Journal of Biogeography (J. Biogeogr.) (2009) 36, 1282–1296

SPECIAL Origin of Mediterranean insular endemics ISSUE in the : integrative evidence from molecular dating and ancestral area reconstruction Guilhem Mansion1*, Federico Selvi2, Alessia Guggisberg1 and Elena Conti1

1Institute of Systematic Botany, University of ABSTRACT Zurich, Zollikerstrasse, Zurich, Switzerland Aim The presence of numerous reliable fossils and the occurrence of many and 2Dipartimento di Biologia Vegetale, Botanica Sistematica, Universita` di Firenze, endemic island make the Boraginales particularly suitable for integrative Firenze, Italy biogeographical studies. In this paper we aim to elucidate the time frame and events associated with the origin of selected endemic to the Mediterranean climate zone. More specifically, we describe and examine the alternative palaeo- and neoendemic hypotheses for their origin. Location Corsica and Sardinia (continental fragment islands) and the (an oceanic island archipelago). Methods Eighty-nine accessions, representing 30 genera from five families ascribed to the Boraginales, were examined for six chloroplast DNA regions. We used an integrative approach including phylogenetic analyses (MrBayes), Bayesian molecular dating (T3 package) with four fossil constraints on nodes, and biogeographical reconstructions (diva) to elucidate the temporal and spatial origins of the Corso-Sardinian and Canary Island endemics. Results Species of endemic to the Canary Islands diverged from their continental sister clade during the Miocene (15.3 ± 5.4 Ma), probably after the rise of the oldest islands (c. 20 Ma). Corso-Sardinian endemics of Borago diverged from their primarily North African sister clade during the late Miocene-Pliocene (c. 6.9 ± 3.6 Ma), well after the initial fragmentation of the islands (c. 30 Ma). Similarly, Corso-Sardinian endemics of diverged from the South African Anchusa capensis during the Pliocene–Pleistocene (c. 2.7 ± 2.1 Ma). Main conclusions The present study reveals an Anatolian origin for Anchusa, Borago and Echium and underlines the importance of the Eastern Mediterranean region as a possible reservoir for evolution in the Mediterranean Basin. For Anchusa and Borago, the divergence from their respective sister clades on the two types of islands post-dated the formation of the islands, thus supporting the neo- endemic hypothesis, whereas the dating results for the origin of Echium endemics *Correspondence: Guilhem Mansion, Institute were less conclusive. of Systematic Botany, University of Zurich, Keywords Zollikerstrasse 107, CH-8008 Zurich, Switzerland. Anchusa, Borago, Canary Islands, continental fragment island, Corsica, Echium, E-mail: [email protected] endemism, oceanic island, Sardinia.

tories for studying processes of speciation because of their INTRODUCTION well-defined boundaries, relative geographical isolation, smal- Since Wallace’s (1902) seminal book Island Life, ‘true islands’ ler area, lower species number and higher proportion of (i.e. portions of land isolated by water from significant endemics compared with continents (Schoener, 1988). landmasses) have been viewed as remarkable natural labora- ‘Continental fragment islands’ form by means of the break-up

1282 www.blackwellpublishing.com/jbi ª 2009 Blackwell Publishing Ltd doi:10.1111/j.1365-2699.2009.02082.x Insular endemism in the Boraginales of continental plates (i.e. through horizontal movements) and islands are expected to harbour a larger number of neoendem- usually harbour a subset of the mainland biota, meaning that ics, owing to the absence of initial connections with continents empty ecological niches are generally rare or absent (Gillespie and the availability of many empty ecological niches (Gillespie & Roderick, 2002; Lomolino et al., 2006; Whittaker & & Roderick, 2002). Overall, the relative composition of Ferna´ndez-Palacios, 2007). In contrast, ‘oceanic islands’ result palaeoendemics and neoendemics in both types of islands from volcanic activity (i.e. from volcanic island formation) and reflects the global history and dynamics of either ‘islands as are initially bare of life, meaning that many empty ecological museums’ or ‘islands as cradles of speciation’ (Cronk, 1992, niches are available for colonization and/or speciation 1997). The distinction between these two categories of islands is (Gillespie & Roderick, 2002; Whittaker & Ferna´ndez-Palacios, fundamental to understanding the origin of their diversity 2007). (proportion of palaeoendemics vs. neoendemics) and has Island endemics can be divided into two main groups: profound implications for conservation (e.g. proportion of palaeoendemics and neoendemics. The distinction between native vs. invasive species, naivete´ vs. invasion). these two types of endemics has been equivocal, as it has relied In the Mediterranean realm (including Macaronesia, i.e. on a number of criteria, including taxonomic, cytological and from north to south, the Azores, Madeira, the Canary Islands phytogeographical evidence (Bramwell, 1972b; Carlquist, and Cape Verde), Corsica–Sardinia and the Canary Islands 1974) or climatic events (Cronk, 1992). For example, island constitute geologically well known and floristically species-rich palaeoendemics can be defined in relation to the climatic representatives of fragment and oceanic islands, respectively. changes that caused the extinction of their closest relatives on In the western Mediterranean Basin, Corsica and Sardinia are the mainland while allowing their survival on islands (Cronk, remnants of the ‘Hercynian belt’, a Palaeozoic mountain chain 1992). By using this criterion, however, the assignment of that extended from Iberia to southern during the early endemics to either group is ambiguous, for it depends on the Oligocene. According to tectonic reconstructions (Alvarez specific climatic event used as the dividing line between et al., 1974; Rosenbaum et al., 2002), an ancestral Corsica– palaeo- and neoendemics. Alternatively, the geological age of Sardinia–Calabria microplate started to rift off from the the islands provides a possible bounded time-scale to allow us Hercynian belt c. 30–28 Ma (million years ago), and rotated to discriminate between palaeo- and neoendemics. We adopt south-eastwards until it collided with the central part of the this criterion here. Apulian microplate (corresponding to the current Italian If a speciation event post-dated the formation of either an peninsula) about 20 Ma. Corsica reached its current position oceanic or a continental island, we categorize the species as a about 16 Ma (Speranza et al., 2002), achieving complete ‘neoendemic’. In this sense, insular neoendemics must have separation from the Sardinia–Calabria microplate. During diverged from their closest relatives in situ. Initially, neo- the late Miocene (c. 6 Ma), the closure and partial desiccation endemics were characterized by geographical and taxonomic of the Mediterranean Sea might have allowed some connec- proximity to their closest extant relatives, and by their derived tions between North , Corsica, Sardinia and Eurasia position in phylogenetic trees (Favarger & Contandriopoulos, (Hsu¨, 1972; Duggen et al., 2003). Finally, Calabria separated 1961; Bramwell, 1972b; Major, 1988; Greuter, 1991). By from Sardinia and collided with southern Italy during the contrast, if the speciation event preceded island formation, Pliocene (Hsu¨ et al., 1977). the species is defined as a ‘palaeoendemic’. In the case of Corsica and Sardinia represent the largest and floristically continental islands, species divergence might have been driven most diverse fragment islands in the Mediterranean Basin by the geological separation of the island from the mainland (Me´dail & Que´zel, 1997). Indeed, 485 out of the 4572 plant (vicariance), and thus the palaeoendemic may represent a species native to Corsica and Sardinia are endemic (c. 10%), surviving element of the ancestral continental biota on the and the Corso-Sardinian (CS) area is currently listed as a island (Gillespie & Roderick, 2002). Alternatively, insular hotspot for plant diversity in the Mediterranean Basin palaeoendemics might have originated ex situ and colonized (Contandriopoulos, 1990; Mariotti, 1990; Gamisan & Jean- the continental or oceanic island after its formation by means monod, 1995; Me´dail & Que´zel, 1997; Bacchetta & Pontecorvo, of long-distance dispersal (LDD). The criteria traditionally 2005). Although the CS flora is very well described in terms of used to identify palaeoendemics have relied primarily on plant communities (e.g. Gamisan, 1991), the origin and geographical and taxonomic isolation (e.g. monotypic genera) affinities of its endemic species are still unclear. Available compared with their closest extant relatives and on their basal cytogeographical and systematic studies suggest that a majority position in phylogenetic reconstructions (Favarger & of taxa endemic to Corsica or Sardinia evolved from a western Contandriopoulos, 1961; Bramwell, 1972b; Major, 1988; Mediterranean stock or invaded these islands from northern Greuter, 1991). Europe, and to a lesser extent from North Africa, the eastern Considering their initial connection with the continental Mediterranean or the Middle East (Cardona & Contandripo- biota and the scarcity of empty ecological niches available for ulos, 1977, 1979; Contandriopoulos & Cardona, 1984; Cont- new colonizers, and assuming enough time since isolation from andriopoulos, 1990; Gamisan & Jeanmonod, 1995). Most of the mainland, fragment islands would be expected to contain a the aforementioned conclusions were drawn from qualitative higher proportion of palaeoendemic than of neoendemic assessments of similarities between floras, rather than from elements (Gillespie & Roderick, 2002). In contrast, oceanic explicit estimates of ancestral areas and times of origin for the

Journal of Biogeography 36, 1282–1296 1283 ª 2009 Blackwell Publishing Ltd G. Mansion et al.

