Laurasian migration explains Gondwanan disjunctions:Evidence from MalpighiaceaeCharles C. Davis†‡, Charles D. Bell§, Sarah Mathews¶, and Michael J. Donoghue§†Department of Organismic and Evolutionary Biology, Harvard University Herbaria, 22 Divinity Avenue, Cambridge, MA 02138;§Department of Ecology andEvolutionary Biology, Yale University, P.O. Box 208106, New Haven, CT 06520; and¶Division of Biological Sciences, 226 Tucker Hall, University of Missouri,Columbia, MO 65211Communicated by Peter H. Raven, Missouri Botanical Garden, St. Louis, MO, March 25, 2002 (received for review November 1, 2001)Explanations for biogeographic disjunctions involving SouthAmerica and Africa typically invoke vicariance of western Gond-wanan biotas or long distance dispersal. These hypotheses areproblematical because many groups originated and diversifiedwell after the last known connection between Africa and SouthAmerica (⬇105 million years ago), and it is unlikely that ‘‘sweep-stakes’’ dispersal accounts for many of these disjunctions. Phylo-genetic analyses of the angiosperm clade Malpighiaceae, com-bined with fossil evidence and molecular divergence-timeestimates, suggest an alternative hypothesis to account for suchdistributions. We propose that Malpighiaceae originated in north-ern South America, and that members of several clades repeatedlymigrated into North America and subsequently moved via NorthAtlantic land connections into the Old World during episodesstarting in the Eocene, when climates supported tropical forests.This Laurasian migration route may explain many other extantlineages that exhibit western Gondwanan distributions.Malpighiaceae contains trees, shrubs, and vines that aredistributed widely in tropical and subtropical forests andsavannas of the Old and New Worlds (ref. 1; Fig. 1). Approxi-mately 85% of the species occur in the New World where theyare pollinated by specialized oil-collecting bees that are absentfrom the Old World. This distribution has invited the develop-ment of alternative theories for the origin and diversification ofMalpighiaceae. Vogel (2) proposed the ‘‘Gondwanian abori-gine’’ hypothesis, in which the current distribution resulted fromthe break-up of western Gondwana (the supercontinent com-prising Africa and South America). This hypothesis implies thatthe Malpighiaceae originated before western Gondwana divided[i.e., ⬇105 mya (3)], as did several included lineages with disjunctdistributions between the New and Old Worlds. In contrast,Anderson (1) and others (4) have favored an ‘‘American colo-nist’’ hypothesis, according to which Malpighiaceae originated innorthern South America, in isolation from Africa, and dispersedeastward across the Atlantic at least twice. This hypothesispredicts that the ages of Malpighiaceae and of divergencesbetween New World and Old World lineages are younger thanthe last known land connection between South America andAfrica.To elucidate the biogeographic history of Malpighiaceae, weused maximum likelihood to estimate the phylogeny for thegroup by using DNA nucleotide data from chloroplast ndhF andnuclear phytochrome C (PHYC) sequences (see Materials andMethods). This data set included sequences from one outgroupand 70 species of Malpighiaceae. To test these biogeographichypotheses, it is necessary to attach a temporal dimension to thephylogeny to infer the timing (i) of the origin of Malpighiaceaeand (ii) of the disjunctions between New and Old World lineages.Although molecular divergence estimates must be viewed withcaution, these approaches may identify a window of time forbranching events that will help select among competinghypotheses.Materials and MethodsOur data set included 71 ndhF sequences and 71 PHYCsequences, which were readily aligned by eye. This data setincluded sequences from one outgroup and 70 species ofMalpighiaceae representing the majority of genera, includingmultiple accessions from morphologically diverse or putativelynon-monophyletic genera. Most ndhF sequences were previ-ously obtained by using the methodology published in Davis,Anderson, and Donoghue (5) and in Davis (6). Two additionalndhF sequences were also generated for this study (GenBankaccession nos. AF500495 and AF500496). The PHYC sequencesare newly generated (GenBank accession nos. AF500522–AF500582), except for AF436794–AF436804 from Davis (6),and were obtained by using previously detailed (7) PCR,cloning, and sequencing procedures. We screened up to fiveclones from several species representing most of the majorlineages within Malpighiaceae (GenBank accession nos.AF500497–AF500521) and found no evidence of duplicationevents within PHYC, consistent with previous findings byMathews and Donoghue (7).To assess the level of congruence between the ndhF and PHYCdata sets, we used the incongruence length difference test (8)implemented inPAUP* V. 4.0B8 (9) for UNIX as the partition-homogeneity test. We used simple taxon addition (saving 10trees per replicate), tree-bisection-reconnection branch-swapping, and heuristic searches with 999 repartitions of thedata. The results (P ⫽ 0.55) indicated that ndhF and PHYC werecongruent and so they were combined for further analysis. Thecombined data set included 1,833 aligned sites. Data matricesanalyzed in this study are available from the first author and fromTreeBASE (http:兾兾www.treebase.org).Phylogenetic analyses by using maximum likelihood wereconducted on the combined data set withPAUP*. Tree searcheswere conducted with 300 random sequence addition replicatesand tree-bisection-reconnection branch swapping. We per-formed 300 bootstrap replicates to assess clade support. Tochoose a model of sequence evolution we performed likelihoodratio tests (10) with likelihood trees generated by using a seriesof models with increasing complexity. The GTR⫹I⫹⌫ modelhad a higher likelihood than other models and was used toevaluate molecular rate constancy.To infer the location of disjunctions between New and OldWorld lineages, ancestral areas were reconstructed by usingdispersal-vicariance analysis (DIVA; ref. 11). DIVA recon-structs ancestral areas by minimizing the number of dispersal andextinction events needed to explain the distribution pattern.Vicariance is the default mode of speciation in DIVA, and suchevents are not counted as steps in identifying optimal solutions;inferred dispersal and extinction events are counted as one stepeach. Our data matrix used to