Aliso: A Journal of Systematic and Floristic Botany Aliso: A Journal of Systematic and Floristic Botany Phylogeny of Polemoniaceae Based on Nuclear Ribosomal Phylogeny of Polemoniaceae Based on Nuclear Ribosomal Internal Transcribed Spacer DNA Sequences Internal Transcribed Spacer DNA Sequences

Nuclear ribosomal internal transcribed spacer (ITS) DNA sequences are used to estimate the phy logeny of 53 members of Polemoniaceae, representing all but two genera of the family. Fitch parsi mony analysis of equal-weighted nucleotide sites result in 1080 minimal-length trees. However, when alignment-ambiguous positions are removed and an II: 10 transition to transversion weighting is im posed only eight trees are found. These data are used to address two issues: I) patterns of diversifi cation in Polemoniaceae, and 2) the circumscription and monophyly of the genus Cilia. Although the monophyly of Polemoniaceae is well supported, relationships inferred among the earliest diverging lineages are altered by character weighting, treatment of indels, and taxon inclusion. In spite of the lack of reliable resolution at the basal nodes, ITS data provide evidence that Cilia, as currently interpreted, is polyphyletic and comprises at least five independent lineages.


INTRODUCTION
The Polemoniaceae are a small, mostly New World, group of between 300 and 340 species (Grant 1959;Willis 1966;Cronquist 1981;Thome 1992;. Evolutionary relationships and classification within this family have long been controversial (Bentham 1845; Gray 1870; Kuntze 1891 ;Brand 1907;Rydberg 1913;Dawson 1936;Wherry 1940;Mason and Grant 1948;Grant 1959;Cronquist 1984;Day and Moran 1986). Although generally considered a "natural" group by most authors, generic and species relationships within the family have remained obscure: the circumscription of species, genera and higher taxonomic groups differing radically from author to author. The most recent family-wide classification of Polemoniaceae is that of Grant (1959), who recognizes 18 genera, placed into five tribes (Table 1). Over the past 30 years, a broad spectrum of data has accumulated bearing on relationships within the family. These include family-and generic-level surveys of palynological (Stuchlik 1967;Taylor and Levin 1975;Chuang et al. 1978;Timbrook 1986;Day and Moran 1986), histological (Carlquist et al. 1984), chemosystematic (Smith et al. 1977;Dean et al. 1978;Harborne and Smith 1978;Harborne 1980;Smith et al. 1982;Wilken et al. 1982;Kiredjiad and Smith 1985), and reproductive data (Grant and Grant 1965;Plitmann and Levin 1983;. Although neither phenetic nor phylogenetic analyses have been conducted on the family using this wide array of morphological and chemical data (but see Wilken and Hartman 1991), recent phylogenetic analyses of nucleotide sequences of the chloroplast gene matK (Steele 1991(Steele , 1992Steele and Vilgalys 1994;Johnson and Soltis 1995 ;Johnson et al. 1996) suggest these tribes are not monophyletic. In light of these data, realignment of Grant's (1959) tribes appears inevitable. This study provides additional insight into classification and evolutionary history of Polemoniaceae from an independent source of evidence. Presented is a molecular phylogenetic study using sequences from the nuclear genes coding for ribosomal RNA, the internal transcribed spacers (ITS 1 and ITS2).

Taxonomic Background
The contrast between various classifications of Polemoniaceae has been reviewed by Grant (1959: 4). There is little consensus among historical classifications. The few general similarities (particularly for classifications since 1900) include: 1) the tendency to place the tropical genera Cantua, Cobaea and Bonplandia in separate tribes apart from the remainder of the so-called "temperate genera" (Brand 1907;1908;Grant 1959); and 2) the inclusion of Linanthus and Leptodactylon in the same tribe with Gilia (Brand 1907;1908;Wherry 1940;Grant 1959) while placing Phlox in a different tribe (Wherry 1940;Grant 1959). A particularly striking disparity among historical classifications in the family is the circumscription of the genus Gilia. These differences range from broad circumscriptions that include nearly all of the "temperate genera" except Polemonium, Phlox, and Collomia (e.g., Bentham and Hooker 1876) to the present treatment by Grant (1959) that segregates Allophyllum, Ipomopsis, Eriastrum, Langloisia, Loeseliastrum, and Navarretia from Gilia.
Grant's currently accepted tribal classification has not found support from recent molecular studies. Steele and Vilgalys (1994) provided evidence from cpDNA matK sequences that some genera included in tribe Polemonieae share more recent common ancestry with genera placed in tribe Gilieae. For example Collomia and Allophyllum (tribe Polemonieae) are more closely related to Gilia and Navarretia (tribe Gilieae) than to either Phlox or Polemonium (tribe Polemonieae). Indeed, Steele and Vilgalys suggest that there is no support for two temperate tribes, rather Polemoniaceae possess a single temperate lineage. These results were not only corroborated by Johnson and Soltis (1995) and Johnson et al. (1996) but due to the increased taxon sampling, evidence was presented that the genus Gilia, as currently circumscribed, is a polyphyletic assemblage. Some members of Gilia (e.g., G. filiformis) are more closely related to genera never associated with tribe Gilieae (e.g., Phlox) than to other members of Gilia.
This controversy over relationships within Polemoniaceae is problematic because Polemoniaceae have been used as a model system for the study of many aspects of character evolution and reproductive biology. The family as well as individual species have been used as primary examples for the pattern and modes of evolution of pollination mechanisms (Grant and Grant 1965;Galen and Kevan 1983; Galen and Stanton 1989), evolution of plant sexual systems (Grant 1959;Schoen 1982aSchoen , b, c, 1983Plitmann andLevin 1983, 1990), hybridization and its evolutionary consequences (Grant 1950(Grant , 1954a(Grant , b, 1957(Grant , 1979Grant and Grant 1960;Day 1965;Coyne 1974;Grant and Wilken 1987), natural selection on floral traits (Ellstrand 1983;Wolf et al. 1986;Ellstrand and Mitchell 1988;Campbell 1989;, and evolution life history (Paige and Whitham 1987a, b). An accurate phylogenetic backdrop is necessary to interpret evolutionary changes in these traits (Donoghue 1989). The chloroplast phylogenies provide an important inroad into understanding the phylogeny of Polemoniaceae. However, a single gene phylogeny may not correspond with an organismal phylogeny (Dwyer et al. 1991;Sanderson and Doyle 1992). A phylogeny based on an independent source of data is both desirable and necessary to either corroborate or refute phylogenetic inferences from the chloroplast genome.
In this paper I provide a preliminary assessment of phylogenetic relationships of Polemoniaceae using DNA sequences of the internal transcribed spacers (ITS 1 and ITS2) of the nuclear ribosomal cistron. Because these spacer sequences are encoded in the nuclear genome, they represent an independent source of phylogenetic information from chloroplast sequences.

