American Journal of Botany 96(10): 1907–1921. 2009.

T HE EVOLUTIONARY ORIGIN OF A SECOND RADIATION OF ANNUAL C ASTILLEJA (OROBANCHACEAE) SPECIES IN SOUTH AMERICA: THE ROLE OF LONG DISTANCE DISPERSAL AND ALLOPOLYPLOIDY 1

David C. Tank 2,3,4 and Richard G. Olmstead3

2 Department of Forest Resources & Stillinger Herbarium, College of Natural Resources, University of Idaho, P.O. Box 441133, Moscow, Idaho 83844-1133 USA; and 3 Department of Biology & Burke Museum of Natural History, University of Washington, Box 355325, Seattle, Washington 98195-5325 USA

Considerable attention has been directed toward understanding the wide gaps in range that are common among many groups of closely related organisms. By placing their biology and geography in a phylogenetic context, we may gain a broader knowledge of the series of historical events that have led to present species distributions. In addition to the North American annuals, a second radiation of annual Castilleja species is in Andean Peru and central Chile. Phylogenetic analyses of chloroplast and nuclear DNA regions revealed a complex history for the origin and diversifi cation of annual Castilleja species in South America. In addition to at least three independent long-distance dispersal events from North America, allopolyploidy has played a signifi cant role in this disjunct radiation. Only C. attenuata occurs in both California and South America, and these results support its recent arrival to central Chile. Two Peruvian species are inferred to be allopolyploids; hybridization between annual lineages derived from inde- pendent long-distance dispersal events from North America gave rise to C. profunda, and hybridization between South American annual and perennial species gave rise to C. cerroana. The relative importance these events are discussed with reference to the observed morphological, ecological, and distributional patterns.

Key words: amphitropical disjuncts; biogeography; Castilleja ; GBSSI; granule-bound starch synthase/invertase; Indian paintbrush; Orobanchaceae; Peru; waxy .

