Phylogeny and Biogeography of Cedrus (Pinaceae) Inferred from Sequences of Seven Paternal Chloroplast and Maternal Mitochondrial DNA Regions

Abstract

Background and Aims

Cedrus (true cedars) is a very important horticultural plant group. It has a disjunct distribution in the Mediterranean region and western Himalaya. Its evolution and biogeography are of great interest to botanists. This study aims to investigate the phylogeny and biogeography of Cedrus based on sequence analyses of seven cytoplasmic DNA fragments.

Methods

The methods used were PCR amplification and sequencing of seven paternal cpDNA and maternal mtDNA fragments, parsimony and maximum likelihood analyses of the DNA dataset, and molecular clock estimate of divergence times of Cedrus species.

Key Results

Phylogenies of Cedrus constructed from cpDNA, mtDNA and the combined cp- and mt-DNA dataset are identical in topology. It was found that the Himalayan cedar C. deodara diverged first, and then the North African species C. atlantica separated from the common ancestor of C. libani and C. brevifolia, two species from the eastern Mediterranean area. Molecular clock estimates suggest that the divergence between C. atlantica and the eastern Mediterranean clade at 23·49 ± 3·55 to 18·81 ± 1·25 Myr and the split between C. libani and C. brevifolia at 7·83 ± 2·79 to 6·56 ± 1·20 Myr.

Conclusions

The results, combined with palaeogeographical and palaeoecological information, indicate that Cedrus could have an origin in the high latitude area of Eurasia, and its present distribution might result from vicariance of southerly migrated populations during climatic oscillations in the Tertiary and further fragmentation and dispersal of these populations. It is very likely that Cedrus migrated into North Africa in the very late Tertiary, while its arrival in the Himalayas would not have been before the Miocene, after which the phased or fast uplift of the Tibetan plateau happened.

INTRODUCTION

Mountain regions between the fortieth parallels of latitude are of interest to biogeographers due to their importance in the survival of many plant species during the Ice Ages (Hewitt, 2004). The Mediterranean is one of 18 world hotspots of biodiversity, where approx. 25 000 plant species, of which 50 % are endemics, are harboured (Cowling et al., 1996; Myers et al., 2000; Scarascia-Mugnozza et al., 2000; Comes, 2004). Phylogeographical studies have provided strong evidence that some refugia were available in montane Mediterranean during the Quaternary, and also shaped the distribution of modern biota (Svenning, 2003; Comes, 2004; Hellwig, 2004). Unfortunately, the origin of the Mediterranean flora is still obscure (Valcárcel et al., 2003). In Asia, the uplift of the Himalayas and the onset of the Asian monsoon changed the world climate (Griffin, 2002; Liu and Yin, 2002), and made the Tibetan plateau, especially its south-eastern part, an important centre for not only the survival of many arctic plants, when they migrated southwards, but also the speciation of various organisms (Axelrod et al., 1996; Myers et al., 2000). It is of great interest to investigate the floristic relationship between the Mediterranean and the Tibetan plateau. Moreover, the significant geological and climatic changes in the two regions during the Cenozoic (Mai, 1989; LePage, 2003; Hellwig, 2004) produced many disjunct distributions (Sanmartín, 2003; Comes, 2004; Hellwig, 2004), and thus give us opportunities to test the dispersal and vicariance theories of biogeography.

Cedrus (true cedar) is one of 11 commonly accepted genera in Pinaceae, first described by Trew in 1757 (Farjón, 2001). It comprises four species with a highly disjunct distribution in circum-Mediterranean and western Himalayas (Farjón, 1990, 2001), i.e. Cedrus deodara (Roxb.) G. Don in the Hindu Kush, Karakoram and Indian Himalayas, Cedrus libani A. Rich. in Turkey, Lebanon and Syria, Cedrus brevifolia (Hook. f.) Henry in Cyprus, and Cedrus atlantica (Endl.) Manetti ex Carriére in North Africa (Algeria, Morocco). The true cedars occurring around the Mediterranean basin were once treated as one species C. libani, including four subspecies, namely ssp. atlantica, ssp. brevifolia, ssp. stenocoma in Turkey and ssp. libani in Lebanon and Syria (Davis, 1965). Some authors divided Cedrus into three species (cf. Scaltsoyiannes, 1998), regarding C. brevifolia as a subspecies of C. libani (Meikle, 1977). The great debate in species delimitation and phylogeny of Cedrus remains unresolved, although previous cytological (Li and Fu, 1995; Bou Dagher-Kharrat et al., 2001), biochemical (Scaltsoyiannes, 1998) and gene flow analyses (Fady et al., 2003) contributed greatly to the interpretation of interspecific relationships of true cedars.

