• age constraints;
  • BEAST;
  • fossil calibration;
  • Magnoliales;
  • molecular dating;
  • phylogeny;
  • r8s;
  • relaxed clock


  1. Top of page
  2. Abstract
  9. Supporting Information

This article addresses the challenges involved in estimating the ages of clades using fossils and DNA sequences. We review the principles and problems of placing fossils in trees of extant taxa and using them to constrain the ages of nodes in molecular dating analyses. Endressinia and Futabanthus provide minimum ages of 112 Mya for the stem lineage and 89 Mya for the crown group of Annonaceae, and the diversity of endosperm ruminations in seeds from the London Clay indicates that the four main clades of Annonaceae had diverged by 50 Mya. Ages inferred using these minimum constraints and a plastid phylogenetic tree for Annonaceae, particularly crown ages of the two main clades (Malmeoideae and Annonoideae), depend on assumptions regarding the pattern of variation in rates of molecular evolution. Our results using methods that assume rate autocorrelation or log-normal distribution of rates suggest that neither assumption fits well the apparently abrupt changes in rates in Annonaceae. Instead of soft-bounded age constraints, we argue for the use of only well-substantiated fossil evidence by means of priors with hard bounds. Thus, we can infer ages that take into account both palaeontological and phylogenetic uncertainty, without confounding the different factors involved. © 2012 The Linnean Society of London, Botanical Journal of the Linnean Society, 2012, 169, 84–116.


  1. Top of page
  2. Abstract
  9. Supporting Information

Phylogenetic methods that simply show the topology of branching of evolutionary lines within groups may be sufficient for many biological purposes, such as the reconstruction of ancestral character states with parsimony or the creation of phylogenetic classifications, but, for other purposes, particularly the reconstruction of biogeographical history or the investigation of the causal relations between evolutionary and geological events, it is necessary or desirable to estimate the actual ages of clades, in terms of the relative or absolute geological timescale. Until recently, the dating of clades was almost exclusively the province of palaeontology, but, with the advent of molecular biology and the concept of a molecular clock (i.e. that DNA sequences diverge at a roughly constant rate through time: Margoliash, 1963; Zuckerkandl & Pauling, 1965), various methods have been proposed and used extensively to date phylogenetic splits based on differences in DNA sequences between groups. These molecular dating methods require some direct or indirect input from fossil or other geological data, but they offer the hope of dating clades that have no fossil record, or in which the fossil record presents problems of interpretation.

In this article, we review fossil and molecular evidence on the ages of clades in Annonaceae, based on both a critical review of the literature and our own original analyses of published molecular data. After presenting methods used in our molecular dating analyses and their results, we discuss first fossil evidence on the ages of clades in Annonaceae, followed by the bases and implications of our own and previous molecular analyses. In this context, we also review general methodological principles for integrating fossils into phylogenetic trees of living organisms (now largely based on molecular data) and the various assumptions, strengths and weaknesses of current molecular dating methods. In addition to providing an introduction to palaeobotanical and molecular dating for students of Annonaceae, this survey may represent a case study that is of more general interest to students of other groups.


  1. Top of page
  2. Abstract
  9. Supporting Information

Fossil and morphological data

Our data on fossils relevant to the dating of Annonaceae are from published sources, which are cited and evaluated in the Discussion. Data on endosperm ruminations, which are of central importance in the evaluation of the fossil seed record, are based on the dataset of Doyle & Le Thomas (1996), with the addition of 33 genera not included or lumped with other taxa in that study, and with ‘Polyalthialongifolia (Sonn.) Thwaites assigned to Enicosanthum Becc. and Ancana F.Muell. assigned to Meiogyne Miq., as discussed in Doyle & Le Thomas (2012). Doyle & Le Thomas (1996) checked most published seed descriptions of the taxa that they considered with original observations on seeds in the Paris herbarium. Scoring of the added genera is based primarily on Van Setten & Koek-Noorman (1992), Chatrou (1998) and Mols (2004). Most problems in scoring involve apparent transitions between spiniform and lamelliform ruminations in Miliuseae Hook.F. & Thomson (sensuChatrou et al., 2012; = the miliusoid clade sensuMols et al., 2004). The spiniform state of Doyle & Le Thomas (1996) includes cases that Van Setten & Koek-Noorman (1992) described as having flattened pegs, as did the spiniform state of Mols (2004). Taxa described as ranging from spiniform to lamellate or becoming lamellate towards the raphe are scored as either spiniform (2) or uncertain (2/3) based on data in Mols (2004); for example, Alphonsea Hook.f. & Thomson is scored as spiniform because the one lamelliform species, A. elliptica Hook.f. & Thomson, is nested within the genus in the phylogenetic analysis of Mols (2004). Treatment of the four genera formerly included in Malmea R.E.Fr. follows Chatrou (1998). Cyathocalyx Champ. ex Hook.f. & Thomson was described by Van Setten & Koek-Noorman (1992) as having flattened pegs, often becoming lamellate towards the raphe, but, according to Wang (2004) and Surveswaran et al. (2010), this taxon consists of two clades, one of which (Cyathocalyx s.s.) has spiniform and the other (Drepananthus Maingay ex. Hook.f.) irregular ruminations, like the related genus Cananga (DC.) Hook.f. & Thomson. In this study Cyathocalyx is treated in the restricted sense. Laurales, the presumed sister group of Magnoliales, which lacks ruminations, has been added as outgroup. The tegminal-chalazal ruminations of Myristicaceae might be treated as a fifth state, but, because this would be an autapomorphy and may or may not be ancestral in that family (Sauquet et al., 2003), we have instead scored Myristicaceae as unknown. MacClade (Maddison & Maddison, 2003) was used to reconstruct the most parsimonious course of evolution of rumination types on the tree found in the analyses of Chatrou et al. (2012), with Annonaceae pruned to the 73 taxa in the present dataset.

Molecular age estimates for Annonaceae in the literature often cannot be compared directly, as they have been used to address different questions and do not use comparable data (numbers of taxa or of DNA sequence characters; Table 1), molecular dating methods or combinations of fossil age constraints (Table 2). For example, Richardson et al. (2004) presented a biogeographical scenario for Annonaceae, thus including a large representative sample of Annonaceae, and used the nonparametric rate-smoothing method (NPRS; Sanderson, 1997). By contrast, the focal issue of Doyle et al. (2004) was the age of crown group Myristicaceae; Annonaceae were represented only sparsely, more in the role of an outgroup, and the Langley–Fitch (LF) and penalized likelihood (PL) methods (Sanderson, 2002a, b) were used. In order to draw direct comparisons between the results found with different methods, given these various factors, we obtained DNA sequence data and phylogenetic tree topologies from four studies: Doyle et al. (2004), Richardson et al. (2004) and Pirie et al. (2006), which included molecular dating analyses, and an earlier version of Chatrou et al. (2012). We performed new analyses as necessary in order to compare results obtained using LF, NPRS and PL with those obtained using BEAST following the procedures described below.

Table 1.  Comparison of taxon and character sampling in studies applying molecular dating techniques in Annonaceae
Doyle et al. (2004)30 taxa: five Annonaceae; 12 Myristicaceae; five other Magnoliales; four LauralesndhF; trnK (including matK); trnT-F: 6845 characters, 2445 variable
Richardson et al. (2004)208 taxa: 202 Annonaceae; five other Magnoliales; one LauralesrbcL; trnL-F: 2848 characters, c. 1110 variable
Pirie et al. (2006)96 taxa: 93 Annonaceae (mainly Malmeoideae); three other MagnolialesrbcL; trnL-F; psbA-trnH: 3696 characters, 1104 variable
This study: spine matrix61 taxa: 58 Annonaceae; two other Magnoliales; one LauralesrbcL, matK, ndhF, atpB-rbcL trnT-L-F, psbA-trnH, trnS-G: 8843 characters, 3777 variable
This study: supermatrix204 taxa: 197 Annonaceae; six other Magnoliales, one LauralesrbcL, matK, ndhF, atpB-rbcL trnT-L-F, psbA-trnH, trnS-G: 8843 characters, 6181 variable
Table 2.  Calibration points used in molecular dating studies of Annonaceae, with corresponding estimates, according to method, for the age of the most recent common ancestor (MRCA) of extant Annonaceae
StudyCalibrationAge of MRCA of Annonaceae (Mya)
  1. LF, Langley–Fitch (clock); Mya, million years ago; NPRS, nonparametric rate smoothing; PL, penalized likelihood.

Doyle et al. (2004)(a) Archaeanthus (Dilcher & Crane, 1984): MRCA Magnoliaceae and Annonaceae 100 Mya69 (LF); 57 (PL)
(b) Lethomasites (Ward et al., 1989): MRCA Magnoliaceae and Annonaceae 120 Mya82 (LF); 69 (PL)
Richardson et al. (2004)(a) Archaeanthus (Dilcher & Crane, 1984): MRCA Magnoliaceae and Annonaceae 98 Mya. Similar results (not reported) obtained using (b) secondary calibration (Wikström et al., 2001): MRCA of Annonaceae and Eupomatiaceae 82 Mya and (c) Maastrichtian seeds (Chesters, 1955): stem node of Annonoideae and Malmeoideae 68 Mya75.57 (NPRS: node age not reported in original publication, thus estimated here based on the same data matrix)
Pirie et al. (2006)Archaeanthus (Dilcher & Crane, 1984): MRCA Magnoliaceae and Annonaceae 98 Mya72.06 (NPRS); 65.60 (PL); 68.36 (multidivtime)
Su & Saunders (2009)(a) Archaeanthus (Dilcher & Crane, 1984): MRCA Magnoliaceae and Annonaceae 98 Mya and (b) Futabanthus (Takahashi et al., 2008): MRCA Annonaceae 89 Mya89.0–90.41 [BEAST 95% posterior probability (PP) interval]
Erkens et al. (2009)(a) Archaeanthus (Dilcher & Crane, 1984): MRCA Magnoliaceae and Annonaceae 98 Mya; (b) Maastrichtian seeds (Chesters, 1955): stem node of Annonoideae and Malmeoideae 68 Mya; (c) secondary calibration (Erkens et al., 2007): crown of Guatteria 11.4 Mya70.3–81.7 (BEAST 95% PP interval, all calibrations)
Couvreur et al. (2011)(a) Endressinia (Mohr & Bernardes-de-Oliveira, 2004): MRCA Magnoliaceae and Annonaceae 115 Mya; (b) Futabanthus (Takahashi et al., 2008): MRCA Annonaceae 89 Mya89.0–93.0 (BEAST 95% PP interval)

DNA sequence data

The different taxon and character sampling represented by these studies is summarized in Table 1. All the DNA sequence data are plastid encoded (and thus effectively represent a single gene tree). The dataset of Chatrou et al. (2012) comprises a densely taxon-sampled two-marker matrix (expanded from that of Richardson et al., 2004) combined with a more sparsely taxon-sampled ‘spine’ of multiple markers, i.e. it is a supermatrix, which includes a high proportion of missing data. We used an earlier version of this matrix in which a small number of taxa subsequently added by Chatrou et al. (2012) were not included. We analysed the densely and sparsely taxon-sampled matrices separately and in combination, and will henceforth refer to these datasets as ‘two-gene’, ‘spine’ and ‘supermatrix’, respectively. For each matrix, the best-fitting substitution model was estimated using ModelTest (Posada & Crandall, 1998) under the Akaike information criterion (Akaike, 1974). Maximum likelihood (ML) branch lengths for these topologies were then calculated, using the ML criterion as implemented in PAUP* (Swofford, 2002). The likelihood ratio test was performed comparing the likelihood of the data given the best-fitting model with and without the constraint of a strict molecular clock. The molecular clock was rejected (P ≤ 0.0001).

Rate smoothing

Rate smoothing under the assumption of a molecular clock was performed using the LF method in the program r8s (Sanderson, 2002b). Sanderson's NPRS (Sanderson, 1997) and PL (Sanderson, 2002a) methods were applied, also as implemented in r8s. Analyses were performed on the single trees with branch lengths derived from the original data. The optimum smoothing parameter for each tree was estimated by multiple rounds of cross-validation, starting with a wide range of values and successively reducing the range of, and increments (cvinc) between, tested values until an optimum was identified among a range of values at cvinc = 0.5. Where cross-validation failed for particular taxa (which was more frequent at low values for the smoothing parameter), these taxa were pruned from the trees and the test was re-run. Error margins (for PL analyses) were estimated by rate smoothing 100 topology-constrained trees with branch lengths derived from bootstrap resampled data (Wikström, Savolainen & Chase, 2001), and summarizing the results with mean values and standard deviations for specified nodes using the ‘profile’ command in r8s. These ranges can be interpreted as confidence limits on branch length estimates, reflecting stochasticity in the sampling of character changes.

Relaxed clock molecular dating

Bayesian relaxed clock phylogenetic analyses are considerably more computationally intensive than rate smoothing methods. We therefore applied BEAST v1.6.1 (Drummond & Rambaut, 2006), under a relaxed clock model that assumes a log-normal distribution of rates, to only one of the matrices analysed, choosing the supermatrix as it represents the largest numbers of both taxa and characters. One unconstrained analysis was performed, in order to test the ability of BEAST to identify the root of Annonaceae given the relaxed clock model (see Discussion). One further, topology-constrained analysis was performed in order to produce age estimates directly comparable with those derived under PL (i.e. with confidence intervals that do not represent topological uncertainty). For the latter, a starting tree with the same topology as used under PL was employed and all topology sampling operators were disabled. This had the further effects of rooting the tree, incorporating the additional evidence from indel characters and reducing the otherwise long convergence times associated with high proportions of missing data in supermatrices such as this (Wiens et al., 2005; Pirie et al., 2008). We compared the results with those of Couvreur et al. (2011), who included a smaller sample of taxa but a similar sequence data matrix, to assess the likely additional impact of topological uncertainty on the confidence intervals. The substitution model was set to GTR + I + G following the ModelTest results, with four categories for the gamma distribution, and a Yule demographic process (pure birth) was assumed. Age constraints were applied as described below. Three independent analyses were run for up to 100 million generations each.

Molecular clock calibration using fossils

From the survey of fossil evidence (see Discussion; absolute ages of units in the relative geological timescale based on Gradstein et al., 2004), we concluded that a relatively small number of fossils might be confidently used as calibration points within Magnoliales, either as fixed or minimum ages for the nodes specified: Archaeanthus Dilcher & P.R.Crane [most recent common ancestor (MRCA) of Magnoliaceae and Annonaceae; 100 Mya], Endressinia B.Mohr & Bernardes-de-Oliveira (also MRCA of Magnoliaceae and Annonaceae; 112 Mya), Futabanthus Ma.Takah., E.M.Friis, Uesugi, Yo.Suzuki & P.R.Crane (crown node of Annonaceae; 89 Mya), Maastrichtian seeds (also crown node of Annonaceae; 68 Mya) and Duguetia sp. (either crown or stem node of Duguetia A.St.-Hil. s.l.; 38 Mya). We compared the results of a number of different analyses using these fossil age constraints (singly and in combination), applying the following methods and datasets:

  • 1
    Using LF, NPRS and PL, with each dataset in turn, we applied Endressinia only as a fixed age. Where previous analyses had been performed using Archaeanthus, rather than Endressinia, to fix the age of the Annonaceae–Magnoliaceae node (variably at 98 or 100 Mya), we rescaled the chronograms instead of re-running the analyses.
  • 2
    Using just PL with the supermatrix, we applied each fossil in turn as a single fixed age calibration and applied three combinations of multiple fossil constraints: (1) Endressinia with Futabanthus, and Endressinia with Futabanthus and Duguetia sp., the latter used to constrain (2) the crown node and (3) the stem node of Duguetia. For the multiple constraints, the deepest node calibrated with the oldest fossil (i.e. Endressinia) was fixed, and the rest of the calibration points were applied as minimum age constraints.
  • 3
    Finally, we applied the combination of Endressinia and Futabanthus as fossil calibrations to the supermatrix using BEAST (the analysis employing topological constraint). For BEAST, we used similar prior distributions as Couvreur et al. (2011) for age constraints: fixing the age of the MRCA of Magnoliaceae and Annonaceae to the age of Endressinia (112 Mya) by means of a uniform prior with minimal bounds, and constraining the crown node of Annonaceae with the age of Futabanthus by means of a further uniform prior (instead of the exponential priors used by Couvreur et al., 2011 and Zhou et al., 2012) with bounds of 89 Mya (a minimum constraint equal to the fossil age) and 112 Mya (a maximum constraint corresponding to the oldest fossil constraining the deepest node in the tree).


