Cheilanthoid Ferns (Pteridaceae: Cheilanthoideae) in the Southwestern United States and Adjacent Mexico-A Molecular Phylogenetic Reassessment of Generic Lines

Cheilanthoids are the most commonly encountered fe rn species of the arid southwest and other xeric habitats throughout the world. Cheilanthes, Notholaena, Pellaea, and Bommeria are the best known southwestern genera, but some authors recognize segregate genera such as Argyrochosma, Aspidotis, Astrolepis, and Pentagramma. Others reject distinctions among some of these genera as artificial , leaving cheilanthoid generic concepts in a state of flux. This unsettled taxonomy is often attributed to morphological homoplasy associated with adaptation to xeric habitats , suggesting the need for new analyses that do not depend on potentially misleading morphology. Nucleotide sequences of the maternally inherited, chloroplast-encoded rbcL gene from 57 species that bear on the relationships of the cheilanthoids of the southwest were c1adistically analyzed under the optimality criterion of maximum parsimony. The results provide new insights into phylogenetic re lationships and generic circumscriptions of these ferns . Mexican Llavea cordifolia is rejected from the cheilanthoids, traditional Cheilanthes, Notholaena, and Pellaea are polyphyletic, and the segregations of A rgyrochosma, Aspidotis, Astrolepis, and Pentagramma are supported. To assess confidence in these conclusions, results of the rbcL-based analys is are compared with those based on ITS sequences of biparentally inherited nuclear ribosomal DNA (nrDNA) for a subset of cheilanthoid taxa. These two data sets yield remarkably congruent topologies at shallower phylogenetic levels, suggesting that previous taxonomic problems in this group may indeed be attributable to diffi culties in interpreting the taxonomic significance of morphological characters . Disagreement at deeper levels of the topologies suggests the need to incorporate data from less rapidly evolving nrDNA regions.


INTRODUCTION
Cheilanthoid ferns are among the most fascinating elements of the flora of the southwestern United States and adjacent Mexico. In part this is because they defy the stereotypic view of ferns as plants of cool, mesic woodlands by thriving in the xeric and semixeric habitats that are at least seasonally characteristic of this region. The apogamous lifestyle (Gastony and Windham 1989) of many of these taxa minimizes the impact of hot, dry conditions on the relatively delicate gametophyte generation because apogamy both shortens the time from spore to sporophyte (Whittier 1970) and eliminates gametophytic dependence on free water for fertilization by motile sperm. Among the features thought to adapt many cheilanthoid sporophytes to such habitats are an indument of wax or insulating scales or trichomes on the leaves, reflexion of the leaf margin to form a false indusium over the sori, and the ability to die back to the rhizome and survive the driest periods in dormant condition. Convergence in adaptive morphology associated with xeric habitats is often credited with responsibility for legendary problems in assessing natural evolutionary lineages in cheilan-thoids and for our consequent inability to develop an acceptable generic taxonomy Tryon 1973, 1982 p. 248;Lellinger 1989 p. 111).
Cheilanthoid ferns constitute subfamily Cheilanthoideae of the homosporous fern family Pteridaceae, a subfamily virtually worldwide in distribution but particularly richly developed in southwestern North America (Tryon and Tryon 1973;Tryon et al. 1990). Prior to newly available molecularly based insights into phylogenetic relationships in this group (Gastony and Rollo 1995), the most authoritative, current, worldwide taxonomy recognized 12 genera of cheilanthoid ferns (Table 1, adapted from Tryon et al. 1990).
Subfamily Cheilanthoideae is difficult to circumscribe morphologically (see key and description in Tryon et al. 1990, pp. 2~31-232 , 240-241). In general, the sori are borne on the distal region of the veins, but in some cases they extend inward from the leaf margin. In most genera, the sori are protected by having parts or all of the margin of the leaflet reflexed over the sori to form a false indusium, and variant states of this character have been used to define the three most frequently encountered cheilanthoid genera in the southwest, Cheilanthes, Pellaea, and Notholaena. In Table I. Synopsis of the generic class ification of che ilantho id ferns (Pteridaceae subfamily Cheilanthoideae) according to the most recent worldwide treatment by Tryon et al. (1990), with the number of species included in the prese nt study given in parentheses. Several species from beyond the southwestern U.S. and adjacent M ex ico were included in a previous smaller data set by Gastony and Ro llo (1995) and are incorporated into the prese nt analysis to provide a broader phylogenetic context for the sou thwestern species.
