The age and biogeography of Citrus and the orange subfamily (Rutaceae: Aurantioideae) in Australasia and New Caledonia†
The authors thank R. J. Bayer for access to unpublished data.
Abstract
The geological history of Australasia, New Caledonia, and Southeast Asia, has been complex, resulting in competing biogeographic hypotheses for taxa found here. Alternative hypotheses—Gondwanan vicariance, rafting terranes, long-distance dispersal—may be distinguished by different predicted divergence times between disjunct sister taxa. Taxa within Rutaceae subfamily Aurantioideae are ideal for testing these hypotheses because of their distributions. Therefore, the ages of Rutaceae and Aurantioideae were estimated using molecular dating. One data set comprised 51 sequences of rbcL and atpB with sampling across rosids and three fossil calibrations: crown Fabales+Fagales+Rosales (>94 Ma), Fabaceae (>51 Ma) and stem Ailanthus, Simaroubaceae (>52 Ma). Another data set comprised 81 Aurantioideae using >8 kb of chloroplast sequence and secondary calibration. Confidence in estimated divergence times was explored by varying the root age, dating method (strict, local, and relaxed clocks), and inclusion of internal calibrations. We conclude that the Rutaceae crown diverged in the Eocene (36.4–56.8 Ma, mean 47.6), whereas the Aurantioideae crown originated in the early Miocene (12.1–28.2 Ma, mean 19.8). This young age suggests that Gondwanan vicariance does not explain the distributions of extant Aurantioideae. Taxa found in New Caledonia may have arrived by separate transoceanic dispersal events.
The advent of molecular dating has added a new dimension to hypothesis testing in biogeography by enabling comparisons between alternative lineage divergence times predicted under different process scenarios (13; 41). For example, a hypothesis of vicariance predicts simultaneous divergence times between sister taxa occurring in isolated areas, coincident with the geological date of isolation. Molecular dating has led to rejection of vicariance in come cases (8; 29) and support in others (8; 12; 28).
Australia, New Guinea, and New Caledonia are now isolated by sea but were once united as part of the supercontinent Gondwana. It has been hypothesized that when New Caledonia drifted away from Gondwana, from c. 80–70 Ma (36), it could have carried populations that subsequently diverged and became vicariant sister taxa occurring in Australia and New Caledonia (32). This hypothesis predicts divergence times older than 70 Ma between disjunct sister taxa in these land masses (32). However, thinning of the rifted continental crust caused New Caledonia to be submerged under the sea from the late Cretaceous to the early Eocene (indicated by marine sediments over much of the island), and then the island was obducted (over-thrust) by ophiolite during the late Eocene (36; 40; 7; 52). Therefore, there was probably no habitable land surface on the island of New Caledonia until c. 35 Ma (36; 40). This likely lack of habitable land leads to the prediction that contemporary terrestrial biota in New Caledonia should have colonized the island by transoceanic dispersal and diverged from their sister taxa in other land masses no earlier than c. 35 Ma (40). A third geological model (31) hypothesizes that islands on the Kenn Plateau, between New Caledonia and Australia, could have acted as a stepping-stone dispersal route during the period of rifting until 35 Ma. Although not conforming to the classical vicariant pattern, this model of vicariance with short-range dispersal has been proposed to explain the distributions of taxa present in these two land masses without the need for long-distance dispersal events that cross wide ocean gaps (31). Under this hypothesis, divergences between disjunct taxa could date to any time in the period between 80 and 35 Ma but are much less likely after this time, when the current wide oceanic gap formed.
Rutaceae consist of c. 160 genera and 1800 species characterized by foliar oil glands and are widely distributed in the tropics. Subfamily Aurantioideae (which includes the Citrus group) has disjunct taxa occurring in Australia, New Guinea, New Caledonia, Asia, and Africa. This distribution makes the subfamily ideal for testing alternative biogeographic hypotheses centered on Australasia (including New Caledonia). The orthodox explanation for disjunct distributions of several members of Aurantioideae in Australia and New Caledonia is vicariance (24; 31), and this hypothesis can be evaluated against the alternatives described by estimating the timing of the divergences between the disjunct taxa. Similarly, to explain transoceanic disjunctions of Aurantioideae between Australasia and Southeast Asia, other authors have proposed testable hypotheses that imply different ages of divergence between taxa (56; 4).
To test these hypotheses, we estimated the age of Rutaceae and of clades within Aurantioideae using fossil-calibrated molecular dating. We performed the molecular dating analysis in two stages: the first used rbcL and atpB chloroplast sequences that were available and alignable for a broad range of eudicot angiosperms and included deep fossil calibration points (11). The second stage used >8 kb of chloroplast sequences available from in-depth sampling within Aurantioideae and a secondary calibration point derived from the first stage.
Initial DNA-based phylogenetic work using chloroplast sequences confirms the placement of the Rutaceae in the Sapindales (20; 5). The phylogenetic relationships among members of the family have not been comprehensively studied in any single analysis, although several studies using molecular data have sampled a reasonable number of taxa. 5 sampled 32 genera in Rutaceae s.l. and found paraphyly among the subfamilies Aurantioideae, Rutoideae, and Toddalioideae, although only Luvunga (which has since been found to be a misidentification: R. J. Bayer, unpublished data) did not group with the remaining Aurantioideae. 54 sampled 23 genera, but found little resolution beyond excluding Dictyoloma and Spathelia (each in their own subfamily) from the core Rutaceae. Similarly, 5 excluded Cneorum (Cneoraceae), Ptaeroxylon (Ptaeroxylaceae) and Harrisonia (Simaroubaceae or Rutaceae), which were placed in a single clade. Hereafter, the sister clade of the Cneorum clade is denoted as Rutaceae s.s., and both clades together as Rutaceae s.l. 44 focused upon Rutoideae and Toddalioideae with 19 genera sampled and found a close relationship between some Toddalia and Zanthoxylum species. Bayer et al. (unpublished data) focused on Aurantioideae with 47 genera sampled and demonstrated the monophyly of this subfamily, but their sample of Toddalia did not group closely with Zanthoxylum (although their sampling in this part of the tree was limited).
The consensus that emerges from these studies is that a sister relationship exists between a clade of five genera (the Cneorum clade) and the remainder of genera traditionally placed in Rutaceae (5; 39). Furthermore, the most recent common ancestor (MRCA) of Rutaceae s.s. can be captured by using Eriostemon, Ruta, and Skimmia, despite some variation among studies in deep relationships within this clade. The MRCA of Rutaceae s.l. can be sampled by adding Cneorum to any of the former three genera.