Sampled Taxa
An attempt was made to include representatives of all of the 19 currently recognized genera of Polemoniaceae (Table 1). Because of the controversy over the limits of the genus Gilia and its relation to Ipomopsis, 13 species of Gilia (members of sections Arachnion, Gilia, Giliandra, Giliastrum, Gilmania, and Saltugilia) and five species of Ipomopsis (members of sections Ipomopsis, Microgilia, Phloganthea [including the Giliopsis group D. In addition, a largely unknown woody member of Polemoniaceae, Gilia scabra (treated by Grant 1959 as a synonym of Linanthus nuttallii ), is also included. Only two genera are not represented. Gymnosteris, with two species, is not included because the DNAs did not sequence well enough for reasonable interpretation (nearly all of the nucleotide sites appeared polymorphic). In addition, the DNA sequence from Huthia coerulea is not included because it is unalignable with other Polemoniaceae sequences.
Selection of an outgroup for Polemoniaceae is problematical. In spite of most present classifications (Cronquist 1981;Takhtajan 1980;Thome 1968Thome , 1992, molecular data from the chloroplast gene rbcL Chase et al. 1993) provide evidence that Polemoniaceae are more closely related to Ericalean groups, usually placed in the Dilleniidae, than to the Asteridae families with which they are traditionally placed: i.e., Convolvulaceae, Hydrophyllaceae, and Solanaceae (but see Anderberg 1992; . In particular, chloroplast sequence data from rbcL  and matK (Johnson et al. 1996) place Fouquieriaceae as the sister group to Polemoniaceae. By contrast, the Chase et al. study of rbcL supports a close relationship between Polemoniaceae and Diapensiaceae. Close relationships between Polemoniaceae and both Diapensiaceae (Linnaeus 1753; Don 1822) and Fouquieriaceae (Humbolt et al. 1823;Engler and Gilg 1924;Abrams 1951;Thome 1977) have been suggested based on morphological evidence. Based on this evidence, I have selected Diapensiaceae and Fouquieriaceae as outgroups. Attempts were made to also include sequences from Solanales, e.g., Nicotiana rustica L. (Venkateswarlu and Nazar 1991); however, these sequences are unalignable.

Collection and DNA Extraction
Material collected in the field was placed on ice for transport to storage at -80°C. Total genomic DNA extracts from leaf material, powdered under liquid nitrogen, used a 2X CTAB protocol, modified from . In some cases a 4X CTAB (with 0'.1 % ~VP-40) extraction buffer was used for species With high polysaccharide content. The DNAs were further purified using cesium chloride/ethidium bromide gradients . The DNAs of Acanthogilia gloriosa and Gilia scabra were extracted from leaf material desiccated using silica gel Chase and Hillis 1991). The DNAs of six taxa were obtained from leaf fragments taken from herbarium collections (Table 1). Subsequently, all DNA samples obtained from herbarium specimens have been verified using freshly collected material, and in no case did the DNA sequences of ITS differ at more than a single, uninformative nucleotide site (Porter unpubl.).