Cheilanthes the reflexed margin is said to be discontinuous, in Pellaea continuous, and in Notholaena the margin is said to be not reflexed over the sori or only slightly so. Unfortunately these characters of leaf mar-gin morphology are so variable as to render these circumscriptions meaningless. Gastony and Rollo (1995) reviewed the frustration of pteridologists from the time of Copeland (1947) who have tried to decipher natural evolutionary lineages that can be circumscribed as genera, subgenera, etc. That history of frustration is synopsized by a statement in the most recent taxonomic summary of cheilanthoid ferns on a worldwide basis: " ... evolutionary lines within the subfamily and the relations of these to other groups in the Pteridaceae are scarcely known" (Tryon et al. 1990, p. 241). Molecular insights into cheilanthoid phylogeny by Gastony and Rollo (1995) suggest that we do not have good generic characters in cheilanthoids because we do not have good genera in cheilanthoids.
Mindful of the attribution of taxonomic problems in cheilanthoids to homoplastic adaptive morphology Tryon 1973, 1982;Lellinger 1989), Gastony and Rollo (1995) used newly available molecular data, presumably unaffected by adaptation to arid habitats, to gain new insight into evolutionary lines within cheilanthoids. They included 26 ingroup taxa in their study of cheilanthoid phylogeny based on nucleotide sequences of the rbcL gene. The present study more than doubles their database to include 53 ingroup taxa. Many of these are directly relevant to the flora of the southwest because they occur in this region. The relevance of others is indirect by providing a more complete context in which to assess relationships among these southwestern species. In addition, this study presents a first attempt to assess confidence in the results of the rbcL data set by comparing a subset of the rbcL results with those of a comparable molecular data set derived from nuclear DNA. Gastony and Rollo (1996) provided a preliminary report of these comparative findings based on analysis of a variant and incompletely aligned rbcLIITS data set for some of the species in the present study.

MATERIALS AND METHODS
This study provisionally follows the taxonomy of the most recent worldwide treatment of cheilanthoid ferns by Tryon et al. (1990), summarized in Table 1. Sporophytes collected directly from nature or grown from spores taken from specimens provided by colleagues were the sources of DNAs used to generate the nucleotide sequences in this study (Table 2), except that the sequence of Cheilanthes (Doryopteris) concolor was supplied by Mitsuyasu Hasebe via GenBank accession U05621. DNA extractions , polymerase chain reaction (PCR) amplifications, sequencing reactions, rbcL primers, and rbcL outgroup selection were as reported by Gastony and Rollo (1995). Primers used in PCR amplifications and sequencing of the internal transcribed spacer (ITS) regions of nuclear ribosomal DNA are presented in Table 3; in addition M13 primers were used in sequencing single-stranded cloned templates as noted in Gastony and Rollo (1995).
All 52 taxa in Table 2 (plus Cheilanthes concolor noted above) were used as the ingroup in the rbcL analysis, including Pityrogramma calomelanos and P. trifoliata which were treated as part of the in group in order to test the affinities of P. triangularis, as discussed below. Maximum parsimony analysis was carried out with PAUP*4.0d54, a test version of the forthcoming new release of PAUP (see Swofford 1993), on a Power Macintosh 85001120 (using TBR branch swapping, MULPARS, and steepest descent). A subset of these species (identified with asterisks in Table 2) was used to generate ITS sequence data. Ingroup and outgroup designation for the ITS analysis is based on the results of the rbcL analysis in Fig. 1, with the clade of species 1-26 (Pellaea andromedifolia to Cheilanthes lanosa) serving as the ingroup and the well defined sister clade of species 27-46 (Notholaena sulph urea to Pityrogramma triangularis) serving as outgroup. From these clades ITS sequences were obtained from in group species 1, 2, 4-14, 16-19, 21, and 23, and the outgroup was represented by species 27, 28, 35, and 38. By using rbcL sequences from this same subset of Fig. 1 species and conducting independent maximum parsimony analyses with PAUP*4.0d54 (with TBR branch swapping, MULPARS, and steepest descent), cladistic analyses of these two molecular data sets were directly comparable. Because of numerous insertions and deletions (indels), the ITS sequences required alignment. This was carried out with Clustal W within the software program SeqPup (Gilbert 1995; < ftp://iubio.bio.indiana.edu/molbio/seqpup/», followed by visual inspection and a few realignments by hand. Two regions (8 base pairs and 9 base pairs, respectively) were excluded from the analyzed ITS sequences because their alignment appeared particularly arbitrary. Indels were not used as characters in the ITS data set. Analyses of the full rbcL data set and of the ITS and rbcL subsets required heuristic searches, and 1000 random addition sequence replicates were used in each analysis to avoid being confined to potential suboptimal islands of most parsimonious trees. Support for clades resolved by the foregoing basic analyses was determined by the bootstrap option of PAUp, using a random addition sequence with TBR, MUL-PARS, and steepest descent sUboptions in effect. Bootstrapping the full rbcL data set used 100 random addition replicates; 500 random addition replicates were used in bootstrapping the ITS subset and its comparable rbcL subset.

