Molecular Phylogenetics, Molecular Evolution, and Patterns of Clade Support in Lithospermum (Boraginaceae) and Related Taxa Author(s): James Isaac Cohen, and Jerrold I Davis Source: Systematic Botany, 37(2):490-506. 2012. Published By: The American Society of Plant Taxonomists URL: http://www.bioone.org/doi/full/10.1600/03634412X635539

BioOne (www.bioone.org) is a nonprofit, online aggregation of core research in the biological, ecological, and environmental sciences. BioOne provides a sustainable online platform for over 170 journals and books published by nonprofit societies, associations, museums, institutions, and presses. Your use of this PDF, the BioOne Web site, and all posted and associated content indicates your acceptance of BioOne’s Terms of Use, available at www.bioone.org/page/terms_of_use. Usage of BioOne content is strictly limited to personal, educational, and non-commercial use. Commercial inquiries or rights and permissions requests should be directed to the individual publisher as copyright holder.

BioOne sees sustainable scholarly publishing as an inherently collaborative enterprise connecting authors, nonprofit publishers, academic institutions, research libraries, and research funders in the common goal of maximizing access to critical research. Systematic Botany (2012), 37(2): pp. 490–506 © Copyright 2012 by the American Society of Plant Taxonomists DOI 10.1600/03634412X635539 Molecular Phylogenetics, Molecular Evolution, and Patterns of Clade Support in Lithospermum (Boraginaceae) and Related Taxa

James Isaac Cohen,1,3 and Jerrold I Davis2 1Department of Biology and Chemistry, Texas A & M International University, 379E Lamar Bruni Vergara Science Center, 5201 University Boulevard, Laredo, Texas 78041, U. S. A. 2Department of Plant Biology, Cornell University, 412 Mann Library Building, Ithaca, New York 14853, U. S. A. 3Author for Correspondence ([email protected])

Communicating Editor: Jennifer Tate

Abstract—Utilizing 10 cpDNA regions and thorough taxon sampling, a phylogeny is reconstructed for Lithospermum and related members of both Lithospermeae, the tribe to which it is assigned, and Boraginaceae. Lithospermum is supported as monophyletic, and the genus is hypothesized to have originated in the Old World, after which there was one colonization of the New World. The heterostylous breeding system is inferred to have originated within Lithospermum either seven times or six times with one loss, and with other independent origins within Lithospermeae. The stability of the 10-region matrix is investigated, as are the number and combination of regions necessary to accurately reconstruct phylogenies. The combination of concatenated regions is important, and the following regions are recommended for future phylogenetic studies of genera of Boraginaceae: the rpl16 intron, matK, psbA-trnH, trnL-rpl32, and trnQ-rps16. The use of these recommended regions is conservative, and it contrasts with most intrageneric studies of Boraginaceae, which often are based on the trnL-trnF spacer and nuclear ribosomal ITS. Keywords—Boraginaceae, cpDNA, heterostyly, Lithospermum, molecular evolution, sampling.

Lithospermum L. (Boraginaceae) comprises approximately seven new genera of Lithospermeae. These efforts cul- 60 species. The genus has a cosmopolitan distribution, with a minated in revisions of both taxa (Johnston 1952, 1953a, b, center of diversity in the southwestern U. S. A. and Mexico, 1954a, b). In these publications, Johnston noted that all of the and with more than half of the species endemic to this region. genera endemic to the New World share a series of nutlet, In the present study, relationships and patterns of character pollen, and floral characters with Lithospermum (Johnston evolution are examined among species of Lithospermum and 1954a, b). Though he hypothesized various relationships their close relatives. Outgroups include representatives of among the genera of the tribe, these were not explicit phylo- more distantly related members of Boraginaceae, and pat- genetic hypotheses. terns of cpDNA evolution are examined within Lithospermum, Until recently, Johnston’s hypotheses remained untested. Lithospermeae, and Boraginaceae. In 2009, Cohen and Davis presented a phylogeny based Linnaeus (1753) described Lithospermum in Species Plantarum, on cpDNA sequence data, including representatives of and he included six species within it. This initial circumscrip- Lithospermum and five of the six genera of Lithospermeae tion encompassed species with smooth, white, lustrous, erect endemic to the New World. This phylogeny provided evi- nutlets, and this Lithospermum-type of nutlet consistently dence that all of these New World genera (Lasiarrhenum has been recognized as characterizing the genus. This has I. M. Johnst., Macromeria D. Don, Nomosa I. M. Johnst., remained the case even as the circumscription of the genus Onosmodium Michx., Perritostema I. M. Johnst., and Psilolaemus has changed, as has occurred frequently. Of the six species I. M. Johnst.) are nested among species of Lithospermum.In that Linnaeus recognized within Lithospermum, most were light of these inferred relationships, the authors included subsequently transferred to other genera, and until recently, all members of these endemic New World genera in only the type species remained in the genus (Cohen and Lithospermum. This inclusion increased the number of species Davis 2009). Since the mid 1700s, taxonomists have consid- in the genus by 18, bringing the total number to approximately erably broadened the circumscription of Lithospermum in three 60 (Cohen and Davis 2009). Two features diagnose this more principal stages. The first major change occurred in the treat- broadly defined Lithospermum: the Lithospermum-type nutlet, ment of Boraginaceae for the Prodromus (de Candolle 1846). and corollas that are not blue or purple (two colors occur in De Candolle described the state of Lithospermum 90 yr after closely related genera). In contrast, species of Lithospermum Linneaus’ original generic circumscription. De Candolle cir- produce corollas that are yellow, yellow-green, orange, or white. cumscribed Lithospermum broadly, and added many new spe- Weigend et al. (2009) also have reconstructed a phylogeny cies to the genus. He included species currently recognized as of Lithospermum and related taxa. Their phylogeny was not as part of Lithospermum as well as others now placed in other, fully resolved as that of Cohen and Davis (2009), and the two closely related genera, such as Buglossoides Moench. and phylogenies conflicted in some respects but both placed the Lithodora Griseb. De Candolle recognized species currently New World genera among species of Lithospermum. placed in Lithospermum through the inclusion of all of those Phylogenies have been reconstructed for only a small num- that were known at the time in the subgenus Eulithospermum ber of genera in Boraginaceae, including Lithodora (Thomas DC. (with the exception of L. chinensis Hook. & Arn., currently et al. 2008; Ferrero et al. 2009), Cerinthe L. (Selvi et al. 2009), a member of Heliotropium L. [Zhu et al. 1995]). Echium L. (Bo¨hle et al. 1996), Anchusa L. (Hilger et al. 2004), During the early to mid 20th century, Johnston (e.g. 1935, Myosotis L. (Winkworth et al. 2002), Lobostemon Lehm. (Buys 1952) studied and reevaluated the taxonomy of Boraginaceae. 2006), Echiochilon Desf. (La˚ngstro¨m and Oxelman 2003) and Throughout his career, Johnston studied Lithospermum and Nonea Medik. (Selvi et al. 2006), but a comprehensive phylog- the tribe to which it is assigned, Lithospermeae. He described eny of Boraginaceae does not exist at present. Difficulties, 35 new species of Lithospermum and described or segregated such as lack of overlap among the taxon samples of different

490 2012] COHEN AND DAVIS: MOLECULAR PHYLOGENETICS OF LITHOSPERMUM 491 studies, have impeded attempts to cobble together a phylog- primers of the cited authors. The only exception was a pair of primers eny of Boraginaceae; therefore, reevaluation of the taxonomy designed by the authors to amplify part of matK for some species of Lithospermum (matK-Lith2F 50 CACGAGTATTGGAATCCTTTTATT 30 of Boraginaceae above the genus-level is problematic, espe- and matK-Lith806R 50 TTGTGTTTCCGAGCCAAAGT 30). PCR mixtures, cially as the family includes many small genera (i.e. with 25 mL in volume, consisted of 67 mM Tris-HCl with 2.1% DMSO and fewer than five species). Through phylogenetic investigation 0.01% TritonX per reaction or 1 +Ex Taq Buffer (Takara Bio Inc., Japan), m m of larger genera, it has been determined that in addition to 2 mM MgCl2, 0.2 mM to 0.25 mM dNTPs, 1 M of primers, 0.125 Lto m m m Lithospermum, other genera, such as Nonea (Selvi et al. 2006), 1 LofTaq polymerase, and 0.1 L to 2.5 L DNA sample, depending on the DNA concentration. Amplifications were performed in an Eppendorf Cryptantha Lehm. ex G. Don (Hasenstab and Simpson 2008), Mastercycler Gradient 5331 thermocycler with the programs and anneal- and Cynoglossum L., are not monophyletic. Although the ing temperatures listed in Cohen and Davis (2009). The PCR products present study focuses on the phylogenetics of Lithospermum were separated on a 1% to 1.5% agarose gel and stained with ethidium and its closest relatives, the outgroup sampling allows for the bromide to determine whether amplification had occurred. Prior to sequenc- ing, some PCR products were purified with the QIAquick PCR purifica- examination of relationships among some of the more dis- tion kit (Qiagen, Germany). tantly related species of Lithospermeae and Boraginaceae. Nuclear ribosomal ITS was examined, as it provides a large number Most of the previous phylogenetic analyses of genera of of informative characters and is used in many phylogenetic studies of Boraginaceae (e.g. La˚ngstro¨m and Oxelman 2003; Hilger genera of Boraginaceae (e.g. La˚ngstro¨m and Oxelman 2003; Thomas et al. 2004; Selvi et al. 2006, 2009; Thomas et al. 2008; Weigend et al. 2008; Weigend et al. 2009). After the initial exploration of this region, it was determined that multiple copies exist in some species of et al. 2009, 2010a) employ only a few DNA regions, usually a Lithospermum and related taxa. Additionally, it proved difficult to gener- combination of one of two cpDNA spacer regions, trnL-trnF ate unambiguous alignments of the sequences of this region for the species or trnS-trnG, and the nuclear ribosomal region ITS. This sam- included in the present study. Given these issues, along with time and pling strategy tends to provide adequate resolution of rela- budgetary constraints and other concerns associated with ITS (see A´ lvarez tionships among genera (e.g. Thomas et al. 2008; Weigend and Wendel 2003), the region was not included in the present study. Rather, a large number of different cpDNA regions infrequently employed et al. 2010a), but resolution of relationships among species for phylogenetic inference were utilized. can prove difficult with this strategy (e.g. Hilger et al. 2004; Sequencing reactions were performed with BigDye 3.1 (Applied Bio- Weigend et al. 2009). Therefore, the present study utilizes systems, California) terminators and locus-specific amplification primers. data from 10 cpDNA regions, one of them protein-encoding Sequencing products were precipitated using a modification of the ethanol/EDTA/sodium acetate method (Applied Biosystems), and and the other nine non-coding. This quantity of cpDNA automated cycle sequencing was performed by the Life Sciences Core sequence data serves three purposes. First, the incorpora- Laboratory Center at Cornell University with an Applied Biosystems tion of multiple regions provides adequate sequence data to (ABI) 3730 DNA Analyzer. Alternatively, sequencing reactions and resolve phylogenetic relationships both within the ingroup, subsequent steps were performed by the Life Science Core Laboratory which includes species of Lithospermum sensu Cohen and Center at Cornell University, using Big Dye terminators and either an ABI 3700 or an ABI 3730. Sequence trace files were compiled, examined, and Davis (2009), and among members of the outgroup, which edited with Sequencher ver. 4.6–4.8 (Gene Codes Corporation, Michigan). includes species of Lithospermeae as well as members of three Sequences were deposited in GenBank (Appendix 1), and the matrix is other tribes of Boraginaceae, Boragineae, Cynoglosseae, available at TreeBASE (study number S11320). and Echiochileae. Second, the examination of this number Alignment, Gap Coding, Phylogenetic Analysis, and Character Evolution— MATRIX OF 10 CPDNA REGIONS—Initial alignments were performed with of cpDNA regions allows for the study of molecular evolu- MUSCLE (Edgar 2004) as implemented by the European Bioinformatics tionary patterns within the genus, tribe, and family. Third, Institute’s MUSCLE server (http://www.ebi.ac.uk/Tools/muscle/index the utilization of 10 cpDNA regions allows us to recom- .html) using the default settings. Subsequent adjustments were made in mend certain cpDNA-region combinations for future phylo- Bioedit ver. 7.0.5.3 (Hall 1999) and Winclada ver. 1.7 (Nixon 2002). genetic studies. Gaps were coded using simple indel coding (Simmons and Ochoterena 2000). Inversions were coded as present/absent. For each inferred inversion, the inverted sequences were recoded with their reverse com- plement, and these characters were included in the analyses (Graham Materials and Methods et al. 2000; Ochoterena 2009; Davis and Soreng 2010). Additionally, unusual nucleotide motifs (Cohen 2011) were coded as present/absent. Taxon Sampling—Sixty-seven species were included in the analyses All characters were weighted equally and treated as unordered (i.e. (Appendix 1). Thirty-seven belong to the ingroup, and this sampling nonadditive). Regions that aligned ambiguously were excluded from represents both the morphological and geographic range of variation all analyses. within Lithospermum sensu Cohen and Davis (2009). The outgroup com- The matrix of 10 cpDNA regions consists of 20% missing data. This prises 30 species from related genera of Boraginaceae: three from percentage largely reflects the use of sequence data from GenBank for Boragineae, seven from Cynoglosseae, one from Echiochileae, and 19 10 species, Buglossoides incrassata (Guss.) I. M. Johnst, Buglossoides tenuiflora from Lithospermeae. The majority of the species were collected from wild DC., Lithospermum cinereum DC., Lithospermum gayanum I. M. Johnst., populations. For these taxa, herbarium specimens were collected and Halacsya sendtneri Do¨rfl., Maharanga emodi DC., Mairetis microsperma (Boiss.) deposited at BH, and leaf tissue was dried and preserved in silica gel for I. M. Johnst., Neatostema apulum (L.) I. M. Johnst., Paramoltkia doerfleri subsequent DNA extraction. Taxa not collected from natural populations (Wettst.) Greuter & Burdet, and Podonosma orientalis (L.) Feinbrun, for were obtained from gardens (i.e. Cornell Plantations, Missouri Botanical which only one or two regions were available. When these 10 species are Garden, Chicago Botanic Garden, and National Botanic Garden of Belgium) excluded from the matrix, the percentage of missing data drops to 5.7%. as leaf samples preserved in silica gel, from the US Department of Agri- The topology of the strict consensus tree, in terms of taxa with more culture as seeds, or as DNA isolations from either the DNA bank of Royal complete coverage, is not affected by the inclusion of these 10 species. Botanic Gardens, Kew or the South African National Biodiversity Institute Both maximum parsimony (MP) and Bayesian inference (BI) phylo- (SANBI). For some species, sequence data for the trnL-trnF spacer and matK genetic analyses were conducted with the same data matrix. The MP were obtained from GenBank. analyses also were undertaken for various subsets of the data matrix DNA Extraction, PCR Amplification, and Sequencing—DNA extrac- as described below. The following search strategy was applied for all tion was performed with dried plant tissue using a modified CTAB MP analyses except where noted: the data were analyzed using TNT extraction method (Doyle and Doyle 1990) that included 2% PVP-40 and (Goloboff et al. 2008), with 1,000,000 trees held in memory, and 1,000 inde- in some cases the addition of 0.5 M glucose in the CTAB extraction buffer. pendent iterations, with 20 trees held per iteration, each includ- The PCR amplifications of the ten cpDNA regions, ndhF-rpl32, psbA-trnH, ing 1,000 parsimony ratchet iterations (Nixon 1999), with 10% probability psbJ-petA, the rpl16 intron, trnK-rps16, trnL-rpl32, trnQ-rps16, ycf6-psbM, of upweighting and 10% probability of downweighting, followed by trnL-trnF (Shaw et al. 2005, 2007), and matK (390F [Cue´noud et al. 2002] 1,000 cycles of tree drifting; afterwards, 100 rounds of tree fusion and and 1710R R [Barfuss et al. 2005]), were performed using the published random sectorial searches were performed (Goloboff 1999a), followed by 492 SYSTEMATIC BOTANY [Volume 37

