Rates of DNA sequence evolution vary among taxa and among genes, and the causes of this variation are many. In some cases, generation time has been shown to be correlated with rates in plants. For example, annual plants can sometimes have higher rates of DNA evolution than perennials 1. In one study, it was shown that long-lived woody grasses exhibit slower rates than short-lived herbaceous ones 2. However, a more extensive set of 33 phylogenetically independent comparisons failed to find a generation time effect for plants in general 3.

A useful method for distinguishing among the potential causes of rate variation is to separately examine nonsynonymous rates (r_N) and synonymous rates (r_S). For example, when r_N increases relative to r_S, relaxation of purifying selection is a possible explanation. However, when r_S increases, but the r_N/r_S ratio is not greatly affected, then an increase in the mutation rate is a possibility. An example of this is Plantago mitochondrial DNA 4. However, a decrease in DNA repair efficiency could also explain such a change. In addition, population processes, such as genetic drift could play a role. Reduced effective population size (N_e) can increase the fixation rate of neutral and slightly deleterious mutations. Thus, if slightly deleterious mutations are common, both r_S and r_N are expected to be higher when N_e is low 56.

One expectation of this drift-based hypothesis is that r_N will vary from protein to protein, as each protein will have different functional constraints and thus a different proportion of slightly deleterious mutations. In contrast, r_S is expected to be similar among proteins when the cause is a change in mutation rate or repair efficiency7.

During founder-effect speciation, genetic drift can be expected to increase the substitution rate, even at silent sites. This has been called the speciation-rate hypothesis 8. For example, speciation rate has been used to explain the difference in non-coding DNA substitution rates between the sister genera Utricularia and Pinguicula in the plant family Lentibulariaceae 9.

Rates can also vary if the underlying mutation rate varies 1011 or if DNA repair is impaired 1213. Although rates themselves are hard to measure, the number of synonymous (d_S) and nonsynonymous (d_N) substitutions can be measured and are used to compare rates and calculate rate ratios.

Epifagus virginiana, a nonphotosynthetic plant, has an increased rate of sequence evolution for plastid DNA in general 14151617 and relative rates tests of the plastid rps2 gene indicate a significant increase for both d_N and d_S. This suggests that purifying selection is at least partially relaxed

and

Phylogenies of the Orobanchaceae and relatives were constructed using maximum parsimony (MP) and maximum likelihood (ML). The MP analysis discovered four most parsimonious trees of 3817 steps, with CI = 0.6275, CI (excluding uninformative characters) = 0.5118, and RC = 0.3888. The strict consensus tree was unresolved as to the position of Schwalbea relative to the Alectra-Orobanche clade, the Bartsia-Melampyrum clade and the Castilleja-Pedicularis clade. It was also unresolved concerning the relationships among the outgroups Mimulus, Kigelia, Hemimeris, Verbascum, Antirrhinum and Veronica.

The ML analysis found two trees, with -ln likelihood values of 24525.88663. The strict consensus of these trees was unresolved, but in a different place, regarding the position of the Cistanche-Epifagus clade. When the MP consensus and the ML consensus were combined into a semistrict consensus tree, a completely resolved tree resulted. This tree is shown in Fig. 1.

All three gene trees exhibited statistically significant rate heterogeneity (p < 0.0005), as assessed by the Kishino-Hasegawa (K-H) test 19. Synonymous and nonsynonymous branch lengths for each of the three genes are shown reconstructed in Fig. 2.

The correlation analyses show that there is a higher correlation of synonymous evolution across genes than nonsynonymous evolution (Fig. 3). They also show that within rps2 and rbcL synonymous and nonsynonymous evolution is poorly correlated, but in matK, the correlation is better (Fig. 4).

Some of the non-photosynthetic plants, (Epifagus, Cistanche, and the Orobanche species) have increases in both synonymous and nonsynonymous rates. Rates are not, however, increased in Boschniakia, Harveya, Hyobanche, Lathaea, Alectra orobanchoides, and Striga gesnerioides.

Synonymous rates vary markedly among taxa. For example, the branches leading to Epifagus are at least two or three times as long as those for most of the photosynthetic taxa. Despite this among-taxon variation, the pattern for each gene is strikingly similar (Fig. 2D, E, F). For example, sister taxa (e.g. Veronica and Antirrhinum, Euphrasia and Panetucellia/Tozzia, Epifagus and Cistanche, O. fasciculata and O. corymbosa) usually show an identical pattern of who is faster than whom. This indicates that the causes of synonymous rate variation are affecting the whole plastid genome in a similar way.

There is even more extensive variation in nonsynonymous rates, both among taxa and among genes. This is not surprising because these genes have different functions and some of the taxa are photosynthetic while others are not. The scale bars in Figs. 2A, B, and 2C show the overall differences in nonsynonymous rates. MatK is much more rapidly evolving than the other two genes for nearly all taxa, but Epifagus' rbcL pseudogene has a similar rate. When looking at taxa across genes, there is much less consistency than with the synonymous rates. There are some big differences, such as the branch lengths for Striga, Cycnium and Schwalbea, and the Euphrasia species. Overall the picture is more complex, as could be expected if there are many causes of variation, differing from taxon to taxon and gene to gene.