PCR Amplification and Sequencing
Sequencing was performed on 39 samples using the method of Sanger et al. (1977). Single-stranded DNAs for both strands of the two internal transcribed spacers of the nuclear ribosomal cistron, ITS 1 and ITS2, were amplified directly by asymmetric polymerase chain reaction (PCR), using a 1 :20 ratio of the primers "ITSS" (S'-GGA AGT AAA AGT CGT AAC AAG G-3') and "ITS2" (S'-GCT GCG TTC TIC ATC GAT GC-3') for the ITS 1 spacer, and the primers "ITS3" (S'-GCA TCG ATG AAG AAC GCA GC-3') and "ITS4" (S'-TCC TCC GCT TAT TGA TAT GC-3') for the ITS2 spacer (White et al. 1990). PCR amplifications, PCR product purification and direct single-stranded DNA seque~cing followed the procedures described by Baldwm (1992), except for the following modifications. The amplification of DNA from herbarium material employed a modified lOx Taq polymerase reaction buffer consisting of 67mM Tris-HCl (pH 8.8), 2 mM MgC1 2 , and 2 mg/ml bovine serum albumin (P~Hibo et al. 1989;). The PCR products were electrophoresed using a I.S% agarose gel (prestained with ethidium bromide) in a 1 X TBE (pH 8.3) buffer, to confirm the single-stranded product and purified using differential filtration in Millipore Ultra-Free-MC microfuge tubes (Millipore UFC-3 THKOO). The purified DNAs were sequenced using the dideoxy chain termination reaction, following the procedures outlined in the TAQuence@> kit (U. S. Biochemical Co.) . Sequencing products were labeled using a-35 SdATP. To reduce base compressions, 7-deaza-dGTP was substituted for dGTP in all of the sequencing reactions. The gels were fixed in S% acetic acid/S% methanol for 4S min and vacuum dried at 80°C for 1 hr. The autoradiographs were exposed for a minimum of 24 hr. All sequences were checked by sequencing the opposite strand.
Sequence data for 14 species were obtained using an Applied Biosystems Model 373A Automated DNA Sequencing System (Perkin Elmer). Template DNAs were prepared by symmetric PCR of the entire ITS region using a 1: 1 ratio of primers ITSS and ITS4. Direct cycle-sequencing of the ITS region followed manufacturers specifications, using the PRISM@> DyeDeoxy@> Terminator Kit (Perkin Elmer) and employed primers ITS2, ITS3, "ITSSI" (S'-AGG TG ACC TGC GGA AGG ATC ATI-3') and "ITS4I" (S'-GGT AGT CCC GCC TGA CCT GG-3'). Primers ITSPSI and ITSP4I are internal to and flanking ITSS and ITS4, respectively. The four primers provide sequences for overlapping fragments that collectively cover both strands of the entire ITS region (ITSl, S.8S and ITS2).

Sequence Alignment
The S6 sequences were truncated to include only ITS 1 and ITS2. Identification of the terminal ends of I!S 1 and ITS2 was based on comparisons with pubhshed sequences for Daucus carota L., Nicotiana rustica, ~icia faba L., species of Astragalus, and representatives of Madiinae (Baldwin 1992(Baldwin , 1993Venkateswarlu and Nazar 1991;Wojciechowski et al. 1993;Yokota et al. 1989). The two spacer regions were aligned separately, using PILEUP (Feng and Doolittle 1987;Higgins and Sharp 1987; see also , a multiple sequence alignment program of the University of Wisconsin Genetics Computer Group (UWGCG) software package. The alignment parameters were varied from GapWeight= 1.0 and GaplengthWeight= 0.1 to GapWeight= S.O and GaplengthWeight= 3.0. With higher gap and length penalty values, fairly large blocks of aligned sequences ar~ p~oduced, but single taxa are sometimes grossly rrusahgned. Low gap and length penalties result in alignments lacking grossly misaligned taxa, but with a greater frequency of gaps. Evenso, alternative alignments were evident. The resulting alignments were both analyzed directly and manually realigned and reanalyzed. The PILEUP aligned sequences were manually realigned to maximize the number of invariant nucleotide positions. The results presented refer to sequences which were realigned manually; however, the conclusions apply to both alignments.

Phylogenetic Analysis
Estimations of phylogenetic relationships, based on combined ITS 1 and ITS2 sequences, were obtained using Fitch parsimony as implemented in PAUP (version 3.0; Swofford 1991). A variety of heuristic approaches  Grant (1959) and collections utilized in the phylogenetic analysis of nrDNA internal transcribed spacer sequences.' The cross (t) indicates those taxa for which herbarium collections were used for DNA extractions. The asterisk (*) indicates those species for which cycle sequencing and automated sequencing procedures were used. Vouchers are cited and housed as indicated. The taxa included to serve as the outgrup are designated as such parenthetically.   was employed to help assure that all of the most parsimonious solutions were obtained. Initially, CLOS-EST addition and TBR (tree bisection-reconnection) swapping was performed. This was followed by 100 replicates of RANDOM addition and TBR branch swapping. Finally, 500 replicates of RANDOM addition were carried out with no swapping, followed by TBR swapping on the resulting set of trees . This should insure that all minimal length tree are found, even if multiple "islands" of equally parsimonious trees exist. The data matrix was analyzed with equal weights at all nucleotide positions and using an 1.1 to 1 transition/transversion bias (based on estimates from Polemoniaceae sequences). Insertions and/or deletions (indels) were coded initially as missing data. Subsequent analyses treated indels as a "fifth nucleotide" (GAPMODE=FIFTHSTATE option in PAUP) and as additional binary characters.