TBR-max, swapping among all most-parsimonious trees until comple- the other consists of species with 80% (any eight of ten) of the regions tion. Additionally, a more abbreviated search strategy was performed: sampled (the 80% matrix). The 100% matrix includes 34 species, and the 10,000 trees held in memory, and one iteration of 1,000 parsimony ratchet 80% matrix includes 53 species. The exclusion analyses were conducted iterations (Nixon 1999), with 10% probability of upweighting and 10% using the abbreviated MP search strategy described above, with ten anal- probability of downweighting, followed by 1,000 cycles of tree drifting, yses performed for each matrix, one for each cpDNA region removed. 100 rounds of tree fusion, random sectorial searches (Goloboff 1999a) and The strict consensus tree for each analysis was compared to that of the TBR-max. Clade support was measured with NONA (Goloboff 1999b) by corresponding matrix with all 10 regions included. To investigate the util- conducting 10,000 jackknife replicates (36% removal probability) (Farris ity of each region, the number of nodes resolved in the consensus tree and et al. 1996). For each replicate, 10 TBR searches were conducted, with one the number of contradictory nodes resolved in the consensus tree were tree held after each replicate, and a total of 100,000 trees held in memory counted and compared. A contradictory node is defined as a node in the for the duration of the entire jackknife resampling. Although some (e.g. consensus tree from an exclusion analysis matrix that conflicts with Felsenstein 2004) have suggested that, in comparison to bootstrapping, the structure of one of the most-parsimonious trees from the corresponding jackknife analyses with less than 50% character removal result in biased 10-region matrix. This method is similar, although not identical, to one estimates of support, others (e.g. Farris et al. 1996; Davis et al. 2004; implemented by Simmons and Miya (2004). Freudenstein and Davis 2010; Simmons and Freudenstein 2011) prefer SUCCESSIVE-INCLUSION ANALYSES—To determine the utility in the recon- support values derived from jackknifing with less than 50% (ca. 37%) struction of phylogenetic relationships of the cpDNA regions with the character removal. These authors provide evidence to favor jackknife greatest number of informative characters as compared to those with analyses with 37% character removal over other measures of support the smallest number informative characters, successive-inclusion (SI) because the resampled characters in jackknife analyses resemble the analyses were performed. These analyses, which were performed in original matrix, which is not necessarily the case for bootstrapping (Davis an MP framework, involve the use of successively larger numbers of et al. 2004; Freudenstein and Davis 2010), and the use of jackknifing individual cpDNA regions in the analyses. We conducted SI analyses by can reduce the risk of unsubstantiated, high support values (Simmons beginning with the cpDNA region with either the greatest or the smallest and Freudenstein 2011). In addition to jackknife values, posterior proba- number of informative characters, and performing a MP phylogenetic bilities were estimated in the BI analyses (see below). analysis. This was followed by the addition of the region with either the For BI analyses, the matrix was divided into two partitions: nucleotide next greatest or next smallest number of informative characters, respec- sequence data and structural characters. The nucleotide sequences were tively, performing a phylogenetic analysis with this matrix, and continu- G analyzed using a GTR + I + model, which, as determined by Modeltest ing until all regions were included. These analyses were performed v3.4 (Posada and Crandall 1998), was the most appropriate model for on the same 100% and 80% matrices that were used in the exclusion these data. The structural characters were analyzed using the binary analyses, with the abbreviated MP search strategy described above, model of MrBayes v3.1.2 (Huelsenbeck and Ronquist 2001; Ronquist and with one exception. For some analyses, 100,000 trees, rather than Huelsenbeck 2003). For the binary model, the coding bias was selected as 10,000 trees, were held in memory. A total of nine analyses were per- variable, as is recommended for this type of data (http://mrbayes.csit formed, one per cpDNA region added, for each type of inclusion analysis .fsu.edu/mb3.1_manual.pdf). Ten separate searches were undertaken (i.e. more to fewer or fewer to more). The total number of nodes and the with MrBayes v3.1.2, and all analyses were run on the servers of the total number of non-contradictory nodes reconstructed in the consensus Computational Biology Service Unit at Cornell University (cbsuapps.tc tree were counted after each analysis, as with the exclusion analyses. .cornell.edu). The following search strategy was employed for all BI anal- These numbers were compared to the respective consensus tree yses: eight chains run for 10,000,000 generations, with the chains sam- reconstructed from the matrix that includes all 10 cpDNA regions. Con- pled every 1,000 generations, and a temperature of 0.27. The first 25% of tradictory nodes were determined in the same manner as for the exclu- the run was treated as burnin, and not used for subsequent calculations of sion analyses. tree statistics. This percentage was determined via observation of the log- For the 100% matrix, the order of the regions from greatest number of likelihood plots. The commands for BI analyses are listed in Appendix 2. informative characters to fewest is rpl16, trnL-rpl32, psbJ-petA, ycf6- A 50% majority rule consensus tree was constructed, and posterior prob- psbM, ndhF-rpl32 (tie), matK (tie), trnK-rps16, trnQ-rps16, psbA-trnH, abilities of nodes were recorded. and trnL-trnF. For the 80% matrix, the order of the regions from greatest For MP trees, multiple tree statistics were calculated. These include number of informative characters to fewest is rpl16, psbJ-petA, matK, consistency indices, matrix information (MI), and total matrix information ycf6-psbM, trnK-rps16, trnL-rpl32, ndhF-rpl32, trnQ-rps16, psbA-trnH, (TMI), and all were calculated after removal of parsimony-uninformative and trnL-trnF. characters. Matrix information was calculated by subtracting the minimum 80% MATRIX JACKKNIFE ANALYSES—To examine the effect that the removal number of steps of the characters in the matrix from the maximum num- or addition of particular cpDNA regions may have on the support of ber of steps of these characters (cf. Farris 1989; Nixon 2002). Total matrix various nodes, 1,000 jackknife replicates were performed for each pertur- information was calculated by the addition of MI to the total number of bation of the 80% matrix (exclusion analyses and SI analyses). Jackknife character states for each character in the matrix (Nixon 2002). Addition- analyses were conducted in TNT (Goloboff et al. 2008), using a 36% ally, for MP trees, steps per informative character (S/IC), matrix informa- probability of removal and the default settings for the traditional search. tion per informative character (MI/IC), and total matrix information per For each analysis, clade support was calculated for the consensus tree of informative character (TMI/IC) were calculated. These three measure- the 80% matrix with all 10 cpDNA regions included. We noted nodes at ments were determined by dividing the number of steps for a region on which support values differed by at least 10% from those of the 80% the MP trees from the matrix with all 10 cpDNA regions, the MI for a matrix with all 10 regions included. region, or the TMI for a region by the number of informative characters ANALYSES OF MATRICES OF FIVE-CPDNA REGIONS—Randomly constructed for the same region. five-cpDNA-region matrices were created and analyzed in a MP frame- CHARACTER EVOLUTION—Morphological characters, such as the presence/ work to compare the utility of five regions, as opposed to 10, in the absence of heterostyly, cleistogamy, and Macromeria-type plants (corollas reconstruction of phylogenetic relationships. Using the 80% matrix of the greater than three centimeters in length, exserted anthers and stigmas, and exclusion and SI analyses, matrices for each cpDNA region were gener- leaves with evident secondary venation) (Cohen 2011), and geographic ated. The matrices of these individual regions were concatenated to con- locations were scored for the 67 species included in the phylogeny (Cohen struct 100 randomly selected five-region matrices. Each matrix was 2011). The ancestral character state reconstructions were mapped onto analyzed using the abbreviated MP search strategy described above, with the MP and BI trees. Fitch optimization (Fitch 1971), as implemented in one exception. Instead of holding 10,000 trees in memory, 100,000 trees Winclada v1.7 (Nixon 2002), was used to investigate the evolution of these were held. characters. Fitch optimization is a conservative method for the reconstruc- The total number of resolved nodes and the number of non-contradictory tion of ancestral character states and results in an estimate of the minimal nodes were counted on the consensus tree from each matrix, and the dif- number of transitions between states. The numbers of gains and losses ference between these two numbers, the number of contradictory nodes, were recorded for each character. also was calculated, as described above. EXCLUSION ANALYSES—A series of exclusion analyses was conducted to compare and evaluate the contribution of each cpDNA region to the structure of the phylogeny. Each exclusion analysis involved the removal of one region, followed by an MP analysis of a matrix consisting of the Results remaining nine regions. Because the full 67-taxon, 10-cpDNA-region matrix included 20% missing data, two data matrices were constructed; Sequence Variation—A total of 10,036 aligned nucleotide one includes species with all 10 regions sampled (the 100% matrix), and sites (Table 1) from 67 species were included in the analyses. 2012] COHEN AND DAVIS: MOLECULAR PHYLOGENETICS OF LITHOSPERMUM 493

Most portions of all regions aligned unambiguously, with the exception of trnQ-rps16. The alignment of this region was ambiguous between species of Cynoglosseae and Echiochileae and those of Lithospermeae; therefore, the (TMI/IC) Total Matrix

Information per alignments for species of these tribes were treated as if they

Informative Character were separate regions, and thus uninformative regarding relationships between these two groups. The mean number of species sequenced for each region was 54, with the greatest number of species, 60, sequenced for the trnL-trnF spacer, and the smallest number of species, 50, sequenced for the Total Matrix psbJ-petA, trnK-rps16, and trnQ-rps16 spacers. The sequence Information (TMI) data yielded 918 parsimony-informative characters, 811 (88%) from nucleotides, and 107 (12%) from gaps, inversions, and unusual nucleotide motifs. The sequenced regions ranged in length from 546 base pairs (bp) in the trnQ-rps16 spacer in

(MI/IC) species of Lithospermeae to 1,390 bp in the ycf6-psbM spacer.