Geographic disjunctions have intrigued biogeographers and disjunctions: “ The new data have, moreover, served to broaden evolutionary biologists and have often been the source of imag- the parameters of taxonomy, at the same time strengthening its inative explanations for the wide gaps in range that are common framework and providing new clues that help to untangle knotty among many groups of closely related organisms. Early work- evolutionary problems. ” ers in plant biogeography (e.g., Axelrod, 1958 ; Florin, 1963 ; Temperate plant groups that are distributed primarily in Raven, 1963 ; Good, 1964 ; Cruden, 1966 ; Thorne, 1972 ) recog- western North America, but are also represented in part in the nized the role that present distributional patterns play in our South American fl ora, are likely to be the result of two basic interpretations of the past history of organisms and the regions distributional pathways. First, there are those montane groups that they occupied. However, to correctly interpret the series of that are more or less continuously distributed at high elevation events that have led to present species distributions, we must across the American tropics and found in Andean South Amer- place the biology and geography of the group of interest in a ica, e.g., Lupinus (Hughes and Eastwood, 2006), Viburnum phylogenetic context. Although he did not use the term phylo- ( Winkworth and Donoghue, 2005 ), Valerianaceae ( Bell and genetic, Wood (1972, pp. 107 – 108 ) recognized that new sources Donoghue, 2005 ), to name a few. This type of distribution is of data brought to plant taxonomy by the then emerging fi eld of likely the result of the continuous or “ stepping stone” migration biosystematics would be essential to understanding biological and diversifi cation of these taxa through the western mountain systems that connect the two continents ( Thorne, 1972 ). In ad- dition, there are other groups that are disjunctly distributed in 1 Manuscript received 9 December 2008; revision accepted 27 May 2009. the temperate regions of both continents. Amphitropical tem- The authors thank J. Ammirati, T. Bradshaw, J. M. Egger, B. Moore, and two anonymous reviewers for critical comments on the manuscript, and J. perate disjunctions are well known between North and South M. Egger, M. Zapata, S. Leiva, A. Sagastegui, and M. Dillon for assistance America, and there are at least 65 genera representing ~130 facilitating fi eldwork in Peru. This research was supported by a Graduate species, species pairs, or species complexes in the Americas Fellowship in Molecular Systematics from the University of Washington that correspond to this distribution ( Raven, 1963 , 1972 ; Thorne, Department of Botany, the Research Award for Graduate Students from the 1972). Most species with this type of distribution are restricted American Society of Plant Taxonomists, the Karling Graduate Student to the mediterranean climates of the Pacifi c coast and are most Research Award from the Botanical Society of America, the Award for frequent between California and Chile ( Raven, 1963 ). Because Graduate Student Research from the Society of Systematic Biologists, the these temperate amphitropical disjuncts are mostly not montane Sigma Xi Grants in Aid of Research from the University of Washington in their distributions, migration via long-distance dispersal has Chapter, and the Giles Award for Graduate Student Field Research from been the preferred hypothesis to explain their disjunct distribu- the University of Washington Department of Botany to D.C.T., the NSF Doctoral Dissertation Improvement Grant DEB-0412653 to R.G.O. for tions (e.g., Raven, 1963, 1972 ; Thorne, 1972). Furthermore, D.C.T., and the NSF grant DEB-0090313 to R.G.O. Raven (1963) concluded that the recurrent pattern of closely 4 Author for correspondence (e-mail: [email protected]) related disjunct species in western North and South America was the result of long-distance dispersal of seeds via migratory doi:10.3732/ajb.0800416 birds. However, disjunct distributions between the hemispheres 1907 1908 American Journal of Botany [Vol. 96 could also arise through a series of short dispersals over time history of this species is poorly understood. Our own observa- following the mostly continuous western American mountains tions suggest that this species may in fact be annual, but may be through the tropics, accompanied by extinction in the tropics able to persist given the right environmental conditions. Al- and diversifi cation in the temperate regions ( Cruden, 1966 ; though the root of this species is robust in some individuals, Thorne, 1972). In this paper, we examine a complex distribu- observation of plants in the fi eld, as well as numerous herbar- tional pattern among annual and perennial Castilleja species ium specimens, has revealed no conclusive evidence of peren- (the paintbrushes) in North and South America that includes niality (i.e., ± woody caudex or old stems from previous years). both widely disjunct annuals and continuously distributed mon- The corolla morphology of C. cerroana is very similar to the tane perennials, and we provide a framework for the origin of annual Castilleja species (i.e., well-developed lower corolla lip the present distributional patterns in the context of a robust phy- with a more or less infl ated pouch that is evidently toothed); logenetic hypothesis. however, the calyx of C. cerroana is more similar to many pe- rennial Castilleja species (Fig. 1C). Members of the Onco- Biology and distribution of South American Castilleja— Cas- rhynchus grade have a calyx that is cleft more or less equally on tilleja comprises approximately 180 mostly perennial species all sides (i.e., subequally cleft; Fig. 1A and 1B), while the ma- found primarily in western North America and Mexico, but with jority of perennial Castilleja species have an unequally cleft about 20 species distributed in Central and South America, in- calyx (i.e., cleft deeper in the front and back than on the sides; cluding a small group of annual species distributed primarily in Fig. 1D). The unequally cleft calyx of C. cerroana is unique Andean Peru. This large and taxonomically complex genus rep- among the group of closely related species in northern Peru; resents the majority of the hemiparasitic subtribe Castillejinae G. however, this feature is also seen in C. laciniata, one of two Don (tribe Pedicularideae Duby, Orobanchaceae Vent.). Within more southerly distributed South American annual species. Castillejinae, perenniality is restricted to Castilleja , where the The remaining South American species assigned to sect. On- large majority of species (~160 spp.) are perennial herbs. Con- corhynchus were all newly described by Chuang and Heckard trary to the commonly held view that annuals are derived from (1992). These four species, C. alpicola , C. peruviana , C. pro- perennials (e.g., Stebbins, 1957 ), phylogenetic relationships in- funda , and C. vadosa , are primarily restricted to the open, high ferred from separate and combined analyses of chloroplast (cp) elevation Jalca habitat (i.e., semiarid Andean grasslands) that and nuclear ribosomal (nr) DNA regions have revealed that the dominates the Peruvian highlands (Fig. 1). Of these species, C. perennial Castilleja species evolved from a grade of annual lin- alpicola is the most restricted in its range and is only known eages in Castillejinae, including the majority of the annual Cas- from the type locality in northern Peru (Chuang and Heckard, tilleja species ( Tank and Olmstead, 2008 ). Because the ancestral 1992 ), where it grows among limestone outcrops at ~4000 m grade of annual Castilleja species represents sect. Oncorhynchus a.s.l. This species is distinguished morphologically from the (sensu Chuang and Heckard, 1991 ) and forms the basis for much other Peruvian annuals in its low stature (mostly < 10 cm tall vs. of the discussion herein, we refer to these annual species as the up to ~50 cm for the other annuals) and compact infl orescence. Oncorhynchus grade for convenience. Castilleja peruviana is distributed in northern Peru and is very The Oncorhynchus grade is comprised primarily of taxa dis- similar morphologically to C. vadosa (Fig. 1B), whose range tributed in western North America, with the majority of species extends south to central Peru. In northern Peru, the distributions geographically centered in California ( Fig. 1 ). In addition to the of these species overlap signifi cantly, and they are often seen North American annual species, a second, albeit smaller, group growing close to each other with no evidence of hybridization of annual Castilleja species occurs in South America with seven (Chuang and Heckard, 1992; D. Tank, personal observation). species distributed in central Chile and Andean Peru (Fig. 1; Morphologically, these two species are distinguished by the Chuang and Heckard, 1992). Castilleja attenuata ( Fig. 1A ) is strongly glandular pubescence and primarily red coloration of the only known diploid ( n = 12) annual species found in South C. peruviana and the mixed nonglandular and short glandular America, and the only species that is known from both western pubescence and rosy-purple coloration of C. vadosa ( Fig. 1B ). North America and South America. This self-compatible spe- The fi nal species described by Chuang and Heckard (1992) , C. cies, distributed from British Columbia to northern Baja Cali- profunda , is found along the Cordillera Occidental from central fornia in North America, is widely disjunct in central Chile, (near Lima) to southern Peru. Castilleja profunda is most simi- where it is found from near sea level to 1000 m a.s.l. and is lar vegetatively to C. laciniata , but can be distinguished from known from at least four provinces (Chuang and Heckard, the latter by its subequally cleft calyx vs. the unequally cleft 1992). The earliest known collection of C. attenuata from Chile calyx of C. laciniata. In addition, the conspicuous deeply re- was made in 1889, and it has been suggested that this species ticulate seed coat of C. profunda is unique among the South may have been only recently introduced to Chile from Califor- American annuals. Although both of these species occur in nia ( Raven, 1963 ; Chuang and Heckard, 1992 ). southern Peru, C. profunda is distributed along the western Castilleja laciniata is known from both central Chile and Cordillera Occidental and C. laciniata is found in the southeast- southeastern Peru and is distributed from sea level to ~500 m ern Department of Cusco ( Chuang and Heckard, 1992 ). a.s.l. in Chile, and above 1700 m a.s.l. in Peru ( Fig. 1 ). Despite Polyploidy in Castilleja is much more common among peren- the wide gap in the distribution of this species and the substan- nial species; however, it is not restricted to the perennials. tial elevational differences between the Chilean and Peruvian Among the western North American annuals of the Onco- populations, no morphological differences have been identifi ed rhynchus grade, C. tenuis is known from both diploid and tetra- between plants of the two regions (Chuang and Heckard, ploid populations and C. brevistyla is a putative allotetraploid 1992 ). (Chuang and Heckard, 1982; Tank and Olmstead, 2008); how- Castilleja cerroana is distributed from extreme southern Ec- ever, the other nine annual species in western North America are uador to central Peru and is usually found above 3000 m a.s.l. diploid ( Fig. 1 ; Chuang and Heckard, 1982 ). In contrast, half of (Fig. 1; Chuang and Heckard, 1992). This species was origi- the strictly South American annuals (including C. cerroana) are nally described as a perennial (Edwin, 1970); however, the life known to be polyploid ( Fig. 1 ; Chuang and Heckard, 1992 ), and October 2009] Tank and Olmstead — South American annual C ASTILLEJA 1909 it is likely that the only diploid annual species in South America With a robust phylogenetic hypothesis based on results from is the disjunct Californian species, C. attenuata . both the chloroplast and nuclear genomes, we may be able to determine the role that long-distance dispersal has played in the Hypothesized origin of South American Castilleja— Al- distribution of annual Castilleja species in South America, as though, the South American annual species are traditionally well as assess the importance of allopolyploidy to the origin classifi ed with the North American species of the Oncorhynchus and diversifi cation of this group of closely related species, es- grade, Chuang and Heckard (1992) hypothesized that the Peru- pecially with regard to the in situ allopolyploid origin of the vian species may be the result of allopolyploid hybridization of Peruvian species, as hypothesized by Chuang and Heckard other South American perennial and annual species. Because (1992) . the Peruvian species were morphologically similar to artifi cial We report the results of phylogenetic analyses of data from hybrids produced between North American annual and peren- both the chloroplast and nuclear genomes. The two noncoding nial species ( Chuang and Heckard, 1992 ), they envisioned a cpDNA regions, trnL - F (Gielly and Taberlet, 1994) and the similar allopolyploid origin for the Peruvian plants. Perennial rps16 intron (Oxelman et al., 1997), as well as the nrDNA inter- Castilleja species in South American are part of an understud- nal transcribed spacer (ITS) and external transcribed spacer ied species complex that includes species from both Central and (ETS) regions (Baldwin et al., 1995; Baldwin and Markos, South America, as well as the type species of the genus, C. fi s- 1998 ) were used to place the South American Castilleja species sifolia L. f. (Chuang and Heckard, 1993). There are four closely sampled here into the existing phylogenetic hypothesis for Cas- related perennial species in Peru (C. fi ssifolia , C. nubigena , C. tilleja ( Tank and Olmstead, 2008 ). In addition, partial sequences pumila , and C. virgata ), and chromosome counts from C. fi ssi- of the nuclear gene encoding granule-bound starch synthase folia , C. nubigena, and C. virgata suggest that members of this (GBSSI or waxy ) were used to investigate the putative allopoly- species complex in South America are largely diploid (n = 12; ploid origin of the Peruvian annuals. Sequences of the waxy Chuang and Heckard, 1993 ). gene have been used extensively to investigate interspecifi c re- Alternatively, the second radiation of annual species in South lationships in a wide range of angiosperm taxa (e.g., Miller et America may be derived from widely disjunct annual ancestors al., 1999 ; Peralta and Spooner, 2001 ; Walsh and Hoot, 2001 ; distributed in western North America. Long-distance dispersal Baumel et al., 2002 ; Small, 2004 ; Levin and Miller, 2005 ; Levin of an annual ancestor from western North America to Andean et al., 2005, 2006 ; Whitson and Manos, 2005 ; Yeung et al., South America, followed by diversifi cation in the southern 2005 ), as well as to reconstruct putative allopolyploid specia- hemisphere, could produce the observed pattern of diversity tion events (e.g., Mason-Gamer, 2001 ; Ingram and Doyle, 2003 ; among the annual Castilleja species. However, the widely dis- Smedmark et al., 2003; Winkworth and Donoghue, 2004; junct distributions between the continents could also arise Harbaugh, 2008 ). through a series of short dispersals over time following the mostly continuous western American mountains through the tropics ( Cruden, 1966 ; Thorne, 1972 ). With the exception of MATERIALS AND METHODS the North American disjunct C. attenuata and the Chilean pop- ulations of C. laciniata, the South American annual Castilleja Taxon sampling— In total, 31 species of Castilleja and three outgroup taxa species are primarily distributed in Andean Peru (Fig. 1), and were used in this study ( Table 1 ). Given the goals presented here, special atten- therefore, do not follow the typical pattern of disjunction ob- tion was given to the annual Castilleja species recognized as section Onco- served between the temperate mediterranean climates of North rhynchus sensu Chuang and Heckard (1991), where 15 of the 16 species were and South America. sampled, including the six annual species found in Andean Peru and central Chile that were treated by Chuang and Heckard (1992). For one of these spe- Because phylogenetic reconstruction produces a bifurcating cies, C. attenuata, accessions from both California and Chile were included. In tree, the reconstruction of hybrid speciation using phylogenetic addition, the sampling included C. cerroana (sect. Pallescentes sensu Chuang methods has been a diffi cult and often paradoxical problem and Heckard, 1992), as well as the three perennial South American species, C. (Funk, 1985; McDade, 1995). Chloroplast DNA is mostly ma- nubigena , C. pumila , and C. virgata . Based on the comprehensive molecular ternally inherited in angiosperms and therefore, will only trace systematic study of subtribe Castillejinae ( Tank and Olmstead, 2008 ), which the maternal history of a hybrid lineage; likewise, the nrDNA included a broad sampling of the taxonomic and geographic diversity of Cas- tilleja, 15 species of Castilleja were chosen to represent the perennial Castilleja repeat, which is homogenized through concerted evolution, clade, including C. ophiocephala , an enigmatic annual species found to be usually is fi xed for one of the parental lineages ( Sang et al., more closely related to perennial species than to members of the Oncorhynchus 1995, 1997 ; Wendel et al., 1995 ; Sang, 2002 ). In the fortunate grade. In addition, three species of Triphysaria were chosen as outgroups, event that the nrDNA repeat is fi xed for the paternal parent of based on the established sister-group relationship of Castilleja and Triphysaria an allopolyploid event, the discordant placement of the putative (Tank et al., 2009; Tank and Olmstead, 2008). Three species, C. arvensis, C. hybrid in cp and nrDNA phylogenies can allow for the recon- lineariloba, and C. ophiocephala, were represented by different accessions in the waxy data set ( Table 1 ). struction of reticulate speciation events in a phylogenetic con- text (e.g., Kim and Donoghue, 2008). Like the nrDNA repeat, Molecular methods —Total genomic DNA for accessions fi rst included in single or low-copy nuclear genes are biparentally inherited; this study ( Table 1 ) was extracted from either silica-gel-dried tissue or herbar- however, they are much less susceptible to concerted evolution ium specimens using the modifi ed 2× CTAB method ( Doyle and Doyle, 1987 ) than large gene families such as nrDNA, and therefore, have an and purifi ed using QIAquick spin-columns following the protocols of the man- increased potential to aid in the identifi cation of allopolyploid ufacturer (QIAGEN, Valencia, California, USA). speciation (reviewed in Sang, 2002). By isolating the homoe- DNA for sequencing the chloroplast (cp) DNA trnL-F and rps16 intron re- ologous loci derived from both parental genomes, an allotetra- gions and the nuclear ribosomal (nr) ITS and ETS regions for accessions fi rst included in this study was generated via polymerase chain reaction (PCR) fol- ploid can be divided into its two genomes, with each homoeolog lowing Tank and Olmstead (2008). Amplifi ed PCR products were purifi ed by tracing its own parental lineage in a standard phylogenetic anal- precipitation from a 20% polyethylene glycol (PEG) solution and washed in ysis (e.g., Sang and Zhang, 1999 ; Doyle et al., 2003 ; Smedmark 70% ethanol prior to sequencing. After repeated attempts, we were unable to et al., 2005 ; Kim et al., 2008 ). obtain PCR product for the ITS region from Castilleja profunda ; however, all 1910 American Journal of Botany [Vol. 96 October 2009] Tank and Olmstead — South American annual C ASTILLEJA 1911 species included in this study are represented by at least three of the four cp and waxy data set. Each analysis consisted of two runs of 5 000 000 generations nrDNA regions sequenced ( Table 1 ). from a random starting tree using the default priors and four Markov chains The primers used for PCR and sequencing of the 3 ′ end of the nuclear gene (using the default heating values) sampled every 100 generations. Because both encoding granule-bound starch synthase (waxy ) were designed from complete the combined cpDNA and nrDNA data set and the waxy data sets consisted of coding sequences available in GenBank from members of Asteridae, including two heterogeneous sections, partitioned analyses were performed (i.e., where Antirrhinum majus (Veronicaceae; GenBank accession AMA6293), Ipomoea the two different sections are modeled under different parameter values). In the batatas (Convolvulaceae; AB071976), and Perilla frutescens (Lamiaceae; partitioned Bayesian analyses, the combined cpDNA and nrDNA matrix was AF210699). Once a number of sequences were generated from members of partitioned by genome, whereas the waxy data set was partitioned correspond- Castillejinae (D. Tank and R. Olmstead, unpublished data), some of the primers ing to exon and intron regions. In these analyses, the individual parameters were optimized for use in the Lamiales ( Fig. 2 ). DNA for sequencing the 3′ end were unlinked across the data partitions. To decrease the chance of obtaining of the waxy gene (exons 7 – 13) used in this study was generated via PCR using false stationarity on a local optimum, we preformed long chains and two inde- the primers waxy-7F and waxy-13R (Fig. 2) under the following conditions: pendent runs for each analysis. Convergence of the chains was determined by 80° C for 2 min; fi ve cycles of 94° C for 1 min, 60° – 64° C for 1 min, and 72° C examining the plot of all parameter values and the – lnL against generation time for 3 min; 30 cycles of 94° C for 30 s, 55° – 60 ° C for 30 s, and 72° C for 1.5 min; using the program Tracer version 1.3 (Rambaut and Drummond, 2004), and and a fi nal extension for 10 min at 72 °C. After repeated attempts, we were un- stationarity was assumed when all parameter values and the – lnL had stabilized able to obtain a PCR product for any portion of the waxy gene from the Chilean for the majority of the run. In addition, the likelihoods of the independent runs accession of Castilleja attenuata , C. laciniata, and C. profunda. Each of these were considered indistinguishable when the average standard deviation of split accessions was sampled from herbarium material, and the quality of the DNA frequencies was < 0.01, as reported by the program MrBayes version 3.1.2 obtained from them was poor. Amplifi ed PCR products were cleaned by gel (Ronquist and Huelsenbeck, 2003). Burn-in trees were discarded and the re- excision using the Wizard SV Gel and PCR Clean-Up System (Promega, Madi- maining trees, and their associated parameter values, were saved. son, Wisconsin, USA), and the cleaned PCR products were cloned using the Nonparametric bootstrapping (Felsenstein, 1985) and Bayesian posterior TOPO TA Cloning kit (Invitrogen, Carlsbad, California, USA). Clones were probabilities (PP) were used to evaluate relative clade support in the phyloge- screened for inserts by PCR directly from the colony using the PCR amplifi ca- netic analyses. Bootstrapping was performed with 1000 replicates, each with 20 tion primers, the PCR products were cleaned by 20% PEG precipitation fol- replicates of stepwise random taxon addition and TBR branch swapping with lowed by washing with 70% ethanol, and any positive clones were sequenced MULTREES off ( DeBry and Olmstead, 2000 ). A majority rule consensus tree with one primer to identify unique sequences. On average, 10 – 20 positive showing all compatible partitions from the resulting posterior distribution of clones were screened using the waxy -11F primer ( Fig. 2 ), and clones that dif- tree topologies was used to recover the Bayesian PP values for each clade. fered by more than 3 bp in a 500-bp region were completely sequenced and The program MrBayes was also used to calculate Bayes factors to compare included in the phylogenetic analyses. To increase the likelihood of sampling the optimal topologies from the separate analyses of the cpDNA and nrDNA all the unique loci and/or alleles that are present in the polyploid South Ameri- data sets (i.e., three independent long-distance dispersal events; see Discussion ) can annual species (including C. cerroana ), we screened 80 – 100 positive clones with that of topologies constrained to contain alternative potential monophyl- from multiple different PCR amplifi cations for each species. etic groups with respect to the number of independent introductions of annual To ensure accuracy, we sequenced both strands of the cleaned PCR products Castilleja lineages from North America into South America (e.g., Rabeling using the DYEnamic ET Terminator Cycle Sequencing Kit (Amersham Biosci- et al., 2008 ). Therefore, we performed additional Bayesian phylogenetic analy- ences, Piscataway, New Jersey, USA) on an ABI 377 DNA sequencer (Applied ses (as described), but in the cpDNA analyses we (1) constrained the monophyly of Biosystems, Foster City, California, USA). The cpDNA trnL/F and rps16 in- all South American Castilleja species (including C. attenuata ) to test the hy- tron regions and the nrDNA ITS and ETS regions were sequenced using the pothesis of a single introduction into South America, and (2) constrained the amplifi cation primers and internal sequencing primers as detailed in Tank and monophyly of all South American Castilleja species except C. attenuata to test Olmstead (2008). The 3′ portion of the waxy gene was sequenced using the the hypothesis of two dispersal events to South America (one for C. attenuata external PCR primers, waxy-7F and waxy-13R, and the seven internal primers and one for the exclusively South American species). Similarly, for the nrDNA as shown on Fig. 2 . Sequence data were assembled and edited for each region analysis, we constrained the monophyly of all South American Castilleja spe- using the program Sequencher version 4.6 (Gene Codes Corp., Ann Arbor, cies (including C. attenuata ) to test the hypothesis of a single introduction into Michigan, USA), and consensus sequences were generated. South America vs. two introductions as inferred from the optimal topology. To calculate the Bayes factors, marginal likelihoods from constrained and uncon- strained Bayesian analyses were estimated as the harmonic mean of the sam- Phylogenetic analyses— Sequence alignments for the fi ve gene regions pled likelihoods ( Nylander et al., 2004 ; Brown and Lemmon, 2007; Rabeling et were performed manually using the program Se-Al version 2.0a11 ( Rambaut, al., 2008 ). The Bayes factor represents the ratio of the marginal likelihoods of 1996). For the two cpDNA and two nrDNA regions, parsimony-informative the two topologies under consideration and was calculated as two times the dif- gaps were coded as presence/absence characters using simple gap coding ( Graham ference between the harmonic mean of the post burn-in log-likelihood of the et al., 2000 ; Simmons and Ochoterena, 2000 ). optimal topology and the constrained topology, following Brown and Lemmon Both the trnL - F and rps16 intron regions are part of the haploid chloroplast (2007). Values > 10 are considered to be very strong evidence favoring the op- genome, and thus, their histories are linked. Likewise, the nrDNA ITS and ETS timal topology ( Kass and Raftery, 1995 ). regions are tightly linked in the rDNA repeat. Therefore, these data were treated as two independent data sets, rather than four (Tank and Olmstead, 2008). Par- simony analyses were conducted on the cpDNA, nrDNA, and combined cp- RESULTS DNA and nrDNA data sets, as well as the individual waxy data set, as implemented in the program PAUP* version 4.0b10 (Swofford, 2002). Heuris- tic searches were performed with 1000 replicates of stepwise random taxon Phylogenetic analyses— The nrDNA ITS and ETS regions, addition and tree-bisection-reconnection (TBR) branch swapping with the as well as the two cpDNA regions, trnL - F and rps16 , aligned MULTREES option in effect, except for analysis of the waxy data set where unambiguously, although numerous short gaps were introduced. MULTREES was off. The program Modeltest version 3.7 ( Posada and Crandall, Alignment of the 3 ′ portion of the waxy gene was more diffi cult. 1998) was used to determine the model of sequence evolution best fit to The seven exon regions aligned unambiguously; however, it each data set, as well as to the waxy exons and introns separately, using the Akaike information criterion. Bayesian phylogenetic analyses were conducted was necessary to introduce numerous large and small gaps using MrBayes v.3.1.2 (Ronquist and Huelsenbeck, 2003) on the cpDNA, throughout the six introns sequenced. Eight short portions of nrDNA, and combined cpDNA and nrDNA data sets, as well as the individual several introns, consisting primarily of microsatellite repeats or