  1. Top of page
  2. Abstract
  9. Supporting Information

Rate smoothing: r8s

A comparison of age estimates for deeper nodes in the phylogenetic tree for Annonaceae, derived using LF, NPRS and PL methods from the data of Doyle et al. (2004), Richardson et al. (2004) and Pirie et al. (2006) and with Endressinia used to fix the MRCA of Magnoliaceae and Annonaceae at 112 Mya, is presented in Figure 1. The smoothing parameter values estimated using cross-validation under PL were between 0.001 and 0.032 across the different datasets. Detailed results derived under PL using Endressinia with Futabanthus as fixed and minimum constraints, respectively, are presented in Figure 2 (the chronogram), Figure 3 (the ratogram) and Tables 3–5. As shown in the PL ratogram and Table 5 (but inferred under both NPRS and PL), rate estimates at deeper nodes in Annonaceae indicate sequentially decreasing rates in Malmeoideae Chatrou, Pirie, Erkens & Couvreur [e.g. given the two-marker dataset: 0.9341 × 10−3 substitutions site−1 Myr−1 at the MRCA of Malmeoideae and Annonoideae Raf. sensuChatrou et al. (2012) (Fig. 2, node K); 0.5912 × 10−3 at the crown node of Malmeoideae (node N); and 0.3807 × 10−3 at the MRCA of Malmeeae Chatrou & R.M.K.Saunders, Maasia Mols, P.J.A.Keßler & S.H.Rogstad, Monocarpia Miq. and Miliuseae (node P)], in contrast with more consistent, higher rates ancestral to Malmeoideae and in Annonoideae and the basal grade of Ambavioideae Chatrou, Pirie, Erkens & Couvreur and Anaxagorea A.St.-Hil. The single universal rates inferred under the LF method for the same datasets fall by contrast within the range of values inferred in Malmeoideae under PL (e.g. 0.3640 × 10−3 given the two-marker dataset).


Figure 1. Summary of molecular dating analyses performed using different methods [Langley–Fitch (LF) (clock); nonparametric rate smoothing (NPRS) and penalized likelihood (PL)], the same calibration point (the age of the fossil Endressinia, fixing the most recent common ancestor of Annonaceae and Magnoliaceae at 112 Mya) and the data presented in three recent studies of Annonaceae. LBC, long branch clade; SBC, short branch clade.

Download figure to PowerPoint


Figure 2. Age estimates for clades using the penalized likelihood (PL) method with all available character data and taxa (supermatrix) and age constraints based on Endressinia and Futabanthus (see text). The identity of terminal taxa is documented in Appendix S1. Precise ages for selected nodes, with corresponding standard deviations, are reported in Table 4.

Download figure to PowerPoint


Figure 3. Ratogram corresponding to the chronogram presented in Figure 2[penalized likelihood (PL) analysis of the supermatrix with age constraints based on Endressinia and Futabanthus, with terminal taxa documented in Appendix S1]. Branch lengths are proportional to evolutionary rate rather than time.

Download figure to PowerPoint

Table 3.  Age estimates for the most recent common ancestor (MRCA; i.e. crown node) of Annonoideae and Malmeoideae as inferred here using (a) the two-gene matrix and (b) the supermatrix, and in three recent studies of Annonaceae employing comparable age constraints (Su & Saunders, 2009; Couvreur et al., 2011; Zhou et al., 2012)
StudyAge of Annonoideae (Mya)Age of Malmeoideae (Mya)
  1. LF, Langley–Fitch (strict clock); Mya, million years ago; NPRS, nonparametric rate smoothing; PL, penalized likelihood; SD, standard deviation.

This study: two-gene matrix (LF; NPRS; PL)63; 58; 5638; 59; 57
This study: supermatrix (PL; SD)68 (4)77 (4)
This study: supermatrix [BEAST; 95% posterior probability (PP) interval]73 (80–65)42 (54–34)
Su & Saunders (BEAST; 95% PP interval)59.6 (70.5–48.1)39.8 (55.1–26.8)
Couvreur et al. (BEAST; 95% PP interval)65.9 (72.4–59.2)32.8 (40–25.8)
Zhou et al. (BEAST; 95% PP interval)68.0 (75.5–60.1)45.6 (59.3–33.0)
Table 4.  Age estimates (millions of years) for selected clades and genera (crown groups), as illustrated in Figure 2, using the supermatrix under penalized likelihood (PL) (assuming rate autocorrelation) and BEAST (assuming log-normal distributed rates) and age constraints based on Endressinia and Futabanthus (see text). Selected clades (bold type) are indentified with the corresponding letters in Figure 2. Point estimates with standard deviations (SDs) based on nonparametric bootstrapping are reported for PL; median values with 95% posterior probability (PP) intervals are reported for BEAST
Eupomatiaceae/Annonaceae (A)1051105109–100
Annonaceae (B)8939398–89
Ambavioideae/Annonoideae/Malmeoideae (C)7748390–76
Ambavioideae (D)5364056–27
Annonoideae/Malmeoideae (K)7547986–72
Annonoideae (J)6847380–65
Annonoideae without Bocageae (R)5956471–56
Annoneae/Monodoreae/Uvarieae (E)4845460–46
Annoneae (F)4444855–41
Bocageae (G)3854757–36
Duguetieae (H)5235063–37
Monodoreae/Uvarieae (I)4244553–37
Uvarieae (L)2933441–26
Xylopieae (M)5535565–44
Malmeoideae (N)6644254–34
Malmeeae/Maasia/Monocarpia/Miliuseae (P)5443137–25
Malmeeae (O)5232733–22
Miliuseae/Monocarpia clade (Q)4842733–22
Oxandra venezuelana/espintana clade3161216–9
Oxandra xylopioides clade187813–4
Polyalthia’ clade F11741014–6
Polyalthia’ clade F21681217–8
Polyalthia’ clade F396913–5
Polyalthia’ clade F4103715–11
Table 5.  Evolutionary rates for selected nodes (indicated in Fig. 2) inferred using the supermatrix under penalized likelihood (PL) (assuming rate autocorrelation) and BEAST (assuming log-normal distributed rates). Rates are in substitutions site–1 Myr–1 × 10–3. Point estimates are reported for PL; median values with 95% posterior probability (PP) intervals are reported for BEAST
NodePL rateBEAST rate
Root (B)1.71.8 (1.0–2.8)
Anaxagorea0.80.5 (0.4–0.7)
Malmeoideae/Annonoideae/Ambavioideae (C)1.61.4 (0.7–2.4)
Ambavioideae (D)0.80.5 (0.3–0.8)
Malmeoideae/Annonoideae (K)1.40.7 (0.2–1.6)
Annonoideae (J)1.41.3 (0.6–2.3)
Annonoideae without Bocageae (R)1.21.3 (0.6–2.3)
Malmeoideae (N)0.90.2 (0.1–0.4)
Malmeeae/Maasia/Monocarpia/Miliuseae (P)0.60.4 (0.1–1.0)
Malmeeae (O)0.40.3 (0.1–0.8)

Relaxed clock molecular dating: BEAST

Convergence and adequate sampling of model parameters were confirmed using Tracer 1.4 (Rambaut & Drummond, 2007). Three topology-unconstrained analyses reached a consistent likelihood plateau after c. 1 million generations and were allowed to continue for a further 99 million. When the post-burnin results were combined, the Effective Sample Sizes (ESS) for all parameters were > 100 (most were > 200). A summary topology is presented in Figure 4. Two of three topology-constrained analyses reached a consistent likelihood plateau after c. 2.5 and 5 million generations, respectively, and were allowed to continue for a further c. 35 million (until the combined post-burnin ESS for all parameters were > 100), after which the post-burnin tree sampled was thinned to leave 10 000 trees from which 95% posterior probability (PP) ranges for node ages were summarized. Detailed results for node ages are presented in Table 4 and Appendix S1 (see Supporting Information). Rates of molecular evolution at selected nodes are presented in Table 5.


Figure 4. Summary of the chronogram obtained from BEAST relaxed clock analysis (maximum clade credibility tree with median branch lengths, relative scale), assuming a log-normal distribution of rates, using the supermatrix without topological constraints. LBC, long branch clade; SBC, short branch clade.

Download figure to PowerPoint

Effects of different data sampling

The effects of different taxon and character sampling on the age estimates (point estimates and confidence intervals derived from bootstrapping) derived using the PL method are presented in Figure 5. Confidence intervals for age estimates given the same taxon sampling overlap, whereas reduced taxon sampling results in more recent ages. Increased character sampling is reflected in narrower confidence intervals, as is increased taxon sampling (given the same character sampling).


Figure 5. Summary of molecular dating analyses performed using the same method (penalized likelihood, PL) and calibration point (Endressinia), but differing numbers of taxa and DNA sequence characters: A, 204 taxa for the markers trnL-F and rbcL only (two-gene matrix); B, 63 taxa represented by rbcL, matK, ndhF, atpB-rbcL, trnT-L-F, psbA-trnH, trnS-G (spine matrix); C, a supermatrix combining both (A) and (B) with sequence data that were unavailable for other than the 63 taxa of (B) coded as missing. Error margins derived from nonparametric bootstrapping are represented as grey bars: dark grey for standard deviations (SDs) and lighter grey for 95% confidence intervals (CI; derived from the SDs). LBC, long branch clade; SBC, short branch clade.

Download figure to PowerPoint

Impact of different fossil constraints

The ages of fossils and the nodes to which they can be related are illustrated in Figure 6 (superimposed on the PL tree calibrated with Endressinia only). Given the different single calibrations (using the PL method with the supermatrix), estimates for the crown node of Annonaceae were c. 68 Mya (Maastrichtian seeds), 77 Mya (Archaeanthus), 86 Mya (Endressinia) and 89 Mya (Futabanthus). When all other fossils are used as constraints, the age of Duguetia sp. substantially predates the estimate for the crown node of Duguetia s.l. (c. 38 Mya as opposed to 20 Mya), but is only slightly older than the estimated age for its stem lineage [i.e. the common ancestor with Fusaea (Baill.) Saff.; 37 Mya]. When Duguetia sp. is used to constrain the crown node of Duguetia s.l., the influence on age estimates for surrounding nodes is considerable, but this impact decreases with increasing distance in the tree. The age of the Duguetia stem node is estimated at 48 Mya, as opposed to 37 Mya (a difference of 11 Myr), but the MRCA of Duguetia and Letestudoxa Pellegr. differs only by 5 Myr (57 Mya as opposed to 52 Mya) and the MRCA of Duguetia and Cymbopetalum Benth. by only 3 Myr (62 Mya as opposed to 59 Mya). When the fossil is used instead to constrain the age of the Duguetia stem node, there is a much smaller influence (38 Mya as opposed to 37 Mya) and little or no impact on the age estimates for deeper nodes.


Figure 6. Summary of the chronogram obtained from penalized likelihood (PL) analysis using the supermatrix, with Endressinia only as a fixed calibration point (the oldest constraint for the deepest node). The ages of other fossils relative to their associated node ages are indicated, including two possible node associations of Duguetia sp.: the crown and stem nodes of Duguetia s.l. (including ‘Pachypodanthium’). LBC, long branch clade; SBC, short branch clade.

Download figure to PowerPoint


  1. Top of page
  2. Abstract
  9. Supporting Information

Fossils and the ages of clades

Before the advent of molecular dating, our knowledge of the age of groups of organisms was derived almost exclusively from the fossil record. Except in pathogens and other organisms evolving on a human timescale, fossils still provide the only direct evidence for the age of the groups to which they belong. Molecular dating techniques can be used to infer the relative lengths of branches (or relative ‘heights’ of nodes) in a phylogenetic tree, but these branch lengths can only be translated into absolute ages by using external evidence, in particular from the fossil record, to fix (‘calibrate’ in the strict sense of many authors) or limit (‘constrain’) the age of one or more nodes. However, the use of fossils for the dating of groups is not as simple as it might seem at first sight. First, the fossil should ideally be placed in the phylogenetic tree of living plants based on an appropriate phylogenetic analysis of the available morphological characters (Doyle & Donoghue, 1993; Crepet, Nixon & Gandolfo, 2004; Doyle & Endress, 2010), a step ignored or glossed over in many studies. In some cases in which a fossil has an especially distinctive synapomorphy of a living group and other characters are consistent, this placement may be performed ‘by eye’, but, in others, a formal analysis using a morphological dataset of living and fossil taxa may be necessary. Second, if this information is to be used in molecular analyses for the dating of other nodes in the tree, it must be used correctly to constrain the ages of the appropriate nodes, and to define the type of constraint in a manner that is consistent with the nature of the uncertainty that is inherent in palaeontological data.

Principles for relating fossils to nodes in a phylogeny

In associating fossils with clades in the tree of living plants, it should be recognized that any living clade can be thought of as having at least two ages (Fig. 7; cf. Doyle & Donoghue, 1993; Doyle et al., 2004). One is the age of the MRCA of all its living members, i.e. the crown group age. Because DNA data are available only for living organisms (except for a few recent fossils), it is evident that all clades seen in molecular phylogenetic trees are crown groups. Another age is the time at which the stem lineage leading to the crown clade separated from the lineage leading to the most closely related living clade, or sister group (the stem group age sensuMagallón & Sanderson, 2001). In addition, there is the time at which a particular synapomorphy of the living clade arose, which may have occurred at any point along the stem lineage. In terms of phylogenetic nomenclature, these ages correspond to potential node-based, total (or stem-based) and apomorphy-based taxa (Cantino et al., 2007).


Figure 7. Placement of fossils (X, Y, Z) in the phylogeny of living organisms in relation to possession of apomorphies (A, B, C, D, E). See text for discussion.

Download figure to PowerPoint

Principles for placing fossils in a tree of living plants can be explained graphically with reference to Figure 7. A fossil can be recognized as a stem relative (a direct ancestor, on the stem lineage itself, or much more likely a side branch from the stem lineage) if it has the derived state (a synapomorphy) of the crown group in at least one of its characters, but a more ancestral (outgroup) state in another (X in Fig. 7, with derived state A but not B). Recognition that a fossil belongs to the crown group may be more difficult than is often recognized. Traditionally, palaeobotanists have assigned fossils that resemble a living group in all their preserved characters to that group, and this may be taken by neobotanists as evidence for existence of the crown group. However, it is also possible that such a fossil is a stem relative attached near the end of the stem lineage, after all the apomorphies in its preserved characters had evolved (such as Z in Fig. 7, with both A and B). Such a plant would not be a member of the crown group, and it might have been both older than the crown group and more primitive than its living members in characters that were not preserved (such as D and E). A fossil can therefore be definitively assigned to the crown group only if it has a synapomorphy of some subgroup of the crown group, such as state C in fossil Y in Figure 7.

