Dated molecular phylogenies indicate a Miocene origin for Arabidopsis thaliana

Abstract

Dated molecular phylogenies are the basis for understanding species diversity and for linking changes in rates of diversification with historical events such as restructuring in developmental pathways, genome doubling, or dispersal onto a new continent. Valid fossil calibration points are essential to the accurate estimation of divergence dates, but for many groups of flowering plants fossil evidence is unavailable or limited. Arabidopsis thaliana, the primary genetic model in plant biology and the first plant to have its entire genome sequenced, belongs to one such group, the plant family Brassicaceae. Thus, the timing of A. thaliana evolution and the history of its genome have been controversial. We bring previously overlooked fossil evidence to bear on these questions and find the split between A. thaliana and Arabidopsis lyrata occurred about 13 Mya, and that the split between Arabidopsis and the Brassica complex (broccoli, cabbage, canola) occurred about 43 Mya. These estimates, which are two- to threefold older than previous estimates, indicate that gene, genomic, and developmental evolution occurred much more slowly than previously hypothesized and that Arabidopsis evolved during a period of warming rather than of cooling. We detected a 2- to 10-fold shift in species diversification rates on the branch uniting Brassicaceae with its sister families. The timing of this shift suggests a possible impact of the Cretaceous–Paleogene mass extinction on their radiation and that Brassicales codiversified with pierid butterflies that specialize on mustard-oil–producing plants.

The most important genetic model in plant biology is Arabidopsis thaliana. It is the first plant to have its entire genome sequenced, and it serves as a key comparison point with other eukaryotic genomes. A. thaliana is diploid and has a small genome distributed on just five chromosomes, considerations in its choice as a model (1). The age of the Arabidopsis crown group (CG), previously estimated at 5.8–3 Mya (2, 3), and of splits within Brassicaceae have been used to understand the pace of evolution in genes affecting self-incompatibility (4, 5), the rate of change in signal transduction and gene expression (6, 7), the persistence of shared chromosomal rearrangements in A. thaliana and Brassica oleracea (8), the tempo of evolution of miRNA sequences (9), the evolution of pierid butterflies specializing in plants that produce mustard oils (10), and the ages of whole-genome duplication (WGD) events giving rise to gene pairs in Arabidopsis (11). As genomes of additional Brassicaceae (e.g., Capsella rubella) and other Brassicales (e.g., Carica papaya) (12) are sequenced, the importance of robust estimates of divergence dates relating these genomes to one another and to the geological record increases substantially.

The accuracy of divergence times inferred from sequence data depends on valid, verifiable fossils to calibrate phylogenetic trees. Previous dates for the origin of Arabidopsis relied on the report of fossil pollen assigned to the genus Rorippa (Brassicaceae) (2, 3). We discovered that this report is not linked to a physical specimen, published image, or description; thus, its validity for calibration of a Brassicaceae phylogeny cannot be evaluated. Reexamination of fossils from the order Brassicales (Table S1) revealed six fossil taxa that are sufficiently documented to serve as age constraints; however, only one has been used in previous estimations of divergence times (13). Among the taxa that had been overlooked is Thlaspi primaevum (Brassicaceae) (14), an Oligocene fossil with angustiseptate winged fruits (Fig. 1A) from the Ruby Basin Flora of southwestern Montana, dated at 30.8–29.2 Mya (15). It may have been overlooked because many of the generic determinations from the Ruby Basin Flora remain to be verified, or there may have been concerns about convergence of winged fruits from unrelated angiosperm families. Although the placement of T. primaevum in Brassicaceae now has been confirmed (16), this fruit form evolved independently multiple times within the family (17). However, among extant Brassicaceae, angustiseptate fruit combined with concentrically striate seeds (Fig. 1 B, C, and E) is unique to species of Thlaspi. We examined the seed chamber of T. primaevum and found seed striations in the same pattern as those of extant Thlaspi seeds (Fig. 1D). Alliaria petiolata is the only other Brassicaceae with striated seeds (18) (Fig. S1), but it has longitudinally oriented striations, and the fruit is latiseptate; latiseptate fruits are ancestral for the clade defined by the coalescence of A. petiolata and Thlaspi arvense (Fig. S1). Thus, T. primaevum is a valid age constraint within CG Brassicaceae, and it was placed at the coalescence of T. arvense and A. petiolata as a minimum age constraint for this split (Fig. 2 and Figs. S2–S4).

