Printer Friendly

Phylogenetics and character evolution in the grass family (Poaceae): simultaneous analysis of morphological and chloroplast DNA restriction site character sets.

I. Abstract

A phylogenetic analysis of the grass family (Poaceae) was conducted using two character sets, one representing variation in 364 mapped and cladistically informative restriction sites from all regions of the chloroplast genome, the other representing variation in 42 informative "structural characters." The structural character set includes morphological, anatomical, chromosomal, and biochemical features, plus structural features of the chloroplast genome. The taxon sample comprises 75 exemplar taxa, including 72 representatives of Poaceae and one representative of each of three related families (Flagellariaceae, Restionaceae, and Joinvilleaceae); Flagellaria served as the outgroup for the purpose of cladogram rooting. Among the grasses, 24 tribes and all 16 subfamilies of grasses recognized by various modem authors were sampled. Transformations of structural characters are mapped onto the phylogenetic hypotheses generated by the analysis, and interpreted with respect to biogeography and the evolution of wind pollination in the grass family. A major goal of the study was to test the monophyly of several putatively natural groups, including Bambusoideae, Pooideae, Arundinoideae, and the "PACC clade" (the latter comprising subfamilies Panicoideae, Arundinoideae, Chloridoideae, and Centothecoideae), as well as to analyze the phylogenetic structure within these groups and others. Several genera of controversial placement (Amphipogon, Anisopogon, Anomochloa, Brachyelytrum, Diarrhena, Eremitis, Ehrharta, Lithachne, Lygeum, Nardus, Olyra, Pharus, and Streptochaeta) also were included, with the goal of determining their phylogenetic affinities. The two character sets were analyzed separately, and a simultaneous analysis of the combined matrices also was conducted. The combined data set also was analyzed using homoplasy-implied weights. Among major results of the combined unweighted analysis were resolution of a sister-group relationship between Joinvillea and Poaceae; resolution of a clade comprising Anomochloa and Streptochaeta as the sister of all other grasses, with Pharus the next group to diverge from the lineage that includes all remaining grasses; and resolution of other taxa often assigned to Bambusoideae s.l. (including Ehrharta and Oryzeae, and excluding a few other taxa as noted) as a paraphyletic assemblage, within which is nested a clade that consists of Brachyelytrum, the PACC clade (including Amphipogon), and Pooideae (including Brachypodieae, Stipeae, Anisopogon, Diarrhena, Lygeum, and Nardus). Within the PACC clade, Aristida is identified as the sister of all other elements of the group; Chloridoideae, Centothecoideae, and Panicoideae are each resolved as monophyletic, the latter two being sister-groups; and the remaining Arundinoid elements constitute a paraphyletic group within which are nested these three subfamilies. Within the Pooideae, four "core tribes" (Bromeae, Hordeeae [i.e., Triticeae], Agrostideae [i.e., Aveneae], and Poeae, the latter including Sesleria) are resolved as a monophyletic group that is nested among the remaining elements of the subfamily (Brachypodieae, Meliceae, Stipeae, Anisopogon, Diarrhena, Lygeum, and Nardus). A second principal goal of the analysis was to identify structural synapomorphies of clades. Among the synapomorphies identified for some of the major clades are the following: gain of a 6.4 kb inversion in the chloroplast genome in Joinvillea and the grasses; reduction to 1 ovule per pistil, gain of a lateral "grass-type" embryo, and gain of an inversion around the gene trnT in the chloroplast genome in the grasses; loss of arm cells in the clade that consists of Brachyelytrum, Pooideae, and the PACC clade; loss of the epiblast and gain of an elongate mesocotyl internode in the PACC clade; gain of proximal female-sterile florets in female-fertile spikelets, gain of overlapping embryonic leaf margins, and gain of Panicum-type endosperm starch grains in the clade that comprises Centothecoideae and Panicoideae; and loss of the scutellar tail of the embryo in Pooideae (in one of two alternative placements of Pooideae among other groups). These findings are consistent with an origin and early diversification of grasses as forest understory herbs, followed by one or more radiations into open habitats, concomitant with multiple origins of C4 photosynthesis and specialization for wind pollination.