Within subfamily Aurantioideae, the study by R. J. Bayer et al. (unpublished data) included all but one genus, Limnocitrus, and therefore presents the most complete phylogenetic picture within the subfamily using molecular data to date (Fig. 1). The deepest division within this phylogeny places most members of Swingle and Reece's (1967) Clauseneae in clade 1, which is sister to clade 2, that includes all members of Citreae and the remaining Clauseneae (Merrillia and Murraya excluding Bergera) (Fig. 1). Clade 1 includes plants from southern and southeastern Asia, southern China, Malesia, and Australia and extends eastward to the western Pacific with one genus (Micromelum) in New Caledonia. Clausena is the only genus to extend westward as far as tropical and southern Africa. Clade 2 has a similar distribution, although two genera are present in New Caledonia (Citrus s.l. [including Oxanthera] and Murraya) whereas four are present in Africa (Citropsis and the closely related Aeglopsis, Afraegle and Balsamocitrus) (56).
Relationships within Rutaceae subfamily Aurantioideae based on over 8 kb of chloroplast sequence redrawn from R. J. Bayer (unpublished data); accession details are also available from the first author on request. Ruta and Toddalia are the only outgroups shown here—the remainder of taxa belong to subfamily Aurantioideae. Thicker branches are supported by at least 80% (parsimony) or at least 0.95 (PP); otherwise, support values are marked near branches (parsimony bootstrap/PP for the three partition Bayesian analyses). Boldfaced numbers are mean dates and 95% credibility intervals based on the secondary calibration using a fixed Citrus-Clausena node age (see Table 3). The Citrus-Clausena age is drawn from the primary calibration and is in boldfaced italics. Clades shown by triangle have additional accessions sampled, with the number of samples shown after the genus name and the length based on the longest terminal branch in that clade (all other branch lengths are proportional to inferred molecular change). Boxed names are taxa in common with the analysis shown in Fig. 2. Dashed arrows point to New Caledonian taxa. Distributions of genera are to the right of names (Af = Africa, As = Asia, SE-As = south-east Asia, Au = Australia, NG = New Guinea, NC = New Caledonia). *Atalantia has a doubtful record from New Guinea (56).
The ages of Rutaceae and genera within the family have been inferred previously using information from fossils (30), biogeographic disjunctions (24) and fossil-calibrated molecular dating (39). 30 describe a late Cretaceous seed fossil from central Europe as Rutaspermum biornatum Knobloch and Mai. 59 attributed an Upper Paleocene seed fossil to Euodia costata (Chandler) Tiffney (although 24, suggests it might belong in Tetradium). Further fossils placed in Rutaceae and assigned to extant genera become common in the Eocene (22). 24: 25) suggested that both Euodia and Melicope date to the Upper Cretaceous (at least 71 Ma), but it is not clear whether crown or stem ages were being estimated. His primary evidence was disjunction between these land masses and that were formerly part of Gondwana (including Australia, Madagascar, and New Caledonia), though he assumed that vicariance resulting from the Gondwanan breakup was followed by some transoceanic dispersal because these genera now overlap in distribution (24). 39 used various fossils to calibrate molecular dates of Sapindales lineages, estimating a crown age for Rutaceae s.l. of 56.5–86.0 Ma using Bayesian methods.
MATERIALS AND METHODS
Data and analyses
The first stage of the analysis was designed to estimate the age of Rutaceae and its major clades using available deep primary calibration points. We concatenated sequences from GenBank of chloroplast rbcL and atpB (with ends trimmed; total of 2694 bp) from 51 taxa representing Rutaceae, other rosids and asterid outgroups. Alignment was by eye and straightforward. Sampling was uneven with greater emphasis on rosid II clade members (2) to compare the molecular-derived dates with fossils described later.
Phylogenies were estimated using a Bayesian Markov chain Monte Carlo (MCMC) search (62) in the program MrBayes 3.1.2 (26), with two parallel runs of 5 M generations and 10 chains each, using the GTR+I+G model. This model was found to be optimal by the program Modeltest 3.7 (45) using both the hierarchical likelihood ratio test (17) and the Akaike information criterion (1). Convergence of the parallel runs was determined by examining the average standard deviation of split frequencies, which fell below 0.01. Clade support was assessed using the posterior probabilities (PP) from the Bayesian analysis and also using parsimony bootstrap analysis (16) with five random addition sequence replicates for each of 200 bootstrap replicates, holding a maximum of 100 trees per random addition sequence replicate.
Relaxed molecular clock dating
To test whether the data were clock-like, we specified the GTR+I+G model and optimal parameter values from Modeltest in a Nexus format sequence file that was analyzed using maximum likelihood in PAUP* 4.10b (57) in a heuristic search using five random addition sequence replicates. The best tree was used for a likelihood ratio test in PAUP* of clocklike behavior of the data. Clock vs. nonclock constraints were applied to the best tree, and the parameter values reestimated with branch lengths to find the likelihood score of each constraint on the best tree.
A number of methods for molecular clock dating are currently available (48). Our preference is for one that estimates uncertainty of the dates by taking into account uncertainty in the topology and branch lengths, one that allows fossil calibrations, one that accommodates rate differences among branches, and one that includes the assumption of autocorrelation. Some methods can account for the uncertainty in the age of fossils used as calibration points (15), but uncertainty in fossil placement on trees is probably a greater source of error (35). As well, a given fossil is rarely, if ever, the oldest representative of its lineage, and as such the true age of the lineage bearing the fossil's synapomorphies is perhaps unknowable. We applied the approach of minimum constraints using penalized likelihood (PL) (50), which is a widely used and well-understood method, to accommodate this latter uncertainty. As we show later, the effect of a key calibration point (the Ailanthus fossil) appears to have a much greater effect on node ages than the other issues raised above.
We combined PL with a Bayesian approach to estimating phylogenetic uncertainty (61) as follows. One hundred trees were taken from the last 200 K generations of one of the MrBayes runs (arbitrarily chosen) and imported into r8s 1.71 (50). Dates with confidence intervals were estimated in PL using two strategies. Strategy 1 used successive searches in which the only constraint was fixing the age of the root (the crown age of Saxifragales+other rosids) to 95 Ma, and thereafter in 10-Ma increments (more detail later). Strategy 2 added the following minimum-age constraints: Fabales+Fagales+Rosales (FFR node) ≥ 94 Ma, and Ailanthus stem ≥ 52 Ma (explained later). Cross validation (49) with the truncated Newton algorithm (50) was used to find the optimal smoothing parameter of 7.9.