Fig. 1.

Fossil and extant angustiseptate winged Thlaspi fruits and striated seeds. (A) T. primaevum fossil from the Ruby Basin Flora, western Montana. Extant T. arvense fruit backlit to show placement of seeds in the two locules (wings) (B) and with a portion of the valve removed to show striated seeds (C). Locules are separated by the replum. (D) Scanning electron micrograph of the fossil seed chamber (indicated by rectangle in A) showing impressions of striated seeds. (E) T. arvense striated seed. S, seed; R, replum; W, wing. (Dotted scale bars, 5 mm; solid scale bars, 1 mm.)

Fig. 2.

Brassicales chronogram inferred using BEAST (28). Clades with >50 species are represented by wedges proportional to species diversity. See figure for key to symbols. Thickened branch leading to Core Brassicales marks an inferred 2- to 10-fold shift in diversification rate. Putative intervals for the α and β WGD are based on refs. 12, 24, and 26. Oli, Oligocene; Pal, Paleocene; Pl, Pliocene; Q, Quaternary.

The other five potentially useful Brassicales fossils are placed outside the Brassicaceae crown group. Capparidoxylon holleisii is a Miocene wood fossil with affinities to extant Capparis, dated at 17–16.3 Mya (Table S1) (19). We explored its use as a constraint within Core Brassicales (Brassicaceae + Cleomaceae + Capparaceae) (SI Materials and Methods). Dressiantha bicarpellata is a Turonian fossil dated at 93.6–89.3 Mya (20), the oldest known putative brassicalean fossil; it provided the single age constraint used by Couvreur et al. (13). However, membership of D. bicarpellata within Brassicales is contentious (21), and thus its validity as an age constraint for estimating divergence times in the order remains unclear. To explore these issues we (i) assessed the impact of using it to constrain four nodes along the backbone of the Brassicales tree, based on results from analyses of combined nucleotide and morphological data (SI Materials and Methods) and (ii) estimated ages in Brassicales without D. bicarpellata. Finally, three species of Akania were used to constrain the ages of nodes within the Bretschneidera/Akania/Tropaeolum clade (Fig. 2). To take full advantage of these age constraints deeper in Brassicales, we expanded our sample of the plastid locus NADH dehydrogenase subunit F (ndhF) and the nuclear locus phytochrome A (PHYA), which have provided a robust phylogenetic framework for the Brassicaceae (17, 22), to include members of the Core Brassicales, giving us a matrix of 179 species (SI Material and Methods and Table S2). We estimated divergence times for trees inferred from each locus and from the combined data.

Most estimates of divergence times in Brassicaceae have assumed that rates of nucleotide evolution are equal across the tree and have been based on either the mutation rate at synonymous sites (2, 3) or the estimated times of genome duplications (23–26). Two recent analyses have allowed rates to be uncorrelated across the tree, but one of these analyses relied on D. bicarpellata as a single age constraint (13), and the other relied on a single secondary age constraint (27). Neither of these analyses took into account uncertainties in the topology of the tree (13, 27). Our analyses used multiple age constraints, allowed rates of nucleotide evolution to be uncorrelated across the tree, and accounted for uncertainty in the phylogenetic hypothesis and in the placement of the fossil age constraints.