¬ Fig. 1. Map showing the approximate distributions of annual and perennial Castilleja species in North and South America, with particular attention given to the South American annual species distributed primarily in Andean Peru (inset). Photographs of representative (A – C) annual and (D) perennial species are shown, including an enlarged inset of the fl ower. (A) C. attenuata (photo: B. Legler), (B) C. vadosa (photo: D. Tank), (C) C. cerroana (photo: D. Tank), (D) C. virgata (photo: D. Tank). 1912 American Journal of Botany [Vol. 96

waxy FJ939170 FJ939171 FJ939184 FJ939185 FJ939186 FJ939187 FJ939188 FJ939189 FJ939204 FJ939205 FJ939206 FJ939207 FJ939219 FJ939220 FJ939143 FJ939144 FJ939145 FJ939146 FJ939151 FJ939152 FJ939223 FJ939224 FJ939210 FJ939211 FJ939212 FJ939213 FJ939214 FJ939197 FJ939198 FJ939199 FJ939200 FJ939166 FJ939167 FJ939168 FJ939169 FJ939190 FJ939191 FJ939192 FJ939193 FJ939221 FJ939222 FJ939215 FJ939216 FJ939153 FJ939154 FJ939201 FJ939202 FJ939203

— —

trnL-F FJ939262 FJ939255 FJ939263 FJ939259

c

rps16 FJ939275 FJ939268 FJ939276 FJ939272

GenBank accession number GenBank accession number FJ939237 FJ939230 FJ939238 FJ939234

FJ939250 FJ939243 FJ939251 FJ939247 EF103709 EF103639 EF103780 EF103858

ITS ETS b — — — —

DNA voucher voucher DNA Tank 01-8 (WTU)Tank EF103694 EF103622 EF103763 EF103841 Tank Tank & Egger 05 – 19 (WTU) Beardsley s.n. (WTU)Beardsley EF103681 EF103605 EF103746 EF103824 Behn s.n. (F) Egger & Tank 1187 (WTU)Tank & Egger EF103715 EF103646 EF103787 EF103865 Egger 550 (WTU) EF103687 EF103612 EF103753 EF103831 Egger & Tank 1185 (WTU)Tank & Egger EF103698 EF103626 EF103767 EF103845 Halse 4905 (WTU) EF103686 EF103611 EF103752 EF103830 Sperling 319 (F) Egger 400 (WTU) EF103683 EF103608 EF103749 EF103827 Egger 623 (WTU) EF103688 EF103614 EF103755 EF103833 Tank & Egger 05-(WTU) & Tank Egger 559 (WTU) EF103689 EF103615 EF103756 EF103834 Egger 555 (WTU) 02-4 (WTU) Tank — EF103606 EF103747 EF103825— Tank Tank 01 – 49 (WTU) — EF103647 EF103788 EF103866 Beardsley 98-7 (WTU)Beardsley EF103684 EF103609 EF103750 EF103828

ambigua auriculata (A. Gray) T. I. Chuang & Heckard & I. Chuang T. (A. Gray) (Benth.) T. I. Chuang & Heckard& I. Chuang T. (Benth.) ex ex L.f. * (Benth.) T. I. Chuang & Heckard & I. Chuang T. (Benth.) (A. Gray) T. I. Chuang & Heckard& I. Chuang T. (A. Gray) Benth. (Benth.) T. I. Chuang & Heckard var. Heckard var. & I. Chuang T. (Benth.) Eastw. var. var. Eastw. (Benth.) T. I. Chuang & Heckard var. Heckard var. & I. Chuang T. (Benth.) * (A. Gray) T. I. Chuang & Heckard & I. Chuang T. (A. Gray) Edwin Hook. & Arn. Hook. & Arn. var. Arn. var. & Hook. Cham. & Schltdl. T. I. Chuang & Heckard& I. Chuang T.