The pattern of matK rate variation is very similar in the synonymous and nonsynonymous figures. This fits well with the fact that it is less constrained overall, as can be seen by comparing the scale bars in Figs. 2A, B, and 2C.

Rates were compared by using two categories at a time and testing for significant differences using likelihood ratio tests. These tests are summarized in Table 2. In general, it was found that the nonphotosynthetic plants have higher synonymous and nonsynonymous rates of change, although when tested separately, it was found that Harveya, Hyobanche and Boschniakia do not have higher synonymous rates.

Purifying selection is relaxed in the nonphotosynthetic plants for all three genes. The test values are as follows. matK: 2×LR = 10.67, df = 1, p = 0.001; rbcL: 2×LR = 31.3, df = 1, p = 2.2 × 10^-8; rps2: 2×LR = 8.56, df = 1, p = 0.00343.

Another way to describe the difference in the pattern of synonymous and non-synonymous rates is to say that the former are more correlated across genes. This can be seen in Fig. 3, which shows plots comparing two genes at a time. In comparisons including rbcL nonsynonymous rates, the data point from the Epifagus pseudogene has been excluded. Its unconstrained evolution is not typical of "nonsynonymous" change and its position on the plot made it an extreme outlier with an enormous influence on the regression line.

For rps2 and rbcL, the synonymous plots are more highly correlated, whereas for matK, which is relatively unconstrained, they are about the same.

The dramatic rate increase observed in Epifagus 14, with branches 5–10 times as long as other taxa, can now be seen to have begun earlier in the history of the Orobanchaceae. It is shown to be composed of increases in both synonymous and nonsynomymous rates. The general pattern is that many of the non-photosynthetic plants, such as Epifagus, Cistanche, and the Orobanche species, have increases in both synonymous and nonsynonymous rates, indicating that both (1) selection is relaxed, and (2) there has been a change in the rate at which mutations are entering the population in these species. However, rate increases are not immediate upon loss of photosynthesis, since we do not see increases in Boschniakia, Harveya, Hyobanche, Lathaea, Alectra orobanchoides, and Striga gesnerioides. This pattern is similar to that found using smaller data sets 1520. Separate analyses of synonymous and nonsynonymous rates give us some insight into potential mechanisms.

The speciation rate hypothesis predicts that more speciose clades should have a faster r_s (and therefore larger d_s) than a species-poor sister group. This was suggested as a cause of rate variation in non-coding DNA in the Lentibulareaceae 9. In this study, not all genera have been sampled, and those that have are often represented by one or two species. In addition, some genera may not be monophyletic (for example, Orobanche is not). Thus, accurate numbers of species cannot be assigned to individual branches or clades. However, a few things can be noted. Euphrasia, with ~170 spp., is clearly more speciose than its sister group, with 3 spp. It has a somewhat faster d_s. However, the Schwalbea lineage, with a single species, has a fairly high d_s. Its position is not certain, but its sister group is probably the Bartsia – Melampyrum clade (>300 spp.), the Castilleja – Pedicularis clade, (>700 spp.), or the union of the two. These groups have do not have dramatically higher d_s values; in fact the Castilleja – Pedicularis clade's value is slightly lower.

Differences in generation time may play some role in the observed rate variation. However, as was found previously 15, the pattern is not clear. Since most of the genera sampled in this study contain both annuals and perennials, it is likely that most branches on the tree actually represent a combination of annual and perennial evolutionary history. However, there are some intriguing details that might merit further study. The clade containing Bartsia, Euphrasia and Melampyrum contains mostly annuals 21 and has some high d_s values, as one would expect from a generation time effect. However, both Euphrasia and Melampyrum contain almost exclusively annuals and have very different rates. Likewise, the large clade containing Boschniakia and Epifagus contains mostly annuals and has an overall high d_s. The perennials Boschniakia and Cistanche have lower d_s than their sister taxa, which also supports the generation time hypothesis, but there is as much variation among categories (perennial, annual) as between categories.

The distinctive pattern of rate increases in Orobanchaceae has at least two causes. It is clear that there is a relaxation of constraint in many (though not all) non-photosynthetic lineages. However, there is also some force affecting synonymous sites as well. At this point, it is not possible to tell whether it is generation time, speciation rate, mutation rate, DNA repair efficiency or some combination of these factors. Clearly, generating additional data from nuclear and mitochondrial genes would help us to more clearly distinguish among these hypotheses. Some of the above-mentioned hypotheses (generation-time, speciation rate) would be expected to affect nuclear and mitochondrial genomes in a similar fashion, whereas factors affecting mutation rate or efficiency of DNA repair would not, as these process involve different, though perhaps overlapping, sets of enzymes in each of the three genomes 222324.

NDY and CWD conceived of and designed the study together. NDY did the sequencing, data analyses and drafted the manuscript. CWD provided the genomic DNA samples and provided the conducive laboratory environment, both physical and intellectual, as well as many suggestions for the manuscript.