Comparative studies of molecular substitution rates between lineages provide insights into the mechanisms that cause evolution of DNA sequences. Under the neutral theory 12 rates of nucleotide substitutions are expected to be equal to rates of mutation, thus a constant rate of nucleotide substitution in homologous DNA sequences should be observed among lineages that share mutation rates. Neutral theory assumes that genetic drift is the primary evolutionary mechanism causing molecular evolution and predicts that rates of sequence change would be both constant over time and independent of the effective population size. Heterogeneity in substitution rates can be explained under neutral theory by either unevenness of mutation rates at individual loci (manifested as locus effects) or correlated mutation rates across all loci within species (manifested as lineage effects). Alternatively, natural selection may cause rate heterogeneity among loci and lineages via purifying selection that reduces the probability of substitution due to functional constraint or through the increased probability of substitution associated with positive natural selection 3456. Identifying causes of rate heterogeneity as well as specific variables that affect underlying mutation and substitution rates is fundamental to understanding the mechanisms that cause evolution of DNA sequences (reviewed in 7).

Differences in generation time could affect substitution rates, causing lineage effects on substitution rates if organisms with shorter generation times experience more mutations per unit of chronological time than organisms with longer generation times. This neutral explanation for rate heterogeneity among lineages is commonly called the generation time hypothesis. Under the generation time hypothesis, lineage-specific heterogeneity in rates of divergence can be explained by differences in the number of germ line cell divisions per unit time among lineages that otherwise share constant mutation rates. Therefore, under the generation time hypothesis substitution rates are expected to be negatively correlated with generation time 589. Generation time effects on synonymous substitution rates have been widely observed at multiple loci for several mammalian species 239101112131415. Generation-time-like effects have also been tested for in organism such as RNA viruses where faster substitution rates were correlated with higher frequencies of replication 16 and in spore-forming bacteria where rates of divergence were not related to spore dormancy 17.

In angiosperms, expected generation time impacts on rates of molecular evolution are not as clear as in animals since plants lack distinct germ and somatic cell lines. Plant cells are totipotent and the number of cell divisions between germination and gamete production can vary from individual to individual and even among parts of a single individual. The generation time hypothesis modified for plants assumes that variation in the frequency of cell replication is correlated with differences in annual/perennial habit. Since annuals have shorter minimum time to first flowering than perennials, it has been assumed that annuals would also experience a higher frequency of cell replication per chronological time and thereby a faster rate of divergence when compared to perennials 18. The generation time hypothesis has been invoked to explain why annual species exhibited higher rates of molecular evolution than perennial species for several nuclear, mitochondrial and chloroplast loci (e.g. 1920212223). However, results from studies that support a generation time effect in plants have two primary limitations 24. First, some studies used multiple non-independent comparisons in their analyses that may lead to statistical difficulties as well as potential phylogenetic bias. Second, the taxa compared were highly divergent so that other evolved differences in addition to generation time could also have caused the rate variation observed. Comparing divergence rates in phylogenetically-independent sets of annual/perennial pairs that are recently diverged can correct for these two pitfalls when testing for a generation time effect in angiosperms 24.

Loci that can be used to estimate divergence rates are limited in the vast majority of angiosperms, which restricts comparisons of substitution rates in multiple independent sets of recently diverged plant taxa. For example, the plant mitochondrial genome exhibits a fast pace of structural evolution but the lowest rate of nucleotide substitutions of all three plant genomes making it especially difficult to obtain sequences in multiple plant lineages with sufficient divergence 1825262728. Universal primers are available for multiple chloroplast regions but, like mitochondrial regions, the utility of these regions is often limited by low sequence divergence at shallow phylogenetic relatedness. Nuclear loci are not widely available in multiple plant lineages since nuclear genomes have variable architecture, abundant multigene families with rapid duplication and loss complicating the identification of orthologous loci 1827. There is also a wide range of substitution rates among nuclear DNA sequences in plants 29, requiring multiple loci in comparative studies to average rates over independent loci.

The internal transcribed spacers (ITS1 and ITS2) of nuclear ribosomal DNA (nrDNA) are the only nuclear DNA markers currently available for comparative tests of the generation time hypothesis in a broad range of recently diverged plant taxa for several reasons. First, ITS regions are universally amplifiable in plants and many plant taxa have been sequenced. Second, ITS regions are highly variable at the nucleotide level. Third, it is commonly believed that ITS multicopy arrays are homogenized by concerted evolution so that intraspecific polymorphism does not complicate estimates of divergence 3031. Moreover, ITS regions have been used extensively in molecular evolution studies of plants such as demonstrating that rates of ITS nucleotide substitution are associated with species diversity 32, reproductive isolation and life history 33, and environmental variables (33; but see 34).