Greenm. (A. Heller) T. I. Chuang & Heckard var. Heckard var. & I. Chuang T. (A. Heller)

Fernald (Benth.) T. I. Chuang & Heckard & I. Chuang T. (Benth.)

1. used in this study. information of plant material that was and voucher Taxa a

campestris exserta densifl ora densifl C. densifl ora C. cusickii C. cerroana C. attenuata C. campestris C. attenuata C. arvensis C. auriculata C. laciniata C. lacera C. exserta C. ambigua C. elmeri Table Castilleja Mutis C. alpicola Taxon C. lineariloba C. lineariloba C. linariifolia C. lasiorhyncha October 2009] Tank and Olmstead — South American annual C ASTILLEJA 1913

waxy FJ939225 FJ939226 FJ939228 FJ939229 FJ939139 FJ939140 FJ939194 FJ939195 FJ939196 FJ939208 FJ939209 FJ939147 FJ939148 FJ939149 FJ939150 FJ939227 FJ939217 FJ939218 FJ939141 FJ939142 FJ939174 FJ939175 FJ939176 FJ939177 FJ939180 FJ939181 FJ939182 FJ939183 FJ939172 FJ939173 FJ939159 FJ939160 FJ939178 FJ939179 FJ939155 FJ939156 FJ939157 FJ939158 FJ939138 FJ939161 FJ939162 FJ939163 FJ939164 FJ939165

—

trnL-F FJ939267 FJ939266 FJ939265 FJ939257 FJ939260 FJ939258 FJ939264 FJ939261 FJ939256 sequences.

waxy

c

rps16 FJ939280 FJ939279 FJ939278 FJ939270 FJ939273 FJ939277 FJ939274 FJ939269 FJ939271 o Index Herbariorum, http://sweetgum.nybg.org/ih/). o Index Olmstead (2008) . Olmstead (2008)

GenBank accession number GenBank accession number FJ939242 FJ939241 FJ939240 FJ939232 FJ939235 FJ939239 FJ939236 FJ939231 FJ939233

FJ939254 FJ939253 FJ939252 FJ939245 FJ939248 FJ939249 FJ939244 FJ939246

— — ITS ETS

b — — — — DNA voucher voucher DNA Egger 1259 (WTU) Egger 1261 (WTU) Egger 1265 (WTU) Tank & Egger 05-0 (WTU) & Tank Tank & Egger 05-3 (WTU) & Tank Tank 01-3 (WTU)Tank EF103682 EF103607 EF103748 EF103826 Egger & Tank 1197 (WTU)Tank & Egger EF103714 EF103645 EF103786 EF103864 Egger 570 (WTU) EF103685 EF103610 EF103751 EF103829 Weberbauer 5416 (F) Weberbauer Colwell s.n. (WTU)Colwell EF103690 EF103616 EF103757 EF103835 Tank & Egger 05-2 (WTU) & Tank Tank 01-2 (WTU)Tank EF103702 EF103631 EF103772 EF103850 Olmstead 01 – 83 (WTU) EF103701 EF103630 EF103771 EF103849 Moran 15394 (JEPS)Soza 1791 (WTU) EF103741 EF103677 EF103819 EF103897— Colwell s.n. (WTU)Colwell EF103710 EF103640 EF103781 EF103859 Tank & Egger 05-4 (WTU) & Tank Colwell s.n (WTU)Colwell EF103712 EF103642 EF103783 EF103861 Tank & Egger 05-4 (WTU) & Tank

subsp. versicolor

subsp. rst used in this study are in boldfaced type; an asterisk after a name indicates a different accession sampled for the type; an asterisk after a name indicates different rst used in this study are boldfaced tenuifl ora tenuifl (Wiggins) Tank & J. M. Egger& Tank (Wiggins) * (Wiggins) Tank & J. M. Egger & Tank (Wiggins) (Jeps.) T. I. Chuang & Heckard var. Heckard var. & I. Chuang T. (Jeps.) Torr. T. I. Chuang & Heckard& I. Chuang T. Bong. Fisch. & C. A. Mey. A. Mey. C. & Fisch. Kunth Benth. var. Benth. var. T. I. Chuang & Heckard& I. Chuang T.

Eastw. (Benth.) T. I. Chuang & Heckard & I. Chuang T. (Benth.) Douglas ex Hook. Douglas ex (Benth.) Wedd. ex Herrera ex Wedd. (Benth.) (Wedd.) Edwin (Wedd.) T. I. Chuang & Heckard& I. Chuang T. (Benth.) T. I. Chuang & Heckard& I. Chuang T. (Benth.)

Rydb. (A. Heller) T. I. Chuang & Heckard & I. Chuang T. (A. Heller) (A. Gray) T. I. Chuang & Heckard & I. Chuang T. (A. Gray) 1. Continued. a

Vouchers are for accessions used in this study (herbarium where deposited in parentheses). Herbarium acronyms are according t are for accessions used in this study (herbarium where deposited parentheses). Herbarium acronyms Vouchers Species or accessions fi and Tank type; all other sequences are from GenBank accession numbers for sequences generated by this study are in boldfaced

a b c

T. pusilla pusilla T. T. eriantha T. gratiosa T. T. versicolor rubicundula

Triphysaria

C. virgata C. vadosa C. tenuis

C. tenuifl ora C. rubicundula C. profunda C. pilosa C. peruviana C. peirsonii C. parvifl ora C. ophiocephala C. ophiocephala C. occidentalis C. nubigena Table C. miniata Taxon C. pumila 1914 American Journal of Botany [Vol. 96 long runs of a single nucleotide, could not be confi dently aligned South American perennial species sampled in this study, C. nu- and were excluded from the analyses (positions 66 – 75, 114 – bigena , C. pumila , and C. virgata , formed a well-supported 126, 138 – 140, 515 – 544, 808 – 822, 1052 – 1082, 1094 – 1114, and clade (PP = 1.0, bootstrap = 71%; Fig. 3 ) nested within the pe- 1465 – 1484 in the alignment). Sequence characteristics for the rennial Castilleja clade and was resolved as the sister group to three data sets used in this study, as well as resulting parsimony a clade of primarily Mexican species. In this analysis, the pe- tree statistics are presented in Table 2 . Trees and data matrices rennial Castilleja clade was only weakly supported (bootstrap = for the combined cpDNA and nrDNA data and the waxy data 57%, PP = 0.83; Fig. 3); however, the more inclusive clade con- are available in TreeBASE ( http://treebase.org , study accession taining the perennials plus C. ophiocephala received strong number S2352). support (PP = 1.0, bootstrap = 100; Fig. 3 ). The separate Bayesian analyses of the individual cpDNA and Castilleja alpicola , C. cerroana , C. laciniata , C. peruviana , nrDNA data, as well as the partitioned cpDNA/nrDNA com- and C. vadosa formed a well-supported monophyletic group bined data set and partitioned waxy exon/intron data each that is part of the Oncorhynchus grade that includes the North achieved apparent stationarity between 250 000 – 750 000 gen- American annuals. Within this South American annual clade, erations; however, because these analyses contained long chains C. alpicola , C. cerroana , C. peruviana, and C. vadosa formed a (5 000 000 generations) and high sample frequencies (100 gen- well-supported clade, and C. laciniata was resolved as the sister erations), a conservative burn-in of 1 000 000 generations was to this group. Castilleja attenuata, the only other annual species used for each analysis. Majority rule consensus trees calculated represented in South America, was separate from the South from the posterior distributions of tree topologies (excluding American clade, and the Californian and Chilean accessions of burn-in trees) were used to obtain the Bayesian PP values. In this species were sister to each other. Figure 3 also shows the the two independent Bayesian analyses that were conducted for discordant position of C. profunda, excluded from the com- each data set, all parameters reached stationarity at the same bined analyses, when the cpDNA and nrDNA data were ana- level. lyzed independently. Although trees resulting from the analyses of the cpDNA Results of the topological hypothesis testing using the Bayes data were not well resolved or well supported, in general, clades factor suggest that these results are robust to alternative topolo- that were supported by both moderate to high parsimony boot- gies. When either the cpDNA or nrDNA topologies were con- strap values (≥ 70%) and signifi cant Bayesian PP values (≥ 0.95) strained so that all of the South American Castilleja species were also recovered in the analyses of the nrDNA data. A no- were monophyletic, including C. attenuata , which occurs in table exception was the placement of the South American an- both North and South America (corresponding to a hypothesis nual species C. profunda . In the cpDNA analyses, C. profunda of a single introduction of an annual Castilleja lineage to South joins the North American annuals C. tenuis and C. lacera to America), the Bayes factor results indicate very strong evidence form a well-supported clade (bootstrap = 96%, PP = 1.0; Fig. 3 ). (i.e., > 10; Kass and Raftery, 1995 ) in support of the optimal By contrast, the nrDNA analyses place C. profunda in a well- topologies (cpDNA Bayes factor = 87.5; nrDNA Bayes factor = supported clade with the South American species C. alpicola , 117.2). Likewise, when the cpDNA topology was constrained C. cerroana , C. laciniata , C. peruviana, and C. vadosa (boot- so that all of the exclusively South American annual Castilleja strap = 94%, Bayesian PP = 1.0; Fig. 3 ). With the exception of species were monophyletic (i.e., C. alpicola , C. cerroana , C. this species, the trees resulting from the separate analyses of the laciniata , C. peruviana, and C. vadosa), corresponding to the cpDNA and nrDNA data sets (not shown) were largely consis- hypothesis of two independent introductions into South Amer- tent. Thus, subsequent analyses of the combined data were done ica (one for C. attenuata and one for the exclusively South with C. profunda excluded. American annual species), the Bayes factor result indicated Figure 3 shows the majority rule consensus tree with mean very strong evidence in support of the optimal topology (Bayes branch lengths resulting from the partitioned Bayesian analysis factor = 58.4). of the combined cp and nrDNA data. Overall, the topology is Figure 4 shows the majority rule consensus tree resulting consistent with results of previous phylogenetic analyses that from the partitioned Bayesian analysis of the waxy sequence included all six genera of Castillejinae and a larger sampling of data. Phylogenetic analysis of the partial sequences of the nu- perennial Castilleja species (Tank and Olmstead, 2008). The clear gene waxy resulted in a gene tree that is consistent with