Results and Discussion

Using Bayesian approaches (28), we estimated the origin of CG Arabidopsis at ≈13.0 Mya [95% highest probability density (HPD): 17.9–8.0] (Fig. 2 and Table 1), considerably older than the frequently cited estimate of 5.8–3 Mya (Table 1) (2, 3). When a penalized likelihood approach (29, 30) was used, the estimate for this node was even older (Table S3). In the subsequent discussion, we focus on estimates from the Bayesian analyses (Table 1 and Table S3). The placement or exclusion of D. bicarpellata had little impact on the age of this node and other nodes in the tree (Table S3). Conversely, the T. primaevum constraint did lead to slightly older ages for most nodes in the tree (Table S4). Nonetheless, even without this constraint within CG Brassicaceae, age estimates were substantially older than previous estimates (Table 1), possibly resulting from our use of multiple constraints, dense taxon sampling (but see ref. 13), from our allowing for phylogenetic uncertainty, alternative placement of fossil calibrations, from the relative completeness of our molecular datasets (cf. ref. 13), or from a combination of these factors. The previous estimate for CG Arabidopsis is primarily Pliocene (5.3–1.8 Mya), whereas the revised age falls within the Miocene (23.03–5.3 Mya) and spans a particularly warm period in recent earth history that included the Middle Miocene climatic optimum (31). Thus, warming may have played a role in the divergence of A. thaliana from other Arabidopsis. Moreover, the much older optimal age estimate of 13.0 Mya suggests that the pace of chromosomal rearrangements (1), divergent gene regulation (6), and the breakdown of self-incompatibility (4) may have proceeded more slowly than has been appreciated.

Table 1.

Comparison of published dates with those estimated in this study using Bayesian inference in BEAST on the ndhF + PHYA Brassicales chronogram

Node age	Koch et al. 2000 (2), 2001 (3) synonymous substitution rate, single fossil constraint (Mya)	Franzke et al. 2009 (27) BEAST, nad4, single secondary constraint (Mya)	Couvreur et al. 2010 (13) BEAST, supermatrix, single fossil constraint (Mya)	This study BEAST, ndhF + PHYA, four fossil constraints (Mya)
Arabidopsis CG	5.8–3	ND	ND	17.9 – 13.0 – 8.0
Brassiceae CG (whole genome triplication)	14	ND	ND	28.3 – 22.5 – 15.6
Lineage I CG	19–13	19.0 – 8.0 – 0.5	36.1 – 27.3 – 18.2	42.8 – 35.6 – 28.5
Lineage II CG	18.1	ND	37.2 – 28.2 – 18.1	37.8 – 30.8 – 23.7
Lineage III CG	ND	ND	29.3 – 21.4 – 14.8	42.7 – 35.5 – 28.3
Arabidopsis–Brassica MRCA	21–16	ND	≈34*	50.7 – 43.2 – 36.6
Core Brassicaceae CG	ND	28.0 – 11.0 – 1.0	42.8 – 32.3 – 20.9	54.3 – 46.9 – 39.4
Brassicaceae CG	60–30	35.0 – 15.0 – 1.0	49.4 – 37.6 – 24.2	64.2 – 54.3 – 45.2
Brassicaceae–Cleomaceae MRCA	ND	35.0 – 19.0 – 1.0	ND	76.5 – 64.5 – 54.4
Core Brassicales CG	ND	ND	ND	83.2 – 71.3 – 59.7

Ages in bold are optimal reconstructions and are bracketed by upper and lower bounds of the 95% HPD interval. CG, crown group; MRCA, most recent common ancestor; ND, not determined.

*Approximate mean date inferred from chronogram and for which HPD data were not available (13).