II. Introduction

In recent years agrostologists have generated several explicit phylogenetic hypotheses concerning major diversification patterns in the grass family (Poaceae). In cases where cladistic analysis was used, these analyses have been based on morphological characters (Baum, 1987; Kellogg & Campbell, 1987); on chloroplast DNA (cpDNA) restriction sites (Davis & Soreng, 1993; Yaneshita et al., 1993); on nucleotide sequences of nuclear ribosomal RNA (rRNA; Hamby & Zimmer, 1988, 1992), genes of the chloroplast genome (Doebley et al., 1990; Cummings et al., 1994; Nadot et al., 1994; Barker et al., 1995; Clark et al., 1995; Duvall & Morton, 1996; Hongping Liang & Hilu, 1996), and the nuclear gene Alcohol-dehydrogenase (Morton et al., 1996); and on various combinations of these character sets (Kellogg & Linder, 1995). Most of these analyses have yielded results consistent with the recognition of two major clades in the family, one of them corresponding to subfamily Pooideae, the second comprising subfamilies Panicoideae, Arundinoideae, Chloridoideae, and Centothecoideae, collectively referred to as the "PACC clade" (Davis & Soreng, 1993). However, there have been substantial differences in the circumscription of Bambusoideae and, within the PACC clade, of various taxa recognized as Arundinoids.

Any discussion of conflicting hypotheses of phylogenetic structure in Poaceae must grapple with the varied hypotheses implicit in a long series of noncladistic classifications and hypotheses of relationships published since the late 1950s (Stebbins, 1956; Reeder, 1957, 1962; Tateoka, 1957; Prat, 1960; Stebbins & Crampton, 1961; Butzin, 1965; Clifford, 1965; Clifford et al., 1969; Sharma, 1979; Caro, 1982; Hilu & Wright, 1982; Watson et al., 1985; Clayton & Renvoize, 1986; Soderstrom & Ellis, 1987; Tzvelev, 1977, 1989; Watson & Dallwitz, 1992), some of which circumscribed groupings similar to the Pooideae and the PACC clade, as later detected by cladistic analyses. Although there are many areas of agreement among these treatments, classification of a subset of taxa in one or the other of these two major clades, aside from the disposition of taxa assigned to Bambusoideae, remains unsettled. Among genera and generic assemblages that have been particularly difficult to place are Ampelodesmos Link, Anisopogon R. Br., Danthoniastrum Holub, Duthiea Hack., Lygeum L., Metcalfia Conert, Nardus L., Pseudodanthonia Bor & C. E. Hubb., Stephanachne Keng, and Stipeae. These so-called orphan genera and tribes usually are placed either in Pooideae or Arundinoideae (compare the classification of Clayton & Renvoize, 1986, with that of Watson & Dallwitz, 1992) or their informal equivalents (Tzvelev, 1977, 1989).

Beyond the problems posed by these orphan taxa, a historical emphasis by systematists on temperate and widespread tropical taxa has resulted in inadequate representation in exemplar-based studies of a third assemblage of grasses, subfam. Bambusoideae s.l. The general neglect of Bambusoideae in higher-level studies is particularly troublesome, for it is a diverse group that has been divided into as many as 13 tribes (Clayton & Renvoize, 1986) or 8 subfamilies (Caro, 1982). Meanwhile, evidence has mounted that an understanding of the earliest diversification of grasses requires broad, unconstrained representation from the Bambusoideae (Davis & Soreng, 1993; Clark et al., 1995).

While the Pooideae and PACC clades have been recognized as monophyletic groups in most cladistic studies, cladistic results concerning Bambusoideae are limited and frequently contradictory. Bambusoideae may be monophyletic (Kellogg & Campbell, 1987; Kellogg & Watson, 1993) or may be paraphyletic or polyphyletic in a number of different ways (Davis & Soreng, 1993; Barker et al., 1995; Clark et al., 1995; Kellogg & Linder, 1995; Duvall & Morton, 1996; Hongping Liang & Hilu, 1996). However, most exemplar-based cladistic analyses conducted to date have included such limited sampling among supposed elements of Bambusoideae and other putatively early-diverging taxa that it is difficult and perhaps premature to make more than general comparisons among them.

In many of the noncladistic classifications and hypotheses cited above, Bambusoideae are recognized as a natural group. Even in some cladistic analyses, Bambusoideae have been constrained as monophyletic (Kellogg & Campbell, 1987) or treated as a natural group (Kellogg & Watson, 1993). Other noncladistic hypotheses, dating back to Warming (1895), Bessey (1917), Bews (1929), Roshevits (1937), and Tateoka (1957), have suggested that Bambusoideae represent an archaic group from which are derived groups now conventionally placed in Pooideae or in the various subfamilies of the PACC clade.