In considering the use of fossils to date molecular trees, it is generally acknowledged that the age of a fossil will always, to a greater or lesser extent, be an underestimate of the age of the lineage to which it can be attached (Benton & Ayala, 2003; Reisz & Muller, 2004; Heads, 2005) or, more precisely, of the subtending node in the tree. Although the discrepancy is often attributed to the fragmentary nature of the fossil record, the degree of age underestimation is dependent on multiple more or less interrelated factors, some of which relate to the necessity of identifying a synapomorphy that is derived within a clade to determine the clade membership. First, there is the time lag between the MRCA of a crown group and the origin of potentially informative morphological differences within it, such as the time between the crown group age and the origin of apomorphy C in Figure 7, or between the stem age and the origin of apomorphy A. Second, there is incomplete preservation, an additional difficulty involved in using morphology to determine where a fossil, as opposed to a living plant, belongs in a phylogenetic tree. Most soft parts of organisms and even entire lineages of soft-bodied organisms are not observed in the fossil record at all. In plants, pollen and leaves are more commonly preserved than flowers, fruits and seeds, but pollen and leaves offer relatively few characters with which they can be placed with precision. This problem is illustrated by the uncertainty on the position of a fossil that has apomorphies A and B, but in which the parts bearing apomorphies D and E are not preserved: whether it represents a stem relative such as Z in which D and E had not evolved, a later stem relative in which D and E had evolved or a member of the crown group. Third, there is the problem of sporadic preservation. This dictates that even if a fossil taxon has an apomorphy that unambiguously relates it to a particular lineage, the chance that the oldest known specimen of this taxon was preserved immediately after the origin of this apomorphy is a fraction of the already small likelihood of its preservation at all. This likelihood will vary greatly between taxa and as a function of abundance and ecology. For example, if the oldest known stem relative of a crown clade belongs to a side line that persisted until after origin of the crown group, it may be younger than the crown node, like X in Figure 7. Because of these factors, a fossil stem relative such as X or Z only gives a minimum age for the node at which the stem lineage of the crown clade separated from its living sister group, and a fossil such as Y that shares an apomorphy with a living subgroup of the crown group provides only a minimum age for the crown node.

Can fossils be used to define maximum age constraints?

This situation can create a dilemma, because many (if not all) molecular dating methods require at least one fixed calibration point or maximum age in the tree, preferably below the group under study. However, the absence of older fossils cannot be interpreted directly as the nonexistence of the lineage at that time. The proposal of fixed or maximum ages based on fossils therefore requires additional assumptions about the sum of the effects of the different factors described above. Thus, if the fossil record of a group is dense, such that it is almost always represented in fossil assemblages, and the whole crown group has a distinctive synapomorphy that is readily preserved and recognized in the fossil state, it may be suggested that the crown group is unlikely to be much older than the oldest known fossil with this synapomorphy. Among angiosperms, the best example is that of tricolpate pollen, which is a unique synapomorphy of the huge eudicot clade (with various later modifications) and has a continuous record soon after its first sporadic appearances near the Barremian–Aptian boundary, c. 125 Mya (Doyle, 1992; Hughes, 1994), although the validity of this example has been questioned, based largely on conflicting molecular dates (e.g. Smith, Beaulieu & Donoghue, 2010). No comparable evidence is available to fix the age of Magnoliales or any other larger clade to which Annonaceae belong. To some extent, the practical problems of assigning maximum ages can be addressed by Bayesian methods, which treat constraints as probability distributions, as discussed below. However, the underlying theoretical problems remain.

Geological evidence for clade ages

An alternative means to derive maximum age constraints for clades, which has not been applied to Annonaceae, is to use geological evidence, for example using the known age of an oceanic island on which a clade is inferred to have diversified to impose a maximum age for its crown group. Older ages would imply multiple independent dispersals to the island plus the extinction of all the closest mainland relatives, a less parsimonious scenario. Contrary to the approach used by Richardson et al. (2001b), and reiterated recently in a review paper by Forest (2009), maximum age constraints should not be applied to stem nodes of island clades. By definition, the stem nodes of such clades predate the inferred dispersal of the clade to the island, and may therefore predate the age of the island itself. Instead, the conservative approach is to use geological evidence for maximum age constraints for crown nodes; this may be contrasted with the use of fossils for minimum age constraints for stem nodes. A further and potentially more serious drawback to the use of ages of islands as maximum age constraints is that oceanic islands often form as chains, each built up successively as a tectonic plate moves over a hot spot in the Earth's mantle, a classic case being the Hawaiian Islands (Baldwin & Sanderson, 1998). The original dispersal might therefore have been to an older island that has since eroded away (Price & Clague, 2002; Heads, 2005). Finally, the use of these or other geological events, such as the closure of the Panama isthmus or the uplift of the Andes, as calibration points for molecular dating would create a logical circularity should the ages estimated be used to test biogeographical hypotheses related to the geological event in question (Cody et al., 2010).

Secondary calibration

Finally, in the absence of direct evidence for age constraints, so-called ‘secondary’ calibration points may offer a means to obtain age estimates. In this case, the tree is calibrated either with an age taken from a different analysis (such as those published in the widely cited study of Wikström et al., 2001) or an absolute rate of molecular evolution inferred from a different group (e.g. Richardson et al., 2001a). This is not an ideal approach, firstly because the error margins associated with the molecular method (see below) inevitably reduce precision. Calibrations or constraints based on clades with higher or lower than average rates of molecular evolution may introduce bias (e.g. the use of Fagales, which show unusually low divergence in rbcL: Sanderson & Doyle, 2001). Furthermore, correct interpretation of the original primary calibration, i.e. whether it was a minimum or maximum age constraint or a fixed age, might be overlooked when the primary data are presented in a different study. In worst case scenarios, the inappropriate treatment of uncertainty in combination with uncritical assessment of dubious primary calibration points can result in the compounding and perpetuation of error, as argued by Benton & Ayala (2003).

Evidence of Annonaceae from the fossil record


The palaeobotanical literature contains a fair number of assignments of fossil leaves to Annonaceae, but few (if any) stand up to critical phylogenetic scrutiny and most are from the early Tertiary, postdating more diagnostic seed and floral fossils discussed below. Although Doyle & Le Thomas (1996) considered the evolution of secondary and tertiary venation in a phylogenetic context, there have been no comprehensive studies aimed at the recognition of diagnostic differences in leaf architecture between Annonaceae and other Magnoliales, or among clades within the family. Annonaceae resemble other Magnoliales in having pinnately veined, entire-margined leaves that are usually elliptical to obovate in shape (i.e. widest near or above the midpoint of the blade), but, consistent with their greater systematic diversity, Annonaceae vary more than the other families in spacing, angle and course (brochidodromous to eucamptodromous) of the secondary veins and in tertiary and finer venation (Klucking, 1986; Doyle & Le Thomas, 1996). More primitive annonaceous leaves, with reticulate tertiary venation, as in the basal genus Anaxagorea (the sister group of all other Annonaceae), are similar to leaves of Eupomatia R.Br. (the sister group of Annonaceae; Doyle & Le Thomas, 1996; Scharaschkin & Doyle, 2006), whereas others with derived percurrent tertiary venation approach some members of Magnoliaceae and Myristicaceae (Doyle & Le Thomas, 1996; Doyle et al., 2004). Dilcher & Lott (2005) stated that the leaves of Magnoliaceae and Annonaceae can be distinguished on the basis of moderate (< 55°) secondary vein angles and closed ultimate venation in Magnoliaceae and higher (> 55°) secondary angles and open ultimate venation in Annonaceae. Many members of Annonaceae have higher secondary angles (illustrated by figures of Anaxagorea and Duguetia in Doyle & Le Thomas, 1996) and, because both Anaxagorea and Eupomatia have high secondary angles (Scharaschkin & Doyle, 2006), this condition may be ancestral for the family. However, other members of Annonaceae have lower secondary angles, comparable with those of Magnoliaceae (e.g. Asimina Adans. and Annickia Setten & Maas in Doyle & Le Thomas, 1996), and so this character is not always sufficient to distinguish between leaves of the two families.

With a few exceptions (e.g. Wolfe, 1977; Roth, 1981; Dilcher & Lott, 2005), most reports of fossil annonaceous leaves are from what has been called the ‘picture matching’ phase of angiosperm palaeobotany, which predated more rigorous and consistent methods of analysis of leaf architectural characters (Hickey, 1973; Dilcher, 1974; Hickey & Wolfe, 1975; Ellis et al., 2009), to say nothing of principles for the appropriate analysis of their phylogenetic significance. Many identifications from this phase are notorious for overlooking equally close or closer matches with other extant taxa, and many have been shown to be incorrect by the analysis of cuticle structure. In a preliminary analysis of venation and epidermal anatomy, Dilcher (1971, 1974) estimated that about 60% of Berry's identifications of Eocene leaves in the Mississippi Embayment were erroneous at the generic and often the familial level. Late Cretaceous leaves from Massachusetts and New York that were described by Hollick (1906) as Guatteria cretacea Hollick are doubtful because of their lack of higher order venation; they also differ from most Annonaceae, including most, although not all, species of Guatteria Ruiz & Pav. (Erkens, 2007), in their lanceolate rather than obovate or elliptical shape. Berry (1916, 1941), Chaney & Sanborn (1933), Sanborn (1935) and Lakhanpal (1969) identified leaves from Eocene and Oligocene floras of the south-eastern USA (Mississippi Embayment) and Oregon as Annona L. Some of these may represent Annonaceae, but they have not been evaluated with modern leaf architectural or cuticle analysis. Similar considerations apply to Berry's identifications of leaves from the Miocene of Central and South America as Annona and Guatteria (Berry, 1918, 1920, 1922a, b, 1923, 1937). Many of these differ from most members of Annonaceae in their asymmetrical and ovate shape.

Probably the most secure leaf record of Annonaceae consists of fossils described from the middle Eocene of the Mississippi Embayment by Berry (1916) as Magnolia leei Knowlton, but reassigned to the Neotropical annonaceous genus Duguetia (as Duguetia sp.) by Roth (1981) and Dilcher & Lott (2005), based on venation, paracytic stomata and the presence of stellate-peltate trichomes. In Annonaceae, such trichomes appear to be restricted to Duguetia, the related African genus Pachypodanthium Engl. & Diels (synonymized with Duguetia by Chatrou, 1998; Chatrou, Koek-Noorman & Maas, 2000) and Meiocarpidium Engl. & Diels (Fries, 1959; Le Thomas, 1969), but Meiocarpidium differs in having conspicuously sunken stomata (Roth, 1981). Elsewhere in Magnoliales, peltate scales occur in Galbulimima F.M.Bailey (Himantandraceae), but this genus differs in its distinctive radial arrangement of stomata below the scale (Roth, 1981). Considering only epidermal characters, the fossil might be a stem relative of Meiocarpidium or Galbulimima that lacked the derived features of the living genera, but its leaf architecture is more like that of Duguetia. Stellate-peltate trichomes were inferred to be a synapomorphy of Duguetia and ‘Pachypodanthium’ in the analysis of Doyle & Le Thomas (1996), which did not include Meiocarpidium. Doyle & Le Thomas (1996) defined their stellate-peltate state as also including tufts of simple trichomes of the sort found in Tetrameranthus R.E.Fr., Uvaria L. and some Annona spp. (Jovet-Ast, 1942; Van Setten & Koek-Noorman, 1986), in order to allow for the possibility that stellate-peltate trichomes were derived from the tufted type, but, judging from their phylogenetic results (and more recent molecular studies), the two types arose independently.

Based on a morphological phylogenetic analysis, Chatrou (1998) concluded that ‘Pachypodanthium’ was nested within Duguetia, but more recent molecular analyses indicate that the two taxa are sister groups (L. Chatrou, pers. comm.). If stellate and peltate trichomes are not distinguished, their presence in the middle Eocene fossil may be a synapomorphy linking it with the Duguetia–‘Pachypodanthium’ clade, which would provide a minimum age of c. 38 Mya for the node connecting Duguetia (including ‘Pachypodanthium’) with its sister group, Fusaea. However, trichomes in Neotropical Duguetia vary from stellate hairs to peltate scales with variously fused radiating cells (Maas, Westra & Chatrou, 2003), whereas ‘Pachypodanthium’ has only stellate hairs (Le Thomas, 1969), which resemble those of Duguetia uniflora (DC.) Mart. (L. Chatrou, pers. comm.). This might suggest that peltate scales are a synapomorphy of the Neotropical clade, or some large portion of it, to the exclusion of ‘Pachypodanthium’. If so, as the fossil has peltate scales consisting of cells fused about one-third of their length, it could be more closely related to Neotropical Duguetia, and 38 Mya would be a minimum age for the crown node of Duguetia (including ‘Pachypodanthium’). However, better information on the systematic distribution of trichome types is needed to decide between these alternatives. Hopefully, more comprehensive surveys of leaf architecture and epidermal anatomy in a phylogenetic framework will allow more accurate assignment of fossil leaves to Annonaceae and to clades within the family.


In the fossil pollen record, there are reports of morphological types now restricted to the two major sister clades that make up most of Annonaceae. These were called the SBC (short branch clade) and LBC (long branch clade) by Richardson et al. (2004) and the malmeoid-piptostigmoid-miliusoid (MPM) and inaperturate clades by Doyle, Bygrave & Le Thomas (2000) and Doyle et al. (2004), but are now formally classified as subfamilies Malmeoideae and Annonoideae (Chatrou et al., 2012).

Reticulate-columellar monosulcate pollen, described by Sole de Porta (1971) from the Maastrichtian (latest Cretaceous) of Colombia as Foveomorphomonocolpites humbertoides, was compared by Muller (1981) with the Malmea tribe of Walker (1971), which corresponds closely to the malmeoid clade of Doyle & Le Thomas (1996), the South American Centred clade of Pirie et al. (2006) and tribe Malmeeae of Chatrou et al. (2012). Muller considered this the oldest accepted pollen record of Annonaceae. The phylogenetic implications of this pollen are sensitive to relationships among the early branches of Malmeoideae, especially the incompletely resolved arrangement of several granular monosulcate genera and the columellar monosulcate genus Annickia, which form a clade treated as tribe Piptostigmateae in the tree of Chatrou et al. (2012). If Annickia is linked with the granular members of Piptostigmateae, or basal in the Malmeoideae, parsimony optimization indicates that columellar monosulcate pollen is basic for the combined Malmeoideae and Annonoideae, but if the granular Piptostigmateae are sister to the rest of Malmeoideae, such pollen may be basic for either the combined Malmeoideae and Annonoideae or all of Malmeoideae except the granular Piptostigmateae. As such pollen is more derived than the granular monosulcate pollen of Anaxagorea and Ambavioideae (Doyle & Le Thomas, 2012), the Colombian fossil may therefore provide a minimum age of c. 68 Mya for either the split between Ambavioideae and the combined Malmeoideae and Annonoideae, or for the crown node of Malmeoideae.

Tetrads of inaperturate, reticulate-columellar grains from the early Eocene of the US Gulf Coast [Annona (?) foveoreticulatus: Elsik, 1974] and the Oligocene of Cameroon (Inaperturotetradites reticulatus: Salard-Cheboldaeff, 1978) have been compared with Annona, but similar pollen also occurs elsewhere in Annonoideae. Because loss of the aperture (or its modification to a round thin area that becomes proximal during development: Tsou & Fu, 2002) is a synapomorphy of Annonoideae, and tetrads arose at several points within this clade (Doyle & Le Thomas, 2012), these tetrads provide a minimum age of c. 50 Mya for the split of Malmeoideae and Annonoideae, but this would probably be a substantial underestimate.


Fossil wood identified with Annonaceae is also known from the Palaeocene–Eocene of England (Polyalthioxylon oldhavenense: Crawley, 2001), the Eocene of Oregon (Annonoxylon bonesii: Wheeler & Manchester, 2002) and more poorly dated but probably post-Eocene horizons in Sudan (Annonoxylon striatum and A. edengense: Boureau, 1950, 1954). These authors made tentative comparisons of these wood types with modern genera. However, although the relationship of the fossils to Annonaceae seems well established, based on the diagnostic combination of characters, such as simple vessel perforations, reduced fibre pits and apotracheal parenchyma bands, a more detailed analysis of the distribution of wood characters in a phylogenetic framework is needed to assess their position within the family. There have been extensive surveys of wood anatomy in Annonaceae (notably Vander Wyk & Canright, 1956), but Doyle & Le Thomas (1996) found that the two characters that seemed most likely to be informative, vessel density and ray width, were highly homoplastic across the family.