Robustness and Sampling
An assessment of the relative support for clades was performed using the decay index (Bremer 1988;Donoghue et al. 1992) and the bootstrap (Felsenstein 1985). Decay values were calculated for one, two and three steps longer than the shortest trees. My purpose is to identify only the weakly supported nodes. Bootstrap values were calculated from 100 replicate Fitch parsimony analyses, as implemented by PAUP 3.0, using heuristic searches based on CLOSEST addition and TBR branch swapping.
The distribution of tree-lengths, based on 10,000 randomly generated trees was evaluated for skewness (gl). A strongly left-skewed tree length distribution has been suggested to indicate that a data set is highly structured and likely contains strong phylogenetic signal (Goodman et al. 1979;Fitch 1984;Werman 1986;Hillis and de Sa 1988;Hillis and Dixon 1989;Hillis and Huelsenbeck 1992;Hillis 1991;Huelsenbeck 1991). However, this test actually assesses whether the gl of a given data set deviates from the expected gl of random data (see also Kallersjo et al. 1992).
To investigate the sensitivity of these data to taxon sampling, an analysis based on resampling terminal taxa was conducted. Lanyon's jackknife is a heuristic tool to explore the effect of taxon sampling on internal branch support and tree structure (Lanyon 1985a, b;Felsenstein 1988;Siddall 1995). Lanyon (1985a) points out that trees generated during this jackknife procedure simulate possible outcomes had the investigator failed to sample a taxon or if extinction had occurred. Clearly, taxon sampling (inclusion) can have profound consequences on assessment of monophyly (Doyle and Donoghue 1987;Gauthier et al. 1988;Donoghue et al. 1989;. The manually realigned data matrix was analyzed by sequentially re-moving each member of the ingroup, one at a time, and performing a parsimony analysis (CLOSEST addition, HOLD= 2, TBR branch swapping) on the resulting data sets (pseudoreplicates). In all cases, only the strict-consensus tree from analyses of each pseudoreplicate was retained. The trees were compared and pseudoreplicate-consensus trees were generated. This was done by first returning the excluded taxon from each pseudoreplicate run back into the pseudoreplicate strict-consensus tree. The excluded taxon was placed in the same position it is found in the strict-consensus tree from the analysis of the entire data set. This is based on the assumption that the analysis of the full data set provides evidence for the taxon's placement. After the excluded taxon is returned to the tree, a consensus of the entire set of trees was obtained, using strict and majority rule consensus procedures.

ITS Sequence Variation
The ITS 1 regions of all representatives of Polemoniaceae, as well as the outgroups, were consistently larger than the ITS2 regions. Within Polemoniaceae the ITS 1 region is generally 249-253 bases. Extremes beyond this range include the sequences of Gilia rigidula at 242 base pairs (bp) and Gilia scabra at 262 bp. The outgroups possessed the largest ITS 1 sequence of this analysis, 267 bp. In Polemoniaceae, percent G+C content of ITS 1 ranged from 42.2% in Cobaea baiurita to 65 .0% in Acanthogilia gloriosa. ITS2 sequences were mostly between 188 and 193 bases, but ranged from 187 bp in Phlox standsburyi and Linanthus nuttallii to 195 in Gilia foetida. The ITS2 region of the outgroups was slightly larger, 196 bp. Percent G+C content of ITS2 in Polemoniaceae varied from 48.3% in Cobaea baiurita to 61.8% in Acanthogilia gloriosa.

Alignment
Both ITS 1 and ITS2 possess several highly conserved regions, flanked by more variable regions (Baldwin et al. 1995). The conserved regions are easily and nearly unambiguously aligned; however, the variable regions become increasingly difficult to align as pairwise distance increases. As a result, alignment of some taxa (i.e., outgroup taxa) should be viewed with skepticism. The variable regions of ITS in Polemoniaceae are frequently associated with short indels of one to a few nucleotides. Sequences of ITS 1 and ITS2, aligned using variable weights in PILEUP produce similar alignments; however, the more variable regions seemed better aligned (i.e., there is a higher frequency of invariant sites) using GapWeight= 1.0 and GaplengthWeight= 0.1 (the aligned sequences us-ing PILEUP are not shown). The manually realigned sequences are used in all of the following comparisons. Even so, the manually aligned sequence data set is not without alternative alignments and alignment in some regions remains ambiguous (Appendix I).
The aligned, combined sequences of ITS 1 and ITS2 required, on average, gaps at 14.39% of the sites of any given sequence. The inclusion of the outgroup was responsible for gaps at 4.53% of these nucleotide positions. The combined data matrix, possesses 529 characters, of which 120 are invariant and 346 (65.28%) are potentially informative. The pair-wise levels of divergence range from 0.18% (between Langioisia setosissima subsp. setosissima and Langloisia setosissima subsp. punctata) to 42.4% (between Diapensia lapponica and Cantua quercifolia).

Phylogenetic Analysis
All of the parsimony analyses of the full, equalweighted data set, with indels treated as missing, resulted in recovery of the same 1080 minimum-length trees. For the data set including only potentially informative sites, the most parsimonious trees were of 1552 steps (tree length including all sites is 1646), with a consistency index (C.I.) of 0.423, retention index (R.I.) of 0.622, and rescaled consistency index (R.c.) of 0.263. The strict consensus tree ( Fig. 1) shows that the primary areas of disagreement in the set of minimal length trees involves 1) relationships involving /pomopsis sonorae and Loeseliastrum matthewsii in the Loeselieae Clade, 2) relationship of Gilia campanulata within the Linanthieae Clade, and 3) relationships between three clades of Gilia species as well as Gilia splendens, within the Gilieae Clade. Aside from the basal nodes, the phylogenetic estimate is not highly sensitive to coding of the indel regions or transitiontransversion weighting. Figure 2 shows the strict consensus trees from analyses that 1) treat each indel position as a fifth character state while all other sites are equally weighted, and 2) remove all indels and alignment-ambiguous positions while imposing a 1.0: 1.1 transition to transversion weighting. In both analyses many fewer trees are recovered, 12 and 8 respectively. The topologies differ primarily in the inferred relationships of Cobaea, Cantua, Bonplandia and Acanthogilia to the remainder of Polemoniaceae. Both of these trees are very similar in structure to the full, equal-weighted analysis.