Discussions of Bambusoideae are complicated by the fact that circumscriptions of the group vary widely. For example, among the genera and tribes omitted from this subfamily by one or more modem authors are Anomochloa, Brachyelytrum P. Beauv., Centotheceae, Diarrheneae, Ehrharteae, Oryzeae, Parianeae, Phaenosperrna Benth., Phareae, Phyllorachideae C. E. Hubb., and Streptochaeta Schrad. When excluded, these orphan taxa are included in other subfamilies (usually Arundinoideae, Chloridoideae, or Pooideae) or are segregated to small subfamilies of their own (e.g., Anomochlooideae Potztal, Streptochaetoideae [C. E. Hubb.] Butzin, Pharoideae L. G. Clark & Judz., Parianoideae [C. E. Hubb.] Butzin, Oryzoideae [Ehrhartoideae Link if Ehrharta is included], and Centothecoideae).

Another complicating factor encountered by grass systematists is that most existing classifications of the grasses are intuitive and idiosyncratic to one degree or another. Because the grass family is so large, groups, once recognized, often are considered to be coherent without objective analyses of data to demonstrate their naturalness. Among numeric methods, phenetic analyses of grass data sets without constrained groups have provided phenograms of the family (Hilu & Wright, 1982; Watson et al., 1985; and others reviewed in Baum, 1987). However, in their dependence on similarity measures, and absence of explicit rooting, phenetic results are logically compromised for inferring phylogeny (Farris, 1980; Watrous & Wheeler, 1981; Wiley, 1981).

One way to achieve broad sampling within large and diverse groups in analyses, cladistic or otherwise, has been to use conglomerate taxa, such as genera, tribes, or subfamilies, as the terminals of analysis (e.g., Hilu & Wright, 1982; Watson et al., 1985; Baum, 1987, and others reviewed therein; Kellogg & Campbell, 1987; Kellogg & Watson, 1993). An advantage in this approach is that all putative groups of interest can be considered concurrently. However, this approach involves theoretical and methodological difficulties (e.g., assumption of monophyly of groups, creation of nonexistent character-state combinations, and adoption of various ad hoc approaches to the treatment of polymorphism within the artificial terminals that are created; Nixon & Davis, 1991; Nixon, 1996). Apart from the methodological compromises inherent in such approaches, it should also be recognized that the monophyly of a group is tested when its constituent elements are treated as separate terminals, and that, in contrast, the practice of "terminal fusion," which results in conglomerate groups that are used as individual taxa, or "composite terminals," assumes rather than tests their unity (Nixon & Davis, 1991; Nixon, 1996; Nixon & Carpenter, 1996). For these reasons we recognize exemplar sampling as a superior alternative to the use of conglomerate terminals. Exemplar sampling uses individual specimens, or individual species, as terminal taxa that serve as representatives of putative higher-level groups.

In most molecular phylogenetic studies, all data for each terminal are drawn from a single DNA isolation and therefore from a single species, if not a single organism; studies of this sort epitomize exemplar sampling. Structural character sets (e.g., morphology, anatomy, palynology) also could employ exemplar sampling, but most published studies draw upon a literature assembled over many years by numerous investigators, and observations of all characters included in a study usually are not available for any but the most economically important species (e.g., Oryza sativa). A major practical difficulty with exemplar-based phylogenetic analyses of large groups, apart from computation, is sampling large subgroups in a depth sufficient to demonstrate their monophyly. At the same time, it is essential that orphan taxa be sampled, as these, like fossils, are often pivotal in phylogenetic reconstruction (Farris, 1976; Rieppel, 1993; Nixon, 1996).

Here we present a study of the phylogenetic structure of the grass family, using exemplar sampling to test the monophyly of the PACC clade, Pooideae, Bambusoideae s.l., and many lesser groupings. We have conducted simultaneous and separate parsimony analyses of two data sets, one comprising a miscellaneous set of "structural characters," representing variation in morphology, anatomy, biochemistry, and structural sequences of the chloroplast genome, the second reflecting cpDNA restriction site variation. Simultaneous analysis, an approach sometimes called "total evidence" or "combined analysis" has been advocated by, among others, Eernisse and Kluge (1993), Kluge and Wolf (1993), Wheeler et al. (1993), Kim and Jansen (1994), Smith and Littlewood (1994), Williams et al. (1994), Chase et al. (1995), and Nixon and Carpenter (1996).

The taxon sample includes representative species of 24 tribes and all 16 subfamilies of grasses recognized by various modem authors, and includes 10 of 13 tribes of Bambusoideae as delimited by Clayton and Renvoize (1986). Three nongrass outgroups also are included (Nixon & Carpenter, 1994); the outgroups represent Flagellariaceae, Restionaceae, and Joinvilleaceae, all of which were included in Poales by Dahlgren et al. (1985). Also included are exemplars from orphan genera and tribes that are postulated to be either among the earliest diverging elements within the family or situated in the critical nexus between the Arundinoideae, Pooideae, and Bambusoideae (Aristida L., Ampelodesmos R. Br., Amphipogon, Anisopogon, Anomochloa, Brachyelytrum, Chasmanthium Link, Diarrhena P. Beauv., Eremitis Doll, Ehrharta Thunb., Lygeum, Nardus, Pharus P. Browne, Stipeae, and Streptochaeta).