Two recent studies compared rates of ITS1 and ITS2 divergence using phylogenetically independent sets of angiosperms differing in life history but reached opposite conclusions about whether differences in life history affect rates of divergence. In the first study, Whittle and Johnston 24 did not find an association between relative rates of nucleotide substitution and annual/perennial life history in 22 species pairs, leading them to conclude that the generation time hypothesis does not apply to angiosperms. In another recent study, clades with a predominantly herbaceous life history exhibited an almost twice-faster average rate of divergence than predominantly long-lived woody clades using 28 independently calibrated absolute rates of ITS nucleotide substitution 35. Both studies consistently did not reject the null hypothesis of constant divergence rates when comparing life histories. Since low statistical power of rate tests was suspected, both papers also treated substitution rate differences qualitatively or categorically (e.g. annual is faster or perennial is faster regardless of the magnitude of the rate difference). These conflicting results mandate further research into whether differences in generation times are correlated with substitution rates in angiosperms.

Given that ITS1 and ITS2 are currently among the only sequences available to test for rate heterogeneity among a wide sampling of plant taxa, it is essential to assess the statistical power of rate heterogeneity tests based on ITS sequences. It is critical to determine the magnitude of rate heterogeneity required to reliably reject the null hypothesis of rate constancy when evaluating whether differences in annual/perennial habit have heterogeneous substitution rates. Low statistical power will result in type II errors (incorrectly failing to reject the null hypothesis of rate constancy) that could lead to an erroneous conclusion that annual/perennial habit is not associated with divergence rates. One main cause of low statistical power is a small number of nucleotide substitutions available to estimate divergence, a common situation when recently diverged species are being compared. Simulations have shown that the power of Tajima's relative rate test 36, distance-based relative rate tests 9, and the maximum-likelihood relative ratio test 6 are all dependent on sequence lengths, the relatedness of the outgroup taxa, and the employment of an appropriate model of nucleotide substitution 3738. The alternative approach of categorical treatment of substitution rate differences in annual/perennial comparisons is based on the assumption that the direction of rate differences would accurately test rate heterogeneity. However, this approach has not yet been subjected to a rigorous power analysis.

In this article, we test whether annual/perennial habit affects rates of divergence by comparing both relative rates of molecular evolution and categorical branch length differences in 16 independent annual/perennial species pairs. ITS1 and ITS2 sequences were obtained from GenBank under strict sampling criteria designed to control for artifacts contributing additional variation in divergence rates that could obscure any rate variation caused by differences in life history. The criteria were that each annual/perennial pair was recently diverged, had at least eight nucleotide changes between taxa, had ITS sequences for two outgroup taxa available, and the ITS sequences were originally obtained from a single PCR amplicon. The power of maximum likelihood relative rate tests was investigated by determining the degree of rate heterogeneity required to reliably reject rate constancy for DNA sequences simulated under average nucleotide substitution parameters of ITS sequences. We also used simulations to assess whether categorical treatment of branch length differences is an appropriate method to test for rate heterogeneity when a relative rate test does not reject rate constancy. In addition, we utilized sequences of three nuclear loci, two chloroplast regions and multiple intra-specific nrDNA ribotypes for the annual Arabidopsis thaliana and two closely related perennials (Arabidopsis lyrata subsp. lyrata and A. lyrata subsp. petrea) to test whether substitution rate differences at the ITS regions were correlated across multiple loci as expected under the generation time hypothesis.

Overall, the null hypothesis of rate constancy for ITS sequences was not rejected in the majority of annual/perennial comparisons based on both maximum-likelihood and Tajima's 1D relative rate tests. When rate constancy was rejected, annuals exhibited higher rates of nucleotide substitution in most cases. Categorical treatment of branch length differences indicated that an excess number of annual species had higher rates of nucleotide substitution. Because these two patterns are expected under the generation time hypothesis for plants, these results support the hypothesis that differences in the annual/perennial habit are associated with rates of molecular evolution in angiosperms.