Full rbcL Data Set
During the first random addition sequence replicate, PAUP found a single most parsimonious tree ( Fig. 1) of 1462 steps with a consistency index (CI) of 0.423, a retention index (RI) of 0.614, and a rescaled consistency index of 0.260. No new minimal trees were found during any of the next 999 random addition sequence replicates, suggesting that local optima insularity (Maddison 1991) is not a problem in this data set and that all of the most parsimonious trees have been found (Swofford 1993, p. 35). Of the 1321 rbcL nucleotide positions used as characters, 862 were constant, 143 were variable but parsimony-uninformative, and 316 were parsimony-informative.
Taxa rejected from the cheilanthoid ingroup.-As noted by Gastony and Rollo (1995) Coniogramme was included in subfamily Cheilanthoideae by Tryon et al. (1990) but was placed well outside the cheilanthoids and sister to all other Pteridaceae plus Vittariaceae in the global analysis of fern phylogeny by Hasebe et al. (1995). This indicates that Coniogramme should be excluded from the cheilanthoids and instead entered as a distant member of the outgroup. In order not to prejudice the results of PAUP's analysis of this much larger data set of cheilanthoids, however, our searches were conducted with Coniogramme alternatively specified as a member of the outgroup or as a member of the cheilanthoid ingroup. Results were identical under both alternatives. Just as shown by Hasebe et al.'s (1995) global rbcL analysis involving many fewer cheilanthoid taxa, Coniogramme (Fig. 1, species 53) is rejected from the cheilanthoids and placed within the outgroup in the present analysis. Similarly, as in the previous analysis with a smaller data set of cheilanthoid taxa (Gastony and Rollo 1995), PAUP was unable to place Llavea (species 52) in the ingroup. Instead Llavea is sister to Coniogramme in the outgroup in a clade strongly supported by 26 synapomorphies and a 100% bootstrap value. This result for Llavea was obtained whether Coniogramme was defined as part of the outgroup or was treated as a member of the cheilanthoid ingroup. Yatskievych et al. (1990) recently reassessed the taxonomy of the genus Pity rogramma, generally placed outside the cheilanthoids in Pteridaceae subfamily Taenitidoideae. They concluded that the P. triangularis complex of the southwestern United States and adjacent Mexico merits recognition as an independent genus Pentagramma, and they formalized a suggestion by Tryon and Tryon (1982 p. 218) that this southwestern complex be transferred to subfamily Cheilanthoideae. To test this new taxonomy, three species of Pityrogramma (including southwestern P. triangularis) were included in the present analysis as elements of the cheilanthoid ingroup. PAUP's analysis could not maintain P. calomelanos and P. trifoliata (species 54, 55) as members of a monophyletic ingroup. Instead it placed them in the outgroup sister to Table 2. Sources of living material of ingroup species and three putative outgroup species of Pityrog ramma used to generate rbcL sequences for thi s study. Outgroup species Adiantum pedatum, Ceratopteris thalictroides, Platyzoma microphylium, Coniogrammejaponica, and Vittaria lineata and ingroup species Cheilanthes concolor were obtained as in Gastony and Rollo ( 1995). Taxonomy follows that of Tryon et al. (1990). Voucher herbarium specimens are deposited at the Indiana Un iversity herbarium (lND) except for Trachypteris pinnata in Brazil (SJRP). Asterisks foll owing the species names identify the subset of species constituting the ingroup (*) and outgroup (**) for the ITS and directly comparable rbcL analyses as explained more fu ll y in the text.   (Benham 1992). , Also vouchered at MO.
Ceratopteris of Pteridaceae subfamily Ceratopteridoideae ( Fig. 1), thereby supporting their traditional placement outside of subfamily Cheilanthoideae. Clarification of their relationship to other members of traditional subfamily Taenitidoideae must await a broader study of that group. These two Pityrogramma species are very strongly united by 50 synapomorpbies and a 100% bootstrap value, but their close association specifically with Ceratopteris should be regarded as only provisional, given the very long autapomorphic branch length of C. thalictroides and the weak bootstrap support for this outgroup clade. Pityrogramma triangularis (species 46), on the other hand, did nest robustly within the cheilanthoid ingroup, just above the basalmost ingroup clade of Bommeria species (47-49) in Fig. 1. This very strongly supports removal of the P. triangularis complex from Pityrogramma to Pentagramma and the placement of Pentagramma in subfamily Cheilanthoideae.