Information per The regions with the greatest number of informative nucleo-

Informative Character tides were the rpl16 intron, with 121 informative nucleotides out of 1,310 bp (11.6% of the total sequence length), the psbJ- petA spacer, with 110 informative nucleotides out of 1,060 bp (MI)

Matrix (11.1% of the total sequence length), and matK,with108infor-

Information mative nucleotides out of 1,299 bp (8.4% of the total sequence length). The least informative regions in terms of total number of characters were the trnQ-rps16, psbA-trnH,andtrnL-trnF

(S/IC) spacers, each of which provided 43 or 44 informative nucleo- Steps per Character Informative tides. However, both psbA-trnH and trnL-trnF have a greater percentage of informative nucleotides, 10.6% and 7.2% respec- tively, than trnQ-rps16, which yields between 2% and 5%

% Total informative nucleotides. Additionally, although ycf6-psbM pro- Characters Informative vides 84 informative nucleotides (almost as many as trnQ- rps16 and trnL-trnF combined), it is one of the longest regions included in the present study, and only 6.2% of the nucleo- Total

Characters tides are informative. Of the ten sampled regions, the ycf6- Informative psbM, ndhF-rpl32,andpsbJ-petA spacers provide the greatest number of structural DNA characters (indels, inversions, and unusual nucleotide motifs), 18, 17, and 15 respectively. Most other regions have fewer than 10 structural DNA characters. Of all the cpDNA regions, the rpl16 intron has the largest MI and TMI, 557 and 981 respectively (Table 1). Although rpl16 Gaps, Inversions, and Unusual Motifs

Number of Informative also has the largest TMI/IC, 7.7, matK yields the largest MI/IC, 4.6. The trnQ-rps16 spacer produces the smallest MI, TMI, MI/IC, and TMI/IC. Phylogenetic Results—MATRIX OF 10 CPDNA REGIONS— Eighty most-parsimonious trees of 1,808 steps (CI - 0.62, Nucleotides % Informative RI - 0.80) were discovered. The strict consensus tree is pro- vided in Fig. 1. The 50% majority rule consensus tree of the BI analyses, with a likelihood of –31,821.6425, is provided in Fig. 2. Topologies of the consensus trees from the MP and Number of Informative Nucleotides BI analyses are nearly identical for both outgroup and ingroup relationships. Relationships among species of Cynoglosseae are well

Aligned resolved and well supported, with all branches having greater Length (bp) than 96% jackknife support (JK) and a posterior probability (PP) of 1. Omphalodes Tourn. ex Moench is monophyletic and

57535050 98151 563 1,06060 1,058 7954 110 4458 973 856 90 10.2 1,390 11.1 10.6 1,299 88 44 84 9.8 108 9.5 17 7.2sister 15 4 6.2 8.4 to 9Mertensia 96 8 125 7 48 18 9 12.5 12.7Roth. 99 11.5 96 51 2.3 The 102 2.1 10.9 2.4 117 10.4two 8.3 431 1.8 7.6 426species 9.1 206 1.9 2.1 1.6 361 of 4.5 1.8 3.4Cynoglossum 304 4.3 176 318 539 3.7 3.2 718 824 3.5 3.1 362 4.6 666 612 7.4 6.6 325 7.5 624 881 6.7 6.4 6.4 6.1 7.5 of taxa Number included in the analyses, C. officinale L. and C. pringlei Greenm., 50 (45 + 5) 546 + 927 27 + 16 5 + 2 3 + 11 57 5.8 + 3.4 1.6 123 2.2 271 4.8 do not form a monophyletic group. Cynoglossum officinale is sister to Lindelofia longiflora (Benth.) Baill., while C. pringlei is sister to Amsinckia tessellata A. Gray. Although only three

1. Summary statistics for 10 cpDNA regions included in analyses. species of Boragineae (Anchusa leptophylla Roem. & Schult.,

Region Symphytum asperum Lepech, and Trachystemon orientalis (L.) intron 55 1,310 121 11.6 6 127 12.2 2.1 557 4.4 981 7.7

Table G. Don) are included in the present analysis, the tribe is

ndhF - rpl32 psbA - trnH psbJ - petA rpl16 trnK - rps16 trnL - rpl32 trnL - trnF trnQ - rps16 ycf6 - psbM matK Mean10 cpDNA Regions 538 10,036 54supported 811 1,004 as 8.1 monophyletic 81 8.3 107 (99% JK, 1 11 918 PP). 9.1 91 2 9.5 3,443 2 3.8 344 6,400 3.7 7 626 6.7 494 SYSTEMATIC BOTANY [Volume 37

Fig. 1. Strict consensus of 80 most-parsimonious trees (L – 1,808 steps, CI/RI – 0.62/0.80) obtained with matrix that includes 67 taxa and 10 cpDNA regions. Numbers above branches are jackknife percentages greater than 50%. Rectangles denote origins of heterostyly. Triangles represent originsof Macromeria-type plants. Circles denote origin of cleistogamy. Diamonds represent species that Johnston and other authors traditionally have not included in Lithospermum (see text for discussion). Species in light gray ellipses are native to North America. Species in dark gray ellipses are native to southern Africa. Species in open ellipses are native to South America. Species without ellipses are native to Eurasia and northern Africa. Boragineae, Cynoglosseae, and Echiochileae are in brackets, but Lithospermeae (Podonosma and its sister group) is not.

Lithospermeae is monophyletic, and the tribe forms a well weak support (< 50% JK, 0.89 PP). A clade including Cerinthe resolved and well supported clade (92% JK, 1 PP). Podonosma L., Mairetis I. M. Johnst., Paramoltkia Greuter, Halacsya Do¨rfl., Boiss. is sister to the rest of Lithospermeae (92% JK, 1 PP). A Neatostema I. M. Johnst., and two species of Lithodora receives clade composed of Echium L., Onosma L., and Maharanga DC. strong support (85% JK, 1 PP). The resolution in this clade is sister to the remainder of the genus, but this clade has differs between the resulting phylogenies of the MP and BI 2012] COHEN AND DAVIS: MOLECULAR PHYLOGENETICS OF LITHOSPERMUM 495

Fig. 2. 50% majority rule consensus tree (Likelihood: -31,821.6425) obtained with matrix that includes 67 taxa and 10 cpDNA regions. Numbers above branches are posterior probabilities greater than 0.9. Boragineae, Cynoglosseae, and Echiochileae are in brackets, but Lithospermeae is not.

analyses (Figs. 1–2). For example, in MP analyses Cerinthe is than 85% JK, while each of three species pairs, L. scabrum resolved as part of a polytomy with the other members of this Thunb. and L. cinereum DC., L. johnstonii J. I. Cohen and clade; however, in the BI analyses, Cerinthe is placed as sister L. exsertum (D. Don) J. I. Cohen, L. multiflorum A. Gray and to the other members of this group. In the results of both L. macromeria (DC.) J. I. Cohen, receives 50% JK or greater. analyses, this seven-species clade is sister to one composed Additionally, L. calcicola B. L. Rob. is supported as sister to of Arnebia Forssk. and Moltkia Lehm. L. leonotis and L. nelsonii with 62% JK. In the BI analyses, all Buglossoides does not form a monophyletic group because of these relationships receive strong support (>0.95 PP), as do B. purpureo-caerulea L. is not a member of the moderately to an additional five clades (Fig. 2). well supported clade (72% JK, 1 PP) that includes the other The MP and BI analyses recover similar relationships three species of the genus sampled for the present study. among members of the ingroup, but the BI analyses result Glandora is not monophyletic according to the structure of in slightly more resolution and support. The only conflicting the present phylogeny; however, the relationships between relationship between the two analyses involves the place- the species of this genus have weak support. Glandora oleifolia ment of L. caroliniense (Walter ex J. F. Gmel.) MacMill. In the (Lapeyr.) D. C. Thomas is resolved as sister to Lithospermum, MP analyses, this species is resolved as sister to a clade that and Glandora diffusa (Lag.) D. C. Thomas is sister to the clade includes L. revolutum and L. calycosum (Fig. 1). However, in that includes G. oleifolia and Lithospermum. the BI analyses, L. caroliniense is sister to a large clade that Both Lithospermum and the New World members of includes 19 species, and this relationship receives strong Lithospermum are resolved as monophyletic, but these two support (0.97 PP) (Fig. 2). Although other differences are clades, along with most others of the ingroup, receive weak noticeable between the MP and BI consensus trees, these support in MP analyses (Fig. 1), but moderate support, 0.99 PP differences involve resolution, not conflict, among species and 0.93 PP respectively, in BI analyses (Fig. 2). Despite weak and clades. to moderate support, in the MP analyses each of six species CHARACTER EVOLUTION—Three morphological characters, pairs, L. officinale L. and L. erythrorhizon Siebold and Zucc., heterostyly, cleistogamy, and Macromeria-type plants, were L. trinervium (Lehm.) J. I. Cohen and L. discolor M. Martens & examined in a phylogenetic context. On the MP and BI con- Galeotti, L. revolutum B. L. Rob. and L. calycosum I. M. Johnst., sensus trees, heterostyly is resolved in Lithospermum as either L. helleri (Small) J. I. Cohen and L. molle Muhl., L. mirabile I. M. originating at least seven times or at least six times with Johnst. and L. mirabile + incisum,andL. nelsonii Greenm. and one loss (Fig. 1, rectangles). Additionally, heterostyly is L. leonotis (I. M. Johnst.) J. I. Cohen, is supported by greater reconstructed as originating independently four other times 496 SYSTEMATIC BOTANY [Volume 37 amongspeciesofBoraginaceae(Fig.1,rectangles).Cleistogamy matrix. Analyses of the matrices with one region excluded evolved independently twice in Lithospermum,andone tended to resolve consensus trees with as much or slightly additional time in Neatostema (Fig.1,circles),amemberof more resolution than analyses of the matrix with all 10 regions Lithospermeae. Macromeria-type plants are restricted to mem- included (Table 2B). In only two instances did the removal of a bers of Lithospermum. This type of plant originated at least region result in the resolution of contradictory nodes. The either six times or five times with one loss (Fig. 1, triangles), removal of each of two regions, trnQ-rps16 and matK,resulted and these patterns of gains and loss are resolved on both the in consensus trees with much more resolution than that with MP and BI trees. all 10 regions included. The analysis of the matrix with the Given the structure of the phylogeny, optimization of the removal of trnQ-rps16 resulted in the resolution of 48 nodes, biogeography of Lithospermum suggests that the genus origi- but six of the 48 nodes were contradictory. In contrast, anal- nated in Eurasia (Fig. 1), with independent dispersal events yses of the matrix with the removal of matK resolves a con- to southern Africa (Fig. 1, dark gray ellipses) and to North sensus tree with 11 more nodes than that with all 10 regions, America (Fig. 1, light gray ellipses). The South American and none of these nodes was contradictory. species (Fig. 1, open ellipse) arose via a colonization event SUCCESSIVE-INCLUSION ANALYSES—In general, consensus trees by an ancestral species from North America. obtained from the initial SI analyses (i.e. with one, two, or EXCLUSION ANALYSES—Analyses of the 100% matrix with three regions) with the successive addition of regions with all 10 regions included yielded 156 MP trees of 630 steps the greatest number of informative characters were more (Table 2A). The consensus resolved 16 nodes, all of which fully resolved than those from the addition of the regions are congruent with the consensus of the 67-taxon matrix. In with the smallest number of informative characters (Fig. 3). general, the matrices with one region removed resolved a However, with the addition of more regions, this trend does consensus with as much or only slightly less resolution than not continue. After the inclusion of four or five regions, with all 10 regions included (Table 2A). The analysis of the approximately the same numbers of nodes are resolved with matrix with ndhF-rpl32 excluded resolved the greatest num- the addition of subsequent regions, regardless of whether the ber of nodes, 28, but also resolved one contradictory node, added regions have the most or the fewest informative char- with L. officinale placed as sister to L. caroliniense (Walter ex acters. This pattern is observed with both the 100% and 80% J. F. Gmel.) MacMill. rather than to the rest of Lithospermum. matrices. The matrices that involve the addition of regions The exclusion of other solitary regions (i.e. nine-region anal- with successively larger numbers of informative characters yses) did not result in support for any contradictory nodes. yield consensus trees with fewer contradictory nodes than do Analysis of the matrix that excludes the rpl16 intron provided those assembled with the addition of regions in the opposite the most resolution, 24 nodes, without supporting any con- order. The successive addition of regions of the 100% matrix tradictory nodes. The exclusion of the ycf6-psbM spacer does not provide as clear a pattern as does the successive resulted in the fewest nodes resolved (15). addition of regions of the 80% matrix (Fig. 3). In both matri- The 80% matrix with all 10 regions included resulted in the ces, the inclusion of 10 cpDNA regions does not result in discovery of 654 MP trees of 1,686 steps, and the consensus consensus trees with the greatest number of nodes resolved; resolved 35 nodes (Table 2B). As with the 100% matrix, the the addition of seven (80% matrix) or eight (100% matrix) consensus tree was congruent with that of the 67-taxon regions does so.

