Molecular evidence of polyphyletism in the plant genus Carum L . ( Apiaceae )

An analysis of internal transcribed spacer (ITS) DNA sequences of the four species of Carum L. (Apiaceae) known in Italy revealed that this genus is polyphyletic. Maximum parsimony with bootstrap resampling, maximum likelihood and Bayesian inference analyses resulted in three distinct clades: Carum carvi L. clustered within tribe Careae Baill. (former Aegopodium clade); Hellenocarum multiflorum (= Carum multiflorum), Carum heldreichii and Carum appuanum clustered within the tribe Pyramidoptereae Boiss.; and H. multiflorum and C. heldreichii formed a well supported clade. Since the sister group of H. multiflorum and C. heldreichii was Bunium elegans the autonomy of Hellenocarum from Carum is confirmed by our study. We also found that C. appuanum clustered separately from the other Carum species, with the closest related species appearing to be Scaligeria moreana but this still had few morphological similarities with C. appuanum.

Evolutionary relationships among genera of Apiaceae, subfamily Apioideae, have been particularly difficult to resolve (Katz- Downie et al., 1999).In the last years many researchers have worked on this group, often finding high incongruence between molecular data and traditional taxonomic schemes.Nevertheless, the taxonomic treatments used in floras and monographs are still often derived from that proposed by Drude (1898).Recent cladistic analyses of molecular data supported the hypothesis that many of Drude's tribal and subtribal taxa are unnatural (Downie and Katz-Downie, 1996;Downie et al., 1998;Kondo et al., 1996;Valiejo-Roman et al., 1998;Downie et al., 2000a;Downie et al., 2000b).
The most used molecular markers in the Apioideae and many other angiosperms (Baldwin et al., 1995) have been the internal transcribed spacer (ITS) region of nuclear rDNA.In order to establish the phylogenetic position of the Italian species of Carum L. (representing most European species) we analyzed the nucleotide sequences of the ITS region as molecular markers.
The genus Carum is an important genus of the family Apiaceae, and contains about 20-30 species from Europe, North Africa and Asia (Hiroe, 1979;Pimenov and Leonov, 1993).Five species are present in Europe (Tutin, 1968), four of them in Italy (Pignatti, 1982).The best known species of this genus is Carum carvi L. (caraway or Persian cumin), which is one of the oldest herbs known (Nemeth, 1998).It is used traditionally as a condiment, oil and drug and, more recently, for the extraction of carvone, a compound which inhibits sprouting in potatoes (Langenberger and Davis, 2002b;Nemeth, 1998).Caraway is also important for honey production (70 to 134 kg ha -1 ) from Canadian behives (Langenberger and Davis, 2002a) and new medicinal uses such as anti-hyperglycemic potential have recently been reported (Eddouks et al., 2004).
Genus Hellenocarum was originally described by Wolff (1927) and after Tan and Sorger (1986) it is weakly delimited from Carum and might perhaps be better recognized at subgeneric rank.The verification of this hypothesis was one of the aims of this paper.
Silica gel preserved samples of leaf tissue were obtained directly in the field for Carum appuanum (Monte Matanna, Apuan Alps, 12 August 2003) and C. heldreichii (Scaffaiolo Lake, Tuscan Appennines 6 August 2003) while a sample of H. multiflorum coming from a population in Gravina di Laterza, southern Italy, was collected on 4 July 2002 and sent to us by the Botanical Garden of Lecce.
Genomic DNA was isolated using a modified Doyle and Doyle (1990) cetyltrimethylammonium bromide (CTAB) extraction protocol in which grinding the tissue in sea-sand and 70% (v/v) isopropanol substituted for the RNase step.Approximately 40 mg of leaf tissue were used for each extraction.The DNA concentrations were estimated by gel electrophoresis on 1% (w/v) agarose gel.For each sample the PCR reactions were carried out with about 10 ng of genomic DNA in a final volume of 50 μL containing 1.25 U of Taq polymerase (Takara) and the 18S sequence primer (5'-CGTAACAAGGTTTCCGTAG) and 26S primer (5'-AGTCCGCCCTGATGGGCGA).The thermal cycling profile consisted of 35 cycles of 1 min at 94 °C, 1 min at 50 °C and 2 min at 72 °C followed a final extension of 7 min at 72 °C.Clear cut single-banded fragments were separated on 1% (w/v) agarose gels.The resulting single-banded amplification products were purified and directly sequenced in both directions using the primers described above and a Perkin Elmer automated sequencer model 310 at the Center for Biotechnological Services (CIBIACI) of the University of Florence.We used asymmetrical PCR cycle sequencing and the BigDye Terminator Ready Reaction Kit (Applied Biosystems).