Strategy 2 was repeated using two other dating methods in r8s—Langley Fitch (LF) (molecular clock) (33) with the truncated Newton algorithm (50) and LF (local clock) with the Powell algorithm (21; 47)—to test for differences that might be due to the assumptions implicit in the PL procedure, although we expected the PL estimate to be the most realistic. Penalized likelihood, simply put, allows many small rate changes across the tree but penalizes large shifts in rate to a degree influenced by the smoothing parameter (49). The LF procedure enforces a clock and is clearly an extreme and inappropriate procedure for these data that have been shown to be nonclocklike (see Results). The Langley Fitch (local rates) method lies somewhere in between by allowing a few rate changes across the tree, but only in clades designated a priori. Rates are kept constant within the predefined rate classes. Three rate classes were designated and applied as follows: one class for the Bataceae+Brassicaceae clade and the Medicago+Glycine (Fabaceae) clade (including stems in both cases), a second class for the Simaroubaceae clade (crown only) and the Saxifragaceae clade (including stem), a third class for the remaining rates. These selections were based on rates estimated from the ML tree using PL with a 115-Ma root and strategy 2 (not shown). Rates were ordered and graphed using Excel (Microsoft) and appeared to describe a sigmoid curve (Appendix S1, see Supplemental Data with online version of this article). Clades with a majority of branches in the lower and upper parts of the curve were given two different rate classes from that of the remainder of branches.
The second stage of the analysis used 200 trees drawn from the stable posterior distribution found by R. J. Bayer et al. (unpublished data) from their Bayesian MCMC analysis of 8 kb of cpDNA (sequences lodged in GenBank) for just the Aurantioideae. Dates were estimated from these using PL, with 95% credibility intervals approximated by using twice the standard deviation around the mean (as per 53). Parsimony bootstrapping was also done as in R. J. Bayer et al. (unpublished data) for branch-support comparison with the PP. The MRCA of the subfamily (defined by the MRCA of Citrus and Clausena) was constrained alternately to the minimum, mean and maximum values of the credibility interval of the PL-estimated ages found in the first analysis. The reason for using these particular secondary constraints depended upon the outcome of the first stage of the analysis (see Discussion, Root age). Cross validation was used to find the optimal smoothing parameter of 5.0.
Fossil calibrations
Perhaps the most relevant fossil for understanding the age of Citrus is a leaf from Italy dated to the Pliocene (19). This fossil has a prominently winged and articulated petiole and is a good match in shape and size to leaves of extant C. aurantium (19). However, the articulation of the petiole occurs throughout subfamily Aurantioideae, in genera distantly related to Citrus (e.g., Glycosmis, Merrillia and Murraya) as well as in more closely related ones (e.g., Pleiospermium, Citropsis and Atalantia: 56). Flattened petioles with a channel formed by the extension of the margin also occur widely in the subfamily (e.g., in Glycosmis and Merrillia), but petioles with wings as prominent as the fossil are more restricted, being found within Citrus and in two other related species, Pleiospermium latialatum (B.E. Pfeil, personal observation; 19) and Citropsis gabunensis, the latter species having leaves that are highly similar to Citrus aurantium (56). We did not impose the Citrus leaf fossil age as a constraint in the analysis because the age (“Pliocene”) is imprecise, ranging from 1.8–5.3 Ma.
Fossil samaroid fruit that share characters with all extant species of Ailanthus have been reported from a horizon from the early Eocene flora of North America (52.2–52.7 Ma), from the early or middle Eocene of eastern Russia and of northeastern China, and from the middle Eocene of Germany (9). Initial analyses showed that support for the relationship of Ailanthus as sister to other Simaroubaceae is high using rbcL (18) and combined rbcL and atpB sequence data (5). Increased sampling of genes and taxa is consistent with the earlier findings, however, with newly sampled genera Castela, Holocantha and Picrasma forming a clade sister to the remaining Simaroubaceae genera, including Ailanthus (6). A conservative minimum age constraint of 52 Ma was applied to the divergence of Ailanthus from Simarouba, following 38. Given that most genera in Simaroubaceae have drupaceous fruits, that both the clade containing Castela, Holocantha, and Picrasma (sister to the remaining genera in Simaroubaceae) and its sister family, Meliaceae (39), lack samaroid fruit, and that the only samaroid fruit in Simaroubaceae are found in Ailanthus, the placement of the constraint clearly lies somewhere between the crown Ailanthus node and the node below it. By placing the constraint on the node below, our analysis errs toward a younger date, although perhaps not by much, given that the crown node is relatively close to the node below it (6).
The minimum constraint of 94 Ma on the FFR node age derives from an unnamed collection from Rhamnaceae (3) cited as “fossil W” by (11), but we applied the age much deeper in the tree because of our limited sampling and uncertainty of phylogenetic reconstruction of nested nodes (e.g., Rosaceae+Fagaceae+Curcurbitaceae). Therefore, the FFR node represents a very conservative application of this age and will tend to underestimate the ages of nodes above.
The rationale for testing successively older fixed-age constraints at the root node was based on the behavior of earlier implementations of nonparametric rate smoothing (NPRS) and PL. Prior to r8s version 1.71, analyses without a fixed root constraint using either NPRS or PL could allow the root to become much older than would be realistic (B. E. Pfeil, personal observation; r8s manual with version 1.71; 50). However, a root extrapolated from a contained younger node may result in an inappropriately young root age. We explored the rates estimated along branches using the differing root ages to determine whether a root older than 95 Ma was more appropriate. The fixed root age was extrapolated from the FFR minimum node age of 94 Ma because it needed to be set slightly older; r8s does not accommodate two hierarchically related nodes with the same age. Therefore, we chose 95 Ma as the youngest fixed age we allowed for the root.