Table 2. Results of exclusion analyses of 100% matrix (A) and 80% matrix (B).

Number of Number of Number of A. Excluded Region MP Trees Length CI/RI Nodes Resolved Contradictory Nodes ndhF-rpl32 4 544 0.59/0.68 28 2 psbA-trnH 148 593 0.59/0.69 16 0 psbJ-petA 227 524 0.61/0.71 16 0 rpl16 intron 3 544 0.58/0.68 24 0 trnK-rps16 425 569 0.59/0.68 16 0 trnL-rpl32 38 546 0.58/0.68 19 0 trnL-trnF 103 603 0.58/0.68 17 0 trnQ-rps16 28 582 0.59/0.68 19 0 ycf6-psbM 206 570 0.58/0.68 15 0 matK 17 578 0.57/0.66 20 0 10 cpDNA Region 156 630 0.29/0.34 20 n/a

Number of Number of Number of B. Excluded Region MP Trees Length CI/RI Nodes Resolved Contradictory Nodes ndhF-rpl32 65 1,479 0.63/0.80 37 0 psbA-trnH 2,635 1,572 0.63/0.80 31 0 psbJ-petA 954 1,437 0.64/0.81 33 1 rpl16 intron 378 1,439 0.62/0.79 36 0 trnK-rps16 148 1,505 0.62/0.79 36 0 trnL-rpl32 363 1,504 0.31/0.40 38 0 trnL-trnF 65 1,602 0.63/0.80 37 0 trnQ-rps16 2 1,593 0.31/0.40 48 6 ycf6-psbM 890 1,530 0.61/0.79 34 0 matK 8 1,493 0.31/0.39 46 0 10 cpDNA Regions 653 1,686 0.31/0.40 35 n/a 2012] COHEN AND DAVIS: MOLECULAR PHYLOGENETICS OF LITHOSPERMUM 497

Fig. 3. Results of successive-inclusion analyses of 100% matrix (A) and 80% matrix (B; see text). Black diamonds represent number of total nodes resolved by successive addition of regions in order of most to fewest informative characters. Dark gray squares denote number of non-contradictory nodes resolved by successive addition of regions in same order. Gray triangles denote number of total nodes resolved by successive addition of regions in order of fewest to most informative characters. Light gray circles represent number of non-contradictory nodes resolved by successive addition of regions in same order.

80% MATRIX JACKKNIFE ANALYSES—Most of the 35 nodes in of five cpDNA regions resolved between 27 and 34 total the consensus of the 80% matrix are supported by more than nodes, with 35 nodes reconstructed from the 80% matrix with 95% JK, although six nodes have JK support less than 50% all 10 cpDNA regions included. However, most of these (Fig. 4). In the jackknife analyses of the exclusion analysis matrices resulted in consensus trees with between 27 and matrices, the removal of one region results in at least three 31 non-contradictory nodes. The median number of nodes (although in one case two) nodes with a minimum of a 10% resolved in the consensus trees of these analyses is 32 total change in JK (either upward or downward) (Table 3). Seven nodes and 29 non-contradictory nodes. The mode is 31 total nodes (11, 12, 22, 25, 27, 29, and 31) are most affected by the nodes and 29 non-contradictory nodes. There is a greater exclusion of one region. In four of the exclusion analyses, a range in the total number of nodes reconstructed than in the change of at least 10% JK occurs at four or more of these nodes. number of non-contradictory nodes resolved. In the SI analyses, the order in which regions are added, Most of the randomly constructed matrices of five cpDNA either greatest number of informative characters to smallest regions resolved at least one node that is contradicted by or vice versa, affects the JK values, even if the order does not the consensus of the 80% matrix with 10 regions included. greatly influence the number of nodes resolved (Fig. 3, and However, 72% of the matrices yielded three or fewer contra- Results of Successive-Inclusion Analyses). When regions are dictory nodes, and only 8% of the matrices reconstructed added in the order of most informative characters to fewest, consensus trees with greater than seven contradictory nodes fewer regions must be added for the JK values to be similar to (Fig. 5). those with all 10 regions included (Table 3). Not unexpect- edly, this is the case for the seven nodes that are most Discussion affected by the removal of one region. ANALYSES OF MATRICES OF FIVE-CPDNA REGIONS—In these Phylogenetics—PHYLOGENETICS OF LITHOSPERMUM—Lithospermum, analyses, the total number of nodes resolved in the consensus as currently circumscribed, is a monophyletic group, albeit tree is usually greater than the number of non-contradictory with only weak to moderate support (Figs. 1–2). The results nodes resolved. These numbers are the same in only 17% of obtained in the present analyses are consistent with those the cases (Fig. 5). Most of the randomly constructed matrices of Cohen and Davis (2009) and Cohen (2011) and provide 498 SYSTEMATIC BOTANY [Volume 37

Fig. 4. Strict consensus of 654 most-parsimonious trees (L – 1,686 steps, CI/RI – 0.62/0.80) obtained with 80% matrix that includes 53 species, each of which has data for at least eight of the 10 cpDNA regions. Numbers above branches are jackknife percentages. Numbers below branches are identifica- tion numbers for nodes (see text). 02 OE N AI:MLCLRPYOEEISO IHSEMM499 LITHOSPERMUM OF PHYLOGENETICS MOLECULAR DAVIS: AND COHEN 2012]

Table 3. Jackknife (JK) values for nodes resolved by 80% matrix and various subsets. Bold numbers in gray boxes denote a decrease in JK values by 10% or more from the values obtained with the complete 80% matrix, and numbers in bold italics in black-lined boxes represent an increase by 10% or more. See Fig. 4 for node numbers. A terminal node unites a terminal pair of species, and a subterminal node is located one branch in from a terminal node on phylogeny.

SI Analyses (regions added in order of greatest to smallest SI Analyses (regions added in order of smallest to greatest Exclusion Matrices (region excluded below) number of informative characters) number of informative characters)

Location 80% ndhF- psbA- psbJ- trnK- trnL- trnL- trnQ- ycf6- psbJ- ycf6- trnK- trnL- ndhF- trnQ- psbA- trnL- psbA- trnQ- ndhF- trnL- trnK- ycf6- psbJ- Node of Node matrix rpl32 trnH petA rpl16 rps16 rpl32 trnF rps16 psbM matK rpl16 petA matK psbM rps16 rpl32 rpl32 rps16 trnH trnF trnH rps16 rpl32 rpl32 rps16 psbM matK petA 1 terminal 100 100 100 100 100 100 100 100 100 100 100 90 100 100 100 100 100 100 100 100 2 99 99 100 100 100 100 100 100 2 subterminal 97 96 95 96 96 97 97 97 96 97 88 28 71 91 93 94 94 95 95 97 059626261727294 96 3 internal 100 100 100 100 100 100 100 100 100 100 100 31 99 89 99 100 100 99 100 100 0848598 98 99 100 99 100 4 subterminal 100 100 100 100 100 100 100 100 100 100 100 53 100 100 100 100 100 100 100 100 361 96 94 94 97 98 100 100 5 terminal 100 100 100 100 100 100 100 100 100 100 100 83 100 100 100 100 100 100 100 100 15 99 100 100 100 100 100 100 100 6 internal 99 96 99 99 99 100 99 99 99 100 89 08099 100 99 99 99 99 99 00 08782656899 99 7 subterminal 100 100 100 100 100 100 100 100 100 100 100 08499 100 100 100 100 100 100 08788100 100 100 100 100 100 8 terminal 99 93 99 98 99 99 99 99 99 98 99 0858597 99 97 99 99 99 0292993 91 89 97 98 99 9 internal 100 100 100 100 100 100 100 100 100 100 100 086100 100 100 100 100 100 100 092 97 100 100 100 100 100 100 10 terminal 86 88 89 84 79 62 73 86 81 84 88 166515586 88 87 89 86 00 72948506379 11 internal 63 66 46 48 56 60 56 65 61 81 25 0 0 83 63 58 47 49 47 65 00 1021312856 12 internal 57 65 55 8 60 44 29 60 46 58 51 0054 51 57 62 53 58 60 0 0 0 0 0 1 10 6 60 13 terminal 62 67 60 48 63 37 27 64 66 63 58 000059 72 64 64 64 0 0 3 0 11 37 37 41 63 14 subterminal 96 96 98 96 95 83 89 90 97 97 99 241187 94 95 94 90 19 25 35 46 59 89 87 78 95 15 terminal 100 100 100 100 100 100 100 100 100 100 100 22 87 100 100 100 100 100 100 100 21 92 100 100 100 100 100 100 100 16 internal 100 100 100 100 100 100 100 100 100 100 100 06398 99 100 99 99 100 100 0 0 11 82 80 94 99 99 100 17 internal 97 98 96 97 86 96 97 98 94 87 95 06292 98 96 96 91 97 98 0 0 37 16 18 16 82 88 86 18 internal 99 99 99 99 99 99 99 99 98 97 98 0547394 99 95 98 99 99 0 3 63 79 83 84 99 99 99 19 internal 47 2 45 44 44 44 55 45 47 45 46 0 44 30 22 11 5 46 46 45 01 258 46 43 44 44 44 20 internal 99 98 99 100 98 97 99 100 99 99 99 0024699 99 99 99 100 00 0758892 96 97 98 21 terminal 99 100 100 99 100 100 99 100 99 99 100 0311787100 98 99 100 100 052 94 97 98 99 100 100 100 22 internal 62 36 37 46 71 72 65 62 59 63 58 0 0 0 0 35 5 32 35 62 002267 69 55 54 59 71 23 terminal 39 13 4 37 42 32 41 38 37 44 38 000041 02038 001845 34 44 41 41 42 24 terminal 99 98 99 99 94 99 99 99 99 99 99 56 89 93 89 99 96 99 99 99 00 0 71 67 83 85 91 94 25 subterminal 43 63 44 16 13 56 39 29 34 43 40 62643 42 51 24 20 29 29 1 0 0 0 0 0 0 0 13 26 terminal 100 100 100 100 100 100 100 100 100 100 100 70 91 88 99 100 100 100 100 100 3277096 100 99 100 100 100 27 subterminal 40 22 42 50 33 23 38 44 31 18 37 000439 16 36 45 44 00 0592343 47 33 28 terminal 100 100 100 100 100 100 100 100 100 100 100 61 92 100 100 100 100 100 100 100 35 21 93 99 99 100 100 100 100 29 subterminal 48 35 49 59 43 51 35 32 19 46 54 000047 11336320 0 123659 55 55 51 43 30 terminal 90 96 90 95 85 83 85 86 93 91 85 23 24 22 23 92 91 88 84 86 29 11 55 38 68 83 84 92 85 31 subterminal 37 34 16 45 37 12 34 34 17 37 48 000043 14 13 20 34 00 812937 44 43 37 32 terminal 99 99 98 99 98 99 99 99 99 93 99 8445897 99 97 98 98 99 0304153446397 98 98 33 subterminal 88 83 86 74 88 87 70 89 92 76 89 010547783 92 87 89 0 0 0 1 51 52 72 73 88 34 terminal 99 99 99 98 99 98 98 99 99 97 98 36 54 84 93 97 99 99 99 99 21 5 3 9 63 84 95 98 99 35 terminal 97 97 97 95 85 98 94 98 98 98 97 55 83 85 89 97 97 97 98 98 0 0 0 0 62 63 63 63 85 500 SYSTEMATIC BOTANY [Volume 37