Fossil seeds related to Annonaceae are more distinctive and extend back into the Cretaceous. The most diagnostic seed characters are the perichalazal ring (where the raphe runs most of the way around the seed as a result of differential growth) and ruminate endosperm formed by ingrowths of both the tegmen and testa (derived from both inner and outer integuments) of the sides of the seed (Corner, 1976; Van Setten & Koek-Noorman, 1992). These testal ruminations contrast with the tegminal and/or chalazal ruminations of Myristicaceae (Doyle et al., 2004). Within Magnoliales, testal ruminations are shared with Eupomatia and Degeneria A.C.Sm., but Annonaceae are the only group with a perichalazal ring. Doyle & Le Thomas (1996) and Doyle & Endress (2000) scored Galbulimima as lacking ruminations, but it is better scored as unknown (?) based on the report of probable reduced ruminations by Doweld & Shevyryova (1998), as in Sauquet et al. (2003) and Endress & Doyle (2009). Perichalazal seeds with ruminate endosperm also occur in Austrobaileya C.T.White in the basal ‘ANITA’ grade, but the ruminations differ in their finely ramified form (Endress, 1980).

Within Annonaceae, ruminations show significant variation among taxa. The two most distinctive types are lamelliform, with four (sometimes two) thin transverse plates, and spiniform, with fine spines penetrating the endosperm like pins in a pincushion, but these intergrade via forms with radially subdivided plates (Van Setten & Koek-Noorman, 1992). Other seeds differ in having thicker and more irregular processes, variously described as stout or woody pegs or plates, here called irregular ruminations; this is the type found in the outgroups Eupomatia and Degeneria. The implications of ruminations for systematic placement of fossil seeds can be evaluated in the context of Figure 8, which shows a parsimony reconstruction of the course of evolution of rumination types on a pruned version of the Bayesian tree of Chatrou et al. (2012).


Figure 8. Composite cladogram of Annonaceae and outgroups based on published molecular analyses, showing the evolution of the endosperm rumination character inferred from parsimony optimization with MacClade (Maddison & Maddison, 2003). See text for sources of data on tree topology and ruminations. ‘Uncertain’ indicates that the taxon was scored as having either of two states (here either spiniform or lamelliform, i.e. 2/3), ‘equivocal’ that the reconstructed state on a branch is ambiguous. Figures of fossil seeds illustrate the three major types of rumination: irregular in Anonaspermum ovale from the London Clay (early Eocene, reproduced from Reid & Chandler, 1933, c. 2×), spiniform in A. punctatum (London Clay, c. 2×) and lamelliform in A. commune (London Clay, c. 1.4× in 14 and 15, 2× in 17) and A. gilbediense (Maastrichtian of Nigeria, reproduced from Chesters, 1955, c. 2×). LBC, long branch clade; SBC, short branch clade.

Download figure to PowerPoint

The oldest reported seeds of Annonaceae are Anonaspermum gilbediense and A. phytoniscus, described from the Maastrichtian (latest Cretaceous) of Nigeria by Chesters (1955; Fig. 8). These seeds have a perichalazal ring, a synapomorphy of Annonaceae as a whole, and lamelliform ruminations, which are characteristic of most members of Annonoideae and some Malmeoideae (Fig. 8). Similar ruminations occur in a seed (Unonaspermum corneri) described by Bonde (1993) from the Deccan intertrappean beds of India, near the Cretaceous–Tertiary boundary. Most members of Malmeoideae have spiniform ruminations, which appear to be ancestral for this clade (Doyle et al., 2000, 2004; Fig. 8), whereas Anaxagorea and most Ambavioideae have thicker, irregular ruminations. Both ingroup topology and outgroup comparison indicate that irregular ruminations are ancestral in Annonaceae, as they are shared with Eupomatia, Degeneria and, probably, Galbulimima (Doweld & Shevyryova, 1998).

Given the distribution of lamelliform ruminations in Annonaceae, their presence in the Nigerian fossils might seem to be evidence for the presence of Annonoideae or some derived member of Malmeoideae, but this is ambiguous. With the combined morphological and molecular tree of Doyle et al. (2000), parsimony optimization indicated that the common ancestor of Malmeoideae and Annonoideae could have had irregular, spiniform or lamelliform laminations, so that the spiniform and lamelliform types could have been derived independently from the irregular type or one from the other (Doyle et al., 2004). Doyle et al. (2004) therefore argued that the Nigerian seeds could only be used as evidence that the combined Malmeoideae–Annonoideae clade had diverged from Ambavioideae by this time (68 Mya). More recently, as noted by Couvreur et al. (2008b), this picture has been further blurred by molecular evidence that the African genus Meiocarpidium is the earliest branching taxon in the ambavioid clade, because this genus is reported to have lamelliform ruminations (Van Setten & Koek-Noorman, 1992). If these ruminations are indeed comparable with those in Annonoideae and Malmeoideae, which should be verified, their presence in Meiocarpidium could mean that lamelliform ruminations originated earlier, with a reversal to irregular in Ambavioideae, or that they originated twice and the fossils are relatives of Meiocarpidium rather than the Malmeoideae–Annonoideae clade (Fig. 8). Given that both scenarios are equally parsimonious, the Nigerian seeds only provide a minimum age for the crown node of Annonaceae as a whole.

The Nigerian seeds are of broader significance for the ecological evolution of angiosperms, as they represent some of the oldest large diaspores in the group. Most Cretaceous angiosperm fruits and seeds are much smaller, and one of the most striking changes in the early Tertiary is an increase in diaspore size, which has been variously linked with climatic or faunal changes at the Cretaceous–Tertiary boundary (Upchurch & Wolfe, 1987; Wing & Tiffney, 1987; Eriksson, Friis & Löfgren, 2000). In representing an exception to the rule for Cretaceous diaspore size, the Nigerian seeds are potentially critical in evaluating the causal factors responsible for these changes.

Better evidence for diversification in Annonaceae is provided by perichalazal seeds with ruminate endosperm from the early Eocene London Clay, described by Reid & Chandler (1933) as 13 new species of Anonaspermum O.M.Ball (see also Collinson, 1983). These show all three main types of rumination: lamelliform (e.g. A. commune, A. pulchrum, A. rotundatum), spiniform (A. punctatum) and irregular (A. ovale, A. complanatum, A. corrugatum), plus intermediate types (Fig. 8). Together, setting aside the possibility that the spiniform seeds are related to Cyathocalyx or the lamelliform seeds to Meiocarpidium, these fossils indicate that Malmeoideae and Annonoideae, as well as Anaxagorea and Ambavioideae, had diverged by c. 50 Mya (Doyle et al., 2004).

The discovery of a fossil fruit containing seeds of the A. ovale type led Chandler (1978) to transfer this species to the extant genus Uvaria, based on the large number of seeds (17 preserved) in two alternating rows and the bipartite vs. quadripartite endosperm ruminations. She contrasted this with the torulose fruits containing one row of seeds in such taxa as Desmos Lour. and Dasymaschalon Dalla Torre & Harms, now known to belong to the same liana clade (Uvarieae Hook.f. & Thomson, sensuChatrou et al., 2012). This fossil is substantially older than the molecular age of Uvaria found by Zhou et al. (2012; stem, 32 Mya; crown, 27 Mya). However, although the data of Van Setten & Koek-Noorman (1992) confirm that such fruits with two rows of seeds are indeed typical of Uvaria (and the genera synonymized with it by Zhou et al., 2010), similar fruits also occur in Fissistigma Griff., Mitrella Miq., Dielsiothamnus R.E.Fr. and Toussaintia Boutique (all Uvarieae; the last two with one row of seeds), and in many other clades of Annonaceae. In addition, the thickness and irregularity of the ruminations in the fossil would be anomalous in Uvarieae and more typical of Ambavioideae, as noted above. Van Setten & Koek-Noorman (1992) also noted variation between bipartite and quadripartite ruminations in many genera other than Uvaria (e.g. Dielsiothamnus). These observations suggest that it would be unwarranted to accept this fossil as a record of Uvaria, or even of Uvarieae as a whole.


The oldest evidence for crown Annonaceae is provided by Futabanthus, a fossil flower described by Takahashi, Friis & Crane (2008) from the early Coniacian of Japan (c. 89 Mya), which has a most likely trimerous perianth, numerous stamens and numerous carpels borne on a flattened receptacle (Fig. 9). Like most Annonaceae, it lacks inner staminodes, which occur in both Anaxagorea and the outgroups Degeneria, Galbulimima and Eupomatia (Endress, 1984). The loss of inner staminodes may therefore be a synapomorphy that links Futabanthus with the clade consisting of Annonaceae other than Anaxagorea. The fact that the stamens have an extended connective apex, like the outgroups, Anaxagorea, most Ambavioideae and the genus Greenwayodendron Verdc. near the base of Malmeoideae (Doyle & Le Thomas, 1996; Doyle et al., 2000), rather than the peltate apex of most members of Malmeoideae and Annonoideae, may place it near the base of the Ambavioideae–Malmeoideae–Annonoideae clade. The absence of inner staminodes also tends to exclude Futabanthus from other positions in the clade consisting of Degeneria, Galbulimima, Eupomatia and Annonaceae.


Figure 9. Scanning electron micrographs of flower of Futabanthus from the Coniacian of Japan (Takahashi et al., 2008): A, view of whole flower; scale bar, 1 mm; B, close-up showing stamen morphology; scale bar, 0.5 mm. Images kindly provided by Masamichi Takahashi.

Download figure to PowerPoint

An alternative hypothesis is that Futabanthus diverged earlier in Magnoliales, near Magnoliaceae, and never had inner staminodes. It differs from Magnoliaceae in lacking an elongate floral receptacle, but, contrary to Takahashi et al. (2008), this does not rule out a relationship, as the elongate receptacle is an autapomorphy of Magnoliaceae that would not be expected in their common ancestor with other Magnoliales (Endress & Doyle, 2009). However, the short length and the close packing of the stamens, as in Annonaceae but not other Magnoliales, are potential synapomorphies that favour a relationship with Annonaceae. Futabanthus therefore provides a minimum age of c. 89 Mya for the crown node of Annonaceae.

Kvaček & Eklund (2003) interpreted a still older flower, Pecinovia annonoides J.Kvaček & H.Eklund from the mid-Cenomanian of the Czech Republic (c. 96 Mya), as possibly related to Annonaceae, based on its probably trimerous perianth and numerous stamens with a globular apical extension of the connective. However, this flower appears to differ from most Annonaceae in being unisexual and having two rather than three whorls of tepals, and the connective extensions are more convex than the peltate apices of most Annonaceae and do not cover the anthers as much. The combination of characters in the pollen, which is small, globose and monosulcate and has finely verrucate sculpture, is also unlike that of any living Annonaceae. Most monosulcate Annonaceae have a smooth, foveolate or reticulate tectum, and the most similar verrucate sculpture is found in Miliuseae, in which the pollen is either inaperturate or disulculate (Doyle & Le Thomas, 2012). An analysis of this fossil in a broader systematic context is needed to evaluate its relationships.

Fossils of other Magnoliales: constraining the stem age of Annonaceae

Fossils representing other Magnoliales are significant for dates in Annonaceae because they provide age constraints for deeper nodes. Most widely cited has been Archaeanthus, from near the Albian–Cenomanian boundary in Kansas (Dilcher & Crane, 1984), which is represented by bilobed leaves, flowers with an elongate receptacle bearing numerous carpels and scars of other floral parts, and detached tepals and stipule-derived bracts, indicating a floral organization essentially identical to that of living Magnoliaceae. Crepet et al. (2004) considered a relationship to Magnoliaceae to be unproven, but it has been confirmed by a morphological phylogenetic analysis of Doyle & Endress (2010), with living taxa constrained to arrangements based on analyses of molecular and morphological data. This indicated that Archaeanthus may be either a stem relative of Magnoliaceae or nested within the family. It therefore provides a minimum age for the common ancestor of Magnoliaceae and Annonaceae, i.e. the crown node of the clade called Magnoliinae, which includes all Magnoliales except Myristicaceae (Sauquet et al., 2003). Its age has been cited as early Cenomanian, c. 98 Mya, but it may actually be latest Albian, in which case 100 Mya would be a more appropriate figure (Gröcke et al., 2006; see discussion in Doyle & Endress, 2010).

Other more fragmentary evidence led Doyle et al. (2004) to suggest that Magnoliales are significantly older than Archaeanthus. One is the monosulcate pollen genus Lethomasites J.V.Ward, J.A.Doyle & Hotton, from the Aptian of Virginia (Ward, Doyle & Hotton, 1989), which has a thin nexine and a granular infratectum, a possible synapomorphy linking it with Degeneria, Galbulimima, Eupomatia and Annonaceae. However, we consider this to be uncertain because of the small number of characters involved and the fact that Lethomasites has a distinctive fossulate tectum not known in living Magnoliales. An isolated laminar stamen containing smooth monosulcate pollen from the Albian of Virginia (dispersed stamen type 1 of Crane, Friis & Pedersen, 1994) is also suggestive of Magnoliales, but again this similarity is based on only a few characters.

More important in terms of both age and more convincing systematic placement is Endressinia, represented by a stem with attached leaves and flowers from the late Aptian of Brazil (Mohr & Bernardes-de-Oliveira, 2004), with poorly preserved tepals, numerous staminodes and numerous plicate carpels (the absence of stamens is probably a result of poor preservation). The late Aptian dating of these deposits has been strengthened by Heimhofer & Hochuli (2010). Mohr & Bernardes-de-Oliveira (2004) suggested that Endressinia was closest to Eupomatia, but the phylogenetic analysis of Doyle & Endress (2010) indicated that it is equally parsimonious to place it anywhere within the clade consisting of Degeneria, Galbulimima, Eupomatia and Annonaceae, or on its stem lineage. Like Archaeanthus, Endressinia therefore provides a minimum age for the crown node of Magnoliinae, but one that is considerably older (c. 112 Mya).

DNA evidence for the ages of clades: the molecular clock

As far back as the 1960s, biologists were investigating the relationship between divergence in molecular data and elapsed time. Margoliash (1963) showed that levels of divergence in cytochrome c were proportional to the time since divergence, based on the fossil record, and, in 1965, Zuckerkandl & Pauling introduced the term ‘molecular evolutionary clock’ to describe this phenomenon (Zuckerkandl & Pauling, 1965). In 1968, Kimura published the ‘neutral theory’, which postulates that most mutations are selectively neutral. The process of mutation is stochastic, occurring randomly rather than at regular intervals, but, if sufficient numbers of mutations are observed, these would be expected to accumulate at an approximately constant rate. In reality, this is often not the case: indeed, it has been argued on theoretical and empirical grounds that differences in rates can be expected given a number of common factors, such as differences in generation time, metabolic rate, species-specific mutation rates and population sizes (Bousquet et al., 1992; Gaut, 1998). In the last 10 years, a body of mainly plastid-encoded DNA sequences has been collected for species of Annonaceae, primarily for use in phylogenetic reconstruction. An increasingly clear picture of phylogenetic relationships at the generic level is emerging (Couvreur et al., 2011; Chatrou et al., 2012), but amounts of molecular change, represented by the lengths of branches leading from the root to the different terminal taxa of the tree, and therefore rates of evolution, have been shown to be highly variable. In other words, the plastid sequences have evolved in a non-clock-like manner (Doyle et al., 2004; Richardson et al., 2004; Pirie et al., 2006), a phenomenon frequently observed in plants in general (Gaut, 1998; and numerous other empirical studies).