For sequence and phylogenetic analysis the resulting ITS sequences were visualized and checked by eye with the CHROMAS 1.43 software (C.McCarthy, School of Biomolecular and Biomedical Sciences, Brisbane, Australia).We performed a BLAST (Altschul et al., 1997) search to exclude the sequencing of any contaminant organism.The new ITS sequences produced during our investigation were deposited in the GenBank, the accession numbers being given in Table 1.Other GenBank sequences were chosen to sample for all the main clades of Umbellifers indicated in previous molecular studies (in particular Katz- Downie et al., 1999).We used as outgroups Oenanthe pimpinelloides and Ligusticum porteri and not some representatives of the Apioid superclade because one of the aims of the study was to test the relationship between Careae and Pyramidoptereae with respect to other groups.Moreover it was difficult a priori to be sure that some Carum representatives would nest outside Pyramidoptereae and Careae.In a previous analysis of the Apioideae subfamily by Katz- Downie et al. (1999), Pyramidoptereae and Careae clustered together with 67% bootstrap support while less than 50% bootstrap support supported the clade formed by the rest of the "Apiod Superclade".A maximum parsimony analysis of Careae+Pyramidoptereae with two representatives of the Apioid superclade as outgroups (data not shown) yielded a strict consensus tree with the same topology as those obtained in this analysis.Optimal multiple alignment was obtained with CLUSTALW 1.81 (Thompson et al., 1994) and checked by eye.Parsimony analysis was performed with PAUP 4.0b1 (Swofford, 1998) for MS DOS operating system.All characters were weighted equally, and character state transitions were treated as unordered.Gaps were treated after Simmons & Ochoterena (2000) and coded with Simple Gap Coding using the Gapcoder software (Young & Healy, 2003).This process codes indels as separate characters in a data matrix, which is then considered along with the DNA base characters in the phylogenetic analysis.The maximum parsimony analysis was done with 100 replicated heuristic searches, using random stepwise addition of taxa, tree bisection reconnection (TBR) branch swapping, and MULTREES in effect.Bootstrap (Felsenstein, 1985) resampling (BS in the trees description) was performed using TBR branch-swapping with ten random taxon entries per replicate and the multrees option in effect with 100 replicates.
A maximum likelihood (Felsenstein, 1981) search approach was carried out using the MrMODELTEST 2.0 program (Nylander, 2004) to evaluate the best likelihood model as settings in a maximum likelihood (ML) phylogenetic analysis in PAUP and for Bayesian Inference with the MrBayes 3.4b4 program (Huelsenbeck, 2001;Huelsenbeck et al., 2002).The maximum likelihood heuristic search was done with 10 random additions and TBR branch swapping, and the command ADDSEQ = ASIS with PAUP.The Bayesian analysis was done using the sequence evolution model indicated by the MrMODELTEST program based on the Akaike information criterion (Akaike, 1974).The Bayesian phylogenetic analysis was used for assessing the robustness of tree topology and the support for clades.The posterior probability of the phylogenetic model was estimated using Markov chain Monte Carlo (MCMC) sampling with the Metropolis-Hastings-Green algorithm.Four chains were run, three heated and one cold, for 10 6 generations and sampled every 100 generations.Following the analysis, the posterior probabilities were checked in the output of Mrbayes to estimate the number of trees that should be discarded as "burn-in".Stationary values were reached at approximately 20,000 generations, so the first 200 trees, or "burn-in" period of the chain, were discarded.Phylogenetic inference is therefore based on those trees sampled after generation 20,000.After the "burn-in" trees were removed from the data set, the remaining trees were used to produce a 50% majority-rule consensus tree (with PAUP) in which the percentage support was considered equivalent to Bayesian posterior probabilities.To test the significance of the difference of less parsimonious trees with respect to the most parsimonious solution, the Papini et al.