CS taxa. Furthermore, the suggested proportion of palaeoen- was not supported by the embedded position of these species demics (32%) to neoendemics (68%) in the flora of Corsica in recent phylogenetic studies (Selvi et al., 2006). Finally, was inferred from the number of monospecific or taxonom- Echium contains c. 60 species found in the Macaronesian ically isolated taxa (Verlaque et al., 1997), rather than being archipelagos, the Mediterranean Basin and the Middle East based on explicit phylogenetic analyses. (Bramwell, 1972a; Bo¨hle et al., 1996). The Canary Islands off the north-western Atlantic coast of Overall, most hypotheses concerning the origin and status Africa are part of the Macaronesian biogeographical region of the above-mentioned insular endemics remain unresolved (Whittaker & Ferna´ndez-Palacios, 2007). Recent geological owing to a lack of rigorous analyses, including molecular surveys have demonstrated the sequential origin of the dating and reconstructions of ancestral areas of distribution. Canaries, starting about 20.2 Ma with the origin of Fuert- The existence of many reliable fossils (Segal, 1966; Gabel, eventura and ‘ending’ about 1.1 Ma with the emergence of 1987; Gottschling et al., 2002), essential for the generation of El Hierro (Guillou et al., 2004). The Canaries exhibit a an absolute and independent time frame for the origin of the Mediterranean climate similar to the one encountered in the respective CS and Canarian endemics through molecular Hercynian Islands (Whittaker & Ferna´ndez-Palacios, 2007) dating methods, and the absence of obvious adaptive traits and constitute the floristically richest islands of Macaronesia, for long-distance dispersal (Selvi et al., 2006) make Boragi- with c. 1700 species and 680 endemic species or subspecies nales especially suitable for an integrative biogeographical (Whittaker & Ferna´ndez-Palacios, 2007; Reyes-Betancort study. et al., 2008). From a phylogenetic point of view, most In this study we used phylogenetic analyses of sequences Canarian plant endemics studied so far have been found to from six chloroplast DNA (cpDNA) markers, Bayesian dating be sister to western Mediterranean clades (Carine et al., methods (with four fossil calibrations) and ancestral area 2004). Several investigated genera, including Argyranthemum reconstructions, in combination with palaeogeographical, (Asteraceae; Francisco-Ortega et al., 1997), Echium (Bora- palaeoclimatic and ecological evidence, to elucidate the time ginaceae; Bo¨hle et al., 1996) and Lotus (Fabaceae; Allan frame and biogeographical events associated with the origin of et al., 2004), exhibit spectacular patterns of island radiation, the selected endemics. More specifically, we ask: (1) but overall only a few phylogenetic studies have incorpo- when did the endemics diverge from their sister clades, and (2) rated a temporal dimension to evaluate the neoendemic vs. where did the endemics come from? We then compare our palaeoendemic origin of Canarian plant endemics (Carine, results with the alternative hypotheses of palaeoendemism and 2005). neoendemism. The first hypothesis predicts that the divergence The Boraginales, with c. 130 genera and 2300 species between the respective endemics and their sister clades pre- (Mabberley, 2000; Gottschling et al., 2001; Langstro¨m& dates plate fragmentation (fragment island endemics) or the Chase, 2002), represent an ideal model system with which to rise of volcanic islands (oceanic island endemics). The second test the spatial and temporal origins of species endemic to hypothesis postulates that phylogenetic divergence post-dates either fragment or oceanic islands. Indeed, seven taxa of the geological events. Anchusa and two of Borago are endemic to Corsica and Sardinia, whereas 27 species of Echium are endemic to the MATERIALS AND METHODS Macaronesian archipelagos, of which 23 are restricted to the Canary Islands (Bramwell, 1972a) (authorities for all species Character and taxon sampling, DNA extraction, used in this study are given in Appendix S1 in Supporting polymerase chain reaction amplification and Information). Anchusa is a moderate-sized , with c. 30 sequencing species occurring in the Mediterranean Basin and the Middle East, and seven species/subspecies restricted to either the A total of 88 new accessions were sequenced for the following coastal (Anchusa crispa subsp. crispa, A. crispa subsp. mariti- six cpDNA regions: (1) the trnL intron; (2) the trnL–trnF ma, Anchusa littorea, Anchusa sardoa) or mountainous intergenic spacer, both included in the tRNALEU intron/spacer (Anchusa formosa, Anchusa capelli, Anchusa montelisana) region; (3) the ndhF gene; (4) the rbcL gene; (5) part of the habitats of Sardinia and Corsica (Bacchetta et al., 2008). trnK intron; and (6) the matK gene, both included in the matK Available hypotheses postulate a Hercynian origin for the region. Plant samples were either obtained from botanical Sardinian endemics of Anchusa (Selvi & Bigazzi, 1998). Borago gardens and herbaria (46 accessions) or collected in the wild consists of four species restricted to north-west Africa (Borago (42 accessions) (see Appendix S1). Selected accessions repre- trabutii, Borago longifolia), Corsica, Sardinia and the Tuscan sented: (1) 30 genera within five families ascribed to the Archipelago (Borago morisiana, Borago pygmaea), and one Boraginales sensu Gottschling et al. (2001, 2004, 2005), namely widespread species, Borago officinalis, distributed in the , Cordiaceae, Ehretiaceae, Heliotropiaceae and Mediterranean region as a weed or cultivated for its horticul- Hydrophyllaceae; (2) several species of , Cryptantha, tural, culinary or pharmacological properties (Selvi et al., and Lithospermum, four genera with reliable fossil 2006). The palaeoendemic status proposed for B. morisiana record (Segal, 1966; Gabel, 1987; Gottschling et al., 2002); (3) and B. pygmaea on the basis of their narrow range and 15 species/subspecies out of the 27 taxa ascribed to Anchusa ecological specialization (Cardona & Contandripoulos, 1979) s. str., including five out of seven taxa endemic to Corsica and

1284 Journal of Biogeography 36, 1282–1296 ª 2009 Blackwell Publishing Ltd Insular endemism in the Boraginales

Sardinia (Bacchetta et al., 2008); (4) four out of five species of positions 295 and 296 of the final alignment, was excluded Borago, including two out of two species endemic to Corsica from the analysis. Gaps were coded with Fastgap (Borchsenius, and Sardinia (Selvi et al., 2006); and (5) 9 out of 29 species/ 2007) and added as binary states at the end of the matrix. The subspecies of Echium endemic to the Canary Islands (Bram- dataset was analysed using Bayesian inference, with a selected well, 1972b; Bo¨hle et al., 1996; Izquierdo et al., 2004) and 3 out model of sequence evolution for each of the six markers (i.e. of 30 Echium species from the mainland. Whereas the trnL intron, trnL–trnF intergenic spacer, ndhF gene, rbcL gene, sampling of the Canarian endemics of Echium encompassed trnK intron, matK gene). The best respective models were their morphological and habit diversity, including annual chosen using the Akaike information criterion, as implemented herbs (Echium bonnetii), monocarpic herbs (Echium auberia- in MrModeltest (Nylander, 2004). A binary model (Lset num), monocarpic rosette trees () and coding = variable) was applied to the coded gaps. Bayesian candelabra trees ( or Echium giganteum), analyses were performed with MrBayes ver. 3.1 (Huelsenbeck the sampling of the mainland Echium taxa represented three & Ronquist, 2001), using one cold and three heated Markov out of four clades strongly supported by a phylogenetic chain Monte Carlo (MCMC) chains run for 5 · 106 cycles, analysis based on a sample of 37 out of 57 species (Bo¨hle et al., sampling trees every 1000 generations, and with a default 1996). The study by Bo¨hle et al. (1996), although omitting any temperature parameter value of 0.2. Trees sampled after explicit temporal estimation, provided sound evidence for the stationarity were used to produce a majority rule consensus monophyly of the Canarian Echium. tree (Fig. 1). Stationarity was determined by plotting the log Sequences of Nicotiana tabacum retrieved from GenBank likelihood (lnL) values against the number of generations (Z00044) were used to delimit the respective boundaries of the (Gen) and established for each parameter using Tracer ver. sequenced regions. Nicotiana was also selected as a rooting 1.4 (Rambaut & Drummond, 2007). Convergence of runs was outgroup, based on previously published phylogenies of asserted when the potential scale reduction factor provided by Boraginales (Gottschling et al., 2001, 2004, 2005; Langstro¨m the sump output approached 1.0, the values of standard & Chase, 2002). DNA extraction, polymerase chain reaction deviations of split frequencies were below 0.001, and the (PCR) amplification and purification, and sequencing followed resulting tree topologies, branch lengths and clade credibility protocols described in Mansion et al. (2008). The sequences of values were similar across runs (Huelsenbeck & Ronquist, primers used for DNA amplification and sequencing are given 2001; Huelsenbeck et al., 2002; Ronquist & Huelsenbeck, in Table 1. 2003). Branch support was estimated with the bootstrap method (Felsenstein, 1985), using paup* (Swofford, 2002). Each bootstrap score (BS) was obtained from 100 replicates Phylogenetic inference with the resample option on, using heuristic searches with 100 DNA sequences were aligned manually using Se-Al (Rambaut, random additions and tree bisection–reconnection (TBR) 1996) and combined in a matrix of 89 accessions (88 new branch swapping. sequenced accessions together with one accession of Nicotiana For the molecular dating analysis and ancestral area as the outgroup) and 4857 aligned characters (89-taxon reconstruction, both requiring fully resolved trees, the initial dataset), including coded gaps. A difficult-to-align region in dataset of 89 accessions (Fig. 1) was reduced by pruning the trnL intron, comprising AT repeats and situated between accessions, but keeping representatives of all major clades

Table 1 General information for the primers used in this study.

