We used PCR to isolate nearly full-length segments of seven mitochondrial genes from Pelargonium hortorum and three genes from P. reniforme. All seven P. hortorum mitochondrial genes show exceptional divergence relative to other angiosperms. This finding suggests that exceptionally rapid mitochondrial evolution – rather than reduced mitochondrial DNA copy number or mitochondrial gene loss/nuclear transfer – is the explanation for the very poor hybridization of heterologous mitochondrial gene probes to P. hortorum total DNA in the survey blots described in Background. Several other lines of evidence support the conclusion that these genes are located in the mitochondrial genome and are functional: 1) Two of the seven genes were tested by Southern blots and shown to hybridize preferentially to purified mitochondrial DNA compared to total DNA from P. hortorum (data not shown). 2) Both genes tested by Northern blots hybridized preferentially to poly (A)- RNA compared to poly (A)+ RNA (data not shown), and all three genes examined for RNA editing are transcribed. 3) Four of the seven genes are invariantly mitochondrially-located in all examined eukaryotes (except in those rare organisms where the gene function has been dispensed with entirely). 4) The five protein genes examined all contain intact open reading frames. 5) In comparisons with homologs from other Geraniaceae (see below), the protein genes all display very low ratios of nonsynonymous to synonymous substitutions, indicating that they are functional and evolving under strong purifying selection.
Because a disproportionate amount of the enhanced divergence of the protein genes was at synonymous sites, we chose to present their divergence graphically in the form of phylogenetic trees constructed from synonymous sites only (Fig. ). The level of divergence in P. hortorum
is generally comparable, at both synonymous sites and all sites for ribosomal RNA genes, to divergences of Plantago rugelii
(Fig. ), which has one of the most divergent mitochondrial genomes in Plantago
] (P. media
is the most divergent Plantago
). For the three genes available, Pelargonium reniforme
is also quite divergent but consistently less so than P. hortorum
(Fig. ). For the eighth mitochondrial gene shown in Fig. , nad1
, only a relatively short sequence was available, and this only from different species of Pelargonium
]. To illustrate synonymous site divergence for this gene, we chose P. tongaense
and P. cotyledonis
, as these are close relatives of P. hortorum
and P. reniforme
, respectively. The less exceptional apparent divergence of nad1
for these two taxa (Fig. ) is probably a function of the very short region analyzed (198 NT in total, and only about 50 synonymous sites) rather than gene-specific differences in substitution rates. Phylogenetic analysis of synonymous sites for five chloroplast and three nuclear genes shows, unlike the mitochondrial situation, no evidence of exceptional divergence in P. hortorum
Figure 1 Extreme divergence of mitochondrial genes in Pelargonium hortorum. Shown are ML trees based on synonymous (dS) sites for protein genes or all sites for rRNA genes. All tree topologies were completely constrained as described in Methods. All dS trees are (more ...)
An expanded data set (Fig. ) was generated for two mitochondrial genes, cox1
and small subunit ribosomal DNA (SSU rDNA), in order to explore the extent and pattern of mitochondrial rate heterogeneity within Pelargonium
and other genera in the Geraniaceae, and to allow comparison with rate variation in other rosids and across eudicots, including the previously described case of rapid mitochondrial evolution in Plantago
]. Seven species of Pelargonium
were examined, as were eight other Geraniaceae species representing the five other genera in the family.
Figure 2 Extensive rate variation in Geraniaceae mitochondrial genes. Shown are ML trees based on synonymous (dS) or nonsynonymous (dN) sites for protein genes or all sites for rRNA genes. All tree topologies were completely constrained as described in Methods. (more ...)
The fifteen Geraniaceae taxa all show enhanced divergence at cox1
synonymous sites compared to all other rosids examined, but to markedly different extents, falling into three divergence groups (Fig. ). The five non-Pelargonium
genera are divergent compared to other rosids, members of Pelargonium
(comprising P. cotyledonis
through P. capitatum
) are much more divergent, and those of the other subgenus, Ciconium
(comprising P. hortorum
through P. candicans
), are even more divergent. The mitochondrial SSU rDNA gene shows much the same pattern, which is striking considering that all sites are included for this gene compared to only synonymous sites for cox1
. The magnitude and overall aspect of enhanced mitochondrial sequence divergence in Geraniaceae – different levels of divergence according to taxonomic group within the family – mirrors the pattern seen within Plantago
(Fig. ) [17
]. Consistent with the results shown in Fig. for Pelargonium hortorum
, levels of chloroplast synonymous site divergence and nuclear SSU rDNA divergence are unexceptional in Geraniaceae (except for somewhat enhanced rbcL
divergence in Erodium
To put the synonymous site divergence data on a quantitative footing, we calculated absolute rates of cox1 synonymous substitutions along all branches in the rosid part of the cox1 trees shown in Fig. . Absolute rates were calculated for each branch by dividing its synonymous branch length by the estimated divergence time of the branch. Divergence times within rosids were calculated using the 4-gene (three chloroplast and one nuclear), 6,226 NT character set used to estimate rosid phylogeny in Fig. . Support for the relationships in this phylogenetic tree is mixed at basal nodes within rosids. Critically, though, support is strong at the node leading to Geraniaceae and throughout the family (96–100% bootstrap support at 13 of 14 nodes; Fig. ).