Can one use data that are not consistent with a strict clock to estimate the ages of clades? A number of approaches have been proposed. One is simply to impose a molecular clock on the data, e.g. as an aspect of a model describing DNA sequence evolution, as used in likelihood-based phylogenetic inference. Any variation in rates between lineages is then taken into account instead by other parameters of the model, such as the differences between rates of change in different bases (such as transitions versus transversions) or across different sites of the sequence alignment (the gamma-distribution shape parameter). Depending on the degree of rate variation, however, the estimated values of these parameters may differ considerably under the (incorrect) assumption of a clock, and may thus no longer represent a good reflection of the underlying process of molecular evolution. As would be expected, when Drummond et al. (2006) simulated sequences both with and without the assumption of a molecular clock, they found that analysing clock-like data using a method assuming a clock produces accurate and precise results, but this is not the case when the data are not clock-like.

A second approach might be to remove those taxa/sequences that deviate most from clock-like behaviour, until those that remain conform to a strict molecular clock (e.g. Bremer, 2000). However, roughly equal branch lengths in a phylogenetic tree do not guarantee that rates were constant along those branches. Lineages that clearly deviate from the molecular clock may reflect rate heterogeneity that is still present, but no longer seen, when these lineages are selectively ignored. In addition, the offending lineages may be those in which we are most interested. Approaches that ignore or attempt to eliminate differences in rates of evolution and assume a molecular clock have therefore generally been supplanted by those that seek to ‘relax’ the clock under one of a number of different assumptions.

Relaxing the molecular clock

One of the biggest challenges to be overcome by ‘relaxed clock’ molecular dating techniques lies in the inextricable link between substitution rate and elapsed time on the branches of a phylogeny (Thorne & Kishino, 2002; Magallón, 2004). A branch of a given length could be the result of a low evolutionary rate over a long period of time or of a high rate over a short time. To tease these factors apart, it is necessary to make assumptions about the expected patterns of rate variation across a phylogeny.

Assumptions and methods in molecular dating

Until recently, the most commonly applied assumption has been that of rate autocorrelation (Sanderson et al., 2004). This means that, overall, the rates of molecular evolution are expected to be more similar between more closely related lineages. Rate autocorrelation can be assumed to apply when most of the variation in rates across a group is caused by inherited factors (Drummond et al., 2006). Under certain conditions, this might not be expected to be the case. Over long timescales, such as those involved in the history of Annonaceae, and particularly with sparser taxon sampling, autocorrelation might break down because of high variation between lineages (Drummond et al., 2006). Although Drummond et al. (2006) found no autocorrelation in a number of datasets, Lepage et al. (2007) suggested that this conclusion may have been the result of limited taxon sampling, and that rate autocorrelated models may fit the data better when taxon sampling is more dense. Rate autocorrelation has been applied in a number of studies of Annonaceae (Doyle et al., 2004; Richardson et al., 2004; Pirie et al., 2005, 2006) using techniques such as NPRS (Sanderson, 1997), PL (Sanderson, 2002a) and certain Bayesian approaches (e.g. multidivtime: Thorne, Kishino & Painter, 1998) (Table 2).

NPRS and PL take as input phylogenetic trees, called phylograms, in which branch lengths represent amounts of change in DNA sequence data. These phylograms are then transformed into ultrametric chronograms, in which all tips are the same distance from the base and the branch lengths are assumed to be proportional to time. This transformation is achieved by allowing rates of evolution to change throughout the tree, such that changes between neighbouring branches are minimized (i.e. autocorrelation). PL incorporates an extra step, the estimation and application of a ‘smoothing parameter’, which limits the degree to which changes are allowed (i.e. it is semiparametric). The more clock-like the tree, the higher the value of the smoothing parameter and the lower the permitted rate variation (Sanderson, 2002a). This converges to the LF clock method, which is equivalent to a PL analysis with a high smoothing parameter value, such that a universal rate is applied across the tree. NPRS, once widely applied, has been shown to perform rather badly (e.g. Linder, Hardy & Rutschmann, 2005; Pirie et al., 2005) and has effectively been usurped by PL. PL can, however, only be applied to trees with a limited number of terminal taxa (a few hundred).

The Bayesian multidivtime method (Thorne et al., 1998; Thorne & Kishino, 2002; not compared in Fig. 1) infers change in evolutionary rate over time following a (parametric) probabilistic model, in which the rate of the parent branch is taken as the mean of a prior distribution within which the rate of the daughter branch is drawn (i.e. it is still a form of rate smoothing). A Markov Chain Monte Carlo (MCMC) procedure is used to obtain PP distributions of rates and ages, from which mean values and credibility intervals can be obtained. Multidivtime was used by Pirie et al. (2006), resulting in similar age estimates to those derived using PL.

A number of the assumptions implemented in the Bayesian program BEAST (Drummond & Rambaut, 2006) differ from those of NPRS, PL and multidivtime. Under the uncorrelated relaxed clock options, instead of assuming that rates are more similar in more closely related organisms, it is assumed that the frequency of rates across the phylogenetic tree as a whole follows a given probability distribution: either an exponential or a log-normal distribution, the latter of which is perhaps more biologically realistic. Applying an exponential distribution assumes that the lowest rates are observed most frequently and that higher rates are progressively more unlikely. By contrast, applying the log-normal distribution assumes that both very low and very high rates of evolution are relatively unlikely and that an intermediate rate will be observed most often. Although the assumptions of exponential or log-normal rate distributions might be seen as alternatives to that of rate autocorrelation, neither actually excludes the possibility that rates might be autocorrelated. It has even been suggested that the relaxed clock models in BEAST might be used to test for evidence of autocorrelation of rates, by means of a summary statistic for rate covariance. BEAST has been used in several recent works on Annonaceae (Couvreur et al., 2008a, 2011; Erkens, Maas & Couvreur, 2009; Su & Saunders, 2009) and Couvreur et al. (2011) reported a value not significantly greater than zero for the posterior distribution of the covariance, which might imply a lack of rate autocorrelation. Such an interpretation of this statistic has, however, been questioned (Ho, 2009), and the evidence should not be regarded as conclusive.

A more recent model for the evolution of evolutionary rates also implemented in BEAST is that of Bayesian Random Local Clocks (RLC; Drummond & Suchard, 2010). Drummond & Suchard (2010) suggested that the numerous small changes in rate inferred under many molecular dating methods may be a result of the way in which rates are modelled, and not actually driven by the data itself. In contrast to relaxed clock models that assume continuously varying evolutionary rates, the RLC approach works under the assumption that the data can be explained by a small number of (possibly large) rate changes. In simple terms, this assumption is implemented by assigning each branch its own rate, whilst placing a strong prior belief in that rate being the same as that of its parent branch (Drummond & Suchard, 2010), by means of the Bayesian Stochastic Search Variable Selection model of Lemey et al. (2009). This model has yet to be used with Annonaceae.

Comparison of dating methods as applied to Annonaceae

What then is the impact of different clock and relaxed clock models on the estimation of ages in Annonaceae? Age estimates derived from the data of Doyle et al. (2004), Richardson et al. (2004) and Pirie et al. (2006) using the LF, NPRS and PL methods are summarized and compared in Figure 1, and the results of the PL and BEAST analyses of the supermatrix are presented in Figure 4 (PL), Appendix S1 (BEAST) and Table 4. Discrepancies are apparent given the same method with different datasets and across each of the methods, as has been observed in other studies (e.g. Bell & Donoghue, 2005).

The most conspicuous area of inconsistency among methods concerns the relative ages of the crown groups of Annonoideae and Malmeoideae: the LF (clock) method and BEAST infer a substantially more recent origin for Malmeoideae than for Annonoideae, whereas age estimates based on NPRS and PL are more nearly equal. The discrepancy between the clock and rate smoothing methods can be explained by the differences in evolutionary rates inferred under the different methods, particularly in this part of the tree. In general, NPRS and PL age estimates are similar, as might be expected from the low optimal smoothing parameter values under PL. Under NPRS and PL, the ancestral rate in Annonaceae is inferred to be relatively high. This is evident from the ratogram in Figure 3 (see also Table 5), estimated using PL (Sanderson, 2002a) with the supermatrix, where the lengths of the branches are proportional to the evolutionary rates. Lower rates in Malmeoideae are accommodated by means of a stepwise decrease, consistent with the assumption of rate autocorrelation. By contrast, the universal rate inferred for the same data under the LF method is relatively low. Thus, the branch subtending the Malmeoideae crown node is smoothed under NPRS/PL to represent less time, but at a higher evolutionary rate, than under LF. As a strict molecular clock does not fit the data, it would seem that the assumption of LF (a universal rate across the tree) is unrealistic. Under these circumstances, a model that allows rates to change might be preferred. However, it is clear that the gradual change in rate that is inferred assuming rate autocorrelation strongly influences the relative ages of these nodes, and that the accuracy of the results is therefore contingent on this assumption being appropriate, which would not be the case if rates changed abruptly.

Both the previous studies using BEAST (Erkens et al., 2009; Su & Saunders, 2009; Couvreur et al., 2011; Zhou et al., 2012) and our own analysis show a large impact of this method on the relative ages of Malmeoideae and Annonoideae. The age estimates for the Malmeoideae crown node are consistently more recent than those for Annonoideae and, in the cases of Couvreur et al. (2011) and Zhou et al. (2012), the confidence intervals for these age estimates do not overlap (Table 3). The same is true with our ages (Table 4 and Appendix S1). These results are rather similar to those inferred assuming (incorrectly) a strict molecular clock (Table 3). The impact extends from deeper nodes (the dramatically different age of the SBC) to more shallow ones, in particular the crown ages of a number of genera of Malmeoideae (e.g. Maasia, Mosannona Chatrou, Pseudoxandra R.E.Fr. and Unonopsis R.E.Fr., which are considerably older under PL; Table 4). However, differences in rates of molecular evolution inferred for the deeper nodes, in particular the stem and root nodes of Malmeoideae (K and N; Fig. 1, Table 5), are clearly decisive in these age discrepancies.

These results have a direct relation to the interpretation of the fossil seed record of Annonaceae. The most critical fossils are the seeds with spiniform ruminations, a state restricted to Malmeoideae, in the London Clay (Reid & Chandler, 1933; Doyle et al., 2004), which is early Eocene (c. 50 Mya). If the LF (clock) or BEAST ages are correct, they would imply that these seeds represent stem relatives of Malmeoideae, rather than members of the Malmeoideae crown group, or conceivably stem relatives of the combined Malmeoideae and Annonoideae clade, if spiniform ruminations originated on the stem lineage of this clade, as is permitted by the analysis in Figure 8. In either case, they would not be directly related to the lineages of Malmeoideae that dispersed out of Africa, the most likely ancestral area of both Malmeoideae and Annonoideae, and are now found in Asia or the Neotropics (Couvreur et al., 2011). They would therefore represent lineages that later became extinct in Eurasia. This scenario can by no means be dismissed, but it might be questioned on parsimony grounds: the required dispersal of a stem relative of Malmeoideae from Africa into Eurasia and its subsequent extinction could be regarded as additional hypotheses that would not be required if the crown group of Malmeoideae was older.

The LF and BEAST results would also be inconsistent with the view of Muller (1981) that the columellar monosulcate pollen species Foveomorphomonocolpites humbertoides (Sole de Porta, 1971) from the Maastrichtian of Colombia (c. 68 Mya) is a record of the ‘Malmea tribe’ of Walker (1971), roughly equivalent to Malmeeae of Chatrou et al. (2012), as this clade is nested in Malmeoideae, which BEAST indicates is much younger. However, in this case, the fossil evidence might be more easily reconciled with the BEAST analysis, as the morphology of the fossil is less diagnostic of Malmeeae. As discussed above, parsimony optimization of pollen characters (Doyle & Le Thomas, 2012) indicates that columellar monosulcate pollen of the fossil type may be ancestral for most or all of Malmeoideae, depending on the position of the African columellar monosulcate genus Annickia, and possibly for the Malmeoideae and Annonoideae clades combined.

A hint as to the accuracy of the BEAST approach in this case may be provided by results that were obtained when outgroups were and were not specified. An important feature of BEAST is that the phylogeny is estimated assuming a model of rate evolution, which roots the resulting chronograms when no outgroups are specified: the method effectively identifies the position of the root on the basis of inferred branch lengths. This aspect of the method has been exploited to identify the rooting for clades in which the outgroup is not known with certainty (e.g. Renner et al., 2008). However, when the outgroup has been identified on the basis of independent data, the performance of BEAST in correctly recovering that rooting can be poor. Annonaceae provide a sobering example: when BEAST is used to infer the phylogeny without first defining the outgroup (by constraining the monophyly of the ingroup), the longer branches in Annonoideae are modelled to represent longer periods of time at lower rates, making Annonoideae older than both the rest of Annonaceae and its magnolialean outgroups (which are thus nested within Annonaceae; Fig. 4). This pathological effect would appear to mirror a general principle set out in Sanderson & Doyle (2001), that topological errors caused by long branch effects (in that example, rooting the angiosperms with Oryza L.) can lead to errors in age estimates: errors in age estimates due to ‘long-branch effects’ can apparently lead to topological errors, too. In the case of Annonaceae, the error is obvious. It may in itself be evidence to suggest that the underlying model for evolutionary rates (in this case, a single log-normal distribution, instead of which, for example, a bimodal distribution reflecting higher rates in Annonoideae might be more appropriate) does not optimally fit the data; an appropriate model should indeed accurately identify the root. It is worth emphasizing that, in cases in which the outgroup (or indeed any aspect of the topology) is unknown, the problem may not be so obvious.

Error associated with relaxing the clock

Part of the controversy regarding molecular dating concerns confusion of uncertainty involved in the calibration(s), as discussed above, with inconsistencies among the results of different molecular dating methods and the published error margins for these results. Molecular dating techniques are subject to a number of sources of error that must be taken into account when assessing the significance of the results (Sanderson & Doyle, 2001). It is useful to clarify here that, when a single node of a phylogenetic tree is fixed to a particular age (e.g. by using the age of a fossil as a fixed calibration), the confidence intervals obtained do not take into account any uncertainty in the calibration, which may of course be considerable. Sources of error associated with relaxing the clock include sampling error (i.e. of individual characters, genes and taxa) and phylogenetic uncertainty, particularly that associated with the parameters of the model used to derive the topology and branch lengths.

Sampling error

Uncertainty caused by the stochastic nature of substitutions in DNA sequence data (termed ‘substitutional noise’ by Sanderson & Doyle, 2001) can be represented using nonparametric bootstrapping, resampling with replacement from the data. We expect error margins of this kind to decrease with increasing amounts of sequence data (although uncertainty may be inevitable, even with a large amount of sequence data; Rannala & Yang, 2007), as is apparent if we compare the chronograms in Figure 5. Each was produced using the PL method with the same calibration point based on Endressinia. Tree (a) represents the results obtained from a dataset of combined rbcL and trnL-F sequences only (following Richardson et al., 2004), whereas (b) and (c) represent the results obtained from matrices of multiple plastid markers (see Methods). The standard deviations and 95% confidence intervals for the estimates based on trnL-F and rbcL sequence data only (a) are considerably wider than those based on all the sequence data (b and c). The same is true if we compare PP ranges for age estimates across the recent studies employing BEAST (Table 3): the study of Couvreur et al. (2011) is based on the largest body of sequence data and produced estimates with the narrowest confidence intervals.

Individual genes can cause large overestimates of ages if paralogous copies of genes are analysed (Sanderson, 2003) (for an example in Annonaceae, see Pirie et al., 2007). Even if no such problematic markers have been sampled, different gene loci may better fit different substitution models, and, if unlinked, may even track different underlying phylogenetic trees because of the various processes that might result in nonbifurcating species trees (e.g. hybridization or lateral gene transfer and coalescent stochasticity, the latter particularly pertinent at low taxonomic levels). These issues are not directly addressed by the methods implemented with the program ‘r8s’ of Sanderson (2002b). Using multidivtime or BEAST, however, it is possible to partition multiple genes/loci with different evolutionary characteristics and apply different substitution models to each partition, thus potentially increasing the realism of the model. Furthermore, BEAST can be used to date nodes given a coalescent process with unlinked gene partitions.