Primer Sequence (5¢fi3¢) Direction Use cpDNA region References trnLc CGAAATCGGTAGACGCTACG F A, S tRNALEU intron/spacer Taberlet et al. (1991) trnLd GGGGATAGAGGGACTTGAAC R S tRNALEU intron/spacer Taberlet et al. (1991) trnLe GGTTCAAGTCCCTCTATCCC F S tRNALEU intron/spacer Taberlet et al. (1991) trnLf ATTTGAACTGGTGCAACGAG R A, S tRNALEU intron/spacer Taberlet et al. (1991) ndhF-972F GTGGTAGAAAGCAACGTGCGACTT F A, S Part of the ndhF gene Olmstead & Sweere (1994) ndhF-2110R TCGGGATCGAACATCAATTGCAAC R A, S Part of the ndhF gene Olmstead & Sweere (1994) rbcL-1F ATGTCACCACAAACAGAAAC F A, S rbcL gene Asmussen & Chase (2001) rbcL-724R CATGTACCTGCAGTAGC R A, S rbcL gene Asmussen & Chase (2001) rbcL-636F GCGTTGGAGAGATCGTTTCT F A, S rbcL gene Asmussen & Chase (2001) rbcL-Rev TCCTTTTAGTAAAAGATTGGGCCGAG R A, S rbcL gene Asmussen & Chase (2001) matk-1F ACTGTATCGCACTATGTATCA F A, S matK gene Sang et al. (1997) matK-590F AAGATGCCTCTTCTTTGCAT F A, S matK gene Sang et al. (1997) matK-1320R GATCCGCTGTGATAATGAGA R A, S matK gene Sang et al. (1997) trnK-2R AACTAGTCGGATGGAGTAG R S matK gene Johnson & Soltis (1995)

F, forward; R, reverse; A, amplification by polymerase chain reaction; S, sequencing.

Journal of Biogeography 36, 1282–1296 1285 ª 2009 Blackwell Publishing Ltd G. Mansion et al.

Figure 1 MrBayes 50% majority rule consensus tree of the Boraginales inferred from the combined 89-taxon dataset (88 new sequenced accessions together with one accession of Nicotiana tabacum as the outgroup; 4857 aligned positions). Bootstrap values are reported above the branches.

1286 Journal of Biogeography 36, 1282–1296 ª 2009 Blackwell Publishing Ltd Insular endemism in the Boraginales

Figure 2 Chronogram of the Boraginales inferred from Bayesian dating analysis of the 73-taxon dataset (4845 aligned positions). F1–F4 indicate minimum age constraints used for calibration, and nodes 1–6 represent the stem and crown nodes of the island endemics, with associated minimum age estimates (Ma) and 95% confidence intervals (grey bars).

Journal of Biogeography 36, 1282–1296 1287 ª 2009 Blackwell Publishing Ltd G. Mansion et al. recovered in the broader analysis (Fig. 2). The reduced dataset, for the imprecision of the fossil age estimates (Magallo´n, comprising 73 accessions and 4845 aligned characters 2004). Overall, the fossil record of the Boraginales comprises (73-taxon dataset), was analysed using Bayesian inference with specimens collected from the Palaeocene to the late Miocene the same settings as above. A list of selected taxa with their (Collison, 1983; Chelebajeva, 1984). The most reliable speci- current distribution is shown in Appendix S2. mens described in the Cordiaceae are fossil from the Asian Eocene (c. 50 Ma), which are very similar to the leaves of the extant Cordia monoica and Cordia myxa (Chelebajeva, Molecular dating analysis and node calibration 1984), thus justifying a minimum constraint of 50 Ma for the Because the likelihood ratio (LR) test comparing the likeli- crown node of the ‘Myxa clade’ of Cordia (Node F1 ‡ 50 Ma; hood scores of the trees resulting from the Bayesian analyses Fig. 2), in agreement with the strategy also adopted by of the 73-taxon dataset with and without an enforced clock Gottschling et al. (2002). The most reliable fossils of Ehretia was rejected (LR = 350, d.f. = 71, P < 0.01), we used the are from the early Eocene formations of the London Clay (c. Bayesian approach implemented in multidivtime (Thorne 50 Ma; Collison, 1983) and resemble the nutlets of the extant et al., 1998; Kishino et al., 2001; Thorne & Kishino, 2002) to East Asian Ehretia acuminata (Reid, 1923), with which they estimate divergence times at the nodes of interest. First, the share essential apomorphies (Gottschling et al., 2002). They maximum likelihood branch lengths of the rooted cladogram can therefore be assigned unequivocally to the ‘Ehretia clade’ and their variance–covariance matrix were estimated using the (Ehretiaceae; Gottschling et al., 2002) (Node F2 ‡ 50 Ma; baseml and estbranches programs in paml ver. 4 (Thorne Fig. 2). Because, the BS value for node F2 was quite low et al., 1998; Yang, 2007), under the F84 + G model of (BS < 50; Fig. 1), an alternative analysis was performed by substitution. Output branch lengths were then used as priors positioning the fossil constraint on the Ehretia crown node (BS in MCMC searches in multidivtime (Thorne & Kishino, 100; Fig. 1). 2002) to find the most likely model of rate change given the Other well-described fossils of Boraginaceae have been sequence data, the chosen topology, four fossil constraints on assigned to the extant Cryptantha and Lithospermum. These nodes and a Brownian motion parameter (nu) that deter- fossils represent segments of fruits or nutlets found in the late mines the permitted rate change between an ancestral and a Miocene strata of South Dakota (Ash Hollow Formation, descendant node. Ogallala Group, c. 10 Ma; Segal, 1966; Gabel, 1987). We Four values for prior distribution, specified in units of therefore assumed a conservative minimum age of 10 Ma for 10 Myr, were set as follows. First, the mean of the prior all of them. The fossil nutlets of C. coroniformis represent the distribution for the age of the ingroup root (rttm) was set to most frequent and reliable fossils of Cryptantha described so eight, with a standard deviation of 0.8 (rttmsd), following the far (Segal, 1966) and closely resemble extant nutlets of estimation of Wikstro¨m et al. (2001) for the age of the C. crassipetala or C. minima. The morphological characteris- Boraginales (c. 80 Ma). Second, the mean of the prior tics of these fossil nutlets are present in most extant Cryptantha distribution for the rate of molecular evolution at the ingroup species but are absent from the sister genera Cynoglossum and root (rrate) was set to 0.005, by dividing the median of the sum Paracaryum (Payson, 1927): the fossil constraint was thus of estimated branch lengths between the ingroup root and the assigned to the crown node of the entire Cryptantha clade tips by the rttm value. Third, the mean of the Brownian (Node F3 ‡ 10 Ma; Fig. 2). Finally, silicified nutlets of the parameter nu (brownmean) was set to 1 per rttm, following extinct Lithospermum dakotense exhibit several features that are Jeff Thorne’s manual instructions. For both rrate and rttm present in extant species of Lithospermum (Gabel, 1987) but values, the corresponding standard deviations of the prior absent in Neatostema (Seibert, 1978). Because Lithospermum is distribution (rratesd and brownmeansd, respectively) were set paraphyletic (Figs 1 & 2), this fossil was used to constrain the equal to the mean so as to reflect our ignorance concerning crown node of the Lithospermum s.l. clade (node F4 ‡ 10 Ma; prior distributions. Finally, we set the bigtime value to 13.0, in Fig. 2). order to reflect the estimated age of the oldest eudicot pollen (c. 125 Ma; Doyle & Donoghue, 1993). The Markov chain was Ancestral area reconstruction run for 100,000 cycles and sampled every 100 cycles, after an initial burn-in period of 100,000 generations. The multidiv- Biogeographical analyses were performed on the fully resolved time analysis was performed twice with different initial Boraginaceae clade inferred from the 73-taxon dataset (Fig. 2; random seed sets, and the output files checked for stationarity BS 100), using the dispersal–vicariance approach implemented of the Markov chain. in the diva software (Ronquist, 1996, 1997). diva is a The transformation of relative ages into absolute ages was parsimony-based method that reconstructs ancestral distribu- performed using four constraints on nodes F1–F4 (Fig. 2). tions for a given phylogeny without assuming any underlying Because, the goal of our study was to compare the time frames area relationships, and favouring a priori processes of vicar- of geological vs. speciation events, we used only palaeobotan- iance and duplication (cost of zero) over dispersal and ical constraints to avoid potential pitfalls of circularity with extinction (cost of one). We performed two ‘extreme’ analyses geological calibration. All fossil constraints were assigned as using either no (option ‘optimize’, favouring vicariance) or minimum, rather than fixed, ages to account (at least in part) maximum (option ‘optimize maxareas = 2’, favouring