Fig. 5. Results of randomly constructed five-region analyses. Histograms depict numbers of occurrences of various results among a sample of analyses. A. Total number of nodes resolved (darker gray bars) and number of non-contradictory nodes resolved (lighter gray bars). B. Number of contradictory nodes. additional details. The species that had been segregated into consistent with the MP and BI trees. Multiple origins of other New World genera are nested among the species of heterostyly in Lithospermum also have been suggested by Lithospermum. The suite of morphological features formerly the results of other analyses (Ferrero et al. 2009; Weigend used to characterize the genus Macromeria (corollas greater et al. 2009; Cohen 2011). In another study that included all than three centimeters in length, exserted anthers and stig- heteorstylous species of Lithospermum, Cohen (2011) hypoth- mas, and leaves with evident secondary venation) originated esized five independent origins, but no losses, of the breeding at least six times or five times with one loss (although despite its system in the genus. In fact, given the current understanding structure, L. oblongifolium has not been included in Macromeria of the phylogeny of Boraginaceae, heterostyly appears to have Greenm.) (Fig. 1, triangles). This contrasts with the situation in evolved independently at least ten times within the family the two species formerly included in Onosmodium, L. molle (Thomas et al. 2008; Ferrero et al. 2009; Cohen 2010). There- Muhl. and L. helleri (Small) J. I. Cohen. These species are fore, despite the complexities of the breeding system, it is sisters, and they are characterized by adaxially glabrous, cam- possible for heterostyly to originate multiple times within a panulate corollas, precociously exserted styles, and leaves genus and a family. with evident secondary venation. Although the present anal- Cleistogamous flowers have been observed in three species ysis indicates that this combination of morphological characters of Lithospermum (Johnston 1952). On the MP and BI phyloge- originated once, only 25% of the species previously included nies (Fig. 1, circles), this character is optimized as having orig- in Onosmodium were sampled. inated twice within the genus, once in L. calycosum and once in Breeding systems within Lithospermum, along with their the ancestor of L. mirabile and L. mirabile + incisum, a result also patterns of evolution, are diverse. Heterostyly is known to recovered by Cohen (2011). Levin (1972) reported cleistoga- occur in eight species of Lithospermum (Johnston 1952), and mous flowers in the heterostylous species L. caroliniense,but the present phylogenetic analyses include all of them. Ances- we have not been able to confirm this observation. Cleistogamy tral character state reconstruction of heterostyly on the MP is infrequent in Boraginaceae, but in addition to Lithospermum, and BI trees indicates that heterostyly originated multiple the breeding system is known to occur in other species, such times within the ingroup, though origins and losses of this as Neatostema apulum (Fig. 1, circles) (Johnston 1953a; Cohen breeding system can be optimized in two different manners. 2011), another species in Lithospermeae, and Cryptantha Heterostyly either originated at least seven times or at least capituliflora (Clos) Reiche (Calvin˜o and Galetto 2003), a mem- six times with one loss (Fig. 1, rectangles). Both scenarios are ber of Cynoglosseae. 2012] COHEN AND DAVIS: MOLECULAR PHYLOGENETICS OF LITHOSPERMUM 501

The phylogenies presented here suggest that Lithospermum placed as sister to L. mirabile in the phylogeny (Figs. 1–2), originated in the Old World and subsequently colonized and this placement is consistent with the structure and form other regions, such as Africa and the New World. The two of the specimen. Future analyses that incorporate nuclear loci southern African species included in the analyses, L. cinereum may help to determine more precisely the relationship of this and L. scabrum, are sisters, and this result is consistent with specimen, and if it is of hybrid origin. the phylogeny of Weigend et al. (2009), in which the five The phylogenies presented here differ in two significant southern African species are part of a monophyletic group. aspects from other recent phylogenies of the genus. First, this In the present analysis, the clade of southern African species phylogeny provides evidence that New World members of is sister to one composed of the Eurasian and New World Lithospermum are monophyletic. Thomas et al. (2008) could members of Lithospermum. The Eurasian species, L. officinale not determine if the New World species are monophyletic and L. erythrorhizon, are members of a clade that is sister to because their phylogeny lacked the necessary resolution. the New World species (Figs. 1–2). Lithospermum officinale and Weigend et al. (2009) and Cohen (2011) resolved Old World L. erythrorhizon, along with the closest relatives of the genus, members nested among New World species. Utilizing vari- Glandora and Buglossoides, have a Eurasian and north African ous DNA regions and taxon combinations, Weigend et al. distribution (Johnston 1952, 1953b, 1954a, b; Greuter et al. (2009) resolved different clades with various combinations 1984). Subsequently, the members of Lithospermum are recon- of North American, Eurasian, and African species as sister to structed as having independently colonized both southern the remainder of the genus. In contrast, the present study, Africa and the New World. In these two areas, the ancestral which includes greater DNA and taxon sampling, and thus is species diversified, giving rise in southern Africa to L. cinereum, a more critical test of phylogenetic relationships, did not L. scabrum, and three other species, and in the New World, resolve any of these clades in this position. The second dif- undergoing many speciation events, which resulted in the ference involves the amount of resolution that occurs, in the majority of the species of the genus. consensus tree, among species of Lithospermum. Other phylo- The present analyses cannot test the monophyly of the five genetic reconstructions (Thomas et al. 2008; Weigend et al. South American species of Lithospermum because it includes 2009) have utilized two or three DNA regions. The data from only one representative from this continent, L. gayanum I. M. this number of DNA regions have not resolved much of a Johnst. Weigend et al. (2009) included most of the South backbone for ingroup relationships, although they have American species in their phylogenetic analyses, and they supported some relationships towards the tips of the tree. In found the group to be monophyletic. If the South American the present analyses many of the ingroup relationships species are monophyletic, then the structure of the present towards the tips of the tree are well resolved, albeit with low phylogeny suggests that three Mexican species, L. distichum support in some instances, as are those of the backbone Ortega, L. oblongifolium, and L. strictum Lehm., are the closest (Figs. 1–2); however, the BI analysis results in greater support relativesoftheSouthAmericanmembersofthegenus for more clades than does the MP analysis. (Figs. 1–2). In a subsequent study, Weigend et al. (2010b) PHYLOGENETICS OF LITHOSPERMEAE—Recently, relationships included L. mediale I. M. Johnst. in their analyses (a species within Lithospermeae have begun to be understood as with a form similar to that of L. distichum and L. gayanum and Thomas et al. (2008), Cohen and Davis (2009), Weigend et al. with a geographic distribution ranging from Guatemala to (2009, 2010b), Cecchi and Selvi (2009), and Selvi et al. (2009) northern South America), and in the resulting phylogeny, have focused on investigations of the intrageneric evolution- the South American species were not resolved as monophy- ary relationships of members of the tribe. The relationships letic. This is because L. mediale was not included in the same presented here (Figs. 1–2) are consistent with those found in clade as the other South American species; however, this other studies of members of Lithospermeae, with two excep- phylogeny had limited taxon sampling, and was based solely tions. In the present phylogeny, neither Buglossoides nor on ITS (see discussion in Materials and Methods regarding the Glandora, the closest relatives of Lithospermum, is resolved as use of ITS for phylogeny reconstruction in Lithospermum). The monophyletic. Buglossoides purpureo-caerulea is not part of the inclusion of L. mediale in future analyses that employ more same clade as the other members of the genus included in the character data may help to determine if the distribution of the present analyses. The phylogenies of La˚ngstro¨m and Chase South American species of Lithospermum originated via long (2002), Thomas et al. (2008), and Weigend et al. (2009) have distance dispersal, if the ancestor of L. mediale initially colonized suggested this relationship, but oftentimes the consensus trees South America and later gave rise to a small radiation of species of their phylogenetic analyses did not resolve the position of in the Andes, or if the South American species of Lithospermum B. purpureo-caerulea as either included or excluded from the arose via multiple, independent colonization events. clade with other members of Buglossoides.Thephylogeniesof One collection included in the present analyses is hypothe- these authors resolve a polytomy between B. purpureo-caerulea sized to be a putative hybrid between L. mirabile and L. incisum. and other members of Buglossoides (but see Ferrero et al. [2009] This collection has been recognized as such because the veg- for evidence supporting a monophyletic Buglossoides). Although etative features resemble those of L. mirabile, while the nutlets in the present phylogeny Buglossoides is not found to be mono- are similar to those of L. incisum. This description, L. mirabile- phyletic, its two sections, Margarospermum (Rchb.)I.M.Johnst. like leaves and L. incisum-like nutlets, is similar to that of and Eubuglossoides I. M. Johnst., may each be monophyletic L. confine I. M. Johnst. Unfortunately, the only specimen of (cf.Weigendetal.2009;Figs.1–2).Therefore,thesesections L. confine that we have seen is the type, and the specimens may be closely related, but not sisters. Johnston (1952) recognized as L. confine are unavailable. The present phylogenies resolve Glandora oleifolia as sister Given this limited amount of material, it is difficult to confi- to Lithospermum, and Glandora diffusa as sister to G. oleifolia dently assign this specimen to this species. and Lithospermum. Even though Glandora is not reconstructed Johnston (1952) hypothesized that L. incisum, L. mirabile, as monophyletic, the present study includes only 33% of the and L. confine are closely related. The putative hybrid was species of the genus. Additional taxon sampling should be 502 SYSTEMATIC BOTANY [Volume 37 conducted prior to a reconsideration of the taxonomy of for the design of molecular systematics studies, but as Buglossoides and Glandora. evidenced by the present study and others (e.g. Mast et al. PHYLOGENETICS OF BORAGINACEAE—Members of four tribes, 2004; Miller et al. 2009), the most informative regions differ Boragineae, Cynoglosseae, Echiochileae, and Lithospermeae, from taxon to taxon. For example, Shaw et al. (2007) hypoth- are included in the present analyses. The tree is rooted esized that of 34 non-coding cpDNA regions, the trnL-rpl32 between Antiphytum floribundum A. Gray, the only member and trnQ-rps16 spacers should provide the greatest number of the Echiochileae (La˚ngstro¨m and Chase 2002) included in of informative characters (including gaps and inversions) for the analyses, and the rest of Boraginaceae (Figs. 1–2). Each angiosperms at the intrageneric level, but we obtained differ- of the other three tribes, Boragineae, Cynoglosseae, and ent results. The trnL-rpl32 spacer was often only the fifth Lithospermeae, is resolved as monophyletic, and the rela- most informative region, while the trnQ-rps16 spacer was tionships among these three tribes are consistent with those always among the three least informative regions. This low from other analyses, such as Mansion et al. (2009) and ranking could be due, in part, to its small length in members Weigend et al. (2010a). of Lithospermeae. In the present analyses, each of two genera of Cynoglosseae The data from the present study suggest that the rpl16 was represented by more than one species. Of these intron and the psbJ-petA spacer would provide the greatest two, Omphalodes is monophyletic, but Cynoglossum is not. number of informative characters for future intrageneric Cynoglossum officinale is reconstructed as sister to Lindelofia studies of Boraginaceae. Shaw et al. (2007) ranked these two longiflora, and both are Old World species. Cynoglossum