In recent years, sequence analysis of multiple genes from different genomes has been successfully used in reconstructing complex phylogenies, meanwhile providing important information to the inference of biogeographical histories of many plant groups (e.g. Baum et al., 1998; Kusumi et al., 2002; Schneeweiss et al., 2004; Winkworth and Donoghue, 2005; Xiang et al., 2005). The combined gene analysis has great advantages in some conifer groups such as Pinaceae (Wang et al., 2000), where chloroplast and mitochondria are paternally and maternally inherited, respectively (Hipkins et al., 1994; Birky, 1995; Mogensen, 1996; Ahuja, 2001). The joint cp- and mt-DNA analysis can detect interspecific gene flow and reticulate evolution. However, no DNA sequence-based phylogeny has been constructed for Cedrus, a very important horticultural plant group. On the other hand, the systematic position of Cedrus in Pinaceae is also quite controversial, like the difficulty in species delimitation of the genus. Based on leaf arrangement, the presence of short shoots, and the petiole and pulvinus form, Cedrus was placed in the subfamily Laricoideae, with Larix and Pseudolarix (Cheng and Fu, 1978; Krüssmann, 1985). This classification was not supported by the subsequent anatomical (Hart, 1987; Frankis, 1988; Wu and Hu, 1997), immunological (Price et al., 1987) and cytological studies (Li, 1995). In particular, Cedrus does not have the same position in molecular phylogenies constructed by different markers such as PCR–RFLP (Tsumura et al., 1995) and DNA sequence (Quinn et al., 2002), and may represent a very distinctive lineage (Wang et al., 2000). The reconstruction of a robust phylogeny of Cedrus comprising all species could also shed some light on the origin of the genus.

To resolve interspecific relationships of plants, sequences of intergenic spacers and gene introns are most widely used because of their fast evolutionary rates (Wang et al., 1999; Wei and Wang, 2003; Shaw et al., 2005). Here, the phylogeny of true cedars based on sequences of seven cytoplasmic DNA fragments, including five from the paternal chloroplast genome and two from the maternal mitochondrial genome, is reconstructed, and the intrageneric classification, divergence times and biogeography of Cedrus are discussed.

MATERIALS AND METHODS

Plant materials

All four species of Cedrus were sampled, and each species was represented by two individuals. To investigate whether the molecular markers we used have infraspecific variation, six individuals of C. libani that are at least 100 m apart were further analysed. Pinus armandii Franchet and Cathaya argyrophylla Chun et Kuang, each represented by one individual, were chosen as outgroups based on molecular phylogenies of Pinaceae (Wang et al., 1998, 2000; Quinn et al., 2002). The origins of materials are shown in Table 1.

Table 1.

Sources of material

Taxa	Origins	Individuals/vouchers
Ingroups
Cedrus atlantica (Endl.) Manetti ex Carriére	Natural History Museum, Budapest, Hungary (cultivated)	2/Folly acb 1905, Debreczy-ca1
Cedrus brevifolia (Hook. f.) Henry	Natural History Museum, Budapest, Hungary (cultivated)	1/Debreczy-cb1
	Cyprus	1/Debreczy-cb2
Cedrus deodara (Roxb.) G. Don	Botanic Garden, Institute of Botany, Beijing, China (cultivated)	2/Wang-cd1, Wang-cd2
Cedrus libani A. Rich.	Elmali, Antalya, Turkey	7/Wang-cl-15
	Taurus Mt, Turkey	1/Debreczy-cl1
Outgroups
Pinus armandii Franchet	Botanic Garden, Institute of Botany, Beijing, China (cultivated)	1/Wang 6102
Cathaya argyrophylla Chun et Kuang	Jinfo Mountain, Nanchuan, Sichuan, China	1/WangXQ94513