1288 Journal of Biogeography 36, 1282–1296 ª 2009 Blackwell Publishing Ltd Insular endemism in the Boraginales

+ (a) Ae Ae-An Onosma graeca (Ae) Onosma taurica (An) – 6 An all but Ir-Ca-Taf-Ame Echium angustifolium (Naf-Ae-Ar-An) (all but Ir-Ca-Taf-Ame) + Ir *Naf-Ib-Co-Sa-Ba-Ap-Ae-Ar-An-Eu-Ca all but Ir-Ca-Taf-Ame (all but Ca-Taf-Ame) 1 Ca Echium wildpretii (Ca) + 10 15.3 Ca Echium auberianum (Ca)

Ca Echium bonnetii (Ca) Echium clade A Ca (Ca) Echium giganteum (Ca) 20 Ma 15 Ma 10 Ma 1 Ma

(b) *Naf-Sa 6.9 *Naf-An Symphytum armeniacum (An) Borago trabutii (Naf) *Naf + 7 + Naf + Sa Borago officinalis (all but An-Ir-Ca-Taf-Ame) 3 Borago morisiana (Sa) Borago pygmaea 1 (Co)

– Co-Sa Borago + Ir Naf Sa (Sa) An clade B + Co Borago pygmaea 2 Phyllocara aucheri (An-Ir) An – 8 + Naf-Ar Ae Hormuzakia aggregata (Naf-Ae-Ar) + 9 * + Naf-Ar (Naf-Ae-Ar) + 2 Anchusella aegyptiaca *Naf-Ib-Co-Sa-Ba-Ap-Ae-Ar-An-Ir (all but Ca-Ame) *Naf-Ib-Co-Sa-Ba-Ap-Ae-Ar-An-Ir – 6 An + Ap-Eu (Ae-Ar-An-Ir) *Ae-An Cynoglottis barrelieri (Ap-Ae-Eu) s.l. An + Ae + Ae-Eu Cynoglottis chetikiana (An) An Anchusa stylosa (Ae-An-Eu) *An (An) + Ib-Eu Anchusa pusilla

Lycopsis arvensis (Ib-Ap-Eu) Anchusa *Ap-Ae-An *Ae + Anchusa cespitosa (Ae) + Sa 5 + Ap-Ae – An Anchusa hybrida (Naf-Co-Sa-Ap-Ae-An) + *Sa-Ae-An Taf Anchusa capensis (Taf) 5 Anchusa formosa (Sa) Taf-Sa Anchusa crispa 4 (Co-Sa) 2.7 Sa + Co 1.2 clade C

30 Ma 21 Ma15 Ma 6 Ma 0 Ma

35 30 25 2015 10 5 0 Time (Ma)

Oligocene Miocene Pliocene Q

(c) Eu Ame Ap Co Ae Ib Ba An

Sa Ir Ca Naf Vicariance Ar Duplication + Dispersal Taf – Extinction

Figure 3 Chronograms with ancestral area reconstructions for the divergence of island endemics from their sister clades: (a) Canarian Echium (clade A, node 1); (b) Corso-Sardinian Borago (clade B, node 3), and Anchusa (clade C, node 5). Nodes with multiple optimal reconstructions are marked by an asterisk; 95% confidence intervals as in Table 2. Contemporary areas of distribution are reported to the right of the chronogram (Naf, North Africa; Ib, Iberian microplate; Co, Corsica; Sa, Sardinia; Ba, Balearic Islands; Ap, Apulian microplate; Ae, Aegean microplate; Ar, Arabian plate; An, Anatolian plate; Ir, Iranian plate; Eu, Eurasian plate; Ca, Canary Islands; Taf, tropical Africa; and Ame, American plate). Symbols: circles, vicariance; diamonds, duplication; +, dispersal; ), extinction; Ma, million years ago. (c) Map showing the areas used for biogeographical reconstructions.

dispersal) area constraints (Zink et al., 2000). We selected 14 et al., 1991; Ro¨gl, 1999; Krijgsman, 2002; Rosenbaum et al., areas (Fig. 3), corresponding to the main continental plates or 2002; Meulenkamp & Sissingh, 2003; Fromhage et al., 2004) microplates identified in tectonic reconstructions (Stampfli and described in detail in Mansion et al. (2008).

Journal of Biogeography 36, 1282–1296 1289 ª 2009 Blackwell Publishing Ltd G. Mansion et al.

RESULTS Table 2 Mean age estimates (Ma) with standard deviations (SD) and 95% confidence intervals for the six nodes of the respective Phylogenetic inference clades A (Echium endemics), B (Borago endemics), and C (Anchusa endemics), as obtained by multidivtime (Thorne & The 89-taxon dataset consisted of the following partitions: Kishino, 2002) analyses. trnL intron = 479 bp; trnL–trnF intergenic spacer = 446 bp; multidivtime ndhF gene = 743 bp, rbcL gene = 1199 bp; trnK intron Node (Ma) = 1014 bp; matK gene = 734 bp; coded indels = 242 binary N Designation Mean (± SD) 95% interval characters. The resulting tree (Fig. 1) included only three polytomies, none of which involved the nodes that are 1 Clade A stem node 15.3 (± 5.4) 6.8–27.6 crucial to estimating the divergence of the island endemics 2 Clade A crown node 9.2 (± 4.3) 3.1–19.6 from their mainland relatives (nodes 1, 3, 5; Fig. 2). The 3 Clade B stem node 6.9 (± 3.6) 2.0–15.7 73-taxon dataset consisted of 4845 aligned positions from 4 Clade B crown node 2.3 (± 1.9) 0.3–7.2 5 Clade C stem node 2.7 (± 2.1) 0.3–8.3 the same partitions as listed above. Because independent 6 Clade C crown node 1.2 (± 1.2) 0.03–4.6 Bayesian and bootstrap analyses of the two matrices yielded congruent topologies and similar branch support (Figs 1 & Ancestral area reconstruction 2) for the resulting 50% majority rule consensus trees, the following description is based on Fig. 1 only (i.e. the Ancestral area reconstruction required a total cost of either 156 89-taxon dataset). (maxareas = 2) or 149 (maxareas = 14) dispersal events. In the The majority rule consensus tree (Fig. 1) defined three optimizations shown on the chronograms of Fig. 3, the nodes strongly supported clades, containing the respective endemics numbered 1, 3 and 5 represent the stem nodes of clades A, B of Echium (clade A, BS 100), Borago (clade B, BS 100) and and C. Both analyses inferred the same ancestral areas for node Anchusa (clade C, BS 100). Clade A is embedded in the Echium 5. Duplication events were inferred for node 3, whereas clade (BS 100) and further divided into two well-supported vicariance events were inferred for nodes 1 and 5. When clades (BS 100 and BS 91). Clade B is included in the Borago multiple area reconstructions were possible for one node clade (BS 100) and is sister to the widespread B. officinalis (BS (marked with an asterisk on Fig. 3), the ancestral areas selected 100). Within clade B, the two accessions of Borago pygmaea at that node were based on geographical contiguity between from Corsica (accession 1) and Sardinia (accession 2) formed a areas during the relevant geological periods. well-supported clade (BS 68), sister to B. morisiana. Finally, clade C remained largely unresolved and showed sister DISCUSSION relationships with the South African Anchusa capensis (BS 86); both were included in a large paraphyletic ‘Anchusa s.l.’ Our integrated approach, based on phylogenetic inference, clade (BS 100). molecular dating estimates and ancestral area reconstructions provides overall support for an Anatolian origin for the stem nodes of the Echium, Borago and Anchusa s.l. clades, which Divergence time estimation appear to have originated mainly during the Middle Oligocene Molecular dating analyses produced the following estimated (Fig. 3a,b), a pattern recently demonstrated for other groups of ages for the nodes of interest: node 1 = 15.3 ± 5.4 Ma (e.g. the Araceae–Areae; Mansion et al., 2008). Keeping (Miocene) and node 2 = 9.2 ± 4.3 Ma (middle–late Mio- in mind (1) the extent of our current sampling, (2) the limits cene) for the stem and crown nodes of clade A (Echium of diva reconstructions (Ronquist, 1996, 1997), and (3) the endemics), respectively; node 3 = 6.9 ± 3.6 Ma (late Mio- fact that the mean ages estimated in this study represent cene–Pleistocene) and node 4 = 2.3 ± 1.9 Ma (Pliocene– minimum ages with sometimes large confidence intervals, we Pleistocene) for the stem and crown nodes of clade B introduce in the following discussion tentative hypotheses (Borago endemics), respectively; node 5 = 2.7 ± 2.1 Ma concerning the origins of clades A, B and C. Finally, based on (Pliocene–Pleistocene) and node 6 = 1.2 ± 1.2 Ma (Pleisto- the present results, we propose a redefinition of the concept cene) for the stem and crown nodes of clade C (Anchusa of insular endemism, taking into account the temporal endemics), respectively (Figs 2 & 3; Table 2). The alternative divergence of the lineages, the respective types of islands analysis, performed by assigning the fossil F2 to the crown (continental versus oceanic), and their geological history. node of Ehretia, gave very similar results for the respective nodes 1–6 (not shown). Based on a recent study of Linder Origin of the island endemics et al. (2005), showing that the age estimate of a test node is not affected by the undersampling of the clade sustained by The Canarian endemics of Echium that node, it seems unlikely that the age of the stem node of the Canarian endemics (node 1) would change significantly Molecular dating estimates and reconstruction of ancestral with increased taxon sampling. areas suggest that clade A (Canarian endemics of Echium;