Linder et al. (2005) demonstrated that NPRS, PL and multidivtime produce older age estimates with increasing density of taxon sampling, up to a particular level when the differences between age estimates with a given method tail off. This effect was most noticeable for nodes distant from the calibration point. NPRS, still widely used at the time, produced by far the most inconsistent results. Pirie et al. (2005), using as an example a phylogenetic tree of Annonaceae, demonstrated this effect using NPRS and further showed that bootstrap measures, as described above, result in error margins that attribute inappropriate precision to these inconsistent results. It is worth noting that with low smoothing parameter values (as appear to be optimal with Annonaceae plastid data), PL behaves similarly to NPRS and appears to be every bit as sensitive to taxon sampling density (contrary to the conclusions of Linder et al., 2005). The effect can be seen in Figure 1 (particularly in NPRS and PL results) and in Figure 5 (using PL), in which increased taxon sampling (supermatrix compared with spine matrix) results in both older and more precise age estimates. The adequacy of the taxon sampling that has been achieved in any phylogenetic study is difficult to define, as it is a function of current taxonomy and the unknown degree of past extinction (Stace, 2005). Some lineages may contain so few surviving species that it is not possible to increase sampling sufficiently for the estimates to stabilize, and it is not necessarily the case that a stabilized result is also an unbiased one.

Phylogenetic and model uncertainty

Although it is hoped that error caused by topological uncertainty will be minimal in age estimates for highly supported nodes, even these nodes are often subject to a small degree of uncertainty. Such uncertainty will influence the associated branch lengths, as will uncertainty in other parts of the topology. An estimation of this error was incorporated by Sanderson & Doyle (2001) by means of a bootstrapping procedure similar to that described for substitutional noise above (and which therefore also incorporates sampling error), but without applying any topological constraints. Bootstrapping is easy to implement, but, as with its application in assessing clade support, interpretation of the significance of the results is not straightforward (Soltis & Soltis, 2003).

Bayesian methodologies offer a different means of incorporating and assessing topological and other aspects of phylogenetic uncertainty. Parameters of interest (including the topology and branch lengths) are sampled in analyses in proportion to their PP. Summaries of PP distributions have an intuitive interpretation: PP of 0.95 (e.g. for clade support, where 95% of the trees sampled in the MCMC include a certain clade) and 0.95 PP credibility intervals (e.g. for node ages) have a 95% chance of being true given that the model and the priors are correct (Ronquist, 2004). This is clearly a different aspect of phylogenetic uncertainty than that measured by bootstrap resampling, and the process by which it is derived also differs substantially, such that the values of each can be expected to differ considerably. The results of our PL and BEAST analyses (Table 4) illustrate such differences: the BEAST 95% PP intervals are consistently wider than those implied by standard deviations from bootstrapping under PL, despite the fact that the topology was constrained in both analyses.

Reconciling relaxed clocks and absolute time constraints

We have described the uncertainty and potential for error commonly encountered when interpreting the fossil record and using DNA sequences to estimate relative node ages. These factors are rather diverse in their nature and therefore not straightforward to interpret when it comes to bringing the two sources of data together to estimate the absolute ages of clades.

Assessing fossils with molecules

Approaches have been proposed that might identify incorrect associations of fossils with nodes when these associations result in differing age estimates. One approach (Near & Sanderson, 2004) employs a cross-validation procedure to compare the effects of different fossil calibrations and identify those with an exceptionally large impact on results. Another (Rutschmann et al., 2007) analyses separately each of the possible combinations of plausible fossil–node associations to determine which combination is the most internally consistent. In cases in which each fossil, when associated with its correct node, underestimates the age of that node to the same degree, these techniques are likely to be effective at identifying and/or eliminating calibration errors. It should be noted, however, that a correctly assigned but older fossil, which in the conservative approach would largely determine the age of a group, would be identified as suspicious by the method of Near & Sanderson (2004) and would not be included in a most consistent set using the technique of Rutschmann et al. (2007) (Marshall, 2008). It therefore makes little sense to apply these methods to groups such as Annonaceae for which the recognized fossil record is sparse and the chance that any fossil will greatly underestimate the age of the clade to which it is correctly assigned is high. Crucially, the validity of fossil cross-validation approaches also relies on the accuracy of the molecular method, which, given some of the issues dealt with above, may not be safe to assume.

Correcting molecular results using fossils

Conversely, it has been argued that more accurate results might be achieved by using fossils to constrain more nodes in the phylogenetic tree (Soltis et al., 2002; Near & Sanderson, 2004; Sanderson et al., 2004; Anderson, Bremer & Friis, 2005). Minimum age constraints that would not influence the result in an accurately constructed chronogram may restrict the degree to which an inaccurately constructed chronogram underestimates the age of a given node, and presumably also other nodes in the same part of the tree. For this purpose, the onus must be placed on the confidence in the age constraints: there is nothing to be gained in correcting a dubious chronogram with a doubtful constraint.

One potentially serious caveat to this approach has yet to be adequately addressed in discussions of molecular dating: the problem that there are generally many more minimum age constraints available than maximum age constraints, which are inevitably more difficult to justify. Although minimum constraints may serve to reduce the number of erroneously recent age estimates, the dearth of maximum age constraints will fail to eliminate error that results in ages that are too old. If there is inaccuracy in the molecular method, this could lead to systematic bias towards older age estimates (the taxon sampling density effect described above being a particularly unfortunate example in this context).

Using ‘soft bounds’ to reflect further knowledge about age constraints

Minimum and maximum age constraints, as applied and discussed so far, follow logically from the simple interpretation of the presence of organisms in the fossil record. They give results that are straightforward to interpret, but may not represent all aspects of our knowledge, for example, with regard to the completeness and continuity of the fossil record of a given group.

Flexible distributions, termed ‘soft bounds’ (Yang & Rannala, 2006), can be applied in a Bayesian approach (e.g. BEAST, Drummond & Rambaut, 2007; multidivtime, Thorne et al., 1998; MCMCTREE, Yang & Rannala, 2006), whereby calibration points can be assigned prior distributions of different shapes to reflect this kind of knowledge (Ho & Phillips, 2009). An exponential prior might be appropriate for fossil calibrations, as it allows the posterior distribution to be older if influenced by other priors in the analysis, but, if not, will be sampled most frequently at the minimum age. A log-normal distribution might be useful when applying secondary calibrations, whereby the distribution can be set to reflect both the error margins of the original estimate and (as is most likely) the fact that the original analysis was calibrated with a fossil and is thus a minimum age.

Determining the shape of these distributions involves additional assumptions that may be difficult to justify (Donoghue & Benton, 2007; Ho & Phillips, 2009). We might have confidence in assuming that the origin of the eudicots did not greatly precede the first appearance of the subsequently continuous fossil record of tricolpate pollen. However, to translate this confidence into a soft-bounded prior is to state explicitly how great is the discrepancy likely to be. In some cases, it may be possible to estimate the probable impact of sporadic preservation in the fossil record (Marshall, 2008), but, as we have discussed, a number of rather different factors are also implicated which in combination may be effectively impossible to model. A further problem is the inconsistent way in which bounds (soft or otherwise) are currently implemented between different Bayesian applications (such that apparently identical priors may result in different joint prior distributions: Inoue, Donoghue & Yang, 2010; Warnock, Yang & Donoghue, 2012). The priors used by Smith et al. (2010), corresponding to an assumption that the true age extends ∼10–15 Myr further back than that of the fossil ‘in most cases’, should, in any case, be regarded as highly arbitrary: this is something we simply cannot know.

Uncertainty, bias and soft bounds

Soft-bounded constraints have nevertheless been credited as ‘the only framework that can account for uncertainties in fossil calibrations’ (Inoue et al., 2010). We would question the validity of compounding these different sources of uncertainty, particularly in the form of soft bounds that permit ages to both pre- and postdate the putative constraint. A single such prior applied to a molecular dating analysis might describe calibration uncertainty adequately if the shape of the distribution could be justified explicitly (although, as we argue above, this is likely to be impossible in most cases). Combinations of two or more such priors, however, have the potential to behave in ways that do not reflect the real nature of the uncertainty underlying the individual constraints (e.g. that some fossils underestimate true ages to a lesser extent than others), but are instead dictated by the complex interaction between data and assumptions overall. Where further assumptions, such as those regarding patterns of variation in evolutionary rate, are violated, soft-bounded priors will reduce the power of the constraints to correct for bias, and result in confidence intervals that are impossible to interpret in terms of any particular source of uncertainty.

We would suggest that, if the prior age constraints are to accurately reflect known probabilities of particular node ages, it will usually be necessary to abandon soft-bounded prior distributions in favour of uniform distributed priors with hard bounds (however wide these might need to be set). Molecular dating analyses calibrated using fossils should, if they are working correctly, deliver age estimates that are too recent. This, we suggest, is the paradigm within which the results should be interpreted and, in the absence of maximum constraints based on other kinds of data, we should abandon the idea that we are attempting to infer, or even approximate, the true ages of clades. By invoking a conservative interpretation in our analyses of both the node with which a given fossil can be associated (indeed, if any) and the interpretation of that fossil strictly as a minimum age constraint, we can hope to obtain accurate minimum ages. Additional assumptions, in the form of more or less arbitrary estimates of unknown or unknowable factors, compromise the confidence that we can have in these results.

Calibrating and dating the phylogeny of Annonaceae

We apply these principles to estimate the age of clades in Annonaceae. The rate of molecular evolution in Annonaceae deviates from a strict molecular clock, but it is not clear which of the models tested to date (rate autocorrelation or log-normal distribution of rates) might best fit the patterns of rate variation (particularly the potential for abrupt changes in rate) that are observed. Indeed, it is conceivable that neither fits well. Bias resulting from taxon sampling density is clearly an issue for some methods (NPRS; PL), although it is not clear what is the nature of this bias (are densely sampled trees too old, or sparsely sampled ones too young?). As our data matrix comprises DNA sequences from the plastid genome only (and these exhibit consistent phylogenetic signals), it is not necessary to apply methods that incorporate errors caused by either sampling of paralogous sequences or lineage sorting artefacts. We apply PL (assuming rate autocorrelation) and BEAST (assuming log-normal distributed rates): the range of results this produces may represent uncertainty in the underlying model.

We have a small number of fossils that can be unambiguously associated with nodes within the ingroup and among the outgroups (other Magnoliales): these represent minimum age constraints for these nodes and cannot be further interpreted to impose any kind of effective maximum constraints by means of probability distributions. There is no geological evidence that might unambiguously be interpreted to provide maximum age constraints. Broader scale (angiosperm-wide) molecular dating analyses have resulted in age estimates in Magnoliales that are either more recent than the fossil evidence implies directly (e.g. Bell, Soltis & Soltis, 2010: 42–71 or 49–76 Mya for the MRCA of Magnoliaceae and Annonaceae) or are otherwise problematic in terms of the fossil record (e.g. Magallón, 2010; Smith et al., 2010), and cannot be used with confidence as secondary calibrations. The ages of fossils relative to the associated node ages estimated using Endressinia (the oldest constraint for the deepest node) only are illustrated in Figure 6. Comparing the alternative constraints on deeper nodes, which have the greatest impact on the ages of younger nodes, Endressinia provides an older minimum age (112 Mya) for the node connecting Annonaceae and Magnoliaceae than does Archaeanthus (100 Mya), and Futabanthus provides an older minimum age for the crown node of Annonaceae (89 Mya) than do the Maastrichtian seeds (68 Mya).

Clearly, both Archaeanthus and the Maastrichtian seeds must therefore be regarded as relatively uninformative with regard to clade ages, being substantially younger than ages for the relevant nodes based on other fossils. This kind of result is to be expected given the fragmentary nature of the fossil record, and it does not say anything about the confidence we should have in any of our fossil–node associations: these are dictated by shared derived characters. However, as Archaeanthus has almost all the preservable synapomorphies of living Magnoliaceae, and may therefore represent a late offshoot of the magnoliaceous stem lineage or even the crown group (Doyle & Endress, 2010), it is unsurprising that it significantly underestimates the age of the subtending node with which it can be associated.

To obtain age estimates within Annonaceae (Fig. 2), we applied a basal fixed age constraint. This is necessary under most molecular dating methods and, as our constraint is based on the age of a fossil, rather than any form of maximum constraint, it means that all resulting ages must be interpreted as minima (assuming the method in question is working correctly). Either Endressinia or Futabanthus could be used to impose such a constraint, but, as Endressinia is associated with a deeper node, its use requires the sampling of additional extant outgroups that are more distantly related to Annonaceae (such as Myristicaceae or Laurales) in order to include in the molecular tree the node representing the MRCA of Annonaceae and Endressinia. The age for the Annonaceae crown node as estimated using Endressinia (86 Mya) is similar to, but slightly more recent than, the minimum age of the node implied by Futabanthus (89 Mya). The use of Futabanthus therefore provides the oldest reliable age estimates within Annonaceae. Endressinia can be used as a fixed calibration, with Futabanthus as an additional minimum constraint (as performed here). However, our results, and those of Su & Saunders (2009), who obtained similar results using Futabanthus in combination with the more recent Archaeanthus in place of Endressinia, suggest that age estimates within Annonaceae are unlikely to differ meaningfully if Futabanthus is used instead as a single fixed calibration and more distant outgroups are omitted from the phylogenetic analyses altogether.

The Eocene leaf fossil Duguetia sp. could provide a minimum age constraint for the crown node of Duguetia if the presence of peltate scales is interpreted as a synapomorphy linking it with Neotropical Duguetia or, alternatively, as a minimum age for the stem node of Duguetia s.l. (including ‘Pachypodanthium’) if peltate and stellate trichomes are not distinguished. Given the level of uncertainty surrounding its node association, we did not include this fossil in the final combined analysis (Fig. 2, Tables 3 and 4). Should its position be clarified, it could have some impact locally in age estimates for Duguetia and related genera, such as Fusaea, and potentially in reducing the error margins associated with these estimates. However, its impact on age estimates for deeper nodes is likely to be marginal.

Uses of dated phylogenetic trees and future prospects

Dated phylogenetic trees open up the possibility of testing whether diversifications in groups occurred within the time slices predicted by biogeographical hypotheses. Molecular dating techniques have been used in Annonaceae to test a number of such hypotheses, for example Gondwanan vicariance (Richardson et al., 2004; Pirie et al., 2006), the geographical origin and dispersal to Asia of Anaxagorea (Scharaschkin & Doyle, 2005), diversification associated with the Andean orogeny (Pirie et al., 2006) or closure of the isthmus at Panama (Erkens et al., 2007), vicariance and dispersal scenarios in tropical Africa (Couvreur et al., 2008a) and South-East Asia (Su & Saunders, 2009), and the speed of apparently rapid species radiations (Erkens, Chatrou & Couvreur, 2012).

The most appropriate interpretation of molecular dating results depends on how the results are to be used. Tests of specific hypotheses are more stringent if the methodology used tends to err in favour of rejection. Assuming that the molecular method is performing correctly, age estimates based on fossil minimum constraints will err towards underestimating the true ages. These data thus represent a stringent test of a hypothesis for which the time frame is recent relative to the ages of the relevant nodes. One example would be the question of whether Pleistocene climatic changes have driven diversification in clades. If the nodes in question are inferred to be pre-Pleistocene, even when the method is likely to have resulted in an underestimation of true ages, the rejection of the hypothesis is robust. Scenarios involving more ancient events, such as Gondwanan vicariance as an explanation for the pantropical distribution of Annonaceae, must be assessed somewhat differently, as minimum age estimates cannot deliver conclusive rejections of hypotheses for which they are too recent. Instead, one might test just how old the relevant nodes would have to be to support the hypothesis, rescale the chronogram (after the analyses) to conform to such ages, and see whether this implies geologically unlikely or impossible ages deeper in the tree.