The simulations reported in this paper supply several insights. First, the simulations showed how often relative rate tests based on ITS-like sequences reject the null hypothesis of rate constancy when rates are in fact unequal. Second, the simulations showed how frequently a categorical comparison of branch lengths detects faster substitution rates even when relative rate tests do not reject the null hypothesis. Third, the simulations provide context for observations of ITS substitution rate homogeneity or heterogeneity reported in earlier studies, in particular, why substitution rates may not have been associated with differences in annual/perennial habit 2435. Because ITS sequences are short and have few diverged sites when compared between recently diverged taxa, relatively low power to reject rate constancy seemed possible. Indeed, the simulations showed that the statistical power to detect rate heterogeneity using ITS sequences is generally low for rate differences less than 3-fold. For simulations based on ITS1-like and ITS2-like nucleotide substitution parameter sets, the relative rate test only achieved an 80% probability of rejecting rate homogeneity with 4.5 or 5-fold rate differences. These power analyses for ITS-like sequences agree with a more general previous study that demonstrated a high type II error for Tajima's 1D test when the DNA sequences compared are short and have few diverged sites 38. The simulations further suggest that categorical treatment of branch length differences is a more powerful indicator of rate heterogeneity at low to moderate substitution rate differences compared to relative rate tests, at least for ITS-like sequences. However, conclusions about the statistical power of the categorical rate comparisons only apply to the average nucleotide substitution model and divergence parameters used in the simulations and may not be a general phenomenon.

The best-documented case of a generation time effect is the 2- to 3-fold faster substitution rate in rodents compared to hominids 127. This well studied example provides some perspective on the magnitude of rate differences we might expect to observe in plants if a generation time effect actually operates. In plants, the magnitude of rate differences between annual and perennial taxa was 2-fold at synonymous sites in the mitochondrial coxI gene 40, 4-fold at both synonymous and non-synonymous sites in the chloroplast rbcL gene 202127, and 2.5-fold at synonymous sites in nuclear Adh loci 20 (Eyre-Walker and Gaut 1997). All of these studies were limited to comparisons of highly divergent annual and perennial species and therefore confounding factors other than differences in generation time might have lead to an overestimation of the impact of annual/perennial habit on substitution rates. In a study focused on phylogenetically independent comparisons, Kay and collaborators 35 found that clades with a predominantly herbaceous life history exhibited divergence rates for ITS sequences almost two times faster than clades with a predominantly long-lived woody life history in 28 phylogenies representing 21 different angiosperm families. The overall substitution rate differences estimated for the annual/perennial species comparisons in this paper were of similar magnitude to the ITS rate differences observed by Kay et al. 35. Here, annuals evolved on average 1.6 times faster rate than perennials when a less divergent outgroup was used, while a slightly lower 1.4-fold average acceleration of annuals was observed with more divergent outgroup taxa. Categorical comparisons for ITS sequences also indicated that faster rates of substitution were correlated with annual habit. Therefore, the ITS results in this paper support a weak substitution rate acceleration for annuals consistent with a generation time effect in plants. We also believe that our sampling methods controlled for rate heterogeneity caused by variables other than annual/perennial habit and helped to better detect a weak generation time effect. This explains in part why our results are distinct from those of Whittle and Johnston 24, even though both studies were based on some of the same ITS data and used similar relative rate tests.

The Arabidopsis sequence data also support a weak annual acceleration in substitution rates. The annual A. thaliana had a significantly faster substitution rate for the chloroplast rbcL locus when compared with A. petraea. In addition, categorical rate comparisons consistently showed a faster substitution rates for A. thaliana for all the nuclear loci and chloroplast regions, even though rate constancy was not rejected by relative rate tests. When the three nuclear loci were combined, A. thaliana had a significantly faster substitution rate than either perennial. The only exception to the pattern of faster divergence rates for A. thaliana were at ITS sequences. The lack of any relative rate tests rejecting rate homogeneity and about half of qualitative comparisons indicating annuals were faster, all suggest that the ITS sequences showed no evidence of a faster substitution rate for A. thaliana. However, the simulation results showed that ITS-like sequences have little power to reject rate constancy when substitution rates are less than 2-fold different. So the pattern of about half of the qualitative rate comparisons showing a faster substitution rate for annuals is consistent with random variation about a mean rate difference of zero. Interestingly, similar results indicating a consistently higher number of synonymous substitutions (but rate constancy was not rejected by Tajima's relative rate test) in A. thaliana than in A. lyrata were observed in five out of six loci using the closely related outgroup species Capsella rubella and Arabidopsis graba 41.