Fig. 2. Diagram of the 3 ′ portion of the waxy gene used in this study. Numbered boxes represent exon regions, and intervening lines represent intron regions. Solid and dashed lines indicate the minimum and maximum lengths, respectively, of each region from sequences generated by this study. Numbers below the exon and intron regions indicate the length, or range of lengths, in base pairs for the individual regions. Arrows above exons indicate the location and direction of PCR and sequencing primers used in this study. Forward and reverse primers are indicated by black and gray arrows, respectively. Forward and reverse primers are numbered for the exon in which they are located and the sequence of each primer is indicated. Primers indicated with an asterisk are optimized for use in the Lamiales (see Methods). October 2009] Tank and Olmstead — South American annual C ASTILLEJA 1915

Table 2. Summary descriptions and parsimony results for sequences included in individual and combined analyses of two chloroplast regions, trnL / F and rps16 , and two nuclear ribosomal regions, ITS and ETS (cpDNA = chloroplast combined, nrDNA = nuclear ribosomal combined).

Description cpDNA nrDNA Combined 3′ waxy

Number of taxa included 79 79 79 66 Sequence characteristics Aligned length 1884 1134 3018 2294 Excluded characters 0 0 0 143 Gaps coded as binary characters 10 12 22 0 Variable sites 108 295 403 918 Pairwise distances (%) 0 – 1.9 0 – 24.4 0 – 24.4 0.2 – 10.6 (exon), 0.1 – 21.2 (intron) Model selection (AIC) TIM+I GTR+I+Γ GTR+I+Γ GTR+ Γ (exon), HKY+Γ (intron) Parsimony analyses Informative sites 43 198 606 579 Number of trees 8 154 48 73 Number of steps 128 509 676 1688 Consistency index (CI) 0.93 0.74 0.73 0.66 Retention index (RI) 0.95 0.84 0.82 0.84 Rescaled consistency index (RC) 0.88 0.62 0.6 0.56 Notes: AIC = Akaike information criterion previous hypotheses of relationships between the annual and small, tissue from multiple individuals was used for DNA ex- perennial Castilleja species (Tank and Olmstead, 2008). The tractions. Thus, the additional sequences may represent allelic backbone relationships among lineages comprising the Onco- diversity in the populations from which they were sampled. Al- rhynchus grade, however, were poorly supported in both the ternatively, these sequences could also be evidence of recent Bayesian and parsimony bootstrap analyses. By contrast, the gene duplications in these species, although this would entail perennial Castilleja clade that did not receive meaningful numerous independent duplication events. Bayesian or parsimony support values in the combined cpDNA As with the results of the combined analysis of cpDNA and and nrDNA analyses, was well supported by the waxy data nrDNA, the gene sequences isolated from the South American (PP = 1.0, bootstrap = 75%; Fig. 4). For the most part, the mul- perennial species, C. nubigena , C. pumila, and C. virgata , were tiple sequences isolated from a species were monophyletic and part of a well-supported clade (PP = 1.0, bootstrap = 99%; Fig. well supported (indicated with a star on Fig. 4 ). However, 4) that was nested within the perennial Castilleja clade. Se- within some clades comprised of waxy sequences from closely quences of the waxy region sampled from the South American related groups of species, the individual species were not mono- species C. alpicola , C. cerroana , C. peruviana, and C. vadosa phyletic (e.g., C. miniata , C. occidentalis , and C. elmeri or C. were recovered in three separate well-supported clades that fell tenuifl ora and C. auriculata ). in the Oncorhynchus grade (Fig. 4). Most of the waxy sequences The number of waxy sequences isolated from an accession from these species, including multiple sequences from each of corresponded well with published chromosome counts for most the four species, were resolved in the South American annual I of the species sampled in this study (i.e., given that the waxy clade ( Fig. 4 ). Two of the four sequences isolated from C. alpi- gene is single copy, a diploid individual should have no more cola and one of the three sequences from C. vadosa form the than two unique sequence types representing the two alleles in South American annual II clade (bootstrap = 100%, PP = 1.0; a heterozygote). For example, two unique waxy sequences were Fig. 4 ). The third clade consists of three of the six sequences isolated from C. arvensis , C. cusickii , C. ophiocephala , C. peir- that were isolated from C. cerroana. Unlike the other waxy se- sonii , and C. virgata ( Fig. 4 ), all of which are only known from quences isolated from C. cerroana , these sequences are nested diploid ( n = 12) populations ( Heckard, 1968 ; Reveal and within the South American perennial sequences and form a Spellenberg, 1976 ; Heckard and Chuang, 1977 ; Spellenberg, well-supported monophyletic group with sequences from the 1986 ; Chuang and Heckard, 1993 ). Despite screening over 20 two perennial species C. nubigena and C. virgata (PP = 1.0, positive waxy clones, we isolated only one waxy sequence from bootstrap = 74%; Fig. 4 ). the inconspicuous South American perennial C. nubigena ( Fig. 4 ). Based on the diminutive corollas that are barely exserted from the calyx, this diploid species (Chuang and Heckard, DISCUSSION 1993 ) is likely autogamous, thus increasing the likelihood that it is also homozygous at the waxy locus. Likewise, six unique Long-distance dispersal and allopolyploidy within the Onco- sequences were isolated from the octoploid C. cerroana , three rhynchus grade — The distribution of Castilleja in South Amer- from the hexaploid C. vadosa, and four sequences were isolated ica is likely to be the result of both long-distance dispersal and from C. tenuifl ora, known from both diploid and tetraploid migration and diversifi cation along the western cordillera con- populations (Heckard and Chuang, 1977; Chuang and Heckard, necting North and South America. Of the approximately 160 1993), suggesting that the accession sampled here is at least perennial Castilleja species, a group of about 20 species is found, tetraploid ( Fig. 4 ). However, there are multiple cases among the more or less continuously, at high elevation in the mountains of diploid species of the Oncorhynchus grade when more than two Central America and Andean South America. The taxonomy of unique waxy sequences were isolated from an accession (C. the Central and South American perennial Castilleja species has campestris , C. exserta , C. lacera , C. lasiorhyncha, and C. been diffi cult (e.g., Weddell, 1857; Pennell, 1920, 1940 , 1953 ; tenuis). However, because all of these accessions were sampled Holmgren, 1978), and the majority of species are recognized as either from herbarium material or from species that are very part of the C. integrifolia- C. fi ssifolia species complex (Chuang 1916 American Journal of Botany [Vol. 96

Fig. 3. Majority rule consensus tree (excluding burn-in trees) with mean branch lengths from the partitioned Bayesian analysis of the combined chlo- roplast (cp) and nuclear ribosomal (nr) DNA data. Branch lengths are proportional to the number of substitutions per site as measured by the scale bar. Numbers above the branches indicate Bayesian posterior probability (PP) values. Numbers below the branches indicate parsimony bootstrap percentages (BP). Relative branch support values are indicated for clades that received BP ≥ 70% and PP values ≥ 0.95. The dashed line indicates the position of Cas- tilleja profunda in the individual analyses of the cpDNA and nrDNA (see text). Bayesian PP and maximum parsimony BP in parentheses indicate relative branch support values from the individual analyses that included C. profunda . Taxa in boldfaced type indicate species or accessions from South America. and Heckard, 1993 ). On the basis of phylogenetic analysis of the waxy and combined cp and nrDNA gene trees (Figs. 3 and 4) multiple cpDNA and nrDNA regions for subtribe Castillejinae, and are sister to a group of meso-American montane species Castilleja most likely originated in and around California ( Tank ( Fig. 3 ), suggesting a stepwise migration and diversifi cation of and Olmstead, 2008). The three South American perennial Cas- perennial Castilleja through montane Central America and An- tilleja species sampled here, C. nubigena, C. pumila, and C. vir- dean South America, as predicted by the more or less continuous gata , are resolved as part of the perennial Castilleja clade in both distribution of related species. October 2009] Tank and Olmstead — South American annual C ASTILLEJA 1917