Based on a survey by Benton & Ayala (2003), agreement between molecular estimates of ages and the fossil record has increased in many groups as fossils have been found and correctly interpreted and molecular dating methods have become more consistent and realistic. This suggests that the stream of studies indicating clade ages outrageously at odds with the fossil record has decreased, although there have been conspicuous recent exceptions, such as analyses indicating a Permian or Triassic origin of crown group angiosperms (Magallón, 2010; Smith et al., 2010). It is important to bear in mind that the most valuable application of molecular dating methods is in obtaining age estimates for clades for which there is no direct evidence from the fossil record and therefore no means to test the result directly. It is therefore imperative that the methods continue to improve, and empirical studies will continue to be valuable in identifying sources of inconsistency and bias.

Future work will no doubt see improved methods applied to the estimation of the age of clades in Annonaceae and greater quantities of data. As we have shown here, additional sequence data reduce the error margins associated with molecular dating methods. However, more stringent tests of the underlying assumptions of the methods are necessary: in particular, the fitting of models assuming rate autocorrelation, log-normal rate distributions or alternatives such as random local clocks to cases such as Annonaceae, in which large shifts in rate are apparent, should be investigated. Finally, the bias inherent in applying only minimum age constraints in chronogram calibration needs to be addressed. Bayesian methodologies, such as BEAST, are rapidly being improved, and it will be interesting to see whether these result in more consistent estimates when applied to different datasets. The greatest improvements, however, are likely to be found in a better interpretation of the fossil record. A comprehensive study of Annonaceae fossils, e.g. a closer examination of seeds from the London Clay for characters in addition to the type of endosperm ruminations (cf. Van Setten & Koek-Noorman, 1992), would be extremely valuable.


  1. Top of page
  2. Abstract
  9. Supporting Information

We thank Lars Chatrou and Hervé Sauquet for providing the molecular data analysed here, and Masamichi Takahashi for kindly providing images of Futabanthus. Helpful critique of the manuscript was provided by Alexandre Antonelli and an anonymous reviewer. Thomas Couvreur, Anita Lendel and Aelys Humphreys are also thanked for constructive comments on earlier drafts, Richard Saunders for discussion of Cyathocalyx, and Mike Sanderson and Brian Moore for discussion of theoretical points.