477
Table 1 -Apiaceae accessions used in our internal transcribed spacer (ITS) sequence study.When a single GenBank (GBAN) accession number is indicated, the whole ITS1-5.8S-ITS2 is intended, otherwise the first accession correspond to ITS1 and the second to ITS2.Species sequenced by the authors are underlined.Herbarium samples are available from the authors.Templeton test (Templeton, 1983) was used as implemented in PAUP.Also distances among the investigated species and those useful for clarifying the phylogenetic relationships were calculated with PAUP with Kimura's settings (Kimura, 1980) and reported in Table 2.The total alignment (ITS1+ITS2) was 469 bp long, plus 69 characters derived from indels coding (simple gaps coding).The Maximum Parsimony analysis showed that 131 characters were constant, 85 parsimony uninformative and 253 parsimony informative.Of the 69 indels-derived characters 45 were parsimony uninformative and 24 parsimony informative.The MrModel test indicated as best fitting model for sequence evolution in the investigated data set the SYM+I+G model with the following PAUP settings: Lset Base = equal; Nst = 6; Rmat = (0.8414 1.9751 1.5984 0.5398 5.9200); Rates = gamma; Shape = 1.1658;Pinvar = 0.0864.We also used the indels data coded as simple gaps with the maximum parsimony analysis.The result was 180 maximum parsimony trees 1359 steps long, CI = 0.480 and RI = 0.671.The analysis without indels produced only 40 maximum parsimony trees 1273 steps long and with CI = 0.458 and RI = 0.641.

Genus species and affiliation
The maximum likelihood tree (score 6522.005) is shown in Figure 1, with Bayesian support above branches.The maximum likelihood tree and the Bayesian tree (majority rule consensus tree of the trees produced by the Bayesian analysis omitting the "burn in" trees) were concordant for the position of the investigated Carum and Hellenocarum species.A strict consensus tree of the maximum parsimony trees (indels analysis) with bootstrap support (BS) above branches is given in Figure 2. All the used phylogenetic methods were concordant about the position of the investigated Carum species.
We used the tribal nomenclature after Downie et al. (2001) particularly for the Careae Baill.and Pyramidoptereae Boiss.tribes, previously called (Downie et al., 1998) the Aegopodium group the Crithmum group respectively.
In our analysis Careae and Pyramidoptereae formed together a monophyletic group both with maximum parsimony and with maximum likelihood criterion, with 84% bootstrapping and 99% Bayesian Support.Also two deletions in positions 56 and 180 (both one bp) characterized this clade.
The two C. carvi accessions clustered together with 100% bootstrap and Bayesian support and were nested in the Careae tribe, this tribe being supported by 91% bootstrapping and 100% Bayesian support and a one bp insertion at position 199.The closest relative of C. carvi was not determinable with parsimony because of polytomy in the consensus tree, while maximum likelihood tentatively identified the closest C. carvi relative as Fuernrohria setifolia.
We placed C. appuanum, C. heldreichii and H. multiflorum resulted in the Pyramidoptereae tribe, monophyly of this tribe being supported by 91% bootstrap support and 100% Bayesian support plus a deletion at position 270-276.The species H. multiflorum and C. heldreichii clustered with 96% bootstrap support and 100% Bayesian support.A 384-386 deletion was common to this tribe and to the Aegopodium.In the genus Aegopodium very variable chromosome counts are known, ranging from 2n = 21-22 to 44 in Aegopodium podagraria (Stepanov and Muratova, 1995) and from 2n = 50 to 2n = 88 in Aegopodium alpestre (Vasil'eva et al., 1994).The position of this genus is of interest since it belonged to the Careae on the basis of the phylogenetic analysis but shared a common insertion with the Pyramidoptereae tribe.
The sister group of Scaligeria moreana was C. appuanum with 92% bootstrap support and 100% Bayesian support.