By contrast, results of the phylogenetic analyses presented consistent with an allopolyploid origin of this species. Given the here indicate that the South American annual Castilleja species, relative positions of the two annual lineages that were involved including C. cerroana whose habit is most likely annual, are the in the allopolyploid origin of C. profunda , the paternal parent of result of at least three separate long-distance dispersal events this putative hybridization event was most likely part of an es- from widely disjunct annual ancestors in temperate North Amer- tablished group of disjunct annual species in South America that ica. The most recent of these dispersal events involved C. at- also played a role in the subsequent diversifi cation of the other tenuata, the only member of the Oncorhynchus grade represented Peruvian annual species (Fig. 3). However, the cpDNA results in both hemispheres, and mirrors the Californian-Chilean dis- imply that the geographic origin of the maternal parent repre- junction pattern that has been attributed to long-distance disper- sents yet another North American annual lineage not known to sal between the two coastal Mediterranean climates (e.g., Raven, exist now in South America. Castilleja profunda is distinguished 1963). The combined cpDNA and nrDNA data confi dently from the other Peruvian annual species primarily by its conspic- placed the Chilean accession of C. attenuata sister to the Califor- uous, deeply reticulate seed coat ( Chuang and Heckard, 1992 ). nian accession of the species ( Fig. 3 ) among a clade of predomi- Seed coat characteristics of the North American members of the nantly Californian annuals, and this result is robust to alternative Oncorhynchus grade have been studied in detail for their taxo- topology testing using the Bayes factor (see Results ). Both nomic signifi cance ( Chuang and Heckard, 1983 ). Of the North Chuang and Heckard (1992) and Raven (1963) suggested that American species, C. tenuis and C. lacera are two of the four the presence of C. attenuata in central Chile may have been the species reported to have deeply reticulate seed coats (Chuang result of human intervention; however, given the frequency of and Heckard, 1983), and this may be the origin of this conspicu- this distributional pattern between the two regions (reviewed in ous feature in C. profunda . Unfortunately, we were unable to Raven, 1963 ; Thorne, 1972 ) and the likelihood of multiple, inde- obtain sequences of the nuclear gene waxy from the low quality pendent long-distance dispersal events of annual Castilleja from DNA isolated from herbarium specimens of C. profunda . North America, it is also possible that C. attenuata was dispersed Castilleja laciniata was resolved as the sister group of the naturally between the two continents. Nevertheless, based on the four annual species (including C. cerroana) distributed in similarity of sequences obtained from the two cpDNA and two northern Peru (Fig. 3). This species is distributed in two dis- nrDNA regions (only two base-pair differences between the two junct populations in South America, occurring in both south- accessions in ~2500 bp of sequence), the introduction of this eastern Peru at high elevation and in central Chile where it is a species into Chile was certainly a recent event, and the phyloge- lowland species. The accession of this species used in this anal- netic evidence ( Fig. 3 ) supports the implication that this disjunct ysis is from Peru, and these data indicate that the Peruvian C. species played no role in the radiation of annual Castilleja in the laciniata shares a common North American ancestry with the Peruvian Andes (Chuang and Heckard, 1992). other Peruvian annuals ( Fig. 3 ). No morphological differences Unlike C. attenuata, the remaining South American annual between the Peruvian and Chilean populations of this species species are primarily distributed in Andean Peru and, therefore, have been identifi ed, and if the two populations indeed repre- do not follow the typical California-Chile disjunction pattern. sent the same species, as suggested by their morphologies, the Despite their montane distribution in South America, phyloge- disjunction between southeastern Peru and central Chile may netic analyses of nuclear and chloroplast DNA regions indicate also have involved a dispersal between the two regions of ~3500 that at least two other independent episodes of long-distance km. Detailed population level sampling within the two disjunct dispersal of annual ancestors from North America were respon- populations will be necessary to determine the direction and sible for the origin and diversifi cation of the annual Castilleja relative timing of this putative dispersal event. species distributed in Andean South America. Phylogenetic analysis of the combined cpDNA and nrDNA data recovered a Allopolyploidy between South American annuals and pe- well-supported clade comprised of the South American annual rennials— On the basis of results of experimental hybridization species that is most closely related to some of the North Ameri- studies involving North American annual members of the On- can annual lineages of the Oncorhynchus grade ( Fig. 3 ). Like- corhynchus grade and North American perennial Castilleja wise, all of the waxy sequences isolated from the South species, Chuang and Heckard (1992) hypothesized that the Pe- American annual Castilleja species were recovered in two ruvian annuals may have had a similar allopolyploid origin. clades (except for the C. cerroana perennial-type sequences, The phylogenetic analyses presented here indicate that long- see Allopolyploidy between South American annuals and pe- distance dispersal of widely disjunct annual ancestors from rennials ) that are part of the grade of waxy lineages belonging North America played a signifi cant role in the diversifi cation of to members of the Oncorhynchus grade (South American an- the annual species in South America and that at least one spe- nual I and II clades, Fig. 4 ). cies is likely an allopolyploid derived from hybridization be- The other long-distance dispersal event was revealed by the tween separate annual lineages of the Oncorhynchus grade. discordant position of C. profunda in the independent phyloge- However, there is also evidence of at least one allopolyploid netic analyses of the cpDNA and nrDNA data sets. Analysis of speciation event between the South American annual and pe- the nrDNA regions confi dently placed C. profunda within the rennial lineages. Sequences of the waxy gene isolated from the clade comprising the other Peruvian annuals ( Fig. 3 ). By con- octoploid species C. cerroana ( n = 48; Chuang and Heckard, trast, analysis of the cpDNA data resolved C. profunda in a clade 1992 ) were differentially resolved in both the perennial Cas- containing the two Californian annual species C. tenuis and C. tilleja clade, as well as the Oncorhynchus grade (Fig. 4). The C. lacera , neither of which are represented in South America cerroana perennial-type sequences were resolved with se- (Fig. 3), and the Bayes factor analysis indicated very strong sup- quences from the perennial species C. virgata and C. nubigena , port for this result vs. the monophyly of the exclusively South while the C. cerroana annual-type sequences were nested American annuals (see Results ). Castilleja profunda is reported within the South American annual I clade ( Fig. 4 ). These two as a hexaploid (n = 36; Chuang and Heckard, 1992 ), and the types of sequences present in C. cerroana provide strong sup- discordant position of this species in the separate analyses is port for the allopolyploid origin of this species, representing the 1918 American Journal of Botany [Vol. 96 October 2009] Tank and Olmstead — South American annual C ASTILLEJA 1919 homoeologous waxy loci derived from the separate annual and As evidenced by the mean branch length estimates resulting perennial parental genomes. from the Bayesian analyses of the combined cpDNA and Among the annual species sampled for the waxy locus in this nrDNA data, as well as the waxy sequence data (Figs. 3 and 4), study, C. cerroana is the only species that combines the ge- there was a marked difference in the amount of sequence diver- nomes of annual and perennial Castilleja species present in gence between the North and South American species of the South America. Although the corolla morphology of C. cerro- Oncorhynchus grade. This is especially evident in the clade ana is more similar to the other annual species included in the comprising the four species primarily distributed in northern waxy analysis, this species is differentiated from the Peruvian Peru, C. alpicola, C. cerroana, C. peruviana , and C. vadosa . annual species by its putative perennial habit and an unequally The conspicuously lower sequence divergence at both chloro- cleft calyx (i.e., cleft more deeply in the front and back than on plast and nuclear loci suggests that the diversifi cation of these the sides), which is more similar to the South American peren- species in northern Peru may represent a recent event in the nial species ( Fig. 1 ). The perennial-type calyx and the reported evolution of the annual Castilleja species in South America. perennial habit of C. cerroana led Chuang and Heckard (1992) Although the majority of waxy sequences isolated from the to place this species tentatively in their sect. Pallescentes, which South American annuals were resolved in the South American was comprised entirely of western North American perennial annual I clade (Fig. 4), an additional lineage comprised of se- species adapted for insect pollination ( Chuang and Heckard, quences from two of the Peruvian species, C. alpicola and C. 1991 ). Discovery of the unique allopolyploid origin of C. cer- vadosa , was also recovered in the waxy gene tree (South Ameri- roana now suggests that these features are the result of hybrid- can annual II clade, Fig. 4 ). This additional lineage may suggest ization, not an ancestry that involved other North American an allopolyploid origin for C. alpicola and C. vadosa from an perennial species (i.e., the C. pallescens group). Of the other additional annual lineage of the Oncorhynchus grade. However, South American species, C. laciniata (not included in the waxy given the putative recent radiation and the largely unresolved analyses) also possesses an unequally cleft calyx. However, nature of the South American annual I clade, it is perhaps more analysis of the cp and nrDNA data did not reveal a pattern con- likely that these sequences represent the incomplete coales- sistent with allopolyploidy for this species ( Fig. 3 ). The fi xation cence of ancestral waxy alleles that were maintained throughout of the nrDNA repeat in an allopolyploid hybrid is a random the diversifi cation of these species in the Peruvian highlands event (Wendel et al., 1995), and although the waxy data revealed (i.e., deep coalescence, sensu Maddison, 1997 ). the allopolyploid origin of C. cerroana, there was no evidence Phylogenetic analyses of the major lineages comprising sub- of hybrid origin of C. laciniata based on the cpDNA or nrDNA tribe Castillejinae have indicated that the large perennial Cas- data alone (Fig. 3). Therefore, it is possible that C. laciniata tilleja clade has undergone a recent diversifi cation in western represents a second annual species that is derived from allopo- North America and Mexico ( Tank and Olmstead, 2008 ), and the lyploidy between annual and perennial lineages in South Amer- results presented here add the South American perennial species ica, but further evidence is needed to test this hypothesis. to this radiation. The diversifi cation of perennial Castilleja lin- eages is closely associated with polyploidy, which may provide Dispersal, polyploidy, and the recent diversifi cation of an- the reproductive isolation necessary to promote diversifi cation nual Castilleja species in South America— The pattern of dis- in this recently derived group (Tank and Olmstead, 2008). Be- junction between the North American annual species and the cause polyploidy is relatively rare among the annual lineages of Andean annuals does not follow the typical coastal distribution the Oncorhynchus grade in North America, perenniality may be of the large majority of temperate disjuncts between western instrumental in facilitating the establishment of polyploid popu- North and South America. In a detailed review of the amphi- lations. However, the radiation of annual Castilleja species in tropical plant groups that have disjunct distributions between the South America suggests that polyploidy has also played an im- two continents, Raven (1963) noted that the disjunctions often portant role in the diversifi cation of these annual species. While mirrored the seasonal routes of migratory birds and were most it is unknown what the relative importance of auto- vs. allopoly- common to groups that occupied relatively open plant communi- ploidy was to the diversifi cation of perennial Castilleja , it has ties. Cruden (1966) discussed the behavior of the numerous mi- been estimated that over half of the perennial species are at least gratory bird groups that move between the temperate regions of in part polyploid ( Heckard, 1968 ; Heckard and Chuang, 1977 ; North and South America and concluded that the migratory Chuang and Heckard, 1993), and many of these species are shorebirds were the most likely candidates for the long-distance known from both diploid and polyploid populations, suggesting dispersal of seeds between the two regions. Although the major- that autopolyploidy may be relatively common within the peren- ity of dispersal events would likely be between the coastal re- nial Castilleja clade ( Mathews and Lavin, 1998 ; Tank and gions of Mediterranean climate, during migrations shorebirds Olmstead, 2008). By contrast, at least two of the six Peruvian are also common in both the seasonally moist areas of the central species were demonstrated to be of allopolyploid origin (C. cer- valley of California and Andean South America, and therefore, roana and C. profunda ), and based on morphology of the calyx, could provide a route for the successful dispersal and establish- C. laciniata may represent a third species of allopolyploid ori- ment between these two primarily open grassland communities. gin. Based on these data, there is no conclusive evidence that