The first split within CG Brassicaceae occurred in the early Eocene, ≈54.3 Mya (95% HPD: 64.2–45.2) (Fig. 2 and Figs. S2–S4). CG Lineages I, II, and III, which contain the majority of Brassicaceae (17, 22), radiated in the mid to late Eocene, from 43.4 to 33.3 Mya (Table 1). Because A. thaliana and B. oleracea (broccoli and related plants) occur in Lineages I and II, respectively, the relatively deep coalescence of these scientifically and economically important species dates to ≈43.2 Mya (95% HPD: 50.7–36.6; Table 1), indicating that conservation of the large chromosomal blocks shared by these two species (8) has persisted for longer than previously thought (2, 3, 13). The B. oleracea genome, along with several close relatives, including Brassica rapa (turnip) and Brassica napus (canola), has been the subject of intense study because of the agricultural importance of the group. The whole-genome triplication event that likely facilitated the origin of these crops was dated previously at 14 Mya (32). Our age estimate for the triplication is centered at 22.5 Mya (95% HPD: 28.3–15.6; Table 1).

Our data suggest that the Core Brassicales stem group (SG) originated around 71.3 Mya (95% HPD: 83.2–59.7), shortly before the end of the Cretaceous, when SG Capparaceae split from SG Brassicaceae and SG Cleomaceae (Fig. 2 and Table 1). Furthermore, using LASER (33), we detected a twofold shift in net diversification rate along the branch to core Brassicales when extinction rates are low or a 10-fold shift when extinction rates are high (Table S5). SG Brassicaceae split from SG Cleomaceae at the Cretaceous–Paleogene boundary (KPB; ≈65.5 Mya) (Table 1 and Fig. 2). Thus, stem members of all three families may have survived the KPB mass extinction event before their subsequent radiations in the Eocene (Fig. 2 and Fig. S4). A pattern of origin before the KPB and radiation afterward has been noted in animal (34) and other plant lineages (35), although the lag time before recovery is greater in Brassicaceae than has been inferred in the other groups (35).

A great deal of interest has centered on the timing of WGD events in the history of A. thaliana, particularly on the most recent of these events, the α WGD, estimated to have occurred between 100 and 20 Mya (11, 36–39), and the β WGD, estimated to have occurred between 235 and 112 Mya (36, 38, 39) (Fig. 2). Recent lines of evidence place the α WGD within Brassicales (12), perhaps within Brassicaceae (26, 27, 36, 40), and place the β WGD within Brassicales (12, 40). These placements have fueled speculations that genome-doubling events are linked with diversification in Brassicaceae (13, 41) and with survival of the KBP mass extinction event (24). These hypotheses are attractive, because WGD may result in the colonization of new habitats (42), major ecological transitions (43), invasiveness (44), and the generation of morphological novelty (45). Our chronogram (Fig. 2 and Fig. S4) provides the framework for testing these hypotheses. For example, if the β WGD occurred around 70 Mya, placing it on the stem lineage of Core Brassicales (Fig. 2), we would expect to find that paralogs stemming from this event are shared by members of Brassicaceae, Cleomaceae, and Capparaceae but not by other families in Brassicales. If a WGD maps to this branch, the locus of the shift in net diversification rate is consistent with a link between genome doubling and survival across the KPB mass extinction and perhaps to subsequent diversification.

Despite the interest in genome doubling and its effects, it is important to consider other factors that might promote diversification. Ehrlich and Raven (46) cited the interaction between Brassicales and pierid butterflies as a prime example of coevolution between plants and insects. Pierids specializing on Brassicales produce nitrile-specifier protein essential for detoxification of glucosinolates, a chemical defense found almost exclusively in Brassicales (10). This key innovation facilitated a host switch onto Brassicales, dated at ≈85 Mya (10, 47), and was followed by the subsequent diversification of pierids (10). The locus of the shift in Brassicales net diversification rate and the timing of the pierid butterfly radiation suggest that they codiversified (Fig. 2 and Fig. S4), supporting a central tenet of coevolutionary theory (46). However, the extent to which diversification in Brassicales may have driven pierid diversification, and vice-versa, requires further study.