Pellaea.-Pellaea is represented in this analysis by 12 species assigned to three sections by Tryon and Tryon (1~82). Species 1-5 and 8-10 in Fig. 1 are traditionally placed in Pellaea section Pellaea. With the exception of South African P. rufa (species 2) and Chilean P. myrtillifolia (not included in this analysis), sec-tion Pellaea is primarily distributed in the southwestern United States and Mexico, although the ranges of some species of this section extend beyond that region (Tryon 1957(Tryon , 1968Tryon and Britton 1958). Pellaea rotundifolia and P. calidirupium (species 6, 7) are New Zealand species of section Pia tyloma. In separate clades in the lower half of Fig. 1 are African P. boivinii (37) and P. calomelanos (45), both of section Holcochlaena. Pellaea in the traditional sense of Tryon et al. (1990) is clearly polyphyletic. As noted by Tryon and Tryon (1982), each section of Pellaea was at one time recognized as a separate genus. This rbcL study suggests that if taxonomy is to reflect phylogeny, at least some of these sections will again require generic recognition. Even largely southwestern Pellaea section Pellaea is polyphyletic. Species 1-5 of this section at the top of Fig. 1 are sister to a clade containing Pellaea section Platyloma (species 6, 7) from New Zealand. The remaining three speci~s of Pellaea section Pellaea (8-10) are sister to Cheilanthes species 11 and 12 recently segregated as the genus Astrolepis, whose distribution ranges from California to Oklahoma and southward along the Cordillera to Argentina, with outlying stations in Georgia and the West Indies (Benham and  , Corrected from the original fungal primer of White et al. (1990) to be plant specific at the sixth (T ~ G) and eighth (A ~ G) positions from the 5' end.
h Corrected from the original fungal primer of White et al. (1990) to be plant specific at the seventeenth (C ~ T) position from the 5' end.
' Corrected from the original fungal primer of White et al. (1990) to be plant specific at the fourth (G ~ A) position from the 5' end. 14 Notholaena pilifera .
Ar; rochosma L...  Table 2. The number of synapomorphies supporting each clade and autapomorphic branch lengths for terminal taxa are indicated above the lines. Numbers below lines represent bootstrap percentages based on 100 replicates. Species are numbered I-57 to facilitate discussion in the text. Taxonomy and nomenclature follow the treatment of Tryon et al. (1990), which presents the most recent comprehensive treatment of cheilanthoids on a worldwide basis. Information at brackets indicates sectional taxonomy within Pellaea. alternative generic taxonomies discussed in the text, and species designated as the outgroup. Note that maximum parsimony was unable to place Llavea cordifolia (52), Pityrogramma calomelanos (54) . or P. trifoliata (55) as members of a monophyletic ingroup, relegating them instead to positions within the outgroup. Windham 1992). These authors thought the base chromosome number of x = 29 common to Astrolepis and Pellaea arose independently in the two groups because it was not previously apparent that Pellaea and Astrolepis are phylogenetic ally closely related. The rbcL phylogeny in Fig. 1 indicates that in this case x = 29 is a synapomorphy resulting from common ancestry.
Argyrochosma and Notholaena.-The entire upper clade discussed above (species 1-12 of Fig. 1) is sister to a group of predominantly southwestern and Mexican species until recently considered species of Notholaena section Argyrochosma (Tryon and Tryon 1982). Windham (1987) segregated these Notholaena species as the genus Argyrochosma, partly because of their unique chromosome base number of x = 27, and he hypothesized that Argyrochosma is the sister group of southwestern Pellaea. This sister relationship is supported by rbcL. Furthermore, rbcL deeply separates these Argyrochosma species from the rest of Notholaena (species 27-31) sensu Tryon and Tryon (1982), robustly supporting Windham's generic taxonomy. Tryon's (1956) revision of Notholaena was based on unpublished work by Charles A. Weatherby that took a broad view of the genus. Tryon and Tryon (1982) removed the first 21 species of Tryon's (1956) revision to Cheilanthes. Among those 21 excised species are C. beitelii and C. cochisensis (Fig. 1, species  11, 12) that this analysis shows to have been correctly removed from Notholaena because they nest with the Pellaea species toward the top of the tree, deeply removed from the Notholaena clade represented by species 27-31. As discussed above, these two species should also be removed from Cheilanthes to Astrolepis as done by Benham and Windham (1992) and Benham (1992). This analysis also shows that another of the 21 excised species (Notholaena aurea (Poir.) Desv.) was correctly removed to Cheilanthes as C. bonariensis (Fig. 1, species 21), as discussed below. Species 27-31 of Fig. 1, considered elements of Notholaena section Notholaena by Tryon and Tryon (1982), form a fairly well supported clade to which three species of Cheilanthes discussed below are successively sister.
