the Bayesian analysis of the trnLF-region (Figs. 1, 2) and the position of Paederota being nested in Veronica in ITS-analyses. This position has been shown to be due to homoplasious C–T transitions (Albach et al. 2004b). Furthermore, Veroniceae, Veronica, and all subgenera of Veronica with the exception of the abovementioned cases are supported with a posterior probability of 1.00 by the Bayesian analysis. Relationships between subgenera remain difficult to discern. All well-supported relationships between subgenera, the conflict between the two DNA regions and differences between parsimony and maximum likelihood have already been discussed in previous studies (Albach 2006; Albach and Chase 2004; Albach et al. 2004b, 2005; Albach and Meudt 2009). The affiliation of V. javanica remains dubious. Among the new sequences generated for this study all species are found in a position expected based on morphology.

Relative Rates in Veronica

The molecular clock hypothesis is rejected for Veronica by a global likelihood ratio test (P < 0.01). When examining the cause of the rejection, annuality proved to be a better explanation than selfing. Annuals have significantly (P < 0.05) higher substitution rates than perennials for both trnLF and ITS (Table 1, top). Selfers have enhanced rates when compared to outcrossers, though the significance is lower.

Table 1 Results of the likelihood ratio tests
Full size table

When only annual species are considered, there is no significant difference between outcrossers and selfers for either marker (Table 1). The same is true when only perennial species are taken into account.

Focusing on selfers only, annual species have significantly enhanced rates when compared to perennial species, both for ITS and for trnLF. When only outcrossers are considered, this effect is not significant, although almost so (P = 0.062) in trnLF (Table 1).

Further evidence that the effect on rate heterogeneity may be stronger for life history as opposed to breeding system comes from directly contrasting likelihoods for the two different constraints, (i) constraining rates of selfers to be lower than those of outcrossers versus (ii) constraining annuals to have lower rates than perennials. No significant difference was apparent for ITS (P = 0.066), but for trnLF the likelihood after constraining life history is significantly lower than the likelihood found under breeding system constraints (P = 0.006; Table 1, bottom). In other words, constraining annuals to have lower rates than perennials is a much more severe a constraint and much less compatible with the data than is constraining rates of selfers to be lower than those of outcrossers. This is a more direct expression of the pattern already apparent in the individual comparisons above: given our data, we can accept with much more confidence that annuals have higher rates than perennials than we can accept that selfers have higher rates than perennials.

Discussion

The results presented here are in accordance with several previous studies demonstrating that annuals have a higher substitution rate than perennials (e.g., Bousquet et al. 1992; Andreasen and Baldwin 2001; Duminil et al. 2008) but contrast with the trend observed in a more general report that found no higher substitution rate in a majority of annuals, evaluating 24 species pairs of annuals and perennials (Whittle and Johnston 2003). In this study, the authors refute the general hypothesis of a correlation in eudicots of life history and substitution rate but not that it may not be valid in specific examples. In fact, the authors found higher substitution rates in 10 out of 24 species pairs. Since they included pairs with non-significant differences in the sign test, it remains to be studied in how far mere stochastic variation in sequences contributed to the conclusion. It would be enlightening to analyze those pairs of Whittle and Johnston (2003), which showed a lower substitution in the annual species, in more detail. Interestingly, a later study that employed much greater datasets from across angiosperms (Smith 2008) reaffirmed the existence of a link between molecular evolutionary rates and life history.

Due to the imperfect correlation of breeding system and life history in Veronica, the genus may be better suited to disentangle the effects of annual life history and selfing on substitution rates than most other flowering plant genera in which the correlation between both is stronger. Nevertheless, even in Veronica, selfers are most frequently annuals and vice versa, and this may explain why a comparison between selfing and non-selfing species across the genus (ignoring life history) yields a significantly lower rate in the latter (Table 1). Conversely, the lack of a difference within annuals and within perennials reveals that selfing has little effect on rates, whereas life history has, regardless of the breeding system (Table 1). Our result confirms theoretical considerations by Ohta and Kimura (1971) and other empirical analyses (e.g., Bousquet et al. 1992; Laroche and Bousquet 1999; Andreasen and Baldwin 2001; Smith 2008) showing that annuals have a higher substitution rate than perennials and that this effect is irrespective of the genome investigated, as previously documented in monocots (Gaut et al. 1996). The results are also in agreement with the lack of an effect of selfing on mutation rate in Amsinckia (Schoen 2005). Furthermore, among the species with the apparently lowest substitution rates [cf. Figs. S1, S2 (supplementary material)], we find all species from the wulfenioid grade, the group of genera in Veroniceae that are consecutive sister groups to Veronica. It may therefore be that a general increase in substitution rate marked the evolution of the genus Veronica followed by an even sharper increase among the annual groups of the genus. However, to further investigate this pattern, a conversion to absolute rates via relaxed clock dating analyses will be indispensable. This is currently hindered by the lack of possible calibration points and reliable extrinsic node age estimates.