pringlei regions as the thirteenth and sixth most informative, respec- is resolved as sister to Amsinckia tessellata,andbothareNew tively. Furthermore, in the present study rpl16, ndhF-rpl32, World species. Given the taxon sampling in the present anal- psbJ-petA, and trnK-rps16 each provide more than 20 informa- yses, it is not possible to determine whether the characters tive characters at the intrageneric level. Shaw et al. (2007, that diagnose Cynoglossum are convergent or ancestral. Through their Fig. 4) included these four regions among the 15 (of 34) future analyses that include greater taxon sampling within most informative regions. In contrast, other regions that Cynoglosseae, it may be possible to better understand the ranked highly for intrageneric studies do not appear to be as evolutionary relationships of Cynoglossum and related genera. useful in Boraginaceae. For example, the trnV-ndhC spacer, 10cpDNA-Region Matrix—Exclusion analyses were con- the third most informative region identified by Shaw et al. ducted to determine if any single region or set of single (2007), provided little information when initially screened for regions is primarily responsible for the results obtained by the present study, and was dropped. analyses of the 10 cpDNA-region matrix. The consensus trees One of the goals of the present study is to determine which from each of the exclusion analyses resolve approximately cpDNA regions are best used in future phylogenetic investi- the same number of nodes as those with all 10 regions gations. The results of the exclusion, SI, and randomly included (Table 2). Therefore, it is concluded that no single constructed 5-cpDNA region analyses helped address this region contributes overwhelmingly to the resolution seen in question. None of the 10 regions provides an overwhelming the 10-region phylogeny. However, the removal of some amount of resolving power in the 10-region matrix, so the regions, such as trnK-rps16 and psbJ-petA, has a greater effect structure of the phylogeny is not primarily the result of one on JK values than does the removal of other regions, includ- or two cpDNA regions. Some regions have a greater number ing trnL-trnF and matK (Table 3). of S/IC, and therefore are less congruent with the phyloge- With the present data set, the number of resolved nodes netic structure of the MP trees. Steps per informative charac- also increases as taxon sampling increases, an important ter does not appear to be strongly linked to the percentage of aspect in the reduction of phylogenetic error (Zwickl and informative characters or the length of a cpDNA region. The Hillis 2002). The consensus of the 80% matrix resolves a S/IC of a region is a useful measure of congruence, but in the greater percentage of nodes than does the 100% matrix (68.6% present study, the results of this statistic demonstrate that it vs. 62.5%). Furthermore, the consensus trees reconstructed is independent of the number of informative characters, the with all 67 taxa and all 10 regions (Figs. 1–2) resolve a greater length of a region, MI, and TMI. percentage of nodes (84.6%) than either of the other two In the SI analyses, the use of five regions, either those that matrix subsets. Additionally, the exclusion matrices derived yield the greatest or smallest number of informative charac- from the 80% matrix tend to yield consensus trees that are as ters, results in consensus trees with approximately the same much or slightly more resolved than that of the consensus phylogenetic relationships as analyses with all 10 regions from the 80% matrix with 10 regions included. This result (Fig. 3). In the present study, the trnL-trnF, psbA-trnH, and differs from that of the exclusion matrices derived from the trnQ-rps16 spacers are the three regions that yield the 100% matrix, which tend to result in consensus trees with fewest informative characters. Each of these regions provides slightly less resolution than the consensus from the 100% ca. 40 informative characters, approximately half as many as matrix with 10 regions included. Moreover, fewer regions the regions with the fourth and fifth fewest informative char- need to be added to the 80% matrix than to the 100% matrix acters. The addition of these fourth and fifth regions to the in order to reconstruct relationships that will not be contra- matrix that includes trnL-trnF, psbA-trnH, and trnQ-rps16 dicted (Fig. 3). Comparisons of these analyses demonstrate that results in a sharp increase in the number of clades resolved greater resolution is obtained when more species are included, in the consensus. This number of clades is approximately despite an increase in the percentage of missing data. equal to that from the analysis of the matrix that includes Chloroplast DNA Sampling in Boraginaceae—Selection of the five regions with the greatest number of informative char- the particular cpDNA regions to include in a phylogenetic acters (Fig. 3), and fewer contradictory nodes are resolved study can affect the number informative characters as well as with addition of regions in order of least informative to most the resolution of the resulting phylogeny. Studies like those informative than in the opposite order. The results of the SI of Shaw et al. (2005, 2007) provide excellent starting points analyses provide evidence that the analysis of a matrix that 2012] COHEN AND DAVIS: MOLECULAR PHYLOGENETICS OF LITHOSPERMUM 503 includes any combination of five cpDNA regions should yield the 10-region matrix, but these consensus trees also resolved relationships that are similar to those obtained from a matrix at least a few contradictory nodes (Fig. 5). Rather than rec- with all 10 cpDNA regions. Additionally, the results of the SI ommend five regions that resolve a large number of nodes, analyses are consistent with those of other studies, such as but at the same time have been observed to support contra- Wortley et al. (2005) and Hardy et al. (2008), that have found dictory nodes and introduce type I error, it seems more pru- that additional sequencing efforts, after a threshold is reached, dent to suggest five regions that will not be contradicted by appear to yield diminishing returns. the addition of more data. This error can be determined only In contrast to the number of nodes resolved, the jackknife with the addition of more sequence data, and collection of analyses of the 80% matrices of the SI analyses provide a these data is not always possible. more complete understanding of the relationship between The use of the cpDNA regions we recommend contrasts character data, resolution, and support. The order in which with recent studies of genera of Boraginaceae (e.g. Thomas regions are added may not influence the number of nodes et al. 2008; Weigend et al. 2009, 2010a, b; Cecchi and Selvi resolved, but it does affect the JK support for nodes. When 2009; Selvi et al. 2009; Ferrero et al. 2009). These studies have regions are added in the order of most informative charac- relied heavily on the trnL-trnF spacer and ITS, a region that ters to least, rather than in the opposite order, fewer cpDNA has been known to result in misleading phylogenies (A´ lvarez regions need to be concatenated to achieve JK support that and Wendel 2003). The trnL-trnF spacer yields fewer infor- is similar to that obtained in analyses of the matrix with all mative characters, less information (both MI and TMI), and ten regions included (Table 3). Although the results of the SI lower JK values than most of the other regions used in the analyses suggest that any combination of five cpDNA regions present study (Tables 1 and 3). Investigators initially may not will yield results similar to those with 10 cpDNA regions, the want to spend time and money sequencing this region; how- results of jackknife analyses provide evidence that support for ever, trnL-trnF can complement the use of other regions. nodes may be similar only if the five regions with the most Analyses of this cpDNA spacer resolve only a small number informative characters are included. After a certain number of nodes, but none of these nodes are contradictory. Further- of regions are added, the JK values of nodes do not seem to more, the inclusion of trnL-trnF can result in increased sup- change, but this number differs depending on the order in port for particular nodes (Table 3). which regions are included. Future studies in Lithospermum will incorporate a greater The results of the SI analyses and jackknife analyses of the number of cpDNA regions as well as include investigations 80% matrix provide information of only limited scope, and of phylogenies based on nuclear DNA. With more cpDNA the results of the randomly constructed five-cpDNA region regions, it will be possible to examine whether the use of matrices yield a better understanding of character sampl- 10 cpDNA regions is sufficient to reconstruct a phylogeny of ing strategies. Analyses of the randomly constructed five- Lithospermum, and whether the results of a phylogeny based on cpDNA-region matrices provide evidence that the amount cpDNA data conflict with those from nuclear DNA sequences. of resolution can vary with the use of different combinations The inclusion of nuclear regions, such as the COSII markers of cpDNA regions (Fig. 5). The five-region analyses indicate of Wu et al. (2006) or members of the pentatricopeptide repeat which regions, when excluded, cause the greatest disparity gene family (Yuan et al. 2009), will allow for an investigation of between the total number of nodes resolved and the number the congruence and conflict of the evolutionary relationships of non-contradictory nodes resolved. The exclusion of vari- reconstructed from different genomes. Additionally, nuclear ous combinations of the following regions results in analyses and cpDNA regions can be concatenated into a combined that yield well resolved consensus trees with multiple (often matrix in which to examine the phylogenetic relationships of greater than five) contradictory nodes: the rpl16 intron, matK, species of Lithospermum and Boraginaceae. psbA-trnH, and trnL-rpl32. The use of these four regions, in Acknowledgments. concert with the trnQ-rps16 spacer (chosen because of its We would like to thank Maria Alejandra Gandolfo-Nixon, Caroline D. Kellogg, Melissa A. Luckow, James S. effect in the exclusion analyses of the 80% matrix), results in Miller, Karl J. Niklas, Andrew L. Hipp, and two anonymous reviewers a consensus tree with 34 nodes resolved, none of which is for helpful comments on this manuscript. Comments at Botany 2010 from contradictory. This is one fewer node than the consensus tree Dick Olmstead and Damon Little were helpful, as these comments pro- from the 80% matrix with all 10 regions included. Further- voked us to further examine the information content of the matrices. This research was supported by grants from the American Society of Plant more, the exclusion of each of these five regions results in Taxonomists, the Harold Moore Jr. Funds, Cornell University’s chapter changes in the JK values of five or fewer nodes (Table 3). This of Sigma Xi, the Cornell University Graduate School, Cornell Univer- contrasts with some of the other quickly evolving regions, sity’s Latin American Studies Program, and Cornell University’s College such as trnK-rps16 and psbJ-petA, in which the JK values of of Agriculture and Life Science’s Andrew W. Mellon Student Research five or more nodes increase or decrease. Therefore, we rec- Grant. Without Shannon C. K. Straub, Marı´a Hilda Flores Olvera, Helga Ochoterena, Socorro Gonzalez, Paty Ledesma Hernandez, Fernando ommend the aforementioned five regions for future phyloge- Alzate, Lucı´aVa´zquez, Janelle M. Burke, and Caroline D. Kellogg, it netic studies of genera of Boraginaceae, especially those that would have been much more difficult to locate and collect these plants. include thorough outgroup sampling. Although other com- In addition, the National Botanic Garden of Belgium, Missouri Botanical binations of cpDNA regions can provide the same amount of Garden, Chicago Botanic Garden, Cornell Plantations, and US Depart- ment of Agriculture provided leaf material or seed, and the Royal Botanic non-contradictory resolution, attempts to find other patterns Garden, Kew and the South African National Biodiversity Institute sup- among these 10 regions, rather than a serendipitous selection plied DNA isolations of some species. of five regions, were unsuccessful. Our recommendation of which five cpDNA regions to sample for future phylogenetic analyses in Boraginaceae Literature Cited is conservative. This is because some of the randomly A´ lvarez, I. and J. F. Wendel. 2003. Ribosomal ITS sequences and plant constructed five-region combinations resulted in consensus phylogenetic inference. Molecular Phylogenetics and Evolution 29: trees with a greater number of nodes resolved than that of 417–434. 504 SYSTEMATIC BOTANY [Volume 37