Cheilanthes sensu lato.-Cheilanthes is a complex genus of 150 or more species and is worldwide in distribution according to its last comprehensive treatment by Tryon et al. (1990). Cheilanthes in this sense is represented by species 11, 12, 17-26, 32-34, and 38-42 in Fig. 1. Species 11 and 12 are those segregated as Astrolepis discussed above, a clearly appropriate generic segregation by Benham and Windham (1992). Species 17-26 are American, largely southwestern, species of Cheilanthes and are discussed below. Cheilanthes species 32-34 and 38-42 are deeply separated from the foregoing in a clade robustly supported by 11 synapomorphies and a 89% bootstrap value. Of these, species 32-34 are American and are with varying degrees of confidence associated with Notholaena (species 27-31). Cheilanthes rigida (32) of Mexico is one of several species whose affinities within Cheilanthes were unclear to Tryon and Tryon (1982). It is sometimes recognized as the genus Cheiloplecton, one of the segregate genera not accepted by Tryon et al. (1990) . Cheilanthes aurea (33) of Mexico, has a yellow wax on the underside of its leaves, not unlike some species of Notholaena with which it shares this clade. Cheilanthes leucopoda (34) of the Edwards Plateau of Texas and of Mexico is another of the several species whose affinities within the genus were unclear to Tryon and Tryon (1982) . A morphological doctoral study by Reeves (1979) placed it into Cheilanthes subgenus Cheilanthes along with C. lanosa (26) in the deeply separated clade above it. The long autapomorphic branch lengths of species 32-34 suggest that the closest affinities of these species are not yet determined and that incorporation of additional species may clarify their affinities. Cheilanthes species 38, 39 are Asian and are segregated by some authors as separate genera Sinopteris and Aleuritopteris, respectively, but their generic segregation from Cheilanthes was not accepted by Tryon et al. (1990). Species 40 is Cheilanthes californica, maintained in Cheilanthes by Tryon et al. (1990), but often recognized as one of the three California species of Aspidotis (Smith 1975). Its association with the two Asian species is very weakly supported in this analysis and will probably be modified by incorporation of additional species. Hawaiian Cheilanthes decora (41) and pantropical C. concolor (42) were transferred to Cheilanthes from their previous placement in Doryopteris (Tryon 1942) by Tryon and Tryon (1982). Their strong association with Doryopteris pedata var. palmata (43) indicates that their true affinity is with Doryopteris rather than with Cheilanthes and that their 1982 transfer should be reversed. Clearly Cheilanthes in the broad sense of Tryon et al. (1990) is polyphyletic and warrants fragmentation into several genera. This CheilantheslDoryopteris alliance (species 41-43) is sister to Trachypteris pinnata (44). Trachypteris is locally Andean with disjunct distributions in Brazil and the Galapagos and with a putatively congeneric species (T. drakeana C. Chr.) in Madagascar (Tryon and Tryon 1982). The pedate form of the mature lamina of Andean T. induta and the typically pedate but heteroblastic and unusually variable lamina architecture of Doryopteris were noted by Tryon and Tryon (1982). The rbcL cladistic relationship of species 41-44 ( Fig. 1) suggests that this leaf architecture may be synapomorphic, although it must be noted that bootstrap support for this clade is quite weak. Tryon and Tryon (1982) also hypothesized a relationship between Doryopteris and Pellaea, particularly citing features of South American Pellaea pinnata (of Pellaea section Ormopteris, not included in this analysis). The cladistic relationship between Pellaea calomelanos (45) and the DoryopterislTrachypteris alliance (41-44) at first suggests some degree of molecular support for their observations. However, as discussed above, Pellaea as circumscribed by Tryon and Tryon (1982) is polyphyletic. Of the Pellaea species examined here, only P. calomelanos (species 45 of Fig. 1) relates to the Doryopteris/Trachypteris alliance, and it is deeply separated from its fellow member of Pellaea section Holcochlaena (P. boivinii, species 37 of Fig. 1) and from Pellaea sections Pellaea and Platyloma (species 1-10 at the top of Fig. 1).