DNA sequences of these materials determined in the present study have been deposited in GenBank under accession numbers DQ983599–DQ983604 (nad1), DQ983605–DQ983610 (nad5), DQ983611–DQ983616 (psbB-psbH), DQ983617–DQ983622 (rpL16), DQ983623–DQ983628 (5′rps12-rpL20), DQ983629–DQ983634 (trnC-trnD) and DQ983635–DQ983640 (trnT-trnF). All vouchers are deposited in PE (Herbarium, Institute of Botany, the Chinese Academy of Sciences, Beijing).

DNA extraction, PCR amplification and sequencing

Total DNA was extracted from silica gel-dried needles using the CTAB method following the protocol of Rogers and Bendich (1988) and used as the template in the polymerase chain reaction. The seven cytoplasmic DNA fragments used were chloroplast trnT-trnF (Wei and Wang, 2003), trnC-trnD (Lee and Wen, 2004), 5′rps12-rpL20, psbB-psbH and rpL16 intron (Shaw et al., 2005), and mitochondrial nad1 intron 2 (Song et al., 2002; Won and Renner, 2003) and nad5 intron 1 (Jaramillo-Correa et al., 2003). These gene regions were amplified using the primers listed in Table 2. The PCR reaction was carried out in a volume of 25 µL containing 5–50 ng of DNA template, 6·25 pmol of each primer, 0·2 mm of each dNTP, 2 mm MgCl₂, and 0·75 U of Taq DNA polymerase. Amplification was conducted in a Tpersonal Thermocycle and T1 Thermocycle (Biometra, Goettingen, Germany). PCR cycles were as follows: one cycle of 4 min at 70 °C, four cycles of 40 s at 94 °C, 20 s at 52–58 °C, and 2–2·5 min at 72 °C, followed by 36 cycles of 20 s at 94 °C, 20 s at 50–56 °C, and 2–2·5 min at 72 °C, with a final extension step for 10 min at 72 °C. PCR products were separated by 1·5 % agarose gel electrophoresis. The band with the right size was cut out and purified using the GFX PCR DNA and Gel Band Purification Kit (Amersham Biosciences, Amersham, UK). Sequencing reactions were performed with the two PCR primers listed above and several internal primers using the DYEnamic Energy Transfer (ET) Terminator Reagent Premix Kit (Amersham Biosciences). The internal primers included c (Wei and Wang, 2003) for the trnT-trnF region, petN2G (5′-CTTGGGCTGCTTTAATGGT AG-3′, forward), psbM2GF (5′-GTAGAGCAGCAATAAATGCAAG-3′, forward), psbM2GR (5′-CTTGCATTTATTGCTGCTCTAC-3′, reverse) and petN3G (5′-ATGGTACGAGGTCCTTCATCC-3′, forward) for the trnC-trnD region, and nad5-IF (5′-GGCTTTAGGGGGCCTTATG-3′, forward) for the nad5 intron 1. petN2G, psbM2GF and psbM2GR were modified from petN2, psbM2 and psbM2R of Lee and Wen (2004), respectively. After precipitation in 95 % EtOH and 3 m NaAc (pH 5·2), the sequencing products were separated on a MegaBACE 1000 automatic sequencer (Amersham Biosciences).

Table 2.

Sequences and references for chloroplast and mitochondrial DNA primers used in the present study