Tree Support
Bootstrap.-The bootstrap 50% majority rule tree (not shown) is structurally identical to the strict consensus insofar as the majority rule tree is resolved. Many of the clades are supported with high bootstrap percentage values (see Fig. 1). Ten clades are supported by values of 95% or greater. Hillis and Bull (1993), suggest that 70% bootstrap percentage values or greater might be a better estimate of a "95% confidence interval" for estimating the correct phylogeny. If this can be generalized to all data (but see Felsenstein and Kashino 1993), another 14 clades fall within the confidence limit.
Some portions of the trees, show little bootstrap support. In general the highest values are associated with the more terminal portions of the phylogeny. The values tend to decrease toward the more basal nodes. The only notable exception is the ancestral node of Polemoniaceae, that possesses a significant value of 99%.
Decay index.-The decay index is here employed to identify those clades that are weakly supported or may be prone to errors in phylogenetic estimation. Decay values shown in Fig. 1 range from one to three. In general, the most weakly supported nodes are the more basal nodes within Polemoniaceae. Again, the only exception to this trend is the ancestral node of Polemoniaceae.
Lanyon's jackknife.-The jackknife procedure required 53 analyses, the consensus of these analyses is shown in Fig. 3. The regions of the phylogeny that are affected by the inclusion of particular taxa are represented as unresolved polytomies. The lack of resolution indicates that a major portion of the phylogenetic estimate can change based solely on which terminal taxa are included in the analysis. However, it is the removal of Acanthogilia and the placement of Cantua that are responsible for much of the lack of resolution (i.e., the Loeselieae Clade). Resolved portions of the consensus tree represent relationships not effected by the current sampling. This includes the Gilieae Clade, the Giliandra clade, the Gilia foetida-G. rigidula clade, the Linanthieae Clade, the /pomopsis longiflora-I. multiflora-I. gunnisonii clade, and the Cobaea scanned-Co biaurita clade.
Skewness of the distribution of random trees.-One of the primary limitations to evaluation of the skewness of the distribution of sets of random trees for a data set of this size is that no 95% confidence limit for skewness values (g 1) has been determined for data sets in excess of 25 terminal taxa (Hillis 1991;Hillis and Huelsenbeck 1992;Hulsenbeck 1991). However, as a conservative test, I have used the 95% confidence limit of the 25-taxon tree for those trees with 25 or more taxa (Hillis and Huelsenbeck 1992).
The distribution of tree length for a set of 10,000 trees based on all 56 sequences is significantly skewed, with a gl statistic of -0.6622. Following Hillis (1991), I removed well-supported clades (as measured by bootstrap values of 95% or greater), except for one  The number below each internode is the decay index for that branch, ranging from dl to d3 . The frequency at which clades occur in 100 bootstrap replications is shown as a percentage, above each internode. Tribal classification of Grant (1959) is indicated by the symbol following each ingroup taxon name e+ = Cobaeeae; • = Cantuaeae; ... = Bonplandieae; D = Polemonieae; • = Gilieae). Intrageneric classification of Gilia is indicated by the numbered brackets (I = sect. Gilia; 2 = sect. Arachnion; 3 = sect. Saltugilia; 4 = sect. Giliastrum; 5 = sect. Giliandra). Gilia scabra is labeled "?" because Grant erroneously considered it synonymous with Linanthus nuttallii.
randomly selected representative, leaving 46 taxa. Another set of 10,000 random trees was generated. Again, the distribution is significantly skewed: gl = -0.4441. This was followed by removal of clades with bootstrap values of 90% or greater, save one representative,leaving 31 taxa, and skewness was determined. Similarly, the distribution remains significantly skewed (gl = -0.4881). This process continued, using 21, 15, 12, II, 10, nine, and eight taxa, each time removing members of the most well supported clades. In all cases, the gl value significantly differs from the expected value of random data. Similar results (Porter unpublished) are obtained from permutation tail probability tests (Faith and Cranston 1991).

DISCUSSION
Sequence data from the two intergenic ITS spacers provide a compelling insight into some aspects of phylogeny. However, the phylogenetic inferences from ITS sequences contains both well-supported conclusions and highly questionable ones. In general, the weakly supported portions of the phylogeny are nodes associated with the earliest diverging lineages within the family. The lack of well-supported relationships at the base of the tree could be due to biological reasons or artifact. Biological reasons could include rapid diversification, low nucleotide substitution rate, or spurious long-branch attraction. Artifacts also could result from the difficulty in aligning the more divergent lineages and the treatment of the large blocks of question marks in the data matrix that result from indels. Fortunately, even with the ambiguity of relationships associated with the earliest diverging lineages, the robust portions of the phylogeny provides an understanding of relationships in Polemoniaceae that directly conflicts with current classification. This insight includes the pattern of diversification of the phlox family as well as issues of monophyly of the genera Gilia, /pomopsis and Linanthus.