1290 Journal of Biogeography 36, 1282–1296 ª 2009 Blackwell Publishing Ltd Insular endemism in the Boraginales

Fig. 1) diverged from its sister clade, comprising Mediterra- terranean Sea (Hsu¨, 1972; Hsu¨ et al., 1977; Krijgsman et al., nean, continental species of Echium, during the Miocene in a 1999; Roca, 2002; Meulenkamp & Sissingh, 2003; Rouchy & large area encompassing the Mediterranean Basin and the Caruso, 2006). A prominent role of land corridors during the Canary Islands (node 1 = 15.3 ± 5.4 Ma; Figs 2 & 3a, MSC has been invoked to partially explain current distribu- Table 2), before undergoing further expansion within tional patterns of beetles (Carmignani et al., 1995; Robertson the Canaries during the middle–late Miocene (node 2 = & Grasso, 1995; Speranza et al., 2002), large mammals (Palmer 9.2 ± 4.3 Ma; Fig. 2, Table 2). This biogeographical pattern is & Cambefort, 2000; Sanmartı´n, 2003) and some floristic congruent with the majority of studies that identified a single elements in Corsica (van der Made et al., 2006), although no origin for Canarian taxa and sister relationships with explicit time frame for the origin of these groups has been Mediterranean taxa, including Argyranthemum (Asteraceae; provided. Francisco-Ortega et al., 1997), Bencomia (Rosaceae; Helfgott To summarize, our study suggests that the phylogenetic et al., 2000), Ixanthus (Gentianaceae; Mansion & Struwe, divergence of the insular endemics of Borago post-dates the 2004), Gonosperminae (Asteraceae; Francisco-Ortega et al., original fragmentation of the CS microplate (30–28 Ma; 2001) and Sideritis (Lamiaceae; Barber et al., 2002). Rosenbaum et al., 2002; Fig. 3b), even at the lower end of Although the current patterns of distribution of the the confidence interval for their minimum age (15.7 Ma; Canarian endemics and their sister clade are mutually exclu- Table 2), thus supporting their neoendemic origin. Earlier sive, and thus technically vicariant (Fig. 3a), the Canary hypotheses of palaeoendemism for the CS endemics of Borago Islands, being of volcanic origin in deep water, were never (Contandriopoulos, 1981, 1990) were based on the marked connected to the African plate (Carracedo et al., 1998). differences in life history and ecological requirements between Therefore, cladogenesis cannot be explained by geological the widespread annual B. officinalis, occurring in xerophytic fragmentation, and over-water dispersal is required to explain and ruderal places, and the endemic perennials B. pygmaea the origin of the insular species of Echium. and B. morisiana, found in shady and humid habitats. Overall, our integrated approach suggests a single coloniza- tion of the Canary Islands by Echium following the emergence The CS endemics of Anchusa of the oldest islands of the archipelago, thus favouring the neoendemic hypothesis. However, age estimates at the lower Our integrative study supports a phylogenetic divergence end of the confidence interval (Fig. 3, Table 2) and the limited between clade C (CS endemics of Anchusa) and A. capensis extent of our current taxon sampling do not completely rule during the Pliocene–Pleistocene (node 5 = 2.7 ± 2.1 Ma; out the possibility of an earlier origin, possibly pre-dating the Figs 2 & 3b, Table 2), before the expansion of the former into appearance of the first island in the archipelago. Sardinia and ultimately Corsica during the Pleistocene (node 6 = 1.2 ± 1.2 Ma; Fig. 2, Table 2). As already mentioned for Borago, the recent colonization of Corsica by A. crispa was The CS endemics of Borago probably facilitated by a drop in sea level during the Combined evidence from dating analyses and diva recon- Pleistocene glacial maxima (Ehlers & Gibbard, 2004) and by structions (Ronquist, 1996) indicates that the common the ability of the seeds to float (occasional thalassochory) and ancestor of the CS endemics of Borago and their widespread germinate after immersion in seawater (Quilichini & Debus- Mediterranean sister B. officinalis expanded from North Africa sche, 2000). into Sardinia some time during the early to middle Miocene The sister relationship between the CS endemics of Anchusa (Fig. 3b). This scenario was chosen from two equally most and A. capensis is somewhat surprising on morphological parsimonious possibilities (not shown), based on palaeogeo- grounds, as the only synapomorphy is represented by the tiny logical maps indicating available land connections between the nutlets with a thin basal ring seam. However, this disjunct respective areas during this time frame (21–15 Ma; Fig. 3b). Mediterranean–South African pattern was partially supported Subsequently, the common ancestor, which broadly occurred by an earlier phylogenetic analysis of Boraginaceae (Hilger throughout North Africa and Sardinia, diverged in situ during et al., 2004) and has been proposed in other plant groups, for the late Miocene–Pleistocene (node 3 = 6.9 ± 3.6 Ma; Figs 2 example Erica (Ericaceae; McGuire & Kron, 2005), Iris and & 3b, Table 2) into B. officinalis, which then dispersed to the Moraea (Iridaceae; Reeves et al., 2001), Centaurium and remaining parts of the Mediterranean region, and the ancestor Chironia–Orphium (Gentianaceae; Mansion & Struwe, 2004), of B. morisiana and B. pygmaea, which later went extinct in and Echium and Lobostemon (Boraginaceae; Bo¨hle et al., 1996), North Africa. Finally, some populations of the endemic although often without the benefit of molecular dating B. pygmaea ultimately dispersed to Corsica during the late estimates (Galley & Linder, 2006). Our current sampling in Pliocene. Anchusa does not allow us to discriminate between the two The expansion of the xerophytic B. officinalis from North routes for plant exchanges between the Mediterranean Basin Africa to Eurasia and the range reduction of the current CS and that may have been available during the endemics of Borago might have been triggered by the Pliocene, namely a central Saharan track via the Ahaggar and establishment of the Messinian salinity crisis (MSC; 5.96– Tibesti mountains and an eastern African track via the Red Sea 5.33 Ma), which resulted in a dramatic drying of the Medi- hills (Wickens, 1976; Que´zel, 1978; Percy & Cronk, 2002).

Journal of Biogeography 36, 1282–1296 1291 ª 2009 Blackwell Publishing Ltd G. Mansion et al.