RESULTS
Using rbcL and atpB, only a single best tree was found following the unconstrained (nonclock) maximum likelihood search (–lnL = 22 924.3, Appendix S1 in online Supplemental Data). A molecular clock constraint found a tree score of –lnL = 23 206.6. The χ2 value of 564.6 with N – 2 = 52 degrees of freedom yields a significant rejection of clocklike behavior of the data (P << 0.01), therefore requiring relaxed-clock methods to determine node ages. In the Bayesian majority rule consensus tree drawn from the stable part of the posterior distribution (Fig. 2), Rutaceae s.s. and Aurantioideae were found to be strongly supported by both parsimony bootstrap (>80%) and PP (>0.95), whereas Rutaceae s.l. (including Cneoraceae) had high PP (0.99) but low (69%) parsimony bootstrap support.
Wider relationships of Rutaceae and other rosid taxa inferred using rbcL and atpB based on the 50% majority rule Bayesian consensus and the GTR+I+G model. Support is at least 80% (parsimony) or at least 0.95 (PP) for thicker branches; otherwise listed in that order. Nodes with two asterisks (**) below are used as age constraints in the molecular dating (see text). GenBank accession numbers for these genes are listed after the taxon names (single numbers refer to whole chloroplast accessions). The family placement of Irvingia is uncertain (formerly part of Simaroubaceae).
PL dates using internal vs. no internal constraints
Clear differences among node ages between the r8s strategy 1 (only the root fixed) and strategy 2 (with internal minima) were seen (Table 1), e.g., with a fixed root of 95 Ma, Rutaceae s.s. mean age is estimated to be 28.3 or 44.8 Ma (strategy 1 and strategy 2, respectively), and the Citrus–Poncirus divergence mean age is 3.3 or 5.5 Ma, respectively.
| Node | Age (strategy 1) | Age (strategy 2) | Age (strategy 1) | Age (strategy 2) | Age (strategy 1) | Age (strategy 2) |
|---|---|---|---|---|---|---|
| Root | 95 | 95 | 105 | 105 | 115 | 115 |
| Fabales-Fagales-Rosales | 70.6 (61.6–78.0) | 94 | 77.7 (67.8–85.9) | 94 | 84.8 (73.9–93.9) | 94.0 (94.0–95.4) |
| Fabaceae | 56.8 (48.0–65.3) | 75.6 (64.6–83.0) | 62.4 (52.7–71.8) | 75.4 (64.5–82.9) | 67.9 (57.3–78.2) | 75.3 (64.5–82.7) |
| Ailanthus stem | 21.5 (16.0–26.9) | 52 | 23.8 (17.7–29.7) | 52 | 26.1 (19.4–32.5) | 52 |
| Rutaceae s.l. (+Cneorum) | 37.5 (27.1–49.3) | 59.0 (50.1–67.4) | 41.5 (30.1–54.4) | 60.8 (51.6–70.0) | 45.5 (33.2–59.5) | 62.7 (53.3–72.7) |
| Rutaceae s.s. (–Cneorum) | 28.3 (21.4–34.5) | 44.8 (34.5–52.5) | 31.3 (23.7–38.2) | 46.1 (35.4–54.5) | 34.4 (26.0–41.8) | 47.6 (36.4–56.8) |
| Aurantioideae (Citrus-Clausena) | 11.7 (7.6–17.0) | 18.6 (11.5–26.0) | 12.9 (8.2–19.1) | 19.2 (11.8–27.0) | 14.1 (8.9–21.3) | 19.8 (12.1–28.2) |
| Citrus-Pleiospermium | 9.6 (6.2–13.8) | 15.4 (9.2–21.4) | 10.6 (6.7–15.6) | 15.9 (9.5–22.3) | 11.7 (7.3–17.5) | 16.4 (9.7–23.1) |
| Citrus-Poncirus | 3.3 (1.5–7.1) | 5.5 (2.2–12.4) | 3.7 (1.6–8.3) | 5.7 (2.3–13.0) | 4.1 (1.8–9.6) | 5.9 (2.4–13.6) |
We increased the fixed root age in 10-Ma increments from 95 to 115 Ma but not beyond, for the following reasons. The penalized likelihood algorithm attempts to minimize rate shifts between branches, but allows shifts so that node ages conform to constraints placed upon the analysis. If we do not enforce internal constraints and shift only the root, we can explore the possibility that an older root age may be consistent with the constraints, while minimizing rate shifts. (Because all fossil constraints are minima, increasing the age of the root will increase internal node ages until eventually they become consistent with internal constraints.) We found that increasing the root age to 115 Ma or older produces internal node ages that approach or exceed the fossil minimum ages in several cases (FFR node, ≥94 Ma, mean estimated at 84.8; Fagaceae-Cucurbitaceae, ≥90 Ma (11), mean estimated at 77.4 Ma (not shown); Oxalidaceae-Malpighiaceae clade, ≥90 Ma (11), mean estimated at 89.5 Ma (not shown); Fabaceae crown, ≥51 Ma (11), mean estimated at 67.9; Pliocene Citrus-Pleiospermium, ≥Pliocene [5.3–1.8 Ma], mean estimated at 11.7), though much younger in one case (Ailanthus stem, ≥52 Ma, mean estimated at 26.1).
However, increasing the root age alone until all ages are equal to or greater than their fossil constraints is only applying rate smoothing to correct for different branch lengths from root to tip. It does not fully exploit fossil constraints as primary evidence for rate changes between nodes. We examined local rates inferred by strategy 2 (constraints applied) and discovered that when the root was fixed at 95 Ma, several branches are inferred to have implausibly high rates (Fig. 3). When the root is fixed at 105 Ma, some branches still have rates that appear to lie outside the main distribution, even with a right skew assumed. When the root is fixed to 115 Ma, no rates look unreasonable (a right skew is still present), even though the internal constraint on the Ailanthus stem nearly doubles the estimated age of that node, with surrounding nodes affected to a lesser degree.
Frequency distribution of local rates (x-axis = changes per site per billion years) inferred using the maximum likelihood tree and the penalized likelihood in the program r8s with all fossil constraints enforced with three different root ages (95, 105 and 115 Ma). Note the break in the x-axis scale between 2.00 and 20.00.
With the root node fixed to 95–115 Ma under strategy 2, internal node ages investigated here do not change by very much (e.g., means in the Fabaceae crown range from 75.6–75.3; in Rutaceae s.s. crown from 44.8–47.6; in Citrus-Poncirus from 5.5–5.9).