Monophyly of Polemoniaceae
The only well-supported node (based on bootstrap, decay and Lanyon's jackknife) in the basal region of the ITS phylogeny is the Polemoniaceae clade. Regardless of differential weighting of indels and/or transitions versus transversions, ITS sequences support the monophyly of Polemoniaceae. There is no support for the treatment of Cobaea as a separate family, Cobaeaceae (Don 1824). However, this study provides at best a weak test of the monophyly of Polemoniaceae. If the phylogeny is treated as unrooted, it can be recognized that a monophyletic Polemoniaceae is the result of the two longest branches (Diapensia and the Fouqueria clade) being united. These two lineages are characterized by nucleotide changes at nearly 30% of all potentially informative sites and between 37% and 50% of potentially informative sites not involving indels, raising the potential for long-branch attraction. This caution is tempered by the concordant support for a monophyletic Polemoniaceae from both chloroplast genes matK (Steele and Vilgalys 1994;Johnson et al. 1996) and ndhF (A. Prather, pers. comm.) and nuclear 18S sequences (as cited in Johnson et al. 1996).

Diversification of Polemoniaceae
The evolution and diversification of Polemoniaceae have been characterized by recent authors (e.g., Grant 1959) as a lineage of tropical origin that has radiated into the temperate regions. Grant postulates the inde-