Barfuss, H. J., R. Samuel, W. Till, and T. F. Stuessy. 2005. Phylogenetic Secre´tariat Med-Checklist, Botanischer Garten & Botanisches Museum relationships in subfamily Tillandsioideae (Bromeliaceae) based on Berlin-Dahlem. DNA sequence data from seven plastid regions. American Journal of Hall, T. A. 1999. BioEdit: a user-friendly biological sequence alignment Botany 92: 337–351. editor and analysis program for Windows 95/98/NT. Nucleic Acids Bo¨hle, U.-R., H. H. Hilger, and W. F. Martin. 1996. Island colonization Symposium Series 41: 95–98. and evolution of insular woody habit in Echium L. (Boraginaceae). Hardy, C. R., P. Moline, and H. P. Linder. 2008. A phylogeny for the Proceedings of the National Academy of Sciences USA 93: 11,740–11,745. African Restionaceae and new perspectives on morphology’s role in Buys, M. H. 2006. A morphological cladistic analysis of Lobostemon generating complete species phylogenies for large clades. Interna- (Boraginaceae). South African Journal of Botany 72: 383–390. tional Journal of Plant Sciences 169: 377–390. Calvin˜o, A. and L. Galetto. 2003. Cleistogamy in the rare Andean peren- Hasenstab, K. E. and M. G. Simpson 2008. Phylogenetic systematics of nial herb Cryptantha capituliflora (Boraginaceae). Plant Systematics and the genus Cryptantha (Boraginaceae). [submitted abstract at Botany Evolution 237: 41–50. 2008, Vancouver, British Columbia, Canada]. de Candolle, A. P. 1846. Boraginaceae. Pp. 1–178 in Prodromus Systematis Hilger, H. H., F. Selvi, A. Papini, and M. Bigazzi. 2004. Molecular system- Naturalis Regni Vegetabilis vol. 10, eds. A. P. de Candolle and A. L. P. atics of Boraginaceae tribe Boragineae based on ITS1 and trnL P. de Candolle. Paris: Fortin, Masson et sociorum. sequences, with special reference to Anchusa s. l. Annals of Botany 94: Cecchi, L. and F. Selvi. 2009. Phylogenetic relationships of the monotypic 201–212. genera Halacsya and Paramoltkia and the origins of serpentine adap- Huelsenbeck, J. P. and F. Ronquist. 2001. MRBAYES: Bayesian inference tation in circum-mediterranean Lithospermeae (Boraginaceae): insights of phylogeny. Bioinformatics 17: 745–755. from ITS and matK DNA sequences. Taxon 58: 700–714. Johnston, I. M. 1935. Studies in the Boraginaceae, XI. Journal of the Arnold Cohen, J. I. and J. I. Davis. 2009. Nomenclatural changes in Lithospermum Arboretum 16: 145–205. (Boraginaceae) and related taxa following a reassessment of phylo- Johnston, I. M. 1952. Studies in the Boraginaceae, XXIII. A survey of the genetic relationships. Brittonia 61: 101–111. genus Lithospemum. With three plates. Journal of the Arnold Arboretum Cohen, J. I. 2010. “A case to which no parallel exists”: The influence of 33: 299–366. Darwin’s Different Forms of Flowers. American Journal of Botany 97: Johnston, I. M. 1953a. Studies in the Boraginaceae, XXIV. A. Three 701–716. genera segregated from Lithospermum. B. Supplementary notes on Cohen, J. I. A phylogenetic analysis of morphological and molecular Lithospermum. Journal of the Arnold Arboretum 34: 1–16. characters of Lithospermum L. (Boraginaceae) and related taxa: Johnston, I. M. 1953b. Studies in the Boraginaceae, XXV. A revaluation of evolutionary relationships and character evolution. Cladistics 27: some genera of the Lithospermeae. Journal of the Arnold Arboretum 34: 559–580. 258–300. Cue´noud, P., V. Savolainen, L. W. Chatrou, M. Powell, R. J. Grayer, and Johnston, I. M. 1954a. Studies in the Boraginaceae, XXVI. Further revalu- M. W. Chase. 2002. Molecular phylogenetics of Caryophyllales based ations of the genera of the Lithospermeae. Journal of the Arnold Arbo- on nuclear 18S rDNA and plastid rbcL, atpB, and matK DNA retum 35: 1–81. sequences. American Journal of Botany 89: 132–144. Johnston, I. M. 1954b. Studies in the Boraginaceae, XXVII. Some general Davis, J. I., D. W. Stevenson, G. Petersen, O. Seberg, L. M. Campbell, J. V. observations concerning the Lithospermeae. Journal of the Arnold Freudenstein, D. H. Goldman, C. R. Hardy, F. A. Michelangeli, M. P. Arboretum 35: 158–166. Simmons, C. D. Specht, F. Vergara-Silva, and M. Gandolfo. 2004. A La˚ngstro¨m, E. and M. W. Chase. 2002. Tribes of Boraginoideae phylogeny of monocots, as inferred from rbcL and atpA sequence (Boraginaceae) and placement of Antiphytum, Echiochilon, Ogastemma variation, and a comparison of methods for calculating jackknife and Sericostoma: A phylogenetic analysis based on atpB plastid DNA and bootstrap values. Systematic Botany 29: 467–510. sequence data. Plant Systematics and Evolution 234: 137–153. Davis, J. I. and R. J. Soreng. 2010. Migration of endpoints of two genes La˚ngstro¨m, E. and B. Oxelman. 2003. Phylogeny of Echiochilon relative to boundaries between regions of the plastid genome in the (Echiochileae, Boraginaceae) based on ITS and morphology. Taxon grass family (Poaceae). American Journal of Botany 97: 874–892. 52: 725–735. Doyle, J. J. and J. L. Doyle. 1990. Isolation of plant DNA from fresh tissue. Levin, D. 1972. Plant density, cleistogamy, and self-fertilization in natural Focus (San Francisco, Calif.) 12: 13–15. populations of Lithospermum caroliniense. American Journal of Botany Edgar, R. C. 2004. MUSCLE: multiple sequence alignment with high 59: 71–77. accuracy and high throughput. Nucleic Acids Research 32: 1792–1797. Linnaeus, C. 1753. Species Plantarum.Stockholm:Impendisdirect. Farris, J. S. 1989. The retention index and the rescaled consistency index. Laurentii Salvii. Cladistics 5: 417–419. Mansion, G., F. Selvi, A. Guggisberg, and E. Conti. 2009. Origin of Medi- Farris, J. S., V. A. Albert, M. Ka¨llersjo¨, D. Lipscomb, and A. G. Kluge. terranean insular endemics in the Boraginales: Integrative evidence 1996. Parsimony jackknifing outperforms neighbor-joining. Cladistics from molecular dating and ancestral reconstruction. Journal of Bio- 12: 99–124. geography 36: 1282–1296. Felsenstein, J. 2004. Inferring phylogenies. Sunderland: Sinauer Associ- Mast, A. R., D. M. S. Feller, S. Kelso, and E. Conti. 2004. Buzz-pollinated ates, Inc. Dodecatheon originated from within the heterostylous Primula subge- Ferrero, V., J. Arroyo, P. Vargas, J. D. Thompson, and L. Navarro. 2009. nus Auriculastrum (Primulaceae): A seven-region cpDNA phylogeny Evolutionary transitions of style polymorphism in Lithodora and its implications for floral evolution. American Journal of Botany (Boraginaceae). Perspectives in Plant Ecology, Evolution and Systematics 91: 916–942. 11: 111–125. Miller, J. S., A. Kamath, and R. A. Levin. 2009. Do multiple tortoises equal Fitch, W. M. 1971. Toward defining the course of evolution: Minimum a hare? The utility of nine noncoding plastid regions for species-level change for a specific tree topology. Systematic Zoology 20: 406–416. phylogenetics in tribe Lycieae (Solanaceae). Systematic Botany 34: Freudenstein, J. V. and J. I. Davis. 2010. Branch support via resampling: 796–804. an empirical study. Cladistics 26: 643–656. Nixon, K. C. 1999. Parsimony ratchet, a new method for rapid parsimony Goloboff, P. 1999a. Analyzing large data sets in reasonable times: Solu- analysis. Cladistics 15: 407–414. tions for composite optima. Cladistics 15: 415–428. Nixon, K. C. 2002. WinClada ver. 1.7. Ithaca, New York: Published Goloboff, P. 1999b. NONA (NO NAME) ver. 2 Tucuma´n, Argentina: by author. Published by the author. Ochoterena, H. 2009. Homology in coding and non-coding DNA Goloboff, P., J. S. Farris, and K. C. Nixon. 2008. TNT, a free program for sequences: a parsimony perspective. Plant Systematics and Evolution phylogenetic analysis. Cladistics 24: 774–786. 282: 151–168. Graham, S. W., P. A. Reeves, A. C. E. Burns, and R. G. Olmstead. 2000. Posada, D. and K. A. Crandall. 1998. Modeltest: testing the model of DNA Microstructural changes in noncoding cpDNA: Interpretation, evo- substitution. Bioinformatics 14: 817–818. lution, and utility of indels and inversions in basal angiosperm Ronquist, F. and J. P. Huelsenbeck. 2003. MRBAYES 3: Bayesian phyloge- phylogenetic inference. International Journal of Plant Sciences 161: netic inference under mixed models. Bioinformatics 19: 1572–1574. S83–S96. Selvi, F., M. Bigazzi, H. H. Hilger, and A. Papini. 2006. Molecular phylogeny, Greuter, W., H. M. Burdet, and G. Long (eds.). 1984. Med-Checklist: A morphology, and taxonomic re-circumscription of the generic com- critical inventory of vascular plants of the circum-Mediterranean countries, plex Nonea/Elizaldia/Pulmonaria/Paraskevia (Boraginaceae-Boragineae). Volume 1: Pteridophyta, Gymnospermea, Dicotyledones (Acanthaceae – Taxon 55: 907–918. Cneoraceae). Gene`ve: Editions des Conservatoire et Jardin botaniques Selvi, F., L. Cecchi, and A. Coppi. 2009. Phylogeny, karyotype evolution de la Ville de Gene`ve, Med-Checklist Trust of OPTIMA, Berlin: and taxonomy of Cerinthe L. (Boraginaceae). Taxon 58: 1307–1325. 2012] COHEN AND DAVIS: MOLECULAR PHYLOGENETICS OF LITHOSPERMUM 505