M. Weinger and L. Zeltner for collecting plant specimens or Types of insular endemism sharing tissue, DNA samples or unpublished results; and the The concept of endemism was introduced by Engler (1882), following herbaria, B, FI, NEU, MO, Z, for loans. We are also who first defined the terms ‘relict’ (i.e. taxa surviving in a indebted to L. Hugot of the Botanical Conservatory of Corsica smaller portion of their former range) and ‘neoendemic’ (i.e. for collecting permits. This study was funded by the Swiss autochthonous taxa newly formed in a defined area). The National Science Foundation grant no. 3100A0-105495 to E.C. relative ages of endemic taxa were deduced from various kinds The Swiss Academy of Natural Sciences and the Georges und of evidence. For example, a palaeoendemic origin for a certain Antoine Claraz-Schenkung funded field work, and the Equal taxon was inferred when related taxa were absent from the Opportunity Office of the University of Zurich funded same or adjacent areas, whereas the occurrence of related taxa consumables. in the area was interpreted as evidence for neoendemism (Wulff, 1950). Cytotaxonomic characteristics were subse- REFERENCES quently used to distinguish between passive (palaeo- and patroendemism) and active (schizo- and apoendemism) Allan, G.J., Francisco-Ortega, J., Santos-Guerra, A., Boerner, E. endemism (Favarger & Contandriopoulos, 1961). Further- & Zimmer, E.A. (2004) Molecular phylogenetic evidence for more, the distinction between palaeo- and neoendemics has the geographic origin and classification of Canary Island been inferred from switches in morphological features (e.g. Lotus (Fabaceae: Loteae). Molecular Phylogenetics and Evo- herbaceous to woody habit) or in breeding system (Bramwell, lution, 32, 123–138. 1972b; Carlquist, 1974) and type of vegetation or substrate Alvarez, W., Cocozza, T. & Wezel, F.C. (1974) Fragmentation (Bramwell, 1972b; Carlquist, 1974). of Alpine orogenic belt by microplate dispersal. Nature, 248, Overall, previous definitions of palaeo- and neoendemics 309–314. have lacked a rigorous spatial and temporal framework Asmussen, C.B. & Chase, M.W. (2001) Coding and noncoding inferred from ancestral area reconstructions and dating plastid DNA in palm systematics. American Journal of Bot- analyses, within which to estimate the place and time of origin any, 88, 1103–1117. of the endemics. Indeed, the distinction between palaeo- and Bacchetta, G. & Pontecorvo, C. (2005) Contribution to the neoendemism has long been plagued by subjective judgements knowledge of the endemic vascular flora of Iglesiente (SW (Bramwell, 1972b; Carlquist, 1974; Cronk, 1992; Vargas, 2007). Sardinia-Italy). Candollea, 60, 481–501. The definitions proposed in this paper rely on unique Bacchetta, G., Coppi, A., Pontecorvo, C. & Selvi, F. (2008) geological events that identify the origin of the islands, rather Systematics, phylogenetic relationships and conservation of than on arbitrarily chosen climatological events of global the taxa of Anchusa (Boraginaceae) endemic to Sardinia impact. Adopting this geological criterion for the distinction (Italy). Systematics and Biodiversity, 6, 161–174. between different types of insular endemics, our analyses Barber, J.C., Francisco-Ortega, J., Santos-Guerra, A., Turner, support a neoendemic origin for the Echium species of the K.G. & Jansen, R.K. (2002) Origin of Macaronesian Sideritis oceanic Canary Islands and for the Anchusa and Borago L. (Lamioideae: Lamiaceae) inferred from nuclear and endemics of the fragmented CS microplate, providing overall chloroplast sequence datasets. Molecular Phylogenetics and support for the assumption of Verlaque et al. (1997) that the Evolution, 23, 293–306. endemics of Corsica–Sardinia are predominantly neoendemics. Bo¨hle, U.-R., Hilger, H.H. & Martin, W.F. (1996) Island col- Other case studies on plant lineages that include insular onisation and evolution of the insular woody habit in Ech- endemics on different types of islands and reliable fossil ium L. (Boraginaceae). Proceedings of the National Academy records (e.g. Rutaceae) are under investigation to test the of Sciences USA, 93, 11740–11745. palaeo- and neoendemic hypotheses in the Mediterranean Borchsenius, F. (2007) FastGap 1.0.8. Available at: http:// realm. 192.38.46.42/aubot/fb/FastGap_home.htm; last accessed 19 Finally, the present investigation highlights the need for November 2008. improved methods of ancestral area inference (Ree et al., Bramwell, D. (1972a) Revision of the genus Echium in Maca- 2005) that allow for the explicit incorporation of a priori ronesia. Lagascalia, 2, 37–115. estimates of the probability of dispersal dependent on the Bramwell, D. (1972b) Endemism in the flora of the distance between landmasses at different temporal scales. Canary Islands. , phytogeography and evolution This would avoid biased reconstructions based on a priori (ed. by D.H. Valentine), pp. 141–159. Academic Press, assumptions of vicariance, as, for example, implemented in London. diva. Cardona, M.A. & Contandripoulos, J. (1977) L’ende´misme dans les flores insulaires me´diterrane´ennes. Mediterranea, 2, 49–77. ACKNOWLEDGEMENTS Cardona, M.A. & Contandripoulos, J. (1979) Endemism and The authors thank: J. M. Ferna´ndez-Palacios and three evolution in the islands of the Western Mediterranean. anonymous referees for constructive comments; M. Endress, Plants and islands (ed. by D. Bramwell), pp. 133–169. Aca- P. Enz, M. Gabel, M. Gottschling, H. Hilger, K. Meyer, A. Vela, demic Press, London.

1292 Journal of Biogeography 36, 1282–1296 ª 2009 Blackwell Publishing Ltd Insular endemism in the Boraginales

Carine, M.A. (2005) Spatio-temporal relationships of the Francisco-Ortega, J., Barber, J.C., Santos-Guerra, A., Febles- Macaronesian endemic flora: a relictual series or window of Herna´ndez, R. & Jansen, R.K. (2001) Origin and evolution opportunity? Taxon, 54, 895–903. of the endemic genera of Gonosperminae (Asteraceae: Carine, M.A., Russell, S.J., Santos-Guerra, A. & Francisco- Anthemideae) from the Canary Islands: evidence from the Ortega, J. (2004) Relationships of the Macaronesian and nucleotide sequences of the internal transcribed spacers of Mediterranean floras: molecular evidence for multiple col- the nuclear ribosomal DNA. American Journal of Botany, 88, onizations into Macaronesia and back-colonization of the 161–169. continent in Convolvulus (Convolvulaceae). American Jour- Fromhage, L., Vences, M. & Veith, M. (2004) Testing alternative nal of Botany, 91, 1070–1085. vicariance scenarios in Western Mediterranean discoglossid Carlquist, S. (1974) Island biology. Natural History Press, New frogs. Molecular Phylogenetics and Evolution, 31, 308–322. York. Gabel, M.L. (1987) A fossil Lithospermum (Boraginaceae) from Carmignani, L., Decandia, F.A., Disperati, L., Fantozzi, P.L., the Tertiary of South Dakota. American Journal of Botany, Lazzarotto, A., Liotta, D. & Oggiano, G. (1995) Relation- 74, 1690–1693. ships between the Tertiary structural evolution of the Galley, C. & Linder, H.P. (2006) Geographical affinities of the Sardinia-Corsica-Provencal Domain and the Northern Cape flora, South Africa. Journal of Biogeography, 33, 236–250. Apennines. Terra Nova, 7, 128–137. Gamisan, J. (1991) La ve´ge´tation de la Corse. Conservatoire et Carracedo, J.C., Day, S., Guillou, H., Rodrı´guez Badiola, E., Jardin botaniques, Geneva. Canas, J.A. & Pe´rez Torrado, F.J. (1998) Hotspot volcanism Gamisan, J. & Jeanmonod, D. (1995) La flore de Corse: Bilan close to a passive continental margin: the Canary Islands. des connaissances, inte´reˆt patrimonial et e´tat de conserva- Geological Magazine, 135, 591–604. tion. Ecologia Mediterranea, 21, 135–148. Chelebajeva, A.I. (1984) The genus Cordia (Boraginaceae) Gillespie, R.G. & Roderick, G.K. (2002) Arthropods on islands: from the Paleogene of Kamchakta (Russian SFSR, USSR) colonization, speciation, and conservation. Annual Review of and adjacent territories. Botanichnyi Zhurnal, 69, 605–615. Entomology, 47, 595–632. Collison, M.E. (1983) Fossil plants of the London Clay. Pale- Gottschling, M., Hilger, H.H., Wolf, M. & Diane, N. (2001) ontological Association, London. Secondary structure of the ITS1 transcript and its applica- Contandriopoulos, J. (1981) Ende´misme et origine de la flore tion in a reconstruction of the phylogeny of Boraginales. de la Corse: mise au point des connaissances actuelles. Plant Biology, 3, 629–636. Bolletino della Societa` Sarda di Scienze Naturali, 20, 187–230. Gottschling, M., Mai, D.H. & Hilger, H.H. (2002) The sys- Contandriopoulos, J. (1990) Spe´cificite´ de l’ende´misme corse. tematic position of Ehretia fossils (Ehretiaceae, Boraginales) Atti dei Convegni Lincei, 85, 393–416. from the European Tertiary and implications for character Contandriopoulos, J. & Cardona, M.A. (1984) Original char- evolution. Review of Palaeobotany and Palynology, 121, acter of the endemic flora in the Balearic Islands. Botanica 149–156. Helvetica, 94, 101–132. Gottschling, M., Diane, N., Hilger, H.H. & Weigend, M. Cronk, Q.C.B. (1992) Relict floras of Atlantic Islands – pat- (2004) Testing hypotheses on disjunctions present in the terns assessed. Biological Journal of the Linnean Society, 46, Primarily Woody Boraginales: Ehretiaceae, Cordiaceae, and 91–103. Heliotropiaceae, inferred from ITS1 sequence data. Inter- Cronk, Q.C.B. (1997) Islands: stability, diversity, conservation. national Journal of Plant Sciences, 165, S123–S135. Biodiversity and Conservation, 6, 477–493. Gottschling, M., Miller, J.S., Weigend, M. & Hilger, H.H. Doyle, J.A. & Donoghue, M.J. (1993) Phylogenies and angio- (2005) Congruence of a phylogeny of Cordiaceae (Boragi- sperm diversification. Paleobiology, 19, 141–167. nales) inferred from ITS1 sequence data with morphology, Duggen, S., Hoernle, K., van den Bogaard, P., Rupke, L. & ecology, and biogeography. Annals of the Missouri Botanical Morgan, J.P. (2003) Deep roots of the Messinian Salinity Garden, 92, 425–437. Crisis. Nature, 422, 602–606. Greuter, W. (1991) Botanical diversity, endemism, rarity, and Ehlers, J. & Gibbard, P.L. (2004) Quaternary glaciations – extinction in the Mediterranean area: an analysis based on extent and chronology. Part I: Europe. Elsevier, Amsterdam. the published volumes of Med-Checklist. Botanika Chroni- Engler, A. (1882) Versuch einer Entwicklungsgeschichte der ka, 10, 63–79. Pflanzenwelt. Engelmann, Leipzig. Guillou, H., Carracedo, J.C., Paris, R. & Torrado, F.J.P. (2004) Favarger, C. & Contandriopoulos, J. (1961) Essai sur l’ende´- Implications for the early shield-stage evolution of misme. Bulletin de la Socie´te´ Botanique Suisse, 77, 383–408. from K/Ar ages and magnetic stratigraphy. Earth and Felsenstein, J. (1985) Confidence limits on phylogenies: an Planetary Science Letters, 222, 599–614. approach using the bootstrap. Evolution, 39, 783–791. Helfgott, D.M., Francisco-Ortega, J., Santos-Guerra, A., Jan- Francisco-Ortega, J., Santos-Guerra, A., Hines, H. & Jansen, sen, R.K. & Simpson, B.B. (2000) Biogeography and R.K. (1997) Molecular evidence for a Mediterranean origin breeding system evolution of the woody Bencomia alliance of the Macaronesian endemic genus Argyranthemum (Rosaceae) in Macaronesia based on ITS sequence data. (Asteraceae). American Journal of Botany, 84, 1595–1613. Systematic Botany, 25, 82–97.