Internally constrained searches: PL vs. LF (strict clock) and LF (local clock)
Despite using different assumptions about how much and how often rates can shift in the phylogeny, both the strict clock and local clock analyses (strategy 2) gave very similar results to PL (strategy 2) in some parts of the tree (Tables 1, 2). Branches closest to internal calibrations that were not exceeded in the searches were not much affected. The Citrus+Poncirus divergence-time estimate was similar across these three internally constrained analyses, with means ranging from c. 5–6 Ma. Citrus+Clausena (i.e., Aurantioideae crown) mean ages ranged from 17.5–19.8 Ma across the same three analyses, whereas a greater difference was seen in Rutaceae s.l. mean ages, which ranged from 53.6–57.5 in LF methods but 59.0–62.7 Ma in PL.
| Node | Age (strategy 2, LF) | Age (strategy 2, LF local) | Age (strategy 2, LF) | Age (strategy 2, LF local) | Age (strategy 2, LF) | Age (strategy 2, LF local) |
|---|---|---|---|---|---|---|
| Root | 95 | 95 | 105 | 105 | 115 | 115 |
| Fabales-Fagales-Rosales | 94 | 94 | 94 | 94 | 94.6 (94.0–100.2) | 94 |
| Fabaceae | 79.2 (70.4–85.1) | 57.2 (51.0–71.1) | 79.7 (71.0–85.5) | 57.2 (51.0–71.0) | 80.6 (71.6–87.1) | 57.3 (51.0–71.0) |
| Ailanthus stem | 52 | 52 | 52 | 52 | 52 | 52 |
| Rutaceae s.l. (+Cneorum) | 53.6 (46.8–62.9) | 53.9 (47.0–63.3) | 55.1 (48.0–64.7) | 55.6 (48.4–65.3) | 56.8 (49.6–66.6) | 57.5 (50.0–67.6) |
| Rutaceae s.s. (–Cneorum) | 39.6 (33.1–46.8) | 39.9 (32.8–46.8) | 40.9 (34.5–48.1) | 41.4 (34.4–48.2) | 42.5 (35.9–49.5) | 43.2 (36.2–49.8) |
| Aurantioideae (Citrus-Clausena) | 17.5 (12.7–22.2) | 17.7 (12.8–22.3) | 18.2 (13.2–23.2) | 18.4 (13.4–23.5) | 19.0 (13.7–24.3) | 19.3 (14.0–24.7) |
| Citrus-Pleiospermium | 14.6 (10.6–18.7) | 14.7 (10.7–19.1) | 15.2 (11.1–19.5) | 15.4 (11.2–20.0) | 15.8 (11.6–20.5) | 16.1 (11.8–21.0) |
| Citrus-Poncirus | 5.0 (2.3–9.0) | 5.1 (2.3–9.1) | 5.3 (2.4–9.3) | 5.3 (2.4–9.4) | 5.5 (2.6–9.7) | 5.6 (2.6–9.9) |
However, a much larger difference was seen in parts of the tree where the fossil minimum ages were exceeded during the analyses. The Fabaceae crown mean ages ranged from 75.3–75.6 Ma in PL, 79.2–80.6 Ma in the LF (strict clock) and 57.2–57.3 Ma in LF (local clock) analyses. In all cases, the 51 Ma minimum age for the Fabaceae crown was exceeded, so does not appear to be influencing rates in these cases (although it may be influencing the lower boundary of the confidence interval in the local clock analysis).
The local clock method, which allows rate shifts in specified parts of the tree, may be overcompensating some real rate shifts. The estimate of the Fabaceae crown age is somewhat younger than that from the strict clock and suggests that the longer branch lengths have been partly reduced when converting molecular change to time. However, the few rate classes applied here (3) allow only large rate steps, which may be the cause of the much younger local clock estimate for the family crown age.
In regard to the Rutaceae s.l. age, the local clock method, however, does not seem to have as large an effect. A local clock rate class was applied to the Ailanthus stem and coincides (in the tree) with the fossil constraint. The constraint was not exceeded in any analysis so is exercising an influence on dates on surrounding nodes. The combined effect may be that the constraint is compensated by the rate shift, leaving surrounding nodes to be dated by a combination of: the nodes below the Ailanthus stem constraint (e.g., Simaroubaceae+Meliaceae+Rutaceae s.l.) being older than 52 Ma; nodes above this but not part of the constraint (e.g., Rutaceae s.l.) being dated by the effect of the constraint and the action of average rates across the tree (clock and local method) or by a more nuanced series of rate changes in PL. In the PL case, autocorrelation will have the effect of producing slower rates in nodes surrounding the constrained Ailanthus stem node, thereby making the Rutaceae s.l. node somewhat older. A greater consistency among methods for nodes within Rutaceae may be explained by the autocorrelation having increased these slow rates to become closer to average rates further up the tree until rates and dates are more in line with the LF dates.
Aurantioideae
The mean, lower, and upper age estimates from PL for the subfamily were used as secondary calibration points to date the more detailed Aurantioideae phylogeny of R. J. Bayer et al. (unpublished data). The reason for using these constraints is given in the next section. The ages found are reported in Table 3 and on Fig. 1.
| Estimated age (Ma) | |||
|---|---|---|---|
| Node | Calibration fixed at lower CI | Calibration fixed at mean | Calibration fixed at upper CI |
| Citrus-Clausena (fixed calibration point at root) | 12.1 | 19.8 | 28.2 |
| Merrillia-Murraya | 5.5 (4.6–6.4) | 9.2 | 13.4 (11.2–15.6) |
| Glycosmis-Micromelum | 11.0 (10.2–11.9) | 18.0 | 25.8 (24.5–27.1) |
| Citrus-Murraya | 9.6 (8.6–10.6) | 15.9 | 23.1 (20.9–25.3) |
| Citrus-Pleiospermium | 7.8 (7.0–8.6) | 12.9 | 18.9 (17.1–20.7) |
| Citrus crown (C. indica-C. sinensis) | 4.3 (3.7–4.9) | 7.1 | 10.3 (8.8–11.8) |
| Citrus-Poncirus | 4.0 (3.2–4.8) | 6.6 | 9.6 (7.7–11.6) |
| Australasian Citrus crown (C. gracilis-Clymenia) | 3.5 (2.9–4.1) | 5.8 | 8.4 (7.1–9.7) |
DISCUSSION
Root age
The choice and phylogenetic placement of fossil constraints are of fundamental importance when determining molecular dates (11; 14). Our analysis of node ages used chloroplast sequences and constraints from three fossils. We found that the fossil ages used as internal constraints suggested that the root (MRCA of Saxifragales+other rosids) was older than the fixed age constraint initially suggested by minimum node age mapping (11). If rate shifts among branches are assumed to vary only within c. one order of magnitude, then the root node may be more reasonably inferred to lie closer to 115 Ma than to the minimum mapped node age of 95 Ma. However, this age might still be an underestimate because of the conservative application of fossil constraints (25; 14).