Carefully designed comparative studies reveal the processes that generate and maintain biodiversity. In this study we identify a major shift in diversification rate that may correlate with survival across and diversification following the KPB mass extinction event, WGD events, and the evolution of pierid butterflies. Our results suggest a number of ways in which our understanding of the history of CG Arabidopsis should be revised, notably that it evolved during a period of warming rather than cooling and that genome structure and developmental processes have been slower to evolve than has been appreciated. The evolutionary history of A. thaliana and its neighborhood described here will allow more precise application of our understanding of this model organism to other flowering plants, land plants, and all eukaryotes.

Materials and Methods

GenBank accession numbers, specifics of the evaluation of published fossils, and detailed methods can be found in SI Materials and Methods.

Phylogenetic Inference.

We used RAxML (48) to infer phylogeny in Brassicales from three alignments. The plastid ndhF alignment comprised 2,067 nucleotide sites representing 170 species; the nuclear PHYA alignment comprised 1,824 nucleotide sites representing 139 species; the combined alignment comprised 3,798 nucleotide sites from 177 species (Table S2). Analyses assumed a general time-reversible model of sequence evolution, with γ-distributed rate heterogeneity; partitions in the combined matrix were allowed to evolve independently. The topology and clade support in the resulting trees are consistent with previously published phylogenetic estimates using ndhF (17) and other plastid (49) and nuclear (3, 22, 50, 51) markers.

Ultrametric Tree and Divergence Date Estimation.

We calculated divergence dates for Brassicales from all three alignments using r8s v. 1.71 (30) and BEAST v. 1.5.3 (28). These methods account in different ways for variation in substitution rates among branches on the tree. We used r8s to explore the impact of placing the Dressiantha constraint at different nodes along the Brassicales backbone; potential placements were inferred in analyses of combined ndhF and morphological data (details are given in SI Materials and Methods). Because the placement of Dressiantha had little impact on age estimates within Core Brassicales (Table S3), we used a single placement in our BEAST analyses.

We allowed BEAST to infer topology, branch lengths, and dates for ndhF and combined data. For the PHYA data, we fixed the tree to that used in the r8s analyses and allowed BEAST to alter branch lengths while inferring dates (SI Materials and Methods). We used a uniform distribution for all three fossil calibrations with the lower hard bound of the distribution set to the youngest age of the fossil (see text, SI Materials and Methods, and Table S1) and the upper hard bound set to the first fossil record for eudicot pollen (125 Mya) (52) for D. bicarpellata and Akania sp. For other fossil calibrations the upper hard bound was determined by the age of other fossils used in the analysis (SI Materials and Methods). BEAST runs of 3 × 10⁷ generations, saving data every 1,000 generations, produced 30,000 estimates of dates under a Yule speciation prior and an uncorrelated relaxed clock (28) for the single-gene datasets. Convergence statistics for each single-gene run were analyzed in Tracer, resulting in 27,000 post–burn-in trees. BEAST runs of 6 × 10⁷ generations, saving data every 1,000 generations, produced 60,000 estimates under a Yule speciation (53) prior and uncorrelated relaxed clock (28) for the combined data. Also, for these data, we resampled at a lower frequency using LogCombiner v. 1.5.3 (28), resulting in a tree file with 30,000 trees. We used TreeAnnotator v. 1.5.3 (28) to produce maximum clade credibility trees from the post–burn-in trees and to determine the 95% probability density of ages for all nodes in the tree (Figs. S2–S4).

Shifts in Net Diversification Rate.

We used LASER v. 2.2 (33) to test for significant shifts in net diversification rates over the history of Brassicales. Clade diversity for each group was determined from the Angiosperm Phylogeny Website (54). We calculated the likelihoods of our phylogenetic data under a model of constant diversification rate over time and under a two-rate model in which the diversification rate is permitted to change (Table S5). Because analyses of diversification rate can be sensitive to extinction, we determined whether our results were sensitive to differences in extinction fraction by reanalyzing the data under low (a = 0) and high (a = 0.99) extinction (Table S5).