¬ Fig. 4. Majority rule consensus tree (excluding burn-in trees) with mean branch lengths from the partitioned Bayesian analysis of the waxy sequence data. Branch lengths are proportional to the number of substitutions per site as measured by the scale bar. Numbers following species names represent clone numbers. An asterisk following the species name indicates a different accession for it than used for the chloroplast and nuclear ribosomal sequence data. Numbers above the branches indicate Bayesian posterior probability (PP) values. Numbers below the branches indicate parsimony bootstrap percentages (BP). Relative branch support values are indicated for clades that received BP ≥ 70% and signifi cant (≥ 0.95) PP values. Monophyletic groups of sequences isolated from one species that received BP ≥ 70% and signifi cant (≥ 0.95) PP values are indicated with a star. Shaded boxes indicate specifi c clades as discussed in the text. 1920 American Journal of Botany [Vol. 96 polyploidy in the remaining three closely related species —C. South American valerians. Organisms, Diversity & Evolution 5 : alpicola, C. peruviana , and C. vadosa — involved hybridization. 147 – 159 . Of these three putative autopolyploid species, chromosome Brown , J. M. , and A. R. Lemmon . 2007 . The importance of data par- counts are only available for C. vadosa (n = 36; Chuang and titioning and the utility of Bayes factors in Bayesian phylogenetics. Heckard, 1992 ); however, given the phylogenetic position and Systematic Biology 56 : 643 – 655 . Chuang , T. I. , and L. R. Heckard . 1982 . Chromosome numbers of the number of unique waxy sequences isolated from C. alpicola Orthocarpus and related monotypic genera (Scrophulariaceae, sub- and C. peruviana ( Fig. 4 ), both species are likely to be at least tribe Castillejinae). Brittonia 34 : 89 – 101 . tetraploid. Furthermore, the distributions of the allopolyploid C. Chuang , T. I. , and L. R. Heckard . 1983 . Systematic signifi cance of cerroana ( n = 48; Chuang and Heckard, 1992), the hexaploid C. seed-surface features in Orthocarpus (Scrophulariaceae subtribe vadosa , and C. peruviana overlap considerably (Fig. 1), and the Castillejinae). American Journal of Botany 70 : 877 – 890 . three species are often found growing in close proximity to each Chuang , T. I. , and L. R. Heckard . 1991 . Generic realignment and syn- other with no evidence of hybridization. This suggests that the opsis of subtribe Castillejinae (Scrophulariaceae – tribe Pediculareae). differential ploidy levels may play a role in the reproductive iso- Systematic Botany 16 : 644 – 666 . lation of these three recently diverged species. However, a de- Chuang , T. I. , and L. R. Heckard . 1992 . New species of bee-pollinated tailed investigation of the distribution of ploidy levels throughout Castilleja from Peru, with a taxonomic revision of South American members of subg. Colacus. Systematic Botany 17 : 417 – 431 . the ranges of the Peruvian annuals, and especially in sympatric Chuang , T. I. , and L. R. Heckard . 1993 . Chromosome numbers of neo- populations, will be necessary to test this hypothesis. tropical Castilleja (Scrophulariaceae, tribe Pediculareae) and their The divergence of the perennial Castilleja clade from its an- taxonomic implications. Annals of the Missouri Botanical Garden 80 : nual ancestors in North America represents a fundamental shift 974 – 986 . in the diversifi cation of Castilleja in western North America Cruden , R. W. 1966 . Birds as agents of long-distance dispersal for dis- (Tank and Olmstead, 2008). The allopolyploid hybridization junct plant groups of the temperate western hemisphere. Evolution between the annual and perennial lineages in Peru that produced 20 : 517 – 532 . C. cerroana is a novel event in the evolution of Castilleja spe- DeBry , R. W. , and R. G. Olmstead . 2000 . A simulation study of re- cies. While allopolyploidy between two North American spe- duced tree-search effort in bootstrap resampling analysis. Systematic cies of the Oncorhynchus grade has been implicated in the Biology 49 : 171 – 179 . C. brevistyla Doyle , J. J. , and J. L. Doyle . 1987 . A rapid DNA isolation procedure origin of the Californian annual ( Chuang and for small quantities of fresh leaf tissue. Phytochemical Bulletin 19 : Heckard, 1982 , 1983 ), there are no reported hybrids between 11 – 15 . North American annual and perennial Castilleja species, de- Doyle , J. J. , J. L. Doyle , and C. Harbison. 2003 . Chloroplast-expressed spite the ability to produce artifi cial hybrids in experimental glutamine synthetase in Glycine and related Leguminosae: Phylogeny, crossing studies ( Chuang and Heckard, 1991 , 1992 ). gene duplication, and ancient polyploidy. Systematic Botany 28 : 567 – 577 . Phylogenetic analyses of chloroplast and nuclear DNA re- Edwin , G. 1970 . New taxa and notes on the Scrophulariaceae of Peru. gions have revealed a complex history for the origin and diver- Phytologia 19 : 361 – 406 . sifi cation of the annual Castilleja species in South America. In Felsenstein , J. 1985 . Confi dence limits on phylogenies— An approach addition to at least three independent long-distance dispersal using the bootstrap. Evolution 39 : 783 – 791 . events (including the recent dispersal of C. attenuata to central Florin , R. 1963 . The distribution of conifer and taxad genera in time and Chile), allopolyploidy involving multiple annual lineages of the space. Acta Horticulturae Bergiani 20 : 121 – 312 . Funk , V. A. 1985 . Phylogenetic patterns and hybridization. Annals of the Oncorhynchus grade, as well as members of the perennial clade, Missouri Botanical Garden 72 : 681 – 715 . has also played a signifi cant role in the disjunct radiation of an- Gielly , L. , and P. Taberlet. 1994 . The use of chloroplast DNA to nual Castilleja in South America. Future research investigating resolve plant phylogenies— Noncoding versus rbcL sequences. the role of dysploidy in sympatric populations of the recently Molecular Biology and Evolution 11 : 769 – 777 . diverged annual species in northern Peru, as well as the putative Good , R. 1964 The geography of fl owering plants. Wiley, New York, New allopolyploid origin and secondary dispersal of C. laciniata be- York, USA. tween southeastern Peru and central Chile, will continue to add Graham , S. W. , P. A. Reeves , A. C. E. Burns , and R. G. Olmstead . to our understanding of the complex interplay between poly- 2000 . Microstructural changes in noncoding chloroplast DNA: ploidy, hybridization, and long-distance dispersal in the diver- Interpretation, evolution, and utility of indels and inversions in basal sifi cation of temperate western American plant groups. angiosperm phylogenetic inference. International Journal of Plant Sciences 161 ( 6 Supplement ): S83 – S96 . LITERATURE CITED Harbaugh , D. T. 2008 . Polyploid and hybrid origins of Pacifi c island sandalwoods (Santalum , Santalaceae) inferred from low-copy nuclear Axelrod , D. I. 1958 . Evolution of the Madro-Tertiary geofl ora. Botanical and fl ow cytometry data. International Journal of Plant Sciences 169 : Review 24 : 433 – 509 . 677 – 685 . Baldwin , B. G. , and S. Markos . 1998 . Phylogenetic utility of the exter- Heckard , L. R. 1968 . Chromosome numbers and polyploidy in Castilleja nal transcribed spacer (ETS) of 18S-26S rDNA: Congruence of ETS (Scrophulariaceae). Brittonia 20 : 212 – 226 . and ITS trees of Calycadenia (Compositae). Molecular Phylogenetics Heckard , L. R. , and T. I. Chuang . 1977 . Chromosome numbers, poly- and Evolution 10 : 449 – 463 . ploidy, and hybridization in Castilleja (Scrophulariaceae) of the Great Baldwin , B. G. , M. J. Sanderson , J. M. Porter , M. F. Wojciechowski , Basin and Rocky Mountains. Brittonia 29 : 159 – 172 . C. S. Campbell , and M. J. Donoghue . 1995 . The ITS region of nu- Holmgren , N. H. 1978 . Castilleja (Scrophulariaceae) of Costa Rica and clear ribosomal DNA: A valuable source of evidence on angiosperm Panama. Brittonia 30 : 182 – 194 . phylogeny. Annals of the Missouri Botanical Garden 82 : 247 – 277 . Hughes , C. , and R. Eastwood . 2006 . Island radiation on a continen- Baumel , A. , M. L. Ainouche , R. J. Bayer , A. K. Ainouche , and M. tal scale: Exceptional rates of plant diversifi cation after uplift of the T. Misset . 2002 . Molecular phylogeny of hybridizing species from Andes. Proceedings of the National Academy of Sciences, USA 103 : the genus Spartina Schreb. (Poaceae). Molecular Phylogenetics and 10334 – 10339 . Evolution 22 : 303 – 314 . Ingram , A. L. , and J. J. Doyle . 2003 . The origin and evolution of Bell , C. D., and M. J. Donoghue. 2005 . Phylogeny and biogeog- Eragrostis tef (Poaceae) and related polyploids: Evidence from nuclear raphy of Valerianaceae (Dipsacales) with special reference to the waxy and plastid rps16 . American Journal of Botany 90 : 116 – 122 . October 2009] Tank and Olmstead — South American annual C ASTILLEJA 1921