A root node age of c. 115 Ma appears consistent with other molecular studies and the fossil calibrations that underpin them. In the study by 34 using 61 chloroplast genes, the age of the rosid node (though Saxifragales, which is sometimes placed as sister to the other rosids, was not sampled) was found to be around 108 Ma, with the closest constrained node in that study being the eudicots, fixed at 125 Ma to coincide with the emergence of tricolpate eudicot pollen (10). 50 explored several phylogenetic constraints and their effect on ages of angiosperm clades. Although sampling was limited, with 18S data and several different constraints (“anthophyte,” “gnetifer,” and “gnepine”), the Fagales-Fabales clade dated at c. 100 Ma and the core eudicots a little over that (50). 27 found that the Saxifragales crown was at least c. 112–120 Ma, the younger of these estimates being consistent with our findings.
Molecular date estimates that are older than estimates from fossils are expected if we assume that many fossils do not represent the earliest occurrence of the taxon bearing the synapomorphies (14), but younger estimates cannot be readily reconciled. We found that the minimum age of the Ailanthus stem was estimated to be much less than the age of the fossil when only the root node was constrained (range of estimates = 19.4–32.5 Ma from strategy 1, vs. fossil constraint = 52 Ma). Either the fossil placement in the phylogeny is wrong, the fossil age is wrong, the phylogenetic estimate (topology and/or branch lengths) is wrong, or there has been a large change in molecular rates around the fossil calibration point. It is also possible that the Ailanthus fossil is correctly placed but that all other fossils are located too deep in the tree.
The fossil placement of Ailanthus fruit seems straightforward because the fossils are nearly identical to fruit of extant taxa and clearly different from those of other extant Simaroubaceae genera (9). It is possible that our conservative placement of the constraint has underestimated ages around this node rather than overestimated them as the unconstrained analysis would imply.
The fossil age is unlikely to be greatly incorrect because fossils nearly as old (early to mid Eocene) as the oldest occurrence are found in three continents; many fossils slightly younger again have been found in many different locations (9). A noteworthy observation is that the uncertainty regarding the fossil age (the horizon containing the Ailanthus fossil is 52.2–52.7 Ma) is dwarfed by the increase in the node age estimate when the fossil constraint is applied—the former is less than 2% of the latter (using the mean node age).
Our phylogenetic estimate does not include all Simaroubaceae genera, but is consistent with other estimates including the same taxa (18; 20; 6). We used over 2.5 kb of sequence data in our analysis and estimated branch lengths using ML (Appendix S2 with online Supplemental Data) and Bayesian (Fig. 2) methods were in close agreement. We explored inferred local rates with different root ages and found that rates are unremarkable with a root age of 115 Ma, even though the Ailanthus fossil constraint doubles the age of the family crown. In short, there seems no reason not to accept the results, i.e., that this fossil constraint is a critical piece of evidence that implies some degree of molecular rate shifting has occurred in this part of the phylogeny. Therefore, we provisionally accept 115 Ma as the best estimate of the root age, and we proceed to estimate divergence times within Rutaceae s.l. using PL and applying the internal constraints described.
The age of Rutaceae
Our findings suggest a minimum age of Rutaceae s.l. (i.e., including Cneorum) from the early Eocene (53.3 Ma) to perhaps late Cretaceous (72.7 Ma) with the mean estimate in the early Paleocene (62.7 Ma) (see Table 1, last column, for these and the following results). The minimum crown age of Rutaceae s.s. is between late Eocene (36.4 Ma) and late Paleocene (56.8 Ma), with the mean estimate in the Eocene (47.6 Ma). Subfamily Aurantioideae dates from at least the mid Miocene (12.1 Ma) to the mid Oligocene (28.2 Ma), the mean estimate lying in the early Miocene (19.8 Ma).
These findings are consistent with some other fossil reports, but not all. Oligocene (c. 23–34 Ma) or younger seeds sufficiently differentiated to closely resemble extant genera Euodia, Phellodendron, and Zanthoxylum (non-Aurantioideae) have been reported from northeastern America (58, 60). These occurrences are younger than and therefore consistent with the minimum crown age found here for the family. The crown of the subfamily Aurantioideae is more recently diverged than those of some other subfamilies, so the probable early Miocene minimum age (mean age 19.8 Ma) suggests stem lineage fossils bearing characters of the other subfamilies may well have been present by this time, which is therefore consistent with the fossils mentioned earlier.
Other Rutaceae fossils of Eocene age have been reported from several continents, including one attributed to extant genus Zanthoxylum (22). However, older fossils assigned to Rutaceae (excluding assignments considered doubtful by the cited authors) go back to the Paleocene and possibly the late Cretaceous, e.g., Rutaspermum biornatum from the Maastrichtian, c. 65–70 Ma (30; 22), which is clearly older than the crown age of the family found here. Rutaspermum is a taxon known only from fossil seeds that disappeared by the late Oligocene—early Miocene (22), and so might represent a stem lineage exemplar. Several characters discussed by 22 are shared by seed of two genera also known from more recent fossils—Toddalia (subfamily Toddalioideae sensu Swingle) and Zanthoxylum (subfamily Rutoideae sensu Swingle): surface ornamentation with pits (probably due to the mesotesta structure), inner tegmen with criss-crossed spiral lignifications, having a flat face (probably due to compression during development). However, in other respects Rutaspermum appears to share two features with Zanthoxylum but not Toddalia: boat-shaped seed and testa with grainy inclusions (of unknown composition) (22), although the former difference may not be especially reliable, as the illustrations provided by Gregor show. Clearly, there are too few characters to reliably place Rutaspermum, but a stem lineage placement or placement closer to Zanthoxylum is not unreasonable.