Explanations for Rate Heterogeneity in Veronica

Among the various hypotheses that have been put forward to explain rates differing between lineages, the generation-time effect is the most often suspected underlying cause for the significant correlation of higher substitution rates with annual life history, e.g., in monocots (Gaut et al. 1992) and in Sidalcea (Malvaceae; Andreasen and Baldwin 2001). In the same way in Veronica, substitution rates are significantly higher in annual species (short generation time) than in perennial species (P < 0.0001; Table 1). When only annuals or perennials are considered, no significant difference between selfing and non-selfing Veronica species is apparent (P ≫ 0.05; Table 1). However, rates in annuals are significantly higher when only selfing or only non-selfing species were taken into account (P < 0.01). One problem with such kind of analysis is that neither perenniality nor selfing or outcrossing can be exactly defined. Perenniality in Veronica and most taxa in which substitution rates has been analyzed indicates only reproduction in several seasons but can denote flowering from the first year after germination or only several years after germination. Likewise, breeding system is also not a binary character and mixed mating may occur in self-compatible species. The comparison between V. triphyllos and V. donii is consequently a comparison between a small-flowered (corolla as large as calyx or slightly larger) and a large-flowered (corolla 4–6-times as large as calyx and as those of the small-flowered species) species rather than truly between a selfer and an outcrosser. In some cases our scoring of breeding system has additionally been guided by pollen–ovule ratios (e.g., V. lycica vs. V. trichadena, Kolf 2008). Even knowing the exact outcrossing rate would not help, because outcrossing rate is variable over time and may have influenced substitution rate differently in the past as today. It is furthermore unlikely that the change in breeding system coincided with the branching event, which means that we compared taxa that exhibited a different breeding system (but also life history) for a variable amount of time. Therefore, analyses like ours can only demonstrate that the cause for substitution rate heterogeneity needs to be searched for rather in a factor associated with the generation time than in the breeding system, but other influences of life history on substitution rate need to be considered as well (Bousquet et al. 1992).

For life history to have a direct effect, one needs to assume that mutations occur predominantly during reproduction (Whittle and Johnston 2003). The fact that cpDNA also shows a generation-time effect, although plastid genomes should evolve independent from the processes involved in reproduction like meiosis, points to a meiosis-independent factor leading to higher substitution rate. Therefore, if the effect of life history on substitution rate is direct, it must be based on the relative time of DNA spent in the germ-line irrespective of whether it is the paternal or maternal germ-line if one assumes that plastid DNA is maternally and nuclear ribosomal DNA is biparentally inherited in Veronica as in most angiosperms. This means that the factor driving higher substitution rates in annuals is independent, or at least on top, of the factor driving higher substitution rates in the paternal when compared to the maternal germ line (Whittle and Johnston 2002). These results however contradict the idea by Klekowski (1989) that the somatic mutation rate scales linearly with generation time but points to a factor that increases somatic mutations in annuals more than in longer-lived plants.

When we look for such effects of life history on substitution rates, various other mechanisms that have been discussed as a cause for lineage effects need to be considered, such as unequal efficiency of DNA repair (Britten 1986). Since repair mechanisms have not been investigated in the plants under study, it is difficult to assess their possible influence. A change in DNA repair efficiency has been shown between distantly related Arabidopsis thaliana and Nicotiana tabacum (Kirik et al. 2000) and within individuals between the male and female germ line (Xu et al. 1998). Differences in DNA repair mechanisms are currently underexplored and cannot be excluded at the moment. Other possibilities include differences in the amounts of cellular mutagens or a higher level of DNA-methylation in annuals but there is currently no data to support either hypothesis. However, a higher substitution rate in annuals is unlikely due to linkage with specialized physiology (e.g., photosynthetic pathway, Gehrig et al. 2003; carnivory, Müller et al. 2006; parasitism, DePamphilis et al. 1997; Nickrent et al. 1998) or different metabolism (Martin and Palumbi 1993; see also Lanfear et al. 2007). A correlation of substitution rate and speciation rate has been suggested by Barraclough and Savolainen (2001) for angiosperms as a whole but is not found within Veronica, in which the most species-rich groups, subgenus Pentasepalae with at least 75 species and the Hebe complex with approximately 150 species seem to have some of the lowest evolutionary rates in the genus Figs. S1, S2 (supplementary material), although a compensatory higher extinction rate in annuals cannot be excluded.

Substitution rates estimated here incorporate the effects of mutation rate and mutation fixation probability. Mutation fixation probability increases with smaller effective population size for neutral mutations. Kimura (1968) and Ohta and Kimura (1971) have shown that a smaller effective population size also increases substitution rates if most substitutions are slightly deleterious but there is no dependence if most substitutions are neutral. Thus, for DNA regions under selection such as coding regions one can differentiate between the effects of mutation rates and mutation fixation probability. However, studies investigating the generation-time effect have either not distinguished between synonymous and non-synonymous sites or used markers that should be close to evolving neutrally such as in our study, in which the only coding DNA is too short to allow any further test. Although there may potentially be other factors besides effective population size that influence mutation rate (e.g., DNA repair system; Kirik et al. 2000), effective population size is the most likely difference between annuals and perennials to influence substitution rate. Unfortunately, effective population size is a complex factor to analyze because apart from effective generation time, actual population size, population growth history, population structure, and outcrossing rate over time would need to be considered. For many parameters, such data are not available and, therefore, an influence of effective population size on the substitution rate cannot be estimated.

There are some theoretical indications that effective population size may be a factor increasing the substitution rate in annuals but not in perennials, because increases in mutation rate are favorable when mutation rate is a limiting factor for adaptation and survival (Sniegowski et al. 2000). This occurs in unstable, changing environments inhabited by annuals and in metapopulations, in which genetic variation is decreased due to bottlenecks (Metzgar and Wills 2000). Metapopulation structure with frequent founder effects is more likely in annuals due to their higher population growth rate (Silvertown and Charlesworth 2001). It is thus conceivable that generation time influences substitution rate by changing effective population size but is not the only factor able to do so. Other factors are known to influence effective population size and could consequently lead to higher substitution rates in these taxa.

In conclusion, among the various possible factors that affect DNA evolutionary rates in Veronica, life history seems more important than breeding system. However, more alternative factors will need to be considered and more taxa will need to be analyzed to generalize the results. Future studies will need to collect detailed information on effective population sizes and test for correlations between these population sizes, states of other biological characters (physiological, morphological, and ecological), and substitution rates in DNA regions from all three plant genomes.