Region	Forward primer	Sequence (5′–3′)	Reserve primer	Sequence (5′–3′)	Reference
cpDNA
psbB–psbH	psbB(F)	TCCAAAAACTGGGAGATCCAAC	psbH(R)	TCAATGGTCTGTGTAGCCAT	Shaw et al., 2005*
rpL16	rpL16F	GCTATGCTTAGTGTGCGACTCGTTG	rpL16R	CCYTTCATTCTTCCCCTATGTTG	Small et al., 1998*
5′rpS12–rpL20	5′rpS12(F)	ATTAGAAATGCAAGACAGCCAAT	rpL20(R)	CGTTTTCGRGCTATA/GTATCC	Shaw et al., 2005*
trnC–trnD	trnC	CCAGTTCGAATCCGGGTGTC	trnD	GGGATTGTAGCTCAATTGGT	Demesure et al., 1995*
trnT–trnF	trnT-F(F)	CATTACAAATGCGATGCTCT	trnT-F(R)	ATTTGAACTGGTGACACGAG	Taberlet et al., 1991*
mtDNA
nad1 intron2	nad1F1	GATCGGCCATAAATGTACTCC	nad1R1	CCCCATATATTCCCGGAGC	Song et al., 2002
nad5 inton 1	nad5-aF	GGAAATGTTTGATGCTTCTTGGG	nad5-bR	CTGATCCAAAATCACCTACTCG	Wang et al., 2000

* Primers that have been modified in this study.

Data analysis

Sequence alignments were made with CLUSTAL X (Thompson et al., 1997) and refined manually for the maximization of sequence homology using BioEdit 5.0.9 (Hall, 1999). Maximum parsimony analysis was performed using PAUP version 4.0b10 (Swofford, 2002) with Pinus armandii and Cathaya argyrophylla as outgroups (Wang et al., 2000). Branch-and-bound searches were conducted with the MULTREES option. Indels in the alignment induced by the ingroups were coded as 1/0 binary characters, and gaps of different lengths were all treated as single events. All character states were specified as unordered and equally weighted. To evaluate relative robustness of the clades found in the most-parsimonious trees, the non-parametric bootstrap analysis (Felsenstein, 1985) was performed with 1000 replicates using the same search settings. The incongruence length difference test (Farris et al., 1995) was conducted, using the partition homogeneity test in PAUP ver. 4.0b10 (Swofford, 2002), to examine the congruence between different datasets. Test settings were 100 random stepwise additions and 1000 replicates of heuristic search with TBR branch swapping using two outgroups.

Owing to the large length variation in the mitochondrial nad1 intron 2, only the combined chloroplast dataset was used to estimate divergence times of Cedrus species. To estimate branch lengths, sequences of the ingroups were fitted using maximum likelihood (ML) models estimated on the topology of the parsimony trees. Optimal ML models for the combined data were selected using Modeltest 3.06 (Posada and Crandall, 1998). To estimate ages of nodes, first the hypothesis of a molecular clock was tested for the dataset using a Likelihood Ratio Test (LRT) that compared likelihood scores of the ML phylogenetic hypothesis with (L1) and without (L0) the molecular clock enforced. Significance was assessed by comparing two times the difference in log likelihoods to a chi-square distribution with n – 2 degrees of freedom, where n was the number of ingroups. When LRT showed that the data did not adhere to a molecular clock, non-parametric rate smoothing (NPRS) (Sanderson, 1997) in the program TreeEdit (http://evolve.zoo.ox.ac.uk/software/TreeEdit/; Rambaut and Charleston, 2002) transformed ML trees into ultrametric topologies. All node heights, sum of the branch lengths from a node to a tip, were transformed from units of molecular change into units of time by calibration to a node of known age. In addition, standard errors for estimates of node age were calculated for two nodes by using the non-parametric bootstrap resampling method (100 replicates) in PAUP*.

Since the NPRS method does not use rate smoothing, penalized likelihood (PL) analysis (Sanderson, 2002) was used additionally to estimate the divergence times of Cedrus. The ages of the nodes in the tree were estimated using penalized likelihood rate smoothing under TN algorithm with the program r8s version 1.7 (Sanderson, 2003, available at http://ginger.ucdavis.edu/r8s). A cross-validation analysis was performed to obtain the most likely smoothing parameter. Standard errors of the divergence times were estimated by a three-step nonparametric bootstrap procedure (Efron and Tibshirani, 1993): (1) 100 data sets were simulated by using the SEQBOOT program in PHYLIP Version 3.6a2 (Felsenstein, 1993); (2) the matrices were imported into PAUP version 4.0b10 and ML trees were generated; (3) the tree file was processed by the program r8s, which summarizes the bootstrap distribution of divergence times for each node that were used to calculate standard errors.