Although the subfamilial placement using Swingle's system implies that Rutaspermum, sharing features with Toddalia and Zanthoxylum, might be near the crown age of the family, recent molecular phylogenetic inference and chemical profiles suggest instead that the latter two genera are quite closely related and their lineages may not represent the deepest coalescence in the family (44). An additional caveat needs to be made regarding Gregor's (1989) observations, which is that the fossil material placed in Rutaspermum and directly cited in that paper (R. chandlerae Collison & Gregor) is all from the middle Eocene (perhaps around 40 Ma), within the crown age found here. The description and pictures of the sole Rutaspermum fossil known from the Cretaceous (R. biornatum Knobloch & Mai) do not mention nor show anatomical features (30) and therefore lack information regarding inclusions within the testa found in Zanthoxylum. That fossil is approximately boat-shaped and has a thick testa and a flattened face (30), but further similarities to Zanthoxylum or Toddalia are hard to ascertain.
Another genus with roughly boat-shaped seed with thick testa is Acronychia (22), which is less closely related to Zanthoxylum than Toddalia is (44). Acronychia also has occasionally anastomosing longitudinal ribs, as does R. biornatum from the Cretaceous. But R. chandlerae and several Zanthoxylum species do not—the former appears to lack ribs, whereas the latter has either fully anastomosing raised ridges on the sclerotesta or short nonanastomosing raised ridges of the sclerotesta (fossils features from figures in 30 and 22; Zanthoxylum features [B.E. Pfeil, personal observation] of CANB herbarium material). Seed from the sister group to Rutaceae s.s. appears to be different, excluding a placement of R. biornatum in that lineage. Dictyoloma and Ptaeroxylon have winged seeds, while Spathelia fruit is indehiscent and winged (5) (http://nt.ars-grin.gov/sbmlweb/OnlineResources/frsdfam/Index.cfm). Cneorum is also indehiscent and has U-shaped seeds not otherwise found in Rutaceae s.l. (http://nt.ars-grin.gov/sbmlweb/OnlineResources/frsdfam/Index.cfm). Harrisonia brownii (B. E. Pfeil, personal observation) and H. abyssinica (55) seed are both more or less smooth and lacking surface ribs.
On balance, the placement of Rutaspermum biornatum from the Cretaceous is unclear and a family stem placement cannot be excluded. The age of this fossil is consistent with this interpretation, being certainly older than the crown minimum age estimate (36.4–56.8 Ma) but not necessarily older than the stem node estimate (53.3–72.7 Ma). This interpretation is also consistent with Gregor's coarse phylogenetic scheme for the family based on fossil seed (Fig. 6 in 22).
The crown age of Rutaceae found here conflicts with an older estimate reported by 39, who used NPRS and Bayesian (Multidivtime) methods. Part of the discrepancy, easily reconciled, is due to the narrower family concept used here (Rutaceae excluding Cneoraceae) vs. the concept used by Muellner et al. (Rutaceae including Cneoraceae). Estimates found here for the more inclusive node age (Rutaceae s.l.) range from 53.3 to 72.7 Ma (mean = 62.7). While Muellner et al. NPRS-based estimates are c. 10–30 Ma older than these, this analysis method may overestimate ages below constrained nodes (B. E. Pfeil, personal observation) and may be unreliable in this case because no node below Rutaceae was fixed in that analysis (39). 39 Bayesian age estimates, however, found Rutaceae s.l. to be 56.5–86.0 (mean 72.9 Ma), which overlaps with our range of estimates, although the means differ by c. 10 Ma.
The crown age of Citrus s.l., estimated here to be 3.7–11.8 (mean 7.1) Ma is broadly overlapping and therefore consistent with the Pliocene (1.8–5.3 Ma) leaf fossil described by 19. However, this fossil age is younger than and therefore also compatible with, deeper positions in the phylogeny of this fossil, such as at the MRCA of Citrus, Citropsis and Pleiospermium—these genera having diverged from Citrus around 7.0–20.7 (mean 12.9) Ma.
The age of subfamily Aurantioideae
A similar discrepancy between our estimates and those of 39 exists regarding the age of subfamily Aurantioideae. Although not reported in their table, 39 Fig. 4 indicates this clade to be c. 30 Ma, whereas we estimate an age range of 12.1–28.2 Ma (mean = 19.8). In general, 39 Bayesian estimate is based on greater taxon sampling outside the subfamily and more internal calibration points (8 vs. 3). However, neither set of internal calibrations is any closer to subfamily Aurantioideae that the Ailanthus stem constraint, identically positioned in both studies.
Our study used both rbcL and atpB for phylogenetic reconstruction and date estimation in the first analysis, whereas 39 used only rbcL for these purposes. Within the subfamily in the first analysis, we used five taxa to their four, but the relationships found in 39 disagree with those from our second analysis, which used more than 8 kb of sequence, with all but one genus sampled (see also Bayer et al., unpublished data). In 39 Fig. 4 relationships are (((Clausena, Citrus) Aegle) Glycosmis)—support not indicated—whereas in the expanded sample, the relationships recovered were ((Citrus, Aegle),(Glycosmis, Clausena)), with high (≥80% bootstrap and ≥0.95 PP) support across models for the Aegle + Citrus clade and high PP for the Clausena + Glycosmis clade, although poor bootstrap support (Fig. 2 and R. J. Bayer et al., unpublished data). The taxa we used here for estimating ages in the first analysis were slightly different (Fig. 2), with relationships (Clausena,(Pleiospermium,(Atalantia,(Citrus, Poncirus)))), which are in agreement with the topology found in R. J. Bayer et al. (unpublished data) and Fig. 1.
It is likely that the phylogeny of the chloroplasts in these taxa is inferred more accurately by R. J. Bayer et al. (unpublished data) than by 39. The incongruent position of Clausena between these trees affects the age estimation, because the Citrus-Clausena clade traverses the deepest node known in the subfamily Aurantioideae, therefore representing the best place to estimate the crown age of the subfamily. However, the likely error in the topology presented in 39 must mean that some characters are optimized incorrectly, with a likely effect on the reliability of the age estimation within the subfamily. Inferred ages differ when topologies differ, e.g., because of different constraints (50) or incongruence (42; 46; 43); topological differences account for at least some of the differences between our results for the age of the subfamily and those of 39. Also, the different methods and underlying assumptions in the two studies could also be contributing to the differing results.
Biogeography
Australia-New Caledonia
While many Rutaceae are distributed widely across the tropics, only three groups of subfamily Aurantioideae are found in New Caledonia. The widespread species Micromelum minutum and Murraya paniculata are found there, as is Murraya crenulata and the endemic group of five species of Oxanthera. Micromelum minutum, Murraya paniculata, and two Oxanthera species have been sampled previously (R. J. Bayer et al., unpublished data), although only Oxanthera specimens were collected from New Caledonia itself. These taxa are clearly supported as part of the subfamily (R. J. Bayer et al., unpublished data), and all are considered indigenous to New Caledonia (37).