B.
Gilia campanulata Acanthogilia glori Bonplandia gemini Fig. 3. Comparison of two parsimony analyses differing in the weighting of gaps and transitions vs. transversions. Tree A is the strict consensus of 12 trees, resulting from the analysis in which no transition/transversion bias is imposed but all gap positions were treated as a fifth nucleotide state (GAPMODE = FIFTHSTATE option in PAUP). The analysis for tree B utilized a 1. 1 to 1.0 transition to transversio n bias but all gap and alignment-ambiguous positions were removed. Tree B is the strict consensus of the resulting eight trees. pendent origin of two temperate lineages, tribes Polemonieae and Gilieae, from different tropical groups (tribes Cobaeeae and Bonplandieae, respectively). This tropical-temperate distinction has persisted. The phylogenetic inference based in matK, presented by Steele and Vilgalys (1994), is suggested to support a tropical origin although the strict consensus tree shows ambiguous relationships between the included genera of tropical distribution. Steele and Vilgalys also propose a single "temperate" lineage, rather than two. Johnson et al. (1996) present a more complete matK phylogeny in which Cantua, Cobaea, and Bonplandia form a "tropical clade." Aside from Acanthogilia, the remaining genera are included in a large "temperate clade. " Given that Acanthogilia is not tropical in distribution, the ancestral condition, if reconstructed using parsimony, is dependent on the distribution of the outgroups. Because Diapensiaceae are temperate in distribution and Fouquieriaceae are subtropical to tropical the ancestral character state assignment is ambiguous.
The phylogeny based on nucleotide sequences from ITS 1 and 2 show Acanthogilia, Cantua, Cobaea, and Bonplandia as a paraphyletic group, positioned as the earliest diverging lineages of the family. This could be interpreted as support for a tropical origin of Polemoniaceae. However, the relationships between these four genera, as inferred by ITS, have been shown to change based on taxon inclusion, character weighting (indels and transition-transversion) and how alignment-ambiguous sites are treated. It is difficult to argue that the ITS phylogeny is error free in this region. Given the uncertainty of relationships at the basal nodes the ancestral character state assignment is also uncertain. The question of a tropical origin for the family remains unanswered. Such determination again requires a more rigorous assessment of sister and outgroup relationships for Polemoniaceae than is presented here or in previous studies (e.g., Chase et al. 1993;Steele and Vilgalys 1994;Johnson et al. 1996).
The ITS spacer sequences support the remaining genera of Polemoniaceae (i.e., Gilia, Ipomopsis, Eriastrum, Langloisia, Loeseliastrum, Loeselia, Allophyllum, Collomia, Navarretia, Phlox, Linanthus, Leptodactylon and Polemonium) as a monophyletic group. This lineage has been referred to as the "Temperate Clade" (Steele and Vilgalys 1994;Johnson et al. 1996). While it is true that the majority of species in this clade occur in temperate regions, such a characterization may obscure the geographic patterns associated with diversification and dispersal within the lineage. This lineage includes several clades that are either wholly or partly tropical in distribution. For example, Loeselia ranges from southernmost Arizona, USA, through Mexico and Central America to Columbia, but all species occur in the tropics of Mexico.
Similarly, Gilia section Giliastrum (exclusive of Gilia tenerrima, G. filiformis, G. campanulata, G. inyoensis i and G. maculata) occurs not only in the temperate southwestern United States, but also the subtropical and tropical latitudes of both Mexico and South America (Peru, Bolivia, and Argentina). These examples serve to point out this clade includes tropical, SUbtropical, temperate, and even boreal lineages. It is not evident where the early diversification took place. Indeed, Grant's (1959) hypothesis of several independent lineages dispersing to and radiating within the temperate regions cannot be ruled out. For convenience I will refer to this clade as the Polemonieae Clade (Fig.   1).
The Polemonieae Clade is comprised of members from three of Grant's (1959) tribes (tribes Gilieae, Polemonieae, and Bonplandieae). None of these three tribes is supported as monophyletic (Fig. 1). Rather, three primary lineages, in addition to the genus Polemonium, make up this clade in all of the most parsimonious trees. The support for relationships between these clades is weak, but the membership of the three lineages is consistent regardless of weighting of transitions vs. transversions or indels and it is corroboration from other molecular phylogenetic studies (Steele and Vilgalys 1994;Prather 1995;Johnson et al. 1996). I refer to these clades as the Loeselieae, Gilieae, and Linanthieae Clades (Fig. 1). Note that the endings are used to distinguish the clades from generic names and are not intended to imply a particular hierarchic rank.
The Loeselieae Clade, which includes Gilia section Giliandra, Gilia section Giliastrum (exclusive of Gilia campanulata, G. inyoensis, G. maculata and G. filijormis), Gilia scabra, Loeselia, Ipomopsis, Langloisia, Loeseliastrum, and Eriastrum, is the sister group to the remaining members of the Polemonieae Clade. The members of this clade have traditionally been included within tribe Gilieae, with the exception of Loeselia, which has been placed in Bonplandieae (Grant 1959). The potential relationship between Loeselia and certain members of tribe Gilieae was recognized by Grant (1959), given his hypothesis that tribe Gilieae has its origin within or near Loeselia. Support, as measured by bootstrap, jackknife and decay, is low for many of the nodes within the Loeselieae Clade. Even so, the phylogenetic inference is highly consistent with that of matK (Johnson et al. 1996). Several noteworthy points can be made concerning the phylogenetic inferences based on ITS sequences. First, the two subspecies of Langloisia setosissima are supported as the sister group of Eriastrum. The sister group of this (the Eriastrum-Langloisia Clade) is Loeseliastrum schottii, supporting Timbrook's (1986) assertion that Langloisia s.s. and Loeseliastrum do not represent a monophyletic group. This represents a direct conflict between ITS and phylogenetic inferences from chloro-Porter ALISO plast matK (Johnson et al. 1996) and trnL intron (Porter, unpubl.) sequences, that support Langloisia and Loeseliastrum as sharing most recent common ancestry. Second, note that /pomopsis is not monophyletic. The placement of /pomopsis tenuifolia as sister to /pomopsis, Eriastrum, Langloisia, and Loeseliastrum is not surprising, given that this member of the Giliopsis group was formerly included in Loeselia (Gray 1876; 1886). /pomopsis sonorae is also not unambiguously more closely related to other /pomopsis species than to other genera. There may be three lineages, representing a grade, included in the current circumscription of /pomopsis. The paraphyly of /pomopsis is corroborated by chloroplast DNA sequence data (Johnson et al. 1996;Porter, unpubl.) and a more detailed ITS analysis of the Loeselieae Clade (Porter, in prep.). Third, the members of Gilia within this clade are neither monophyletic (although, Gilia section Giliandra is monophyletic) nor closely related to those Gilia associated with the type (i.e., Gilia section Gilia, type = G. laciniata R. & P.). In fact, Gilia section Giliastrum, as circumscribed by Grant (1959) is polyphyletic. In addition, Gilia scabra, considered by Grant (1959: 140) synonymous with Linanthus nuttallii, is suggested to share common ancestry with Loeselia. The Gilieae Clade includes Allophyllum, Collomia, Navarretia, Gilia sections Saltugilia, Arachnion, Gilia, and one member of section Giliastrum. Historically, Allophyllum has been variously treated within Collomia (Bentham 1845; Gray 1870), close to Collomia in tribe Polemonieae (Grant and Grant 1955;Grant 1959),or included within Gilia Brand 1907;Mason and Grant 1948;Cronquist 1984), in tribe Gilieae. Navarretia has been of equally uncertain affiliation: included within Gilia by Gray (1870); allied with Collomia, Cantua and Gymnosteris in the Collomia tribe (Wherry 1945); or within Tribe Gilieae, near Langloisia and Leptodactylon (Grant 1959). It is perhaps not surprising that ITS sequences support a close relationship between Navarretia, Collomia, AIlophyllum and a portion of what is now considered Gilia. Even so, this is a striking noncorrespondance with traditional taxonomy because it not only segregates only a portion of Gilia and allies it with Navarretia (both members of Grant's tribe Gilieae), but also strongly supports the relationships between this group and two members of tribe Polemonieae, Allophyllum and Collomia. The monophyly of the Gilieae Clade, supported here by ITS sequences, is both well supported in terms of bootstrap values and Lanyon's jackknife analysis and also corroborated by sequence data from the chloroplast gene matK (Steele 1992;Steele and Vilgalys 1994;Johnson and Soltis 1995). The correspondence between ITS and matK gene phylogenies and the relative robustness of this portion of the ITS phylogeny strongly supports the monophyly of this group.
Perhaps the most intriguing of the hypothesized relationships based on ITS sequence data involves the Linanthieae Clade. This clade is comprised of the genera Linanthus, Leptodactylon and Phlox, as well as four species of Gilia currently included in section Giliastrum. The Linanthieae Clade is well supported but very inconsistent with current classification. Although there is a long history associating Linanthus and Leptodactylon together (Gray 1870;Brand 1907;Wherry 1940;Grant, 1959) and they are currently both included in tribe Gilieae, Phlox (currently in tribe Polemonieae) has not been considered closely related. While floral morphology of Leptodactylon has been compared with that of Phlox (Gray 1870), no direct relationship was espoused. It is noteworthy that Linanthus, Phlox and Leptodactylon all possess pantoporate pollen grains Stuchlik 1967;Taylor and Levin 1975), very similar leaf flavonoid and phenolic chemistry (Smith et al. 1977;Dean et al. 1978;Smith et al. 1982), and possess opposite leaves, an unusual trait in Polemoniaceae.
The Linanthieae Clade is also intriguing because four species of Gilia (G. campanulata, G. inyoensis, G. filiformis, and G. maculata) are also members. These species have historically been closely aligned (see for example Brand 1907); however, most treatments prior to 1989 maintained G. maculata in Linanthus, while the remaining species were retained in Gilia sect. Giliastrum.  has cautiously argued that, in spite of the unique traits possessed by G. maculata, there is sufficient morphological similarity between Gilia campanulata, G. inyoensis, G. filiformis and G. maculata to include them in the same genus, i.e., Gilia. ITS sequence data, presented here, supports in essence Patterson's proposition; however, this group of species is related not to members of Gilia, but to Linanthus, Leptodactylon and Phlox. It is also important to note that ITS sequences do not support the monophyly of these species.