Kass , R. , and A. Raftery . 1995 . Bayes factors. Journal of the American Ronquist , F. , and J. P. Huelsenbeck . 2003 . MrBayes 3: Bayesian Statistical Association 90 : 773-795 . phylogenetic inference under mixed models. Bioinformatics 19 : Keck , D. D. 1927 . A revision of the genus Orthocarpus. Proceedings of 1572 – 1574 . the California Academy of Science Series 4 16: 517 – 571. Sang , T. 2002 . Utility of low-copy nuclear gene sequences in plant phy- Kim , S. , and M. J. Donoghue . 2008 . Incongruence between cpDNA logenetics. Critical Reviews in Biochemistry and Molecular Biology and nrITS trees indicates extensive hybridization within Eupersicaria 37 : 121 – 147 . (Polygonaceae). American Journal of Botany 95 : 1122 – 1135 . Sang , T. , D. J. Crawford , and T. F. Stuessy . 1995 . Documentation of Kim , S. , S. E. Sultan , and M. J. Donoghue . 2008 . Allopolyploid spe- reticulate evolution in peonies (Paeonia ) using internal transcribed ciation in Persicaria (Polygonaceae): insights from a low-copy nu- spacer sequences of nuclear ribosomal DNA: Implications for bio- clear region. Proceedings of the National Academy of Sciences, USA geography and concerted evolution. Proceedings of the National 105 : 12370 – 12375 . Academy of Sciences, USA 92 : 6813 – 6817 . Levin , R. A. , and J. S. Miller . 2005 . Relationships within tribe Lycieae Sang , T. , D. J. Crawford , and T. F. Stuessy . 1997 . Chloroplast (Solanaceae): Paraphyly of Lycium and multiple origins of gender di- DNA phylogeny, reticulate evolution, and biogeography of Paeonia morphism. American Journal of Botany 92 : 2044 – 2053 . (Paeoniaceae). American Journal of Botany 84 : 1120 – 1136 . Levin , R. A. , N. R. Myers , and L. Bohs . 2006 . Phylogenetic relationships Sang , T. , and D. M. Zhang. 1999 . Reconstructing hybrid speciation using among the “ spiny solanums ” (Solanum subgenus Leptostemonum , sequences of low copy nuclear genes: Hybrid origins of fi ve Paeonia spe- Solanaceae). American Journal of Botany 93 : 157 – 169 . cies based on Adh gene phylogenies. Systematic Botany 24 : 148 – 163 . Levin , R. A. , K. Watson , and L. Bohs . 2005 . A four-gene study of evo- Simmons , M. P. , and H. Ochoterena . 2000 . Gaps as characters in se- lutionary relationships in Solanum section Acanthophora. American quence-based phylogenetic analyses. Systematic Biology 49 : 369 – 381 . Journal of Botany 92 : 603 – 612 . Small , R. L. 2004 . Phylogeny of Hibiscus sect. Muenchhusia (Malvaceae) Maddison , W. P. 1997 . Gene trees in species trees. Systematic Biology based on chloroplast rpl16 and ndhF , and nuclear ITS and GBSSI se- 46 : 523 – 536 . quences. Systematic Botany 29 : 385 – 392 . Mason-Gamer , R. J. 2001 . Origin of North American Elymus (Poaceae: Smedmark , J. E. E. , T. Eriksson , and B. Bremer . 2005 . Allopolyploid Triticeae) allotetraploids based on granule-bound starch synthase evolution in Geinae (Colurieae: Rosaceae): Building reticulate species gene sequences. Systematic Botany 26 : 757 – 768 . trees from bifurcating gene trees. Organisms, Diversity & Evolution Mathews , S. , and M. Lavin . 1998 . A biosystematic study of Castilleja 5 : 275 – 283 . crista-galli (Scrophulariaceae): An allopolyploid origin reexamined. Smedmark , J. E. E. , T. Eriksson , R. C. Evans , and C. S. Campbell . Systematic Botany 23 : 213 – 230 . 2003 . Ancient allopolyploid speciation in Geinae (Rosaceae): McDade , L. A. 1995 . Hybridization and phylogenetics. In P. C. Hoch and Evidence from nuclear granule-bound starch synthase (GBSSI) gene A. G. Stephenson [eds.], Experimental and molecular approaches to sequences. Systematic Biology 52 : 374 – 385 . plant biosystematics, 305– 331. Missouri Botanical Garden Press, St. Spellenberg , R. 1986 . Chromosome number reports 90. Taxon 35 : 197 . Louis, Missouri, USA. Stebbins , G. L. 1957 . Self fertilization and population variability in the Miller , R. E. , M. D. Rausher , and P. S. Manos . 1999 . Phylogenetic higher plants. American Naturalist 91 : 337 – 354 . systematics of Ipomoea (Convolvulaceae) based on ITS and waxy se- Swofford , D. L. 2002 . PAUP*: Phylogenetic analysis using parsimony quences. Systematic Botany 24 : 209 – 227 . (*and other methods), version 4b10. Sinauer, Sunderland, Massachusetts, Nylander , J. A. A. , F. Ronquist , J. P. Huelsenbeck , and J. L. Nieves- USA. Aldrey . 2004 . Bayesian phylogenetic analysis of combined data. Tank , D. C. , J. M. Egger , and R. G. Olmstead . 2009 . Phylogenetic clas- Systematic Biology 53 : 47 – 67 . sifi cation of subtribe Castillejinae (Orobanchaceae). Systematic Botany Oxelman , B. , M. Liden , and D. Berglund . 1997 . Chloroplast rps16 34 : 182 – 197 . intron phylogeny of the tribe Sileneae (Caryophyllaceae). Plant Tank , D. C. , and R. G. Olmstead . 2008 . From annuals to perenni- Systematics and Evolution 206 : 393 – 410 . als: Phylogeny of subtribe Castillejinae (Orobanchaceae). American Pennell , F. W. 1920 . Scrophulariaceae of Colombia. Proceedings of the Journal of Botany 95 : 608 – 625 . Academy of Natural Sciences of Philadelphia 72 : 136 – 188 . Thorne , R. F. 1972 . Major disjunctions in the geographic ranges of seed Pennell , F. W. 1940 . Scrophulariaceae. In R. E. Woodson and R. W. plants. Quarterly Review of Biology 47 : 365 – 411 . Schery [eds.], Contributions toward a fl ora of Panama IV. Annals of Walsh , B. M. , and S. B. Hoot . 2001 . Phylogenetic relationships of the Missouri Botanical Garden 27: 338 – 341. Capsicum (Solanaceae) using DNA sequences from two noncoding Pennell , F. W. 1953 . Scrophulariaceae. In J. A. Steyermark [ed.], Botanical regions: The chloroplast atpB-rbcL spacer region and nuclear waxy explorations in Venezuela III. Fieldiana: Botany 28: 516 – 520. introns. International Journal of Plant Sciences 162 : 1409 – 1418 . Peralta , I. E. , and D. M. Spooner . 2001 . Granule-bound starch syn- Weddell , H. A. 1857 . Castilleja , Chloris Andina, 118– 119. Chez P. thase (GBSSI) gene phylogeny of wild tomatoes (Solanum L. sec- Bertrand, Paris, France. tion Lycopersicon Mill. Wettst. subsection Lycopersicon ). American Wendel , J. F. , A. Schnabel , and T. Seelanan . 1995 . Bidirectional in- Journal of Botany 88 : 1888 – 1902 . terlocus concerted evolution following allopolyploid speciation in cot- Posada , D. , and K. A. Crandall . 1998 . Modeltest: Testing the model of ton (Gossypium ). Proceedings of the National Academy of Sciences, DNA substitution. Bioinformatics 14 : 817 – 818 . USA 92 : 280 – 284 . Rabeling , C. , J. M. Brown , and M. Verhaagh . 2008 . Newly discov- Whitson , M. , and P. S. Manos. 2005 . Untangling Physalis (Solanaceae) ered sister lineage sheds light on early ant evolution. Proceedings of from the physaloids: A two-gene phylogeny of the Physalinae. Systematic the National Academy of Sciences, USA 105 : 14913 – 14917 . Botany 30 : 216 – 230 . Rambaut , A. 1996 . Se-Al: Sequence alignment editor [computer pro- Winkworth , R. C. , and M. J. Donoghue . 2004 . Viburnum phylog- gram]. University of Edinburgh, Edinburgh, UK. Available at website eny: Evidence from the duplicated nuclear gene GBSSI . Molecular http://tree.bio.ed.ac.uk/software/seal/ . Phylogenetics and Evolution 33 : 109 – 126 . Rambaut , A. , and A. J. Drummond . 2004 . Tracer [computer program]. Winkworth , R. C. , and M. J. Donoghue . 2005 . Viburnum phylogeny University of Edinburgh, Edinburgh, UK. Available at website http:// based on combined molecular data: implications for taxonomy and tree.bio.ed.ac.uk/software/tracer/ . biogeography. American Journal of Botany 92 : 653 – 666 . Raven , P. H. 1963 . Amphitropical relationships in the fl oras of North and Wood , C. E. 1972 . Morphology and phytogeography: The classical ap- South America. Quarterly Review of Biology 38 : 151 – 177 . proach to the study of disjunctions. Annals of the Missouri Botanical Raven , P. H. 1972 . Plant species disjunctions: A summary. Annals of the Garden 59 : 107 – 124 . Missouri Botanical Garden 59 : 234 – 246 . Yeung , K. , J. S. Miller , A. E. Savage , B. C. Husband , B. Igic , and J. Reveal , J. L. , and R. Spellenberg . 1976 . Miscellaneous chromosome R. Kohn . 2005 . Association of ploidy and sexual system in Lycium counts of western American plants III. Rhodora 78 : 37 – 52 . californicum (Solanaceae). Evolution 59 : 2048 – 2055 .