Whichever estimate of the age of subfamily Aurantioideae is accepted (ours c. 20 Ma or Muellner's c. 30 Ma), both estimates suggest that all the above lineages must have arrived in New Caledonia by long-distance dispersal and not by rafting on New Caledonia as it began to break away from the remainder of East Gondwana in the late Cretaceous (36), nor by stepping stone dispersal across the Kenn Plateau, which did not exist after 35 Ma (31). Even the upper confidence limit of our age estimate (28.2 Ma) rules out the latter hypotheses. The samples of Oxanthera from New Caledonia are nested within a clade of other Australasian Citrus (Fig. 1 and R. J. Bayer et al., unpublished data) that we estimate to be 2.9–9.7 (mean 5.8) Ma, or late Miocene to mid Pliocene (Table 3). These results are consistent with geological evidence suggesting that New Caledonia did not have suitable habitat for terrestrial flora between the late Cretaceous and the end of the Eocene.
Furthermore, we have estimated that the divergence between Murraya and its nearest relative occurred around the early Miocene (c. 9 Ma), whereas that between Micromelum and its nearest relative was around the early Miocene (c. 18 Ma). With further sampling within the latter two genera, the time of divergence between New Caledonian taxa and their closest relatives could be found to be much younger. Swingle and Reece's (1967) hypothesis regarding the evolution of Murraya paniculata speculated that this “species is very widely distributed and doubtless extremely old. However, it has evolved at a very slow rate, as is shown by the small morphologic differences that distinguish it from var. ovatifoliolata (noted below), which has been separated from the other forms of the species for many millions of years ever since Australia lost all land connections in the Eocene period” (56). Our findings suggest a simpler explanation for the limited differences found between the varieties—that they have not been separated for very long at all, certainly much more recently than the Eocene.
Australia-Asia
Australia and New Guinea have not been directly connected by land to Asia since the breakup of Pangea, more than 130 Ma (36). Therefore, ancestors of angiosperm taxa shared across Wallace's Line must have either rafted northwards on terranes (23) or dispersed across a narrowing oceanic gap from c. 20 Ma, as Australasia drifted north and the Australian craton margin contacted the southeast Asian craton margin (36; 23). Despite the latter possibility, some authors have suggested ancient isolation between genera of Aurantioideae in these land masses. For example, with regard to Microcitrus and Eremocitrus, Australasian genera previously considered closely related to but distinct from Asian Citrus, it was hypothesized that “[t]hese remarkable citrus fruits are extremely interesting, in that they show how evolution has proceeded in regions isolated as Australia and New Guinea have been during the last twenty or thirty million years since they were cut off from all other land masses” (56). However, we have dated the crown of Citrus, including Australasian-distributed plants formerly placed in Clymenia, Eremocitrus, Microcitrus, and Oxanthera, at a maximum of 11.8 Ma (mean 7.1 Ma). This age strongly implicates transoceanic dispersal as responsible for the Australia–Asia disjunction in the clade.
56, with their narrower circumscription of Citrus limited to Asian species, considered that land mass to be the origin of the genus. However, 4 proposed an alternative hypothesis and suggested that Citrus may have migrated from Australasia to Asia. They accepted Swingle and Reece's (1967) hypothesized age of Citrus and their assertion that the Australasian genera might be “primitive,” and then constructed a hypothesis of endozoochorous dispersal aided by the movement of terranes, especially that which formed the Halmahera Islands and the eastern Philippines (4), which shifted from near the northeastern edge of what was to become New Guinea to their current positions (23). However, our estimated age of the Australasian clade (mean, 5.8 Ma; upper credibility interval 9.7 Ma) makes speculation about terrane movement between the Eocene (37 Ma) and the present largely immaterial because the group appears to be much younger than these events. The simplest explanation is that the Australasian clade, nested within a group of Asian and southeast Asian lineages (both within Citrus s.l. and within the clade that includes all of the Citreae genera) with two African exceptions (see Introduction), has almost certainly dispersed from west to east no earlier than the late Miocene.
Similarly, Hartley's (2001) hypothesis that the distributions of Euodia and Melicope on different Gondwanan fragments, suggesting that these genera date to upper Cretaceous, is also rejected by our findings. Although all fossil calibrations in principle imply only minimum ages, a lack of older fossils that we can confidently use to calibrate phylogenies in current knowledge also sets a boundary on the maximum ages of lineages for which we have evidence. We have found no evidence that the crown of Rutaceae s.s. is as old as the Cretaceous, but instead it dates only to the late Paleocene. This finding means that no genus within the family is consistent with this hypothesis.
Conclusion
Our molecular dating analysis of the Rutaceae has rejected some published biogeographic hypotheses by finding estimated divergence times that are younger than those predicted. These hypotheses are (1) that Eudoia and Melicope diverged in the late Cretaceous, before the break-up of eastern Gondwana (24); (2) that transoceanic disjunctions in Murraya paniculata are ancient (pre-Eocene) and vicariant (56); (3) that Citrus s.l. in Australia-New Guinea is vicariant in origin and has been isolated for 20–30 Ma (56); and (4) that Citrus migrated into SE Asia from Australasia-New Guinea via drifting terranes (4). We have also found that taxa occurring in New Caledonia, including Murraya, Micromelum, and the endemic genus Oxanthera, are not vicariant in origin, but instead result from at least three separate transoceanic dispersal events. Geological evidence suggests that New Caledonia was not habitable until c. 35 Ma, and consistent with this, none of these dispersal events is estimated to have occurred before 28 Ma, and all probably occurred after 20 Ma. Accepting that all fossil-calibrated dates are minimum estimates (25; 14), we deliberately erred toward making older estimates by calibrating the root of the tree using the oldest feasible date for the root (Saxifragales+other rosids: 115 Ma), and by using the maximum (oldest) age estimate for hypothesis testing. Even with this conservatism, we have comfortably ruled out the hypotheses described earlier, given their predicted timing. Rutaceae subfamily Aurantioideae have radiated and dispersed multiple times across transoceanic gaps between Australia, southeast Asia and New Caledonia probably no earlier than the late Oligocene.