Southwestern Cheilanthes.-The largely southwestern species of Cheilanthes included in this analysis (species 17-26) form their own clade that is fairly strongly supported by nine synapomorphies and a 84% bootstrap value. Species 18-20 have been recognized as part of the "alabamensis group" (Windham and Rabe 1993) in part because of their synapomorphic base chromosome number of x = 29, whereas x = 30 is typical of Cheilanthes. Cheilanthes allosuroides (17; widespread on dry, rocky slopes in Mexico) is very strongly grouped with the three species of the alabamensis group (18-20) in this analysis. We are unaware of a well-documented chromosome count for C. allosuroides but anticipate that it may be based on 29 rather than 30, given its rbcL relationship with the alabamensis group seen here. Its potential membership in the alabamensis group is supported by the observation of Mickel and Beitel (1988) that it shares morphological features with C. notholaenoides and C. cucullans, which are species of the alabamensis group according to Reeves (1979). Cheilanthes bonariensis (21), ranging from Arizona to Texas and southward to South America and the West Indies, was known as Notholaena aurea until it was transferred to Cheilanthes by Tryon and Tryon (1982). The wisdom of this transfer is endorsed by rbcL's placement of species 21 deeply within Cheilanthes. Mexican Cheilanthes intramarginalis (22) is interpreted by some as a member of the segregate genus Mildella (Hall and Lellinger 1967), but rbcL (Fig. 1) indicates that that taxonomy would render this group of southwestern Cheilanthes paraphyletic. Species 23-25 are southwestern and Mexican species treated by Reeves (1979) as members of Cheilanthes subgenus Physapteris, a monophyletic group in this analysis. Cheilanthes lanosa, of Texas and the eastern U.S., was said by Reeves (1979) to represent Cheilanthes subgenus Cheilanthes, the subgenus to which he also attributed C. leucopoda (34) of the Notholaena clade. According to rbcL (Fig. 1) these two species cannot represent a monophyletic subgenus.
Bommeria.-Bommeria species (47-49) constitute the basalmost clade in the cheilanthoids. Gastony and Rollo (1995) noted that rbcL strongly supports the transfer of Hemionitis elegans to Bommeria originally proposed by Ranker (1990) and Ranker and Haufier (1990). The present analysis adds Hemionitis palmata and Bommeria hispida (the only southwestern species of Bommeria north of Mexico) to the species of these genera analyzed by Gastony and Rollo (1995). Hemionitis elegans remains robustly nested within Bomme ria (species 47-49), forming a clade supported by 18 synapomorphies and 100% bootstrap support whereas H. palmata and H. levyi (species 35, 36) form their own very strong grouping in a deeply separated clade.

Assessing Confidence in the rbcL Phylogeny
Does analysis of rbcL nucleotide sequences provide the long-sought key to understanding phylogenetic lineages and generic circumscriptions in cheilanthoid ferns, or does rbcL simply replace our unsatisfactory grasp of cheilanthoid phylogeny based on morphology with a different but equally impaired view based on molecular data? This question needs to be addressed because of a potential problem with phylogenetic reconstruction based on chloroplast DNA (cpDNA). Although the lineages of sexually reproducing species are based on biparental inheritance, cpDNA phylogenies, including those based on the cpDNA rbcL gene, are maternal, biparental, or paternal depending on how the chloroplast genome is inherited. Gastony and Yatskievych (1992) reviewed the paths of cpDNA inheritance in higher plants and demonstrated that the chloroplast genome is maternally inherited in cheilanthoid ferns. Work with angiosperms has shown that hybridization, introgression, and lineage sorting can yield erroneous reconstructions of species lineages based on the maternally inherited chloroplast genome (Rieseberg and Soltis 1991; Doyle 1992). The frequency of allopolyploid hybrid speciation in extant ferns is well known (Wagner 1954(Wagner , 1973Wagner and Wagner 1980), and it is reasonable to infer that such speciation has occurred in the past as well. This has led some to interpret the high chromosome numbers of ferns relative to those of angiosperms as a legacy of paleopolyploidy (reviewed in Gastony 1991). Reticulate evolution resulting from paleopolyploidy would not be tracked by the matrilineal chloroplast genome. Consequently cpDNA gene trees in ferns may be highly discordant with species trees based on nucleus-encoded biparentally inherited characters such as morphology.
One might therefore ask why we find unexpected rbcL phylogenetic relationships in cheilanthoids, such as the separation of southwestern Pellaea section Pel-laea into two clades with one clade sister to Pellaea section Platyloma from New Zealand and the other clade sister to Astrolepis (Fig. 1). These may occur 1) because species relationships based on biparentally inherited, nucleus-encoded morphology are correct but matrilineal rbcL phylogenies do not track species relationships, or 2) because traditional morphological characters are homoplastic (perhaps resulting from adaptation to xeric habitats) or have been misinterpreted, yielding misleading information about species relationships, whereas rbcL is more insightful.