Recent divergence of annual/perennial taxa is an advantage when attempting to infer the possible causes of rate heterogeneity because it reduces the number of evolutionary changes that distinguish the taxa in addition to annual/perennial habit. Unfortunately, that advantage may come at the cost of statistical power to detect potential rate heterogeneity. Recent divergence also means that few substitutions have occurred in the two taxa being compared so that the number of nucleotide changes will be small. The Arabidopsis data further suggest that statistical power to compare annual/perennial substitution rates is limiting. The individual Arabidopsis nuclear loci did not show significant rate heterogeneity. However, the larger sample of changes in the three loci combined showed the approximately 1.4-fold rate difference between annual and perennial was significantly faster for the annual. Since we did not distinguish among synonymous and nonsynonymous sites in the Arabidopsis sequences, the significant rate difference is an average across all types of nucleotide sites and reflects the net substitution rate of neutral sites and any sites influenced by positive or negative selection.

In addition to the low power of ITS sequences to test relative rate hypotheses, ITS sequences may have other limitations that further hamper their ability to detect rate heterogeneity. There is the possibility that the ITS1 and ITS2 sequences might be subject to different selective pressures 42 resulting in region-specific rates of ITS substitution when natural selection is stronger than genetic drift. The ITS sequence data in this paper indicated a pattern of annual taxa exhibiting faster substitution rates that was consistent between both ITS1 and ITS2 regions. Such a pattern is not expected if ITS1 and ITS2 regions experience locus-specific selection pressures. In addition, it has been hypothesized that incomplete concerted evolution could independently affect rates of molecular evolution at either ITS1 or ITS2 30. Reports of multiple nrDNA haplotypes within individuals are becoming increasingly common [e.g. 434445464748495051] suggesting that complete concerted evolution should not always be assumed for ITS sequences. The different intraspecific nrDNA ribotypes used in the Arabidopsis annual/perennial comparisons did in fact change the perception of substitution rates between ITS1 and ITS2 regions. Either ITS1 or ITS2 was observed to have the faster substitution rate for annuals depending on the nrDNA ribotypes used in the Arabidopsis annual/perennial comparison. Thus, the Arabidopsis ITS data suggest that estimates of substitution rates may depend on the nrDNA ribotype employed in comparisons. If selection pressures or polymorphism dynamics have a greater impact on estimates of substitution rates than does a weak generation time effect, any acceleration in the substitution rate of annuals will be difficult to detect with ITS.

The underlying biological mechanisms that might cause an acceleration of substitution rates in annuals are still unclear 1835, although life history features that influence the number of rounds of DNA replication per unit of calendar time are capable of altering the relative substitution rate when mutation rates are constant. Identifying the underlying cause or causes of rate heterogeneity is difficult because variables such as the combined effects of organism size and temperature on metabolic rate 5253, the influx of environmental energy that may lead to mutation 54, and mating system 55 are potentially confounded with differences in generation time. Annual or perennial habit may itself have a variable relationship to the generation time pertinent to substitution rates. For example, many perennials are able to flower in their first year like annuals while other perennial species may require many years until first flower. The total range of possible generation times in plants is very large since some woody perennials may live for thousands of years. The species in this study all fall at the short generation time end of this range since they are either annuals or short-lived perennials. Therefore, the suggestion of a weak annual/perennial substitution rate difference in our study may not apply if plant groups containing perennials with longer lives and greater time to first flower were compared.

ITS substitution rates in 16 phylogenetically-independent comparisons of annual and perennial taxa and from the combination of nuclear loci and chloroplast genome regions in annual and perennial Arabidopsis suggest a modest rate acceleration of less than 2-fold in annuals. These results support an association between rates of nucleotide substitution and annual/perennial habit in plants as expected under the generation time hypothesis. Separately, simulations showed that relative rate tests employing ITS-like sequences are not expected to be powerful enough to reject rate homogeneity when substitution rate differences are small. Given that the power of ITS sequences to test for generation time effects is very low, the conclusion by Whittle and Johnston 24 that no annual/perennial effect on substitution rates exits seems unwarranted. The small substitution rate differences observed here and in other studies points out that testing the generation time hypothesis among closely related plant species will require multiple loci to achieve sufficient power, as was the case in the now classic examples of animal generation time effects. While their availability in many plant taxa facilitates phylogenetically-independent comparisons, ITS sequences by themselves are not likely to be a powerful tool to test hypotheses involving substitution rate heterogeneity. Further studies with greater statistical power have to be carried out before drawing a definitive conclusion about patterns of relative substitution rate heterogeneity in annual and perennial plants and its possible causes.

The four authors jointly conceived and designed the study. DSH and OFP assembled the data and performed the analyses. JMB and MBH coordinated the study. MBH and DSH wrote the manuscript. All authors read, edited and approved the final manuscript.