Journal of Biogeography 36, 1282–1296 1293 ª 2009 Blackwell Publishing Ltd G. Mansion et al.

Hilger, H.H., Selvi, F., Papini, A. & Bigazzi, M. (2004) Mansion, G. & Struwe, L. (2004) Generic delimitation and Molecular systematics of Boraginaceae tribe Boragineae phylogenetic relationships within the subtribe Chironiinae based on ITS1 and trnL sequences, with special reference to (Chironieae: Gentianaceae), with special reference to Cen- Anchusa s.l. Annals of Botany, 94, 201–212. taurium: evidence from nrDNA and cpDNA sequences. Hsu¨, K. (1972) When the Mediterranean dried up. American Molecular Phylogenetics and Evolution, 32, 951–977. Scientist, 227, 27–37. Mansion, G., Rosenbaum, G., Schoenenberger, N., Bacchetta, Hsu¨, K., Montadert, L., Bernoulli, D., Cita, M.B., Garrison, G., Rossello, J.A. & Conti, E. (2008) Phylogenetic analysis R.E., Kidd, R.B., Meleries, F., Muller, C. & Wright, R. (1977) informed by geological history supports multiple, sequential History of the Mediterranean Salinity Crisis. Nature, 267, invasions of the Mediterranean Basin by the angiosperm 399–403. family Araceae. Systematic Biology, 57, 269–285. Huelsenbeck, J.P. & Ronquist, F. (2001) MRBAYES: Bayesian Mariotti, M. (1990) Floristic connections between the Corso- inference of phylogenetic trees. Bioinformatics, 17, 754–755. Sardinian domination and the Ligurian area. Atti dei Huelsenbeck, J.P., Larget, B., Miller, R.E. & Ronquist, F. (2002) Convegni Lincei, 85, 429–448. Potential applications and pitfalls of Bayesian inference of McGuire, A.F. & Kron, K.A. (2005) Phylogenetic relationships phylogeny. Systematic Biology, 51, 673–688. of European and African Ericas. International Journal of Izquierdo, I., Martı´n, J.L., Zurita, N. & Arechavaleta, M. (2004) Plant Sciences, 166, 311–318. Lista de especies silvestres de Canarias. Hongos, plantas y Me´dail, F. & Que´zel, P. (1997) Hot-spot analysis for conser- animales terrestres. Consejerı´a de Polı´tica Territorial y Medio vation in the Mediterranean basin. Annals of the Missouri Ambiente, Gobierno de Canarias, Spain. Botanical Garden, 84, 112–127. Johnson, L.A. & Soltis, D.E. (1995) Phylogenetic inference in Meulenkamp, J.E. & Sissingh, W. (2003) Tertiary palaeoge- Saxifragaceae sensu stricto and Gilia (Polemoniaceae) using ography and tectonostratigraphic evolution of the Northern matK sequences. Annals of the Missouri Botanical Garden, and Southern Peri-Tethys platforms and the intermediate 82, 149–175. domains of the African–Eurasian convergent plate boundary Kishino, H., Thorne, J.L. & Bruno, W.J. (2001) Performance of zone. Palaeogeography, Palaeoclimatology Palaeoecology, 196, a divergence time estimation method under a probabilistic 209–228. model of rate evolution. Molecular Biology and Evolution, 18, Nylander, J. (2004) MrModeltest 2.1. Evolutionary Biology 352–361. Centre, Uppsala University, Sweden. Krijgsman, W. (2002) The Mediterranean: Mare Nostrum of Olmstead, R.G. & Sweere, J.A. (1994) Combining data in Earth sciences. Earth and Planetary Science Letters, 205, 1–12. phylogenetic systematics: an empirical approach using three Krijgsman, W., Hilgen, F.J., Raffi, I., Sierro, F.J. & Wilson, D.S. molecular data sets in the Solanaceae. Systematic Biology, 43, (1999) Chronology, causes and progression of the Messinian 467–481. salinity crisis. Nature, 400, 652–655. Palmer, M. & Cambefort, Y. (2000) Evidence for reticulate Langstro¨m, E. & Chase, M.W. (2002) Tribes of palaeogeography: beetle diversity linked to connection–dis- (Boraginaceae) and placement of Antiphytum, Echiochilon, junction cycles of the Gibraltar strait. Journal of Biogeogra- Ogastemma and Sericostoma: a phylogenetic analysis based phy, 27, 403–416. on atpB plastid DNA sequence data. Plant Systematics and Payson, E.B. (1927) A monograph of the section Oreocarya of Evolution, 234, 137–153. Cryptantha. Annals of the Missouri Botanical Garden, 14, Linder, H.P., Hardy, C.R. & Rutschmann, F. (2005) Taxon 211–358. sampling effects in molecular clock dating: An example from Percy, D.M. & Cronk, Q.C.B. (2002) Different fates of island the African Restionaceae. Molecular Phylogenetics and Evo- brooms: contrasting evolution in Adenocarpus, Genista, lution, 35, 569–582. and Teline (Genisteae, Fabaceae) in the Canary Islands Lomolino, M.V., Riddle, B.R. & Brown, J.H. (2006) Biogeog- and Madeira. American Journal of Botany, 89, 854– raphy, 3rd edn. Sinauer, Sunderland, MA. 864. Mabberley, D.J. (2000) The plant book, 2nd edn. Cambridge Que´zel, P. (1978) Analysis of the flora of Mediterranean and University Press, Cambridge. Saharan Africa. Annals of the Missouri Botanical Garden, 65, van der Made, J., Morales, J. & Montoya, P. (2006) Late 479–534. Miocene turnover in the Spanish mammal record in relation Quilichini, A. & Debussche, M. (2000) Seed dispersal and to palaeoclimate and the Messinian Salinity Crisis. Palae- germination patterns in a rare Mediterranean island ogeography, Palaeoclimatology, Palaeoecology, 238, 228–246. endemic (Anchusa crispa Viv., Boraginaceae). Acta Oeco- Magallo´n, S.A. (2004) Dating lineages: molecular and pale- logica International Journal of Ecology, 21, 303–313. ontological approaches to the temporal framework of clades. Rambaut, A. (1996) Se-Al: Sequence Alignment Editor. Avail- International Journal of Plant Sciences, 165, S7–S21. able at: http://tree.bio.ed.ac.uk/software/seal; last accessed 19 Major, J. (1988) Endemism: a botanical perspective. Analytical November 2008. biogeography: an integrative approach to the study of animal Rambaut, A. & Drummond, A.J. (2007) Tracer v1.4. Available and plant distribution (ed. by A.A. Myers and P.S. Giller), at: http://tree.bio.ed.ac.uk/software/tracer; last accessed 19 pp. 117–146. Chapman & Hall, London. November 2008.