Although fossil and geological event can both be selected as a calibration point for time estimation, the latter as the calibration point may prove to be circular. The earliest identifiable cedar fossil dates back to the Paleocene (Blokhina, 1998), and therefore a minimum date of 54.8 Myr was applied to the node connecting Cedrus to the outgroup in the present study.

RESULTS

The five cpDNA fragments, psbB-psbH, rpL16 intron, 5′rps12-rpL20, trnC-trnD and trnT-trnF, are 404–409, 799, 705, 2253–2259 and 1336 bp in length, respectively (Table 3). No intraspecific sequence variation was detected. In the trnC-trnD region, three 5-bp gaps, one shared by C. libani, C. brevifolia and C. atlantica, one shared by C. deodara and C. atlantica and another one unique to C. deodara, were found. In addition, a 5-bp gap occurred in the psbB-psbH region of C. atlantica that included a 256-bp coding sequence. The partition-homogeneity test indicated that data sets of these cpDNA fragments were significantly congruent (P = 0·996). The combined cpDNA data set had 5875 characters, of which 697 were variable and 198 were parsimony-informative.

Table 3.

Summary of sequence variation in cp- and mt-DNA regions of the genus Cedrus

Region	Ingroups length (bp)	Alignment length (bp)	Substitution sites (informative)		Indels in ingroups (informative)
			Ingroups + outgroups	Ingroups
psbB-psbH	404–409	415	24 (4)	0 (0)	1 (0)
rpL16	799	824	82 (30)	1 (0)	0 (0)
5′rpS12-rpL20	705	706	52 (16)	1 (0)	0 (0)
trnC-trnD	2253–2259	2513	368 (90)	13 (2)	3 (1)
trnT-trnF	1336	1417	171 (58)	3 (0)	0 (0)
cp regions combined	5497–5508	5875	697 (198)	18 (2)	4 (1)
nad1 intron 2	530–852	622*	54 (8)	8 (0)	2 (2)
nad5 intron 1	1252	1266	59 (28)	10 (1)	0 (0)
mt regions combined	1782–2104	1888	113 (36)	18 (1)	2 (2)

* Some regions that could not be aligned have been removed.

The two mtDNA fragments, nad1 intron 2 (partial sequence) and nad5 intron 1, are 530–852 and 1252 bp in length, respectively (Table 3). No intraspecific sequence variation was detected. In the nad1 intron 2, two deletions, relative to C. deodara and outgroups, were found. One of 116 bp occurred in C. atlantica, and the other one of 322 bp was shared by C. libani and C. brevifolia. The partition-homogeneity test indicated that data sets of the two mtDNA regions were fully congruent (P = 1). The combined mtDNA data set was 1888 bp in length after removing some sequences that could not be aligned, including 113 variable sites, of which 36 were parsimony-informative.

Parsimony analysis of the combined cpDNA dataset generated only one most-parsimonious tree with a tree length of 743, a consistency index (CI) of 0·9960 and a retention index (RI) of 0·9851 (tree not shown). Phylogenetic analysis of the combined mtDNA dataset also found single most-parsimonious tree (tree length = 122, CI = 0·9590, RI = 0·8780) (tree not shown). A further partition-homogeneity test between cpDNA and mtDNA data sets suggested that they were completely congruent (P = 1). Parsimony analysis of the combined cp- and mt-DNA data set yielded a single most-parsimonious tree (tree length = 865, CI = 0·9908, RI = 0·9671) topologically the same as the cp- and mt- DNA trees, and all clades of the tree were robustly supported by the bootstrap analysis (Fig. 1). It was found that the Himalayan cedar C. deodara diverged first, and then the North African species C. atlantica separated from the common ancestor of C. libani and C. brevifolia, two species from the eastern Mediterranean area.

Fig. 1.

The single most-parsimonious tree obtained from sequence analysis of combined cp- and mt-DNA regions (length = 865, CI = 0·9908, RI = 0·9671). Numbers above the branches denote branch lengths and bootstrap values (in parentheses), respectively. Numbers below the branches are bootstrap values above 50 % yielded in the cp- (left) and mt- (right) DNA trees, respectively. The present distribution of each species was indicated on the right.