We can test these alternative explanations of unexpected findings without relying on potentially homoplastic morphological data by determining whether results based on matrilineal rbcL agree with those based on biparentally inherited, nucleus-encoded molecular data. The best nucleus-encoded molecular data source at present is nuclear ribosomal DNA (nrDNA). Because of their utility in angiosperms at the suprageneric level (Baldwin 1992) we analyzed the internal transcribed spacer (ITS) regions (ITS 1 and ITS2) of nrDNA as a source of nuclear molecular data directly comparable to the data from rbcL. The structure of the 18S-26S nrDNA repeat unit including the ITS regions is illustrated by Baldwin (1992). In cheilanthoid ferns ITS 1 is about 338 bp long and ITS2 is about 351 bp long. Concerted evolution renders the many copies of these genes homogeneous within individuals (Hillis et al. 1991), rendering them phylogenetically useful (Sanderson and Doyle 1992).
The cheilanthoid ingroup rbcL phylogeny in Fig. 1 provides a well defined hypothesis of rbcL relationships that can be tested with ITS data. The major clade from Pellaea andromedifolia to Cheilanthes lanosa (species 1-26) contains several of the unexpected taxon positionings noted earlier, and 19 of its species were used as a test ingroup from which ITS sequences were determined ( Table 2). The well-defined sister clade from Notholaena sulphurea to Pityrogramma (Pentagramma) triangularis (27-46) provided the four species used as an outgroup (Table 2). Parallel analyses of this subset of taxa, based respectively on rbcL and ITS, yielded results that are directly comparable and permit us to determine whether molecular data from maternally and biparentally inherited genomes generate concordant or discordant conclusions about cheilanthoid phylogeny.
Subset analyses.-The rbcL analysis of this subset of species yielded a single most-parsimonious tree ( Fig. 2A) in the first replicate. During the next 999 random addition sequence replicates, no new minimal trees were found, suggesting that local optima insularity is not a problem in this rbcL subset. This tree is 412 steps long (CI = 0.631 and RI = 0.690), and its topology is precisely the same as in the rbcL tree in Fig. 1 except that fewer species are included in Fig.  2A. Of the 1321 rbcL characters used, 1093 were constant, 85 variable characters were parsimony-uninformative, and 143 were parsimony-informative.
The ITS analysis of this same subset of species also yielded a single most-parsimonious tree (Fig. 2B) in the first replicate. No new minimal trees were found during the next 999 random addition sequence replicates, suggesting that local optima insularity is also not a problem in this data set. This tree is 961 steps long (CI = 0.633 and RI = 0.553). Of the 771 ITS characters used, 355 were constant, 171 variable characters were parsimony-uninformative, and 245 were parsimony-informative.
Direct rbcL vs.  shows that both the rbcL and ITS data sets find the same terminal clades and in some cases the same clades at one or two levels deeper. Both data sets group the top four species of Pellaea section Pellaea (P. andromedifolia to P. breweri) in exactly the same way. Both data sets group Pellaea rotundifolia and P. calidirupium of section Pia tyloma. Both identically group the remaining three species of Pellaea section Pellaea (P. cordifolia to P. notabilis) , the two species of CheilantheslAstrolepis, the three species of NotholaenalArgyrochosma, and the five species of southwestern Cheilanthes allosuroides to C. chip inquensis; and both group southwestern Cheilanthes into identical sub clades of three species and two species, respectively. In the outgroup, both data sets group the two Notholaena species into a single clade that is sister to the remaining two outgroup species. Thus there is remarkable concordance between results with the chloroplast-based and the nucleus-based data sets.