1294 Journal of Biogeography 36, 1282–1296 ª 2009 Blackwell Publishing Ltd Insular endemism in the Boraginales

Ree, R.H., Moore, B.R., Webb, C.O. & Donoghue, M.J. (2005) A Seibert, J. (1978) Fruchtanatomische Untersuchungen an likelihood framework for inferring the evolution of geo- Lithospermeae (Boraginaceae). J. Cramer, Vaduz. graphic range on phylogenetic trees. Evolution, 59, 2299–2311. Selvi, F. & Bigazzi, M. (1998) Anchusa L. and allied genera Reeves, G., Chase, M.W., Goldblatt, P., Rudall, P., Fay, M.F., (Boraginaceae) in Italy. Plant Biosystems, 132, 113–142. Cox, A.V., Lejeune, B. & Souza-Chies, T. (2001) Molecular Selvi, F., Coppi, A. & Bigazzi, M. (2006) Karyotype variation, systematics of Iridaceae: evidence from four plastid DNA evolution and phylogeny in Borago (Boraginaceae), with regions. American Journal of Botany, 88, 2074–2087. emphasis on subgenus Buglossites in the Corso-Sardinian Reid, E.M. (1923) Nouvelles recherches sur les graines du system. Annals of Botany, 98, 857–868. Pliocene infe´rieur du Pont-de-Gail (Cantal). Bulletin de la Speranza, F., Villa, I.M., Sagnotti, L., Florindo, F., Cosentino, D., Socie´te´ Ge´ologique de France, 23, 350–355. Cipollari, P. & Mattei, M. (2002) Age of the Corsica-Sardinia Reyes-Betancort, J., Santos Guerra, A., Guma, I., Humphries, rotation and Liguro-Provencal Basin spreading: new C. & Carine, M. (2008) Diversidad, rareza, evolucio´ny paleomagnetic and Ar/Ar evidence. Tectonophysics, 347, 231– conservacio´n de la flora ende´mica de las Islas Canarias. 251. Anales del Jardı´n Bota´nico de Madrid, 65, 25–45. Stampfli, G., Marcoux, J. & Baud, A. (1991) Tethyan margins Robertson, A.H.F. & Grasso, M. (1995) Overview of the late in space and time. Palaeogeography, Palaeoclimatology, Pal- Tertiary–Recent tectonic and palaeo-environmental devel- aeoecology, 87, 373–409. opment of the Mediterranean region. Terra Nova, 7, 114–127. Swofford, D.L. (2002) PAUP*: phylogenetic analysis using par- Roca, E. (2002) The Northwest-Mediterranean basin (Valencia simony (*and other methods). Version 4. Sinauer Associates, Trough, Gulf of Lions and Liguro-Provencal basins): Sunderland, MA. structure and geodynamic evolution. Peri-Tethys Memoir 6. Taberlet, P., Gielly, L., Pautou, G. & Bouvet, J. (1991) Uni- PeriTethyan rift/wrench basins and passive margins (ed. by versal primers for amplification of three noncoding regions P.A. Ziegler, W. Cavazza, A.F.H. Robertson and S. Crasquin- of chloroplast DNA. Plant Molecular Biology, 17, 1105–1109. Soleau), pp. 671–706. Editions du Muse´um, Paris. Thorne, J.L. & Kishino, H. (2002) Divergence time and evo- Ro¨gl, F. (1999) Mediterranean and Paratethys. Facts and lutionary rate estimation with multilocus data. Systematic hypotheses of an Oligocene to Miocene paleogeography Biology, 51, 689–702. (short overview). Geologica Carpathica, 50, 339–349. Thorne, J.L., Kishino, H. & Painter, I.S. (1998) Estimating the Ronquist, F. (1996) DIVA version 1.1. Computer program and rate of evolution of the rate of molecular evolution. manual available by anonymous FTP from Uppsala Uni- Molecular Biology and Evolution, 15, 1647–1657. versity (ftp://uu.se or ftp://systbot.uu.se). Vargas, P. (2007) Are Macaronesian islands refugia of relict Ronquist, F. (1997) Dispersal–vicariance analysis: a new ap- plant lineages? A molecular survey. Phylogeography of proach to the quantification of historical biogeography. Southern European refugia – evolutionary perspectives on the Systematic Biology, 46, 195–203. origins and conservation of European biodiversity (ed. by Ronquist, F. & Huelsenbeck, J.P. (2003) MrBayes 3: Bayesian S. Weiss and N. Ferrand), pp. 297–314. Springer, Dordrecht. phylogenetic inference under mixed models. Bioinformatics, Verlaque, R., Me´dail, F., Que´zel, P. & Babinot, J.-F. (1997) 19, 1572–1574. Ende´misme ve´ge´tal et pale´oge´ographie dans le Bassin Rosenbaum, G., Lister, G.S. & Duboz, C. (2002) Reconstruction Me´diterrane´en. Geobios, 30, 159–166. of the tectonic evolution of the western Mediterranean since Wallace, A.R. (1902) Island life, 3rd edn. Macmillan, London. the Oligocene. Journal of the Virtual Explorer, 8, 107–130. Whittaker, R.J. & Ferna´ndez-Palacios, J.M. (2007) Island bio- Rouchy, J.M. & Caruso, A. (2006) The Messinian Salinity geography: ecology, evolution, and conservation, 2nd edn. Crisis in the Mediterranean basin: a reassessment of the data Oxford University Press, Oxford. and an integrated scenario. Sedimentary Geology, 188, 35–67. Wickens, G.E. (1976) The flora of the Jebel Marra ( Sang, T., Crawford, D.J. & Stuessy, T.F. (1997) Chloroplast Republic) and its geographical affinities. Kew Bulletin DNA phylogeny, reticulate evolution, and biogeography of Additional Series, 5, 1–368. Paeonia (Paeoniaceae). American Journal of Botany, 84, Wikstro¨m, N., Savolainen, V. & Chase, M.W. (2001) Evolution 1120–1136. of the angiosperms: calibrating the family tree. Proceedings of Sanmartı´n, I. (2003) Dispersal vs. vicariance in the Mediter- the Royal Society B: Biological Sciences, 268, 2211–2220. ranean: historical biogeography of the Palearctic Wulff, E.V. (1950) An introduction to historical plant geography Pachydeminae (Coleoptera, Scarabaeoidea). Journal of Bio- (translated from the Russian by E. Brissenden). Chronica geography, 30, 1883–1897. Botanica, Waltham, MA. Schoener, A. (1988) Experimental island biogeography. Ana- Yang, Z. (2007) PAML 4: a program package for phylogenetic lytical biogeography: an integrative approach to the study of analysis by maximum likelihood. Molecular Biology and animal and plant distribution (ed. by A.A. Myers and P.S. Evolution, 24, 1586–1591. Giller), pp. 483–512. Chapman & Hall, London. Zink, R.M., Blackwell-Rago, R.C. & Ronquist, F. (2000) The Segal, R. (1966) Taxonomic study of the fossil species of the shifting roles of dispersal and vicariance in biogeography. genus Cryptantha (Boraginaceae). The Southwestern Natu- Proceedings of the Royal Society B: Biological Sciences, 267, ralist, 11, 205–210. 497–503.

Journal of Biogeography 36, 1282–1296 1295 ª 2009 Blackwell Publishing Ltd G. Mansion et al.

SUPPORTING INFORMATION

Additional Supporting Information may be found in the online version of this article: Appendix S1 Species sampled, voucher information and GenBank accessions of DNA sequences analysed in this study. Appendix S2 Distribution areas for the species used for diva analysis. Please note: Wiley-Blackwell is not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

BIOSKETCH

Guilhem Mansion is a post-doctoral researcher at the University of Zurich. His main research interests involve identifying the evolutionary processes that have shaped plant distribution in different Mediterranean climate zones.

Editor: Jose´ Marı´a Ferna´ndez-Palacios

Part of this study was presented at the international congress Origin and Evolution of Biota in Mediterranean Climate Zones: an Integrative Vision, held in Zurich on 14–15 July 2007.

1296 Journal of Biogeography 36, 1282–1296 ª 2009 Blackwell Publishing Ltd