When an LRT compared likelihood scores under F81 (the best model given by Akaike Information Criterion), with and without the clock assumption on the ML tree obtained from the combined chloroplast data, the molecular clock was rejected in favour of the null hypothesis of no clock (δ = 2·62614, d.f. = 2, P > 0·05). Thus, a time-scaled NPRS ultrametric tree estimated under TreeEdit, using the topology with initial branch lengths estimated under maximum likelihood (Ribera et al., 2004), was generated and shown in Fig. 2. On the basis of a crown group age estimate of 54·8 Myr, molecular clock estimates suggested the divergence between the North African C. atlantica and the eastern Mediterranean clade comprising C. libani and C. brevifolia at 18·81 ± 1·25 Myr and the split between C. libani from Turkey, Lebanon and Syria and C. brevifolia from Cyprus at 6·56 ± 1·20 Myr. In the PL analysis, divergence times of the above nodes were estimated to be 23·49 ± 3·55 and 7·83 ± 2·79 Myr, respectively.

Fig. 2.

Cedar phylogeny and time scale of evolution determined by NPRS based on the combined chloroplast DNA data, with a calibration point of 54·8 Myr at the root node (indicated by a star) where Cedrus diverged from the outgroups. Standard errors for estimates of node ages are given for two nodes.

DISCUSSION

Intrageneric classification and phylogeny of Cedrus

Based on morphological and anatomical characteristics, all taxonomists accept the Himalaya cedar C. deodara, but classification of the circum-Mediterranean cedars is quite controversial (Davis, 1965; Meikle, 1977; Farjón, 1990, 2001; Scaltsoyiannes, 1998). Davis (1965) treated the circum-Mediterranean cedars as a single species, C. libani. Tutin et al. (1964) and Meikle (1977) recognized the North African cedar as a separate species C. atlantica based on its downy young twigs and small female cones. According to a combination of morphological characters such as angle of branch, cone size, male strobili length, leaf length, number of leaves in pseudo-whorls, together with native distributions, Farjón (1990, 2001) divided the circum-Mediterranean cedars into three closely related species, i.e. C. atlantica, C. brevifolia and C. libani. The external morphology is often affected by biotic and environmental factors (Davis, 1965; Meikle, 1977; Farjón, 1990; Pijut, 2000), and many characters vary continuously among cedar species, which could be responsible for the difficulty of intrageneric division of Cedrus.

Through terpene composition analysis, Canard et al. (1997) found a clear differentiation between C. libani and C. atlantica. This finding is further supported by the isozyme analysis of Scaltsoyiannes (1998), in which the dendrogram of genetic distances of 21 cedar populations comprises five distinct groups corresponding to C. libani ssp. libani, C. libani ssp. stenocoma, C. libani ssp. brevifolia, C. deodara and C. atlantica, respectively. Furthermore, fluorochrome banding patterns indicate that C. atlantica could be recognized as a separate species, and C. brevifolia as a subspecies of C. libani, although the three species are closely related (Bou Dagher-Kharrat et al., 2001). Based on open- and controlled-pollinations, as well as analyses using nuclear and cytoplasmic markers, Fady et al. (2003) found that gene flow occurs naturally between Cedrus taxa in plantation forests, and there are no strong pre-zygotic reproductive isolating barriers among Mediterranean cedars, when sympatric. Therefore, they proposed that Mediterranean Cedrus should be considered as units of a single collective species comprising two regional groups, North Africa and the Middle East.

The distinctness of C. deodara and the close relationships among the Mediterranean cedars are further corroborated by the present cp- and mt-DNA analyses. We found that the Himalaya cedar C. deodara has a basal position, and C. atlantica is sister to the clade comprising C. libani and C. brevifolia (Fig. 1). The tree topology, together with branch length, also supports North African C. atlantica as a separate species, and the reciprocal monophyly of Cedrus species could be the result of geographical isolation, which prevents interspecific gene flow from occurring.