Polyphyletism in Carum 479
Figure 1 -Maximum likelihood tree with Bayesian support reported above branches.
The alternative phylogenetic hypothesis tested with the Templeton test were either that all Carum species grouped together or all Carum species in the Pyramidoptereae tribe grouped together.The hypothesis that all Carum species grouped together was rejected by our data because the resulting tree was 1477 steps long (118 more than the maximum parsimony tree).Both alternative hypotheses were significantly different after the Templeton test.These results, based on the phylogenetic analyses executed with different criteria on all the four species of the genus Carum L. (Apiaceae) presently recognized in Italy revealed that Carum is polyphyletic.Maximum parsimony with bootstrap resampling and maximum likelihood analysis and Bayesian inference analyses agreed in indicating three distinct clades for this genus.Kimura's distances indicated the same result.The conflicts in taxonomic treatment of Apiaceae between traditional treatments (Drude, 1898 and derived treatments) and molecular data has been quite common in the recent literature.For instance Downie et al. (2000b), showed that of 16 genera of which more than one species was examined, 11 were not monophyletic.Therefore the polyphyly of Carum L. is a new finding but not particularly surprising in the context of most recent phylogenetic analyses of the Apiaceae.
The type species of the genus, Carum carvi L., clustered within the Careae Baill.tribe (also called the Aegopodium clade after Downie et al., 2001), as previously observed by other authors.We also found that C. heldreichii, C. appuanum and H. multiflorum clustered within the Pyramidoptereae Boiss.tribe (the Crithmum group in Downie et al., 2001).The first two species formed a well supported clade with 96% bootstrapping and 100% Bayesian support.The autonomy of the genus Hellenocarum in relation to Carum is confirmed and enforced.Further sampling in Bunium and allied genera are necessary to ascertain if C. heldreichii is to be assigned to Hellenocarum or rather to consider both H. multiflorum and C. heldreichii as belonging to the genus Bunium with which (at least with Bunium elegans) these two species formed a clade supported by 74% bootstrapping and 100% Bayesian support and a common insertion at position 461-462.
The chromosome number in H. multiflorum is 2n = 20 (personal observation by the authors and Brullo et al., 1995) and the same number (Favarger, 1973, under Carum carvifolium) has been found in Italy for C. heldreichii.Vasil'eva et al. (1985) found 2n = 17, 18 in Bunium elegans, but other Bunium species such as B. bulbocastanum (Verlaque and Filosa, 1992) and B. cylindricum (Sheidai et al., 1996) proved to have the 2n = 20 as seen in Hellenocarum.The genus Bunium shows very variable chromosome numbers along a descending dysploidy line starting from 2n = 22 to 2n = 12 (Vasil'eva et al., 1985). After Vasil'eva et al. (1985) B. elegans belongs to a group formed also by Bunium simplex and Bunium paucifolium, the chromosome number 2n = 18 found in B. elegans and B. simplex would have arisen by dysploidy starting from the 2n = 20 found in B. paucifolium.This opinion on the origin of chromosome numbers in Bunium supports or, at least, does not contradict the molecular data in indicating that H. multiflorum and C. heldreichii may be close at least to some species belonging to the genus Bunium.
The second test with C. appuanum, C. heldreichii, H. multiflorum grouped together resulted in a 1378 steps tree (19 more than the maximum parsimony tree) and a significant difference after the Templeton test.The position of Carum appuanum was different from all the other Carum species considered here, with the closest species related to C. appuanum appearing to be Scaligeria moreana.
After Bechi et al. (1997) and Garbari (1970) C. appuanum has 2n = 22 chromosomes while in the genus Scaligeria only the chromosome number of Scaligeria stewartiana is known: 2n = 20-24 (Kour et al., 1992).Other species of the tribe Pyramidoptereae having chromosome numbers ranging from 2n = 16 in Lagoecia cuminoides (Baltisberger, 1991) to 2n = 18 in Trachyspermum ammi (Sehgal and Abbas, 1994) and 2n = 20 in Crithmum maritimum (Ruiz de Clavijo, 1990).For the genus Elaeosticta, a chromosome count 2n = 22 was found in Elaeosticta paniculata (Vasil'eva et al., 1993) and Elaeosticta glaucescens (Nazarova and Ghukasyan, 2004).The taxonomic position of Carum appuanum needs revi-   As shown in recent reviews (Alvarez and Wendel, 2003) caution is necessary in studying phylogeny with ITS data.A wide sampling of sequences of Apiaceae is available on the GenBank from previous studies.Possible limitations to the present study might also arise from the incomplete sampling in the genus Carum and other closely allied genera that could possibly reduce the resolution of the phylogenetic inference.

Figure 2 -
Figure 2 -Strict consensus tree of the 180 maximum parsimony trees (analysis with indels) with bootstrap support above branches.

Table 2 -
Kimura distances calculated with PAUP among the investigated species belonging to the Careae and Pyramidoptereae.
The ITS1 length was 214 bp in H. multiflorum, C. appuanum, and C. heldreichii while it reached 215 bp in the two C. carvi accessions.The ITS2 was 215 bp long in C. appuanum, 219 in C. heldreichii and 220 in H. multiflorum, while it reached 223 bp in the two accessions of C. carvi.
sion since it clustered away from all the other Carum species but together with Scaligeria moreana.After a preliminary observation of samples of the genus Scaligeria in the Herbarium Centrale Italicum (in Florence, Italy) and descriptions in most recent floras morphological features do not appear to indicate a sufficient similarity to assert that these two species should belong to the same genus, hence Carum appuanum might need a new autonomous generic status on the basis of the ITS data.In conclusion certainly C. appuanum does not belong to Carum L. s. s. but further analyses with molecular and morphological data in Scaligeria and other Pyramidoptereae genera are necessary to propose a new taxonomic position for this species.