Shaw, J., E. B. Lickey, J. T. Beck, S. B. Farmer, W. Liu, J. Miller, K. C. I. M. Johnst., 19792083/PI 296073, Natl. Bot. Gard. Belgium and USDA, -, Siripun, C. T. Winder, E. E. Schilling, and R. L. Small. 2005. The FJ827384, FJ827357, FJ827434, FJ827411, FJ827475, FJ827312, FJ827453, -, tortoise and the hare II: Relative utility of 21 noncoding chloroplast FJ827287; Cerinthe major L., Cohen 91, Cultivated in New York, JF488876, DNA sequences for phylogenetic analysis. American Journal of Botany JF488904, JF488933, JF488959, JF488986, JF489012, JF489035, JF489062, 92: 142–166. JF489086, JF489114; Cynoglossum amabile Stapf & Drummond, Cohen 89, Shaw, J., E. B. Lickey, E. E. Schilling, and R. L. Small. 2007. Comparison of Cultivated in New York, JF488859, JF488886, JF488916, JF488944, whole chloroplast genome sequences to choose noncoding regions JF488969, -, -, JF489046, JF489067, JF489097; Cynoglossum pringlei Greenm., for phylogenetic studies in angiosperms: the tortoise and the hare III. Cohen 219, “Michoacan, Mexico”, JF488868, JF488895, -, -, JF488978, -, -, American Journal of Botany 94: 275–288. JF489054, -, JF489106; Echium vulgare L., Cohen 212, Cornell Plantations, Simmons, M. P. and H. Ochoterena. 2000. Gaps as characters in sequence- FJ827257, FJ827364, FJ827341, FJ827417, FJ827391, FJ827457, FJ827293, -, based phylogenetic analyses. Systematic Biology 49: 369–381. FJ827317, FJ827269; Glandora diffusa (Lag.) D. C. Thomas, Chase 6063, RBG Simmons, M. P. and M. Miya. 2004. Efficiently resolving the basal clades Kew, FJ827258, FJ827365, FJ827342, -, FJ827392, -, FJ827294, -, -, FJ827270; of a phylogenetic tree using Bayesian and parsimony approaches: a Glandora oleifolia (Lapyer.) D. C. Thomas, “Chase 34889, 34890”, RBG Kew, case study using mitogenomic data from 100 higher teleost fishes. JF488878, JF488906, JF488935, JF488961, JF488988, JF489014, JF489037, Molecular Phylogenetics and Evolution 31: 351–362. JF489064, JF489088, JF489116; Halacsya sendtneri Do¨rfl., Genbank, Simmons, M. P. and J. V. Freudenstein. 2011. Spurious 99% bootstrap and EU919618, -, -, -, -, -, -, EU044885, -, -; Lindelofia longiflora Baill., 19981513- jackknife support for unsupported clades. Molecular Phylogenetics 95, Natl. Bot. Gard. Belgium, JF488860, JF488887, JF488917, JF488945, and Evolution 61: 177–191. JF488970, -, JF489021, JF489047, JF489068, JF489098; Lithodora hispidula Thomas, D. C., M. Weigend, and H. H. Hilger. 2008. Phylogeny and (Sibth. & Sm.) Griseb., Chase 34888, RBG Kew, JF488874, JF488902, systematics of Lithodora (Boraginaceae-Lithospermeae) and its affin- JF488931, JF488958, JF488984, JF489010, JF489033, JF489060, JF489084, ities to the monotypic genera Mairetis, Halacsya and Paramoltkia JF489113; Lithodora zahnii (Heldr. ex Hala´csy) I. M. Johnst., Chase 34891, based on ITS1 and trnLUAA - sequence data and morphology. Taxon RBG Kew, JF488873, JF488901, JF488930, JF488957, JF488983, JF489009, 57: 79–97. JF489032, JF489059, JF489083, JF489112; Maharanga emodi DC., Genbank, Weigend, M., M. Gottschling, F. Selvi, and H. H. Hilger. 2009. -, -, -, -, -, -, -, FJ763269, -, -; Mairetis microsperma (Boiss.) I. M. Johnst., Marbleseeds and gromwells–Systematics and evolution of Lithospermum Genbank, EU919620, -, -, -, -, -, -, FJ763257, -, -; Mertensia virginica Link, and allies (Boraginaceae tribe Lithospermeae) based on molecular Cohen 89, “New York, U. S. A.”, JF488869, JF488896, JF488925, JF488952, and morphological data. Molecular Phylogenetics and Evolution 52: JF488979, JF489004, JF489027, -, JF489070, JF489107; Moltkia petraea (Tratt.) 755–768. Griseb., 2000 1260-54, Natl. Bot. Gard. Belgium, JF488875, JF488903, Weigend, M., M. Gottschling, F. Selvi, and H. H. Hilger. 2010a. Fossil JF488932, -, JF488985, JF489011, JF489034, JF489061, JF489085, -; and extant western hemisphere Boragineae, and the polyphyly Neatostema apulum (L.) I. M. Johnst., Genbank, EU919627, -, -, -, -, -, -, of “Trigonotideae” Reidl (Boraginaceae: Boraginoideae). Systematic FJ763262, -, -; Omphalodes cappadocica DC, 1995-3595, Missouri Bot. Gard., Botany 35: 409–419. JF488871, JF488898, JF488927, JF488954, JF488981, JF489006, JF489029, Weigend, M., M. Gottschling, H. H. Hilger, and N. M. Nurk. 2010b. Five JF489056, -, JF489109; Omphalodes verna Moench, 19830183, Natl. Bot. new species of Lithospermum L. (Boraginaceae tribe Lithospermeae) Gard. Belgium, JF488870, JF488897, JF488926, JF488953, JF488980, in Andean South America: Another radiation in the Amotape- JF489005, JF489028, JF489055, -, JF489108; Onosma stellulata Waldst. & Huancabamba zone. Taxon 59: 1161–1179. Kit., 1992 1317-39, Natl. Bot. Gard. Belgium, -, FJ827376, -, FJ827428, Winkworth, R. C., J. Grau, A. W. Robertson, and P. J. Lockhart. 2002. The FJ827403, FJ827468, FJ827305, FJ827449, FJ827328, -; Paramoltkia doerfleri origins and evolution of the genus Myosotis L. (Boraginaceae). Molec- (Wettst.) Greuter & Burdet, Genbank, EU919630, -, -, -, -, -, -, EU044886, -, ular Phylogenetics and Evolution 24: 180–193. -; Podonosma orientalis (L.) Feinbrun, Genbank, -, -, -, -, -, -, -, FJ763307, -, -; Wortley, A. H., P. J. Rudall, D. J. Harris, and R. W. Scotland. 2005. How Symphytum asperum Lepech., Cohen 221A, Chicago Bot. Gard., JF488872, much data are needed to resolve a difficult phylogeny? Case study JF488899, JF488928, JF488955, JF488982, JF489007, JF489030, JF489057, -, in Lamiales. Systematic Biology 54: 697–709. JF489110; Trachystemon orientalis (L.) G. Don, 1978-1795, Missouri Bot. Wu, F., L. A. Mueller, D. Crouzillat, V. Pe´tiard, and S. D. Tanksley. 2006. Gard., -, JF488900, JF488929, JF488956, -, JF489008, JF489031, JF489058, -, Combining bioinformatics and phylogenetics to identify large sets of JF489111— single-copy orthologous genes (COSII) for comparative, evolution- Ingroup: Lithospermum calcicola B. L. Rob, Cohen 191, “Nuevo Leo´n, ary and systematic studies: A test case in the eusterid plant clade. Mexico”, JF488857, JF488884, JF488914, JF488942, JF488967, JF488996, Genetics 174: 1407–1420. JF489020, -, JF489075, JF489095; Lithospermum californicum A. Gray, “Cohen Yuan, Y. W., C. Liu, H. E. Marx, and R. G. Olmstead. 2009. The pentatri- 43, 157”, “California and Oregon, U. S. A.”, FJ827264, FJ827371, FJ827348, copeptide repeat (PPR) gene family, a tremendous resource for plant FJ827423, FJ827398, FJ827463, FJ827300, FJ827445, FJ827323, FJ827276; phylogenetic studies. The New Phytologist 182: 272–283. Lithospermum calycosum (J. F. Macbr.) I. M. Johnst., Cohen 197, “Nuevo Zhu, G., H. Riedl, and R. V. Kamelin. 1995. Boraginaceae. Pp. 329–427 Leo´n, Mexico”, JF488863, JF488890, JF488920, JF488948, JF488973, in Flora of China vol. 16 (Gentianaceae through Boraginaceae), JF488999, JF489024, JF489050, JF489079, JF489101; Lithospermum canescens eds. Z. Y. Wu and P. H. Raven. Beijing, China and St. Louis, Lehm., Cohen and Straub 12, “Ohio, U. S. A.”, FJ827261, FJ827368, Missouri: Science Press and Missouri Botanical Garden Press. FJ827345, FJ827420, FJ827395, FJ827460, FJ827297, FJ827442, FJ827320, Zwickl, D. J. and D. M. Hillis. 2002. Increased taxon sampling greatly FJ827273; Lithospermum caroliniense MacMill., Cohen 6 and 11, “Indiana, reduces phylogenetic error. Systematic Biology 51: 588–598. U. S. A.”, FJ827260, FJ827367, FJ827344, FJ827419, FJ827394, FJ827459, FJ827296, FJ827441, FJ827319, FJ827272; Lithospermum cinereum DC., Appendix 1. List of species included in analyses, along with collec- Genbank, -, -, -, -, -, -, -, FJ763295, -, -; Lithospermum cobrense Greene, tion information and GenBank accession numbers. Data are presented “Cohen 78, 145, 203”, “Arizona and Texas, U. S. A.”, FJ827262, FJ827369, in the order: taxon, collection, location, matK, ndhF-rpl32, psbA-trnH, FJ827346, FJ827421, FJ827396, FJ827461, FJ827298, FJ827443, FJ827321, psbJ-petA, rpl16 intron, trnK-rps16, trnL-rpl32, trnL-trnF, trnQ-rps16, and FJ827274; Lithospermum discolor M. Martens & Galeotti, “Cohen 216, 243”, ycf6-psbM. “Michoacan and Guerrero, Mexico”, JF488858, JF488885, JF488915, JF488943, JF488968, -, -, JF489045, JF489076, JF489096; Lithospermum Outgroup: Amsinckia tessellata A. Gray, W627115, USDA, JF488867, distichum Ort., “Cohen 192, 202”, “Nuevo Leo´n and D. F., Mexico”, JF488894, JF488924, JF488951, JF488977, JF489003, JF489039, JF489053, JF488854, JF488881, JF488911, JF488939, JF488964, JF488993, JF489017, JF489069, JF489105; Anchusa leptophylla Roem. & Schult., 2002 1090-96, JF489042, JF489072, JF489092; Lithospermum erythrorhizon Siebold & Zucc., Natl. Bot. Gard. Belgium, FJ827256, FJ827363, FJ827340, -, FJ827390, -, Cohen 173, Cornell Plantations, -, FJ827380, -, FJ827431, FJ827407, FJ827472, FJ827292, -, -, -; Antiphytum floribundum A. Gray, Cohen 227, “Michoacan, FJ827309, -, FJ827332; Lithospermum exsertum (D. Don) J. I. Cohen, Cohen 244, Mexico”, JF488852, JF488879, JF488909, JF488937, JF488962, JF488991, -, “Guerrero, Mexico”, -, FJ827387, FJ827360, FJ827437, FJ827414, FJ827478, JF489040, JF489066, JF489090; Arnebia benthamii (Wall ex. G. Don) I. M. FJ827314, -, FJ827337, FJ827290; Lithospermum flavum Sesse´ &Moc.,Cohen Johnst., Chase 34887, RBG Kew, JF488853, JF488880, JF488910, JF488938, 226, “Michoacan, Mexico”, -, FJ827388, FJ827361, FJ827438, FJ827415, JF488963, JF488992, JF489016, JF489041, JF489071, JF489091; Buglossoides FJ827479, FJ827315, -, FJ827338, FJ827291; Lithospermum gayanum I. M. incrassata (Guss.) I. M. Johnst., Genbank, -, -, -, -, -, -, -, FJ763255, -, -; Johnst., Genbank, -, -, -, -, -, -, -, FJ763297, -, -; Lithospermum helleri (Small) Buglossoides purpureo-caerulea (L.) I. M. Johnst., Chase 6055, RBG Kew, J. I. Cohen, Cohen 132, “Texas, U. S. A.”, -, FJ827382, -, FJ827432, FJ827409, JF488877, JF488905, JF488934, JF488960, JF488987, JF489013, JF489036, FJ827473, -, -, FJ827334, FJ827285; Lithospermum johnstonii J. I. Cohen, JF489063, JF489087, JF489115; Buglossoides tenuiflora (L. f.) I. M. Johnst. Cohen 218, “Michoacan, Mexico”, -, FJ827385, FJ827358, FJ827435, Genbank, EU599675, -, -, -, -, -, -, EU599939, -, -; Buglssoides arvensis (L.) FJ827412, FJ827476, FJ827313, FJ827454, FJ827335, FJ827288; Lithospermum 506 SYSTEMATIC BOTANY [Volume 37 latifolium Michx., Cohen 87, “New York, U. S. A.”, JF488866, JF488893, revolutum B. L. Rob., Cohen 199, “San Luis Potosı´, Mexico”, FJ827259, JF488923, JF488950, JF488976, JF489002, JF489026, JF489052, JF489082, FJ827366, FJ827343, FJ827418, FJ827393, FJ827458, FJ827295, FJ827440, JF489104; Lithospermum leonotis (I. M. Johnst.) J. I. Cohen, Cohen 195, FJ827318, FJ827271; Lithospermum rosei (I. M. Johnst.) J. I. Cohen, Cohen “Nuevo Leo´n, Mexico”, FJ827266, FJ827373, FJ827350, FJ827425, FJ827400, 207, “Durango, Mexico”, FJ827268, FJ827375, FJ827352, FJ827427, FJ827465, FJ827302, -, FJ827325, FJ827278; Lithospermum macromeria J. I. FJ827402, FJ827467, FJ827304, FJ827448, FJ827327, FJ827280; Lithospermum Cohen, “Cohen 141, 151”, “Arizona, U. S. A.”, -, FJ827377, FJ827353, -, ruderale Douglas ex Lehm., Cohen 34, “Utah, U. S. A.”, JF488856, JF488883, FJ827404, FJ827469, FJ827306, FJ827450, FJ827329, FJ827281; Lithospermum JF488913, JF488941, JF488966, JF488995, JF489019, JF489044, JF489074, matamorense DC., Cohen and Straub 68, “Texas, U. S. A.”, JF488855, JF489094; Lithospermum scabrum Thunb., LHMS 2125, SANBI and JF488882, JF488912, JF488940, JF488965, JF488994, JF489018, JF489043, Genbank, -, JF488908, -, -, JF488990, FJ827480, -, FJ763293, -, -; JF489073, JF489093; Lithospermum molle Michx., 2003-0671, Missouri Bot. Lithospermum strictum Lehm., Cohen 225, “Michoacan, Mexico”, -, Gard., -, FJ827383, FJ827356, FJ827433, FJ827410, FJ827474, FJ827311, FJ827389, FJ827362, FJ827439, FJ827416, -, FJ827316, FJ827456, FJ827339, -; FJ827452, -, FJ827286; Lithospermum multiflorum Torr. ex A. Gray, “Cohen Lithospermum trinervium (Lehm.) J. I. Cohen, Cohen 228, “Michoacan, Mex- 49, 64”, “Utah and Arizona, U. S. A.”, FJ827263, FJ827370, FJ827347, ico”, -, FJ827386, FJ827359, FJ827436, FJ827413, FJ827477,, FJ827455, FJ827422, FJ827397, FJ827462, FJ827299, FJ827444, FJ827322, FJ827275; FJ827336, FJ827289; Lithospermum tuberosum DC., Cohen 108, “Georgia, Lithospermum nelsonii Greenm., Cohen 184, “Nuevo Leo´n, Mexico”, U. S. A.”, -, FJ827378, FJ827354, FJ827429, FJ827405, FJ827470, FJ827307, -, FJ827265, FJ827372, FJ827349, FJ827424, FJ827399, FJ827464, FJ827301, FJ827330, FJ827282; Lithospermum tubuliflorum Greene, M. Gonzalez 4001, FJ827446, FJ827324, FJ827277; Lithospermum notatum (I. M. Johnst.) J. I. “Durango, Mexico”, JF488865, JF488892, JF488922, -, JF488975, JF489001, -, Cohen, Cohen 188, “Nuevo Leo´n, Mexico”, -, FJ827381, -, -, FJ827408, -, JF489081, JF489103; Lithospermum viride Greene, Cohen 86, “Texas, JF489015, FJ827310, FJ827451, FJ827333, FJ827284; Lithospermum oblongifolium U. S. A.”, JF488862, JF488889, JF488919, JF488947, JF488972, JF488998, Greenm., Cohen 201, “D. F., Mexico”, -, JF488907, JF488936, -, JF488989, -, JF489023, JF489049, JF489078, JF489100— JF489038, JF489065, JF489089, JF489117; Lithospermum obovatum J. F. Macbr., Cohen 208, “Durango, Mexico”, -, FJ827379, FJ827355, FJ827430, Appendix 2. Commands for MrBayes analyses. FJ827406, FJ827471, FJ827308, -, FJ827331, FJ827283; Lithospermum officinale L., Cohen 171, Cornell Plantations, FJ827267, FJ827374, FJ827351, FJ827426, begin MrBayes; charset molecular = 1-9869; charset structural = 9870- FJ827401, FJ827466, FJ827303, FJ827447, FJ827326, FJ827279; Lithospermum 10024; partition favored = 2: molecular, structural; set partition = favored; mirabile Small, “Cohen 83, Cohen and Straub 35”, “Texas, U. S. A.”, lset applyto=(1) nst=6 ngammacat=4 rates=invgamma; lset applyto=(2) JF488861, JF488888, JF488918, JF488946, JF488971, JF488997, JF489022, coding=variable; set autoclose=yes; mcmc nruns=1 ngen=10000000 JF489048, JF489077, JF489099; Lithospermum mirabile X incisum, Cohen and printfreq=1000 samplefreq=1000 temp=0.2700 swapfreq= 1000 nchains=8 Straub 64, “Texas, U. S. A.”, JF488864, JF488891, JF488921, JF488949, savebrlens=yes burnin=2500000; sumt burnin=2500; sump burnin= JF488974, JF489000, JF489025, JF489051, JF489080, JF489102; Lithospermum 2500; end;