Gilia is not monophyletic
The genus Gilia has remained, over the last 100 years, a recurring taxonomic problem, the circumscription of which has changed radically. Mason and Grant (1948) correctly point out that all of the herbaceous genera of the Polemoniaceae, with the exception of Polemonium and Phlox, have historically been placed in Gilia. One of the most broad interpretations of the genus was that of Asa Gray (1886). While recognizing that Gilia was "certainly a polymorphous ... genus" (Gray 1870: 262), he included the currently recognized genera Langloisia, Loeseliastrum, Gymnosteris, Leptodactylon, Linanthus, Navarretia, /pomopsis and Er-iastrum within Gilia. The most recent classification of the family (Grant 1959) has maintained all these segregate genera. However, as noted above, Cronquist (Cronquist 1959;Cronquist 1984) has included /pomopsis and Allophyllum within Gilia.
ITS sequence data provide insight into this conflict. The monophyly of representatives of Gilia sects. Gilia (here represented by G. capitata and G. tricolor), Arachnion (represented by G. crassifolia, G. ochroleuca and G. jiavocincta), Saltugilia (represented by G. stellata and G. scopulorum), and Giliastrum , sensu Grant 1959 (represented by G. tenerrima) is supported (Fig.  2). Conspicuously absent from this clade are Gilia scabra and members of Gilia sects. Giliastrum (represented by G. foetida, G. rigidula, G. latifolia, G. campanulata, G. inyoensis, G. filiformis, and G. maculata), and Giliandra (represented by G. caespitosa, G. subnuda, G. pentstemonoides, G. mcvickerae, G. hutchinsifolia and G. nyensis). All of these samples, except G. campanulata, G. inyoensis, G. filiform is, and G. maculata (in the Linanthieae Clade), are within the Loeselieae Clade. The most parsimonious placement of anyone of these species into the Gilieae Clade requires between 27 and 51 steps (ACCTRANS reconstruction). Moreover, if representatives of Gilia sensu Grant (1959) are constrained to be monophyletic, three minimal-length trees are found of tree length 1693, 47 steps longer than the unconstrained analysis. The inclusion of representatives of /pomopsis into Gilia (as proposed by Cronquist) does not ameliorate this situation. If /pomopsis is constrained to be included in Gilia, sensu Cronquist, the eight trees found are 59 steps longer than the unconstrained analyses. Including additional taxa to make the representatives of Gilia monophyletic-if one considers the minimal-length ITS trees-will result in a group as morphologically diverse as Gray's circumscription of Gilia. If this is considered an unacceptable treatment (i.e., considering the Polemonieae Clade as a single genus, Polemonium) , then the genus Gilia is must be recognized in the more strict sense (i.e., G. tenerrima and its the sister group), it will be necessary to remove those clades that do not share common ancestry.

Conclusions
Sequences of the two internal transcribed spacers, ITS 1 and ITS2, have been demonstrated to provide reliable phylogenetic estimates for portions of Polemoniaceae. The bootstrap, decay index, Lanyon's jackknife, and distribution of the tree length of random trees provide support for much of the phylogenetic inference. These include the Giliandra clade, Gilia foetida-G. rigidula clade, the Allophyllum-Gilia clade, the Linanthus-Phlox-Leptodactylon clade, and the /pomopsis gunnisonii-/. longiflora-l. multiflora clade. Un-fortunately, resolution of relationships among most of the major lineages, representing the basal radiation of the family, is suspect. Even given this limitation, ITS data unambiguously fail to support the current classification of Polemoniaceae that recognizes two temperate tribes, Polemonieae and Gilieae (Table 1). Rather, all of the genera, exclusive of Bonplandia, Cantua (and presumably Huthia), Cobaea, and Acanthogilia, are within a single clade, with members of Grant's tribes, Polemonieae and Gilieae variously related. Further, it has been shown that ITS sequence data does not support Gilia, as currently circumscribed, as a monophyletic group. The taxa now treated as Gilia represent at least five independent lineages, and potentially more (Fig. 1).