Figure 3 Multigene phylogeny of Geraniaceae and other rosids. Shown is a ML tree based on a 6,226 NT alignment comprising three chloroplast genes (atpB, matK, rbcL) and the nuclear SSU rDNA. Numbers on branches are ML bootstrap values. The tree is rooted using (more ...)
A chronogram (time-based tree) of rosids is shown in Fig. , with RS, the absolute rate of cox1 synonymous substitutions, marked on each branch in units of substitutions per site per billion years (SSB units). The dearth of inferred synonymous site changes on the relatively short branches at the base of rosids is reflected in the RS = 0 estimates on most of these branches. RS is comparably low (0.31 SSB) on the H-L branch leading to the common ancestor of Geraniaceae and Crossosoma as throughout the rest of the non-zero rosid branches (RS = 0.14–0.74 SSB). On the next four branches leading from the H-L branch towards the "top" of tree, RS becomes progressively higher, from 0.3→2.4→4.1→30→38 SSB, before diminishing on subsequent branches. There is no reason to think that rate changes should coincide in timing with cladogenetic events. Given this, and wishing to minimize the number of inferred rate changes, we can very provisionally model these branch-wise rate increases with a minimum of two actual rate changes: an ~10-fold increase in RS (from 0.3 to 4 SSB) in the common ancestor of the Geraniaceae (L-M branch) and a further ~10-fold increase (from 4 to 38) in the common ancestor of Pelargonium (N-U; Fig. ). Thus, we provisionally infer at least (see Discussion) a 100-fold overall increase in RS from near the base of rosids to the period of putatively fastest evolution in Pelargonium.
Figure 4 Variable rates of synonymous substitutions in Geraniaceae cox1 genes. Shown is a chronogram based on the topology of Fig. 3. Nodes are labeled A-Z (also see Additional File 1). Shown above each branch are cox1 RS values (RS = absolute rate of synonymous (more ...)
The terminal branches leading to all examined species of Pelargonium have estimated RS values considerably below the genus' peak inferred value of 38 SSB. Likewise, the terminals leading to all non-Pelargonium Geraniaceae have RS values considerably lower than 4 SSB. In fact all of this latter set of terminals have values comparable to the inferred ancestral rosid value of roughly 0.3 SSB, as do even some of the Pelargonium terminals (Fig. ). Thus, there must have been multiple decreases in the synonymous substitution rate, some of considerable magnitude, following its early increases. We are confident of at least three rate decreases in the Geraniaceae: one each in the two subgenera of Pelargonium and one in the common ancestor of the clade comprising Erodium through Sarcocaulon. Furthermore, Fig. raises the possibility of additional rate decreases, e.g, in P. hortorum, in Erodium, and in Hypseocharis.
The large standard errors on many of the RS values within the family (Fig. ) pose a particular problem in formulating rate-change scenarios with any certainty. But what is clear is that there have been multiple increases and decreases in the rate of mitochondrial synonymous substitutions within the Geraniaceae, and that the overall magnitude of these changes has been quite large.
Divergence at nonsynonymous sites in cox1
was also analyzed phylogenetically and is plotted both at the same scale as synonymous site divergence (Fig , middle top) and at a five-fold expanded scale (Fig. , middle bottom). Although replacement sites also show enhanced divergence in Pelargonium
and other Geraniaceae relative to other rosids, the effect is considerably muted compared to the pattern of synonymous site divergence. Absolute rates of nonsynonymous substitutions (RN
) were calculated within the rosids and are given in Additional File 1
, as are RN
values. Among the very lowest RN
values are those on the branches with the highest RS
= 0.02 for the N-U branch of RS
= 30, and RN
= 0.03 for the U-X branch of RS
One notable feature of the very rapidly evolving mitochondrial DNAs of mammals is a pronounced (ca. 20-fold) transition/transversion (ti/tv) bias in favor of transitions [19
]. The most divergent Geraniaceae cox1
genes, in Pelargonium
, show only a modest increase in ti/tv (= 1.91 averaged across all within-Pelargonium
comparisons) relative to all non-Geraniaceae (ti/tv = 0.82). Base composition and the overall mutational spectrum are not significantly different in Pelargonium
than in other plant lineages (data not shown).
Although only limited data are available, there seems to be relatively little RNA editing of mitochondrial genes in the Geraniaceae (Table ). We directly assessed RNA editing, by cDNA sequencing, for three mitochondrial genes in P. hortorum
. Only a single, C→U edit site was found among the three genes, and very few edits were predicted (Table ) for the various other sequenced Geraniaceae mitochondrial genes using a sensitive program [20
] for predicting C→U editing of angiosperm mitochondrial genes. In comparison, other rosids (Crossoma
in Table ), as well as the caryophyllid Beta
and the grasses Oryza
, generally have much more RNA editing of these same five protein genes, with many edit sites conserved among angiosperms.
Observed and predicted number of RNA editing sites