Biogeography of Cedrus

Distribution patterns of organisms are greatly influenced by climatic and geological changes as well as their evolutionary histories. Climate tolerance is a crucial factor for plant distribution, in particular cold and aridity for trees (Pither, 2003; Svenning, 2003). From the beginning of the Tertiary, the global climate changed from extremes of expansive warmth with ice-free poles to extremes of cold with massive continental ice-sheets and polar ice caps, including several warm periods during the mid-Paleocene to early Eocene, and mid-Miocene (Zachos et al., 2003). The climatic oscillation provided opportunities for biota dispersal and subdivision (Mai, 1989; LePage, 2003; Sanmartín, 2003; Spicer et al., 2003; Comes, 2004; Hellwig, 2004; Vasiliev et al., 2004).

Farjón (1990) proposed that the highly disjunct distribution of apparently very closely related taxa of Cedrus could come from contraction of a formerly more extensive range of possibly one or two species, and a possible migration route that lay across the so called Thetys Fold Belt, a series of late Tertiary mountain uplifts along the brink of the old Thetys Sea, once extending from south-east Asia to north-west Africa. However, true cedars have reliable fossil records from the Oligocene of western Kazakhstan, the Miocene of south-west Europe, the Pliocene of south-east Europe, and the early Pleistocene of the Ahaggar Massif in the central Sahara (Florin, 1963; Farjón, 1990), in addition to fossil wood from the Paleocene of Kamchatka in Russian Far East (Blokhina, 1998, 2005). The fact that the fossils are progressively younger from north to south suggests an origin of Cedrus in the high latitude area of Eurasia. Its present distribution in several isolated regions could result from vicariance of southerly migrated populations during climatic oscillation in the Tertiary and further fragmentation and dispersal of these populations. This inference is robustly supported by the molecular phylogenies constructed in this study (Fig. 1), in which the Himalaya cedar is sister to the clade comprising Mediterranean cedars that was further divided into two subclades, i.e. the North African subclade and the subclade consisting of C. libani and C. brevifolia, two species from the eastern Mediterranean area.

The ancestor of Mediterranean cedars might have reached south Europe in the Miocene, which is supported by the fossil evidence mentioned above and the molecular clock estimate of 23·49 ± 3·55 Myr (PL) to 18·81 ± 1·25 Myr (NPRS) for the divergence between the North African C. atlantica and the eastern Mediterranean clade comprising C. libani and C. brevifolia (Fig. 2). The eastward dispersal of this lineage could be finally stopped by aridity in Asia Minor (Wulff, 1944; Griffin, 2002; Liu and Yin, 2002). On the other hand, biological communications between Europe and north Africa could have occurred through the Iberian connection during the Messinian (7–5 Mya) as a consequence of the desiccation of the Tethys, the Italian connection and the Arabian connection during the Miocene (Mai, 1989; Fici, 1991; Geraads, 1998; Krijgsman, 2002; Sanmartín, 2003; Hellwig, 2004). The last two connections, however, were unavailable to Cedrus for the presence of Saharan and Arab-Syrian deserts (Griffin, 2002; Liu and Yin, 2002; Sanmartín, 2003). It is very likely that Cedrus migrated into North Africa in the very late Tertiary, although C. atlantica was isolated from the eastern Mediterranean clade as early as the Miocene. The arrival of Cedrus in the Himalayas should not have been before the Miocene, whenafter the phased or fast uplift of the Tibetan plateau happened as suggested by recent studies, although there are a lot of debates on the time of the uplift (Dettman et al., 2003; LePage, 2003; Spicer et al., 2003). This inference is also corroborated by fossil pollen records from the late Miocene and Pliocene (8.10–4.07 Mya) of the central Loess Plateau in China (Ma et al., 2005) and from the Pliocene of the central Himalayas (Hsü et al., 1973).

It is obvious that more fossil evidence is still needed to retrieve the place of origin of Cedrus. In addition, the present molecular clock estimation of divergence times of true cedars is based on the fossil wood from the Paleocene, the earliest one recorded (Blokhina, 2005). The time values obtained in the present study could be younger than the real divergence times of the group, considering that Cedrus represents a deep or basal branch in the phylogeny of Pinaceae, a family with an origin dating back to the early Cretaceous (Wang et al., 2000). To investigate in detail the differentiation process and dispersal routes of Cedrus, it may be necessary to conduct a phylogeographical study on all of the four cedar species in the future, with extensive population sampling.