The chloroplast and nuclear data sets significantly do not agree, however, with regard to the deeper relationships of the clades they found in common: how the two elements of Pellaea section Pellaea, and Pellaea section Platyloma, and CheilantheslAstrolepis, and NotholaenalArgyrochosma, and southwestern Cheilanthes relate to one another. Since both data sets find the same terminal clades, the terminal clades in common can be collapsed to single summary lines in Fig. 3 in order to facilitate comparisons at the deeper levels of the rbcL and ITS trees. For example, in Fig.  3 the top line of each tree summarizes the identical clade of four species (Pellaea andromedifolia, P. rufa, P. glabella var. missouriensis, and P. breweri) found by both data sets. By comparing the relative positions of these collapsed summary clades, one easily sees how the deeper level placements of these identical clades differ in the rbcL tree versus the ITS tree. Bootstrap support is indicated below the lines in Fig. 3. The inferred phylogenetic relationships among the summary clades differ sharply in the two trees. For example, in the rbcL tree of Fig. 3A the clade of Pellaea section Platyloma (P. rotundifolia to P. calidirupium) is sister to Pellaea section Pellaea 's clade of Pellaea andromedifolia to breweri with a 94% bootstrap confidence value. In the ITS tree of Fig. 3B, however, Pellaea section Platyloma is sister to Pellaea section Pellaea's clade of Pellaea cordifolia to notabilis with a mere 45% bootstrap value. In the rbcL tree the NotholaenalArgyrochosma clade is sister to the Pellaea-to-CheilantheslAstrolepis clade with a 99% bootstrap value, whereas in the ITS tree, Notholaenal Argyrochosma is sister to only CheilantheslAstrolepis with a lesser 76% bootstrap value. Based on rbcL, the Cheilanthes allosuroides to C. chipinquensis clade is sister to everything else in the ingroup with a 100% bootstrap value ( Fig. 2A) whereas based on ITS this Cheilanthes clade is sister to two Pellaea clades with a mere 22% bootstrap value (Fig. 2B, 3). Both data sets indicate that southwestern Pellaea section Pellaea is polyphyletic.
The rbcL gene is well known to be relatively slowly evolving and well suited to studies at generic and familial levels of divergence in ferns (Hasebe et al. 1995 ;Gastony and Ungerer 1997) and at even deeper levels in angiosperms (Chase et al. 1993). Sequence divergence values for rbcL for the subset of taxa used in Fig. 2A ranged from 0.5% to 5.5% for the ingroup taxa and from 0.5% to 7.1 % for the entire subset of ingroup plus outgroup taxa (Table 4).
ITS, on the other hand, evolves relatively rapidly and in ferns is seen here to be particularly useful within genera or subdivisions of genera, although it is also useful at deeper levelsin angiosperms (Baldwin 1992). Sequence divergence values for the two ITS regions for the subset of taxa used in Fig. 2B ranged from 1.1 % to 20.1 % for the ingroup taxa and from 1.1 % to 23.0% for the entire subset of ingroup plus outgroup taxa (Table 4). Combined ITS 1 and ITS2 sequence divergences for Compositae subtribe Madiinae (Baldwin 1992) ranged from 0.5% to 15.0% for the ingroup taxa and from 0.5 % to 20.5 % for the entire set of ingroup plus outgroup taxa.
Conclusions.-As a result of this first analysis of ITS regions of ferns , we now know that many ITS regions appear to be only arbitrarily and unreliably aligned at supra generic levels in these ferns . Thus the inability of ITS to resolve the same deeper clades as rbcL, and the very weak bootstrap support for the ITS data at these deeper levels is now not surprising. ITS simply is not an appropriate molecule for use at these deeper phylogenetic levels in ferns. The concordance between biparentally inherited ITS and maternally inherited rbcL at less deep levels of the phylogeny, how- Table 4 . Sequence divergence values for the subset of taxa in Fig 2A, B. Numbers above diagonal are percentage differences for rbcL. Numbers below diagonal are percentage differences for the a li gned dataset of ITS I plus ITS2. Taxon names are abbreviated from the full names in Table 2 as follows: the abbreviated generic name is followed by the first four letters of the specific epithet (e.g., "e. allo" = "Cheilanthes allosuroides"). ever, generates confidence that rbcL is tracking real species phylogenies, not simply matrilineal gene phylogenies. Maternally inherited rbcL data and biparentally inherited nonmorphological nuclear data do agree at divergence levels where both rbcL and the ITS regions of nuclear ribosomal DNA evolve at appropriate rates. Remarkable congruence is found for terminal and slightly deeper clades based on both data sets. Moreover, rbcL's unexpected finding that southwestern Pellaea section Pellaea is polyphyletic-that it is broken into two deeply separated clades-is also found by nuclear ribosomal DNA. Support for clades at deeper levels of the phylogeny, where the two data sets do not agree, is much stronger for rbcL than for ITS.
One may therefore conclude that 1) at appropriate levels of divergence, matrilineal chloroplast-encoded rbcL is concordant with biparental nucleus-encoded molecular data in cheilanthoids, 2) unexpected rbcL findings are attributable to shortcomings of traditional morphology-based studies, and 3) rbcL does reliably track species phylogeny, enabling us to begin to unweave the tangled web of generic lineages and phylogenetic relationships in cheilanthoid ferns. To resolve the deeper levels of cheilanthoid phylogeny with nuclear molecular data, future attention will focus on the I8S region of mDNA already applied by Wolf (1995) to ferns of the Dennstaedtiaceae and on other potentially informative regions.