  1. Top of page
  2. Abstract
  9. Supporting Information
  • Akaike H. 1974. A new look at the statistical model identification. IEEE Transactions on Automatic Control 19: 716723.
  • Anderson CL, Bremer K, Friis EM. 2005. Dating phylogenetically basal eudicots using rbcL sequences and multiple fossil reference points. American Journal of Botany 92: 17371748.
  • Baldwin BG, Sanderson MJ. 1998. Age and rate of diversification of the Hawaiian silversword alliance (Compositae). Proceedings of the National Academy of Sciences of the United States of America 95: 94029406.
  • Bell CD, Donoghue MJ. 2005. Dating the Dipsacales: comparing models, genes, and evolutionary implications. American Journal of Botany 92: 284296.
  • Bell CD, Soltis DE, Soltis PS. 2010. The age and diversification of the angiosperms re-revisited. American Journal of Botany 97: 12961303.
  • Benton MJ, Ayala FJ. 2003. Dating the tree of life. Science 300: 16981700.
  • Berry EW. 1916. The lower Eocene floras of southeastern North America. U.S. Geological Survey Professional Paper 91: 1481.
  • Berry EW. 1918. The fossil higher plants from the Canal Zone. Bulletin of the U.S. National Museum 103: 1544.
  • Berry EW. 1920. Miocene fossil plants from northern Peru. Proceedings of the U.S. National Museum 55: 279294.
  • Berry EW. 1922a. Tertiary fossil plants from Costa Rica. Proceedings of the U.S. National Museum 59: 169185.
  • Berry EW. 1922b. Tertiary fossil plants from Venezuela. Proceedings of the U.S. National Museum 59: 553579.
  • Berry EW. 1923. Miocene plants from southern Mexico. Proceedings of the U.S. National Museum 62: 127.
  • Berry EW. 1937. Tertiary plants from Venezuela. Proceedings of the U.S. National Museum 83: 335360.
  • Berry EW. 1941. Additions to the Wilcox flora from Kentucky and Texas. U.S. Geological Survey Professional Paper 193-E: 8399.
  • Bonde SD. 1993. Unonaspermum corneri gen. et sp. nov., an annonaceous seed from the Deccan intertrappean beds of India. Journal of the Indian Botanical Society 72: 251253.
  • Boureau E. 1950. Étude paléoxylologique du Sahara (XII). Sur un Annonoxylon striatum n. g., n. sp., des couches de Tamaguilel (Sahara soudanais). Bulletin de la Société Géologique de France 5: 393397.
  • Boureau E. 1954. Étude paléoxylologique du Sahara (XX). Sur un Annonoxylon edengense n. sp., des couches post-éocènes du Sud-Ouest de l'Adrar Tiguirirt (Sahara Soudanais). Bulletin du Muséum National d'Histoire Naturelle, Série 2 26: 286291.
  • Bousquet J, Strauss SH, Doerksen AH, Price RA. 1992. Extensive variation in evolutionary rate of rbcL gene sequences among seed plants. Proceedings of the National Academy of Sciences of the United States of America 89: 78447848.
  • Bremer K. 2000. Early Cretaceous lineages of monocot flowering plants. Proceedings of the National Academy of Sciences of the United States of America 97: 47074711.
  • Cantino PD, Doyle JA, Graham SW, Judd WS, Olmstead RG, Soltis DE, Soltis PS, Donoghue MJ. 2007. Towards a phylogenetic nomenclature of Tracheophyta. Taxon 56: 822846.
  • Chandler MEJ. 1978. Supplement to the Lower Tertiary Floras of Southern England, part 5. Tertiary Research Special Paper 4: 147.
  • Chaney RW, Sanborn EI. 1933. The Goshen flora of west central Oregon. Carnegie Institution of Washington Publication 439: 1103.
  • Chatrou LW. 1998. Changing genera: systematic studies in Neotropical and West African Annonaceae. PhD Thesis, Utrecht University.
  • Chatrou LW, Koek-Noorman J, Maas PJM. 2000. Studies in Annonaceae XXXVI. The Duguetia alliance: where the ways part. Annals of the Missouri Botanical Garden 87: 234245.
  • Chatrou LW, Pirie MD, Erkens RHJ, Couvreur TLP, Neubig KM, Abbott JR, Mols JB, Maas JW, Saunders RMK, Chase MW. 2012. A new higher-level classification of the pantropical flowering plant family Annonaceae informed by molecular phylogenetics. Botanical Journal of the Linnean Society 169: 540.
  • Chesters KIM. 1955. Some plant remains from the Upper Cretaceous and Tertiary of West Africa. Annals and Magazine of Natural History 12: 489504.
  • Cody S, Richardson JE, Rull V, Ellis C, Pennington RT. 2010. The Great American Biotic Interchange revisited. Ecography 33: 326332.
  • Collinson ME. 1983. Fossil plants of the London Clay. London: Palaeontological Association.
  • Corner EJH. 1976. The seeds of the dicotyledons. Cambridge: Cambridge University Press.
  • Couvreur T, Chatrou L, Sosef M, Richardson J. 2008a. Molecular phylogenetics reveal multiple Tertiary vicariance origins of the African rain forest trees. BMC Biology 6: 54.
  • Couvreur TLP, Pirie MD, Chatrou LW, Saunders RMK, Su YCF, Richardson JE, Erkens RHJ. 2011. Early evolutionary history of the flowering plant family Annonaceae: steady diversification and boreotropical geodispersal. Journal of Biogeography 38: 664680.
  • Couvreur TLP, Richardson JE, Sosef MSM, Erkens RHJ, Chatrou LW. 2008b. Evolution of syncarpy and other morphological characters in African Annonaceae: a posterior mapping approach. Molecular Phylogenetics and Evolution 47: 302318.
  • Crane PR, Friis EM, Pedersen KR. 1994. Palaeobotanical evidence on the early radiation of magnoliid angiosperms. Plant Systematics and Evolution Supplement 8: 5172.
  • Crawley M. 2001. Angiosperm woods from British Lower Cretaceous and Palaeogene deposits. Special Papers in Palaeontology 66: 1100.
  • Crepet WL, Nixon KC, Gandolfo MA. 2004. Fossil evidence and phylogeny: the age of major angiosperm clades based on mesofossil and macrofossil evidence from Cretaceous deposits. American Journal of Botany 91: 16661682.
  • Dilcher DL. 1971. A revision of the Eocene flora of southeastern North America. The Palaeobotanist 20: 718.
  • Dilcher DL. 1974. Approaches to the identification of angiosperm leaf remains. Botanical Review 40: 1157.
  • Dilcher DL, Crane PR. 1984. Archaeanthus: an early angiosperm from the Cenomanian of the Western Interior of North America. Annals of the Missouri Botanical Garden 71: 351383.
  • Dilcher DL, Lott TA. 2005. A middle Eocene fossil plant assemblage (Powers Clay Pit) from western Tennessee. Bulletin of the Florida Museum of Natural History 45: 143.
  • Donoghue PCJ, Benton MJ. 2007. Rocks and clocks: calibrating the Tree of Life using fossils and molecules. Trends in Ecology and Evolution 22: 424.
  • Doweld AB, Shevyryova NA. 1998. Carpology, seed anatomy and taxonomic relationships of Galbulimima (Himantandraceae). Annals of Botany 81: 337347.
  • Doyle JA. 1992. Revised palynological correlations of the lower Potomac Group (USA) and the Cocobeach sequence of Gabon (Barremian–Aptian). Cretaceous Research 13: 337349.
  • Doyle JA, Bygrave P, Le Thomas A. 2000. Implications of molecular data for pollen evolution in Annonaceae. In: Harley MM, Morton CM, Blackmore S, eds. Pollen and spores: morphology and biology. Kew: Royal Botanic Gardens, 259284.
  • Doyle JA, Donoghue MJ. 1993. Phylogenies and angiosperm diversification. Paleobiology 19: 141167.
  • Doyle JA, Endress, PK. 2000. Morphological phylogenetic analysis of basal angiosperms: comparison and combination with molecular data. International Journal of Plant Sciences (6 Supplement) 161: S121S153.
  • Doyle JA, Endress PK. 2010. Integrating Early Cretaceous fossils into the phylogeny of living angiosperms: Magnoliidae and eudicots. Journal of Systematics and Evolution 48: 135.
  • Doyle JA, Le Thomas A. 1996. Phylogenetic analysis and character evolution in Annonaceae. Bulletin du Muséum National d'Histoire Naturelle, Section B, Adansonia 18: 279334.
  • Doyle JA, Le Thomas A. 2012. Evolution and phylogenetic significance of pollen in Annonaceae. Botanical Journal of the Linnean Society 169: 190221.
  • Doyle JA, Sauquet H, Scharaschkin T, Le Thomas A. 2004. Phylogeny, molecular and fossil dating, and biogeographic history of Annonaceae and Myristicaceae (Magnoliales). International Journal of Plant Sciences 165: S55S67.
  • Drummond AJ, Ho SYW, Phillips MJ, Rambaut A. 2006. Relaxed phylogenetics and dating with confidence. PLoS 4: e88.
  • Drummond AJ, Rambaut A. 2006. BEAST. 1.6.1 edn. Available at: (Last accessed February 2012).
  • Drummond AJ, Rambaut A. 2007. BEAST: Bayesian evolutionary analysis by sampling trees. BMC Evolutionary Biology 7: 214.
  • Drummond AJ, Suchard MA. 2010. Bayesian random local clocks, or one rate to rule them all. BMC Biology 8: 114.
  • Ellis B, Daly DC, Hickey LJ, Johnson KR, Mitchell JD, Wilf P, Wing SL. 2009. Manual of leaf architecture. Ithaca, NY: Cornell University Press.
  • Elsik WC. 1974. Characteristic Eocene palynomorphs in the Gulf Coast, U.S.A. Palaeontographica Abteilung B 149: 90111.
  • Endress PK. 1980. The reproductive structures and systematic position of the Austrobaileyaceae. Botanische Jahrbücher für Systematik 101: 393433.
  • Endress PK. 1984. The role of inner staminodes in the floral display of some relic Magnoliales. Plant Systematics and Evolution 146: 269282.
  • Endress PK, Doyle JA. 2009. Reconstructing the ancestral angiosperm flower and its initial specializations. American Journal of Botany 96: 2266.
  • Eriksson O, Friis EM, Löfgren P. 2000. Seed size, fruit size, and dispersal systems in angiosperms from the Early Cretaceous to the Late Tertiary. American Naturalist 156: 4758.
  • Erkens RHJ. 2007. From morphological nightmare to molecular conundrum. Phylogenetic, evolutionary and taxonomic studies on Guatteria (Annonaceae). PhD thesis, Utrecht University .
  • Erkens RHJ, Chatrou LW, Couvreur TLP. 2012. Radiations and key innovations in an early branching angiosperm lineage (Annonaceae; Magnoliales). Botanical Journal of the Linnean Society 169: 117134.
  • Erkens RHJ, Chatrou LW, Maas JW, van der Niet T, Savolainen V. 2007. A rapid diversification of rainforest trees (Guatteria; Annonaceae) following dispersal from Central into South America. Molecular Phylogenetics and Evolution 44: 399411.
  • Erkens RHJ, Maas JW, Couvreur TLP. 2009. From Africa via Europe to South America: migrational route of a species-rich genus of Neotropical lowland rain forest trees (Guatteria, Annonaceae). Journal of Biogeography 36: 23382352.
  • Forest F. 2009. Calibrating the Tree of Life: fossils, molecules and evolutionary timescales. Annals of Botany 104: 789794.
  • Fries RE. 1959. Annonaceae. In: Engler A, Prantl K, eds. Die natürlichen Pflanzenfamilien, ed. 2, Band 17a II. Berlin: Duncker & Humblot, 1171.
  • Gaut BS. 1998. Molecular clocks and nucleotide substitution rates in higher plants. In: Hecht MK, MacIntyre RJ, Clegg MT, eds. Evolutionary biology. New York: Plenum Press, 93120.
  • Gradstein FM, Ogg JG, Smith AG, Agterberg FP, Bleeker W, Cooper RA, Davydov V, Gibbard P, Hinnov L, House MR, Lourens L, Luterbacher HP, McArthur J, Melchin MJ, Robb LJ, Shergold J, Villeneuve M, Wardlaw BR, Ali J, Brinkhuis H, Hilgen FJ, Hooker J, Howarth RJ, Knoll AH, Laskar J, Monechi S, Powell J, Plumb KA, Raffi I, Röhl U, Sanfilippo A, Schmitz B, Shackleton NJ, Shields GA, Strauss H, Dam JV, Veizer J, Kolfschoten TV, Wilson D, eds. 2004. A geologic time scale 2004. Cambridge: Cambridge University Press.
  • Gröcke DR, Luduigson GA, Witzke BL, Robinson SA, Joeckel RM, Ufnar DF, Ravn RL. 2006. Recognizing the Albian–Cenomanian (OAE1d) sequence boundary using plant carbon isotopes: Dakota Formation Western Interior Basin, USA. Geology 34: 193196.
  • Heads M. 2005. Dating nodes on molecular phylogenies: a critique of molecular biogeography. Cladistics 21: 6278.
  • Heimhofer U, Hochuli PA. 2010. Early Cretaceous angiosperm pollen from a low-latitude succession (Araripe Basin, NE Brazil). Review of Palaeobotany and Palynology 161: 105126.
  • Hickey LJ. 1973. Classification of the architecture of dicotyledonous leaves. American Journal of Botany 60: 1733.
  • Hickey LJ, Wolfe JA. 1975. The bases of angiosperm phylogeny: vegetative morphology. Annals of the Missouri Botanical Garden 62: 538589.
  • Ho SYW. 2009. An examination of phylogenetic models of substitution rate variation among lineages. Biology Letters 5: 421424.
  • Ho SYW, Phillips MJ. 2009. Accounting for calibration uncertainty in phylogenetic estimation of evolutionary divergence times. Systematic Biology 58: 367380.
  • Hollick A. 1906. The Cretaceous flora of southern New York and New England. U.S. Geological Survey Monograph 50: 1219.
  • Hughes NF. 1994. The enigma of angiosperm origins. Cambridge: Cambridge University Press.
  • Inoue J, Donoghue PCJ, Yang Z. 2010. The impact of the representation of fossil calibrations on Bayesian estimation of species divergence times. Systematic Biology 59: 7489.
  • Jovet-Ast S. 1942. Recherches sur les Anonacées d'Indochine. Anatomie foliaire. Répartition géographique. Mémoires du Muséum National d'Histoire Naturelle 16: 125308.
  • Kimura M. 1968. Evolutionary rate at the molecular level. Nature 217: 624626.
  • Klucking EP. 1986. Leaf venation patterns. Vol. 1. Annonaceae. J. Cramer: Berlin.
  • Kvaček J, Eklund H. 2003. A report on newly recovered reproductive structures from the Cenomanian of Bohemia (Central Europe). International Journal of Plant Sciences 164: 10211039.
  • Lakhanpal RK. 1969. The Rujada flora of west central Oregon. University of California Publications in Geological Sciences 35: 165.
  • Le Thomas A. 1969. Annonacées. In: Aubréville A, ed. Flore du Gabon. Paris: Muséum National d'Histoire Naturelle, 1371.
  • Lemey P, Rambaut A, Drummond AJ, Suchard MA. 2009. Bayesian phylogeography finds its roots. PLoS Computational Biology 5: e1000520.
  • Lepage T, Bryant D, Philippe H, Lartillot N. 2007. A general comparison of relaxed molecular clock models. Molecular Biology and Evolution 24: 26692680.
  • Linder HP, Hardy CR, Rutschmann F. 2005. Taxon sampling effects in molecular clock dating: an example from the African Restionaceae. Molecular Phylogenetics and Evolution 35: 569.
  • Maas PJM, Westra LYT, Chatrou LW. 2003. Flora neotropica monograph 88: Duguetia. New York: New York Botanical Garden.
  • Maddison DR, Maddison WP. 2003. MacClade 4: analysis of phylogeny and character evolution. version 4.06. Sunderland, MA: Sinauer Associates.
  • Magallón S. 2004. Dating lineages: molecular and paleontological approaches to the temporal framework of clades. International Journal of Plant Sciences 165: S7S21.
  • Magallón S. 2010. Using fossils to break long branches in molecular dating: a comparison of relaxed clocks applied to the origin of angiosperms. Systematic Biology 59: 384399.
  • Magallón S, Sanderson MJ. 2001. Absolute diversification rates in angiosperm clades. Evolution 55: 17621780.
  • Margoliash E. 1963. Primary structure and evolution of cytochrome C. Proceedings of the National Academy of Sciences of the United States of America 50: 672.
  • Marshall CR. 2008. A simple method for bracketing absolute divergence times on molecular phylogenies using multiple fossil calibration points. American Naturalist 171: 726742.
  • Mohr BAR, Bernardes-de-Oliveira MEC. 2004. Endressinia brasiliana, a magnolialean angiosperm from the Lower Cretaceous Crato Formation (Brazil). International Journal of Plant Sciences 165: 11211133.
  • Mols JB. 2004. From Miliusa to Miliuseae to miliusoid: identifying clades in Asian Annonaceae. PhD Thesis, University of Leiden.
  • Mols JB, Gravendeel B, Chatrou LW, Pirie MD, Bygrave PC, Chase MW, Kessler PJA. 2004. Identifying clades in Asian Annonaceae: monophyletic genera in the polyphyletic Miliuseae. American Journal of Botany 91: 590600.
  • Muller J. 1981. Fossil pollen records of extant angiosperms. Botanical Review 47: 1142.
  • Near TJ, Sanderson MJ. 2004. Assessing the quality of molecular divergence time estimates by fossil calibrations and fossil-based model selection. Philosophical Transactions of the Royal Society B: Biological Sciences 359: 14771483.
  • Pirie MD, Chatrou LW, Erkens RHJ, Maas JW, Van der Niet T, Mols JB, Richardson JE. 2005. Phylogeny reconstruction and molecular dating in four Neotropical genera of Annonaceae: the effect of taxon sampling in age estimations. In: Bakker FT, Chatrou LW, Gravendeel B, Pelser PB, eds. Plant species-level systematics: new perspectives on pattern and process. Regnum Vegetabile 143. Liechtenstein: A. R. G. Gantner Verlag, 149174.
  • Pirie MD, Chatrou LW, Mols JB, Erkens RHJ, Oosterhof J. 2006. ‘Andean-centred’ genera in the short-branch clade of Annonaceae: testing biogeographic hypotheses using phylogeny reconstruction and molecular dating. Journal of Biogeography 33: 3146.
  • Pirie MD, Humphreys AM, Galley C, Barker NP, Verboom GA, Orlovich D, Draffin SJ, Lloyd K, Baeza CM, Negritto M, Ruiz E, Cota Sanchez JH, Reimer E, Linder HP. 2008. A novel supermatrix approach improves resolution of phylogenetic relationships in a comprehensive sample of danthonioid grasses. Molecular Phylogenetics and Evolution 48: 11061119.
  • Pirie MD, Vargas MPB, Botermans M, Bakker FT, Chatrou LW. 2007. Ancient paralogy in the cpDNA trnL-F region in Annonaceae: implications for plant molecular systematics. American Journal of Botany 94: 10031016.
  • Posada D, Crandall KA. 1998. Modeltest: testing the model of DNA substitution. Bioinformatics 14: 817818.
  • Price JP, Clague DA. 2002. How old is the Hawaiian biota? Geology and phylogeny suggest recent divergence. Proceedings of the Royal Society of London, Biological Sciences 269: 24292435.
  • Rambaut A, Drummond AJ. 2007. Tracer. 1.4 edn. Available at: (Last accessed February 2012).
  • Rannala B, Yang Z. 2007. Inferring speciation times under an episodic molecular clock. Systematic Biology 56: 453466.
  • Reid EM, Chandler MEJ. 1933. The London Clay flora. London: British Museum (Natural History).
  • Reisz RR, Muller J. 2004. Molecular timescales and the fossil record: a paleontological perspective. Trends in Genetics 20: 237241.
  • Renner SS, Grimm GW, Schneeweiss GM, Stuessy TF, Ricklefs RE. 2008. Rooting and dating maples (Acer) with an uncorrelated-rates molecular clock: implications for North American/Asian disjunctions. Systematic Biology 57: 795808.
  • Richardson JE, Chatrou LW, Mols JB, Erkens RHJ, Pirie MD. 2004. Historical biogeography of two cosmopolitan families of flowering plants: Annonaceae and Rhamnaceae. Philosophical Transactions of the Royal Society B: Biological Sciences 359: 14951508.
  • Richardson JE, Pennington RT, Pennington TD, Hollingsworth PM. 2001a. Rapid diversification of a species-rich genus of Neotropical rainforest trees. Science 293: 22422245.
  • Richardson JE, Weitz FM, Fay MF, Cronk QCB, Linder HP, Reeves G, Chase MW. 2001b. Rapid and recent origin of species richness in the Cape flora of South Africa. Nature 412: 181183.
  • Ronquist F. 2004. Bayesian inference of character evolution. Trends in Ecology and Evolution 19: 475481.
  • Roth JL. 1981. Epidermal studies in the Annonaceae and related families. PhD Thesis, Indiana University.
  • Rutschmann F, Eriksson T, Salim KA, Conti E. 2007. Assessing calibration uncertainty in molecular dating: the assignment of fossils to alternative calibration points. Systematic Biology 56: 591608.
  • Salard-Cheboldaeff M. 1978. Sur la palynoflore maestrichtienne et tertiaire du bassin sédimentaire littoral du Cameroun. Pollen et Spores 20: 215260.
  • Sanborn EI. 1935. The Comstock flora of west central Oregon. Carnegie Institution of Washington Publication 465: 128.
  • Sanderson MJ. 1997. A nonparametric approach to estimating divergence times in the absence of rate constancy. Molecular Biology and Evolution 14: 12181231.
  • Sanderson MJ. 2002a. Estimating absolute rates of molecular evolution and divergence times: a penalized likelihood approach. Molecular Biology and Evolution 19: 101109.
  • Sanderson MJ. 2002b. R8s version 1.71, 1.7 edn. Distributed by the author. Davis, CA: Section of Evolution and Ecology, University of California.
  • Sanderson MJ. 2003. Molecular data from 27 proteins do not support a Precambrian origin of land plants. American Journal of Botany 90: 954956.
  • Sanderson MJ, Doyle JA. 2001. Sources of error and confidence intervals in estimating the age of angiosperms from rbcL and 18S rDNA data. American Journal of Botany 88: 14991516.
  • Sanderson MJ, Thorne JL, Wikström N, Bremer K. 2004. Molecular evidence on plant divergence times. American Journal of Botany 91: 16561665.
  • Sauquet H, Doyle JA, Scharaschkin T, Borsch T, Hilu K, Chatrou LW, Le Thomas A. 2003. Phylogenetic analysis of Magnoliales and Myristicaceae based on multiple data sets: implications for character evolution. Botanical Journal of the Linnean Society 142: 125186.
  • Scharaschkin T, Doyle JA. 2005. Phylogeny and historical biogeography of Anaxagorea (Annonaceae) using morphology and non-coding chloroplast sequence data. Systematic Botany 30: 712735.
  • Scharaschkin T, Doyle JA. 2006. Character evolution in Anaxagorea (Annonaceae). American Journal of Botany 93: 3654.
  • Smith SA, Beaulieu JM, Donoghue MJ. 2010. An uncorrelated relaxed-clock analysis suggests an earlier origin for flowering plants. Proceedings of the National Academy of Sciences of the United States of America 107: 58975902.
  • Sole de Porta N. 1971. Algunos géneros nuevos de polen precedentes de la Formación Guaduas (Maastrichtiense–Paleoceno) de Colombia. Studia Geologica, Salamanca 2: 133143.
  • Soltis PS, Soltis DE. 2003. Applying the bootstrap in phylogeny reconstruction. Statistical Science 18: 256267.
  • Soltis PS, Soltis DE, Savolainen V, Crane PR, Barraclough TG. 2002. Rate heterogeneity among lineages of tracheophytes: integration of molecular and fossil data and evidence for molecular living fossils. Proceedings of the National Academy of Sciences of the United States of America 99: 44304435.
  • Stace CA. 2005. Plant taxonomy and biosystematics—does DNA provide all the answers? Taxon 54: 999.
  • Su Y, Saunders R. 2009. Evolutionary divergence times in the Annonaceae: evidence of a late Miocene origin of Pseuduvaria in Sundaland with subsequent diversification in New Guinea. BMC Evolutionary Biology 9: 153.
  • Surveswaran S, Wang RJ, Su YCF, Saunders RMK. 2010. Generic delimitation and historical biogeography in the early-divergent ‘ambavioid’ lineage of Annonaceae: Cananga, Cyathocalyx and Drepananthus. Taxon 59: 17211734.
  • Swofford DL. 2002. PAUP*: phylogenetic analysis using parsimony (*and other methods), version 4. Sunderland, MA: Sinauer Associates.
  • Takahashi M, Friis EM, Crane PR. 2008. Floral evidence of Annonaceae from the Late Cretaceous of Japan. International Journal of Plant Sciences 169: 908917.
  • Thorne JL, Kishino H. 2002. Divergence time and evolutionary rate estimation with multilocus data. Systematic Biology 51: 689702.
  • Thorne JL, Kishino H, Painter IS. 1998. Estimating the rate of evolution of the rate of molecular evolution. Molecular Biology and Evolution 15: 16471657.
  • Tsou C-H, Fu Y-L. 2002. Tetrad pollen formation in Annona (Annonaceae): proexine formation and binding mechanism. American Journal of Botany 89: 734747.
  • Upchurch GR, Wolfe JA. 1987. Mid-Cretaceous to Early Tertiary vegetation and climate: evidence from fossil leaves and woods. In: Friis EM, Chaloner WG, Crane PR, eds. The origins of angiosperms and their biological consequences. Cambridge: Cambridge University Press, 75105.
  • Van Setten AK, Koek-Noorman J. 1986. Studies in Annonaceae. VI. A leaf anatomical survey of genera of Annonaceae in the Neotropics. Botanische Jahrbücher für Systematik 108: 1750.
  • Van Setten AK, Koek-Noorman J. 1992. Studies in Annonaceae. XVII. Fruits and seeds of Annonaceae: morphology and its significance for classification. Bibliotheca Botanica 142: 1101.
  • Vander Wyk RW, Canright JE. 1956. The anatomy and relationships of the Annonaceae. Tropical Woods 104: 124.
  • Walker JM. 1971. Pollen morphology, phytogeography and phylogeny of the Annonaceae. Contributions from the Gray Herbarium 202: 1133.
  • Wang RJ. 2004. Systematics and phylogeny of Cyathocalyx (Annonaceae). PhD Thesis, University of Hong Kong.
  • Ward JV, Doyle JA, Hotton CL. 1989. Probable granular magnoliid angiosperm pollen from the Early Cretaceous. Pollen et Spores 33: 101120.
  • Warnock RCM, Yang Z, Donoghue PCJ. 2012. Exploring uncertainty in the calibration of the molecular clock. Biology Letters 8: 112114.
  • Wheeler EA, Manchester SR. 2002. Woods of the middle Eocene Nut Beds flora, Clarno Formation, Oregon, USA. IAWA Journal Supplement 3: 1188.
  • Wiens JJ, Fetzner JW, Parkinson CL, Reeder TW. 2005. Hylid frog phylogeny and sampling strategies for speciose clades. Systematic Biology 54: 778807.
  • Wikström N, Savolainen V, Chase MW. 2001. Evolution of the angiosperms: calibrating the family tree. Proceedings of the Royal Society of London, Biological Sciences 1482: 22112220.
  • Wing SL, Tiffney BH. 1987. Interactions of angiosperms and herbivorous tetrapods through time. In: Friis EM, Chaloner WG, Crane PR, eds. The origins of angiosperms and their biological consequences. Cambridge: Cambridge University Press, 203224.
  • Wolfe JA. 1977. Paleogene floras from the Gulf of Alaska region. U.S. Geological Survey Professional Paper 997: 1108.
  • Yang Z, Rannala B. 2006. Bayesian estimation of species divergence times under a molecular clock using multiple fossil calibrations with soft bounds. Molecular Biology and Evolution 23: 212226.
  • Zhou L, Su YCF, Chalermglin P, Saunders RMK. 2010. Molecular phylogenetics of Uvaria (Annonaceae): relationships with Balonga, Dasoclema and Australian species of Melodorum. Botanical Journal of the Linnean Society 163: 3343.
  • Zhou L, Su YCF, Thomas DC, Saunders RMK. 2012. ‘Out-of-Africa’ dispersal of tropical floras during the Miocene climatic optimum: evidence from Uvaria (Annonaceae). Journal of Biogeography 39: 322335.
  • Zuckerkandl H, Pauling L. 1965. Evolutionary divergence and convergence in proteins. In: Bryson G, Vogel HJ, eds. Evolving genes and proteins. New York: Academic Press, 97166.

Supporting Information

  1. Top of page
  2. Abstract
  9. Supporting Information

Appendix S1. Chronogram inferred using the supermatrix under BEAST (assuming log-normal distributed rates). The topology is identical to that of Figures 2 and 3, but with the (abbreviated) names of all terminal taxa included. Accession details are documented in Chatrou et al. (2012).

BOJ_1234_sm_Appendix1.pdf1229KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.