We find that both character sets, analyzed separately or simultaneously, support the monophyly of a major clade that consists of the Pooideae, the PACC clade (as sister taxa), and a series of less inclusive taxa that are often placed within the Bambusoideae but here are placed within one or the other of these two major groups; the larger clade that includes Pooideae and the PACC clade is itself nested among tribes that are often assigned to Bambusoideae. Although the phylogenetic relationships supported by the present study are largely consistent with the classification of Clayton and Renvoize (1986), significant differences exist, and we note these and suggest appropriate alterations to the classification. Among the taxa placed within the Pooideae are Anisopogon and Diarrhena. The PACC clade includes all sampled elements of Panicoideae, Arundinoideae (except Anisopogon), Chloridoideae, and Centhothecoideae, but subfam. Arundinoideae is not resolved as monophyletic, and Aristideae is resolved as sister of the rest of the PACC clade. Thus, the various elements conventionally placed in Arundinoideae need to be reclassified, and we suggest that the subfamily be abandoned altogether. Results of the present analysis suggest that Bambusoideae, by any but the most restrictive circumscription, are a non-monophyletic assemblage. Monophyly of the Bambusoideae requires, at the minimum, the exclusion of seven tribes. Anomochloeae and Streptochaeteae should be placed in Anomochlooideae; Phareae in Pharoideae; Oryzeae with Ehrharteae in Ehrhartoideae (i.e., Oryzoideae); Diarrheneae in Pooideae; and Brachyelytreae either in Pooideae or in a small subfamily closely related to Pooideae and the PACC clade. The precise placement of Brachyelytreae remains tentative, for there is conflicting support for its placement as sister of Pooideae or sister of PACC plus Pooideae.

The discussion of suprageneric taxonomy in the present paper is affected by several nomenclatural innovations and name changes, due to priority of date (verified by J. Reveal, pers. comm.), which have been established or uncovered since the publication of Genera Graminum (Clayton & Renvoize, 1986). Subfamily Pharoideae was newly published (Clark & Judziewicz, 1996). Ehrhartoideae Link has priority over Oryzoideae Burmeist. or Bambusoideae Asch. & Graeb. when Ehrharta is included in the same subfamily as Oryza or Bambusa, respectively. Bambusoideae also is predated by Oryzoideae, but in the present work we apply the former name to a narrowly circumscribed group that includes neither Oryza nor Ehrharta; we use the term "Bambusoideae s.l." to refer to a variety of inclusive (and demonstrably paraphyletic) circumscriptions of this subfamily. Agrostideae Bertch. & J. Presl has priority over Aveneae Dumort. when Agrostis is included in the latter tribe; Chlorideae Bertch. & J. Presl has priority over Cynodonteae Dumort. when Chloris Sw. is included in that tribe; Hordeeae Bertch. & J. Presl has priority over Triticeae Dumort. when Hordeum L. is included in that tribe; and Sacchareae Bertch. & J. Presl has priority over Andropogoneae Dumort. when Saccharum L. is included in that tribe. Table I and most figures use the Clayton and Renvoize classification, unchanged, to facilitate reference to other contemporary literature, but the above changes in tribal names are noted in Table I. In the text, names with nomenclatural priority are used (except Bambusoideae, as noted above), or are cross-referenced if the text specifically refers to a group name employed by other authors.

III. Methods

A. TAXON SAMPLE

The exemplar species sampled for the present analysis are listed in Table I, in which the grasses are arranged according to the classification by Clayton and Renvoize (1986). Voucher specimens are housed at BH unless otherwise noted in the table. To assist the reader in evaluating the selection of exemplar taxa in the study, the number of species, genera, and tribes (if any), are reported for subfamilies and tribes (Table I). We sampled 72 grass species representing 24 tribes and including all 16 subfamilies proposed in various modem treatments of the family. Because reticulate origins might distort the phylogenetic history between structural characters susceptible to recombination and the linearly inherited chloroplast genome, we excluded genera from the present analysis if there was reasonable evidence from our larger analyses of their origin through intertribal hybridization (Soreng et al., 1990; Davis & Soreng, 1993, and unpubl. data).

Three genera, from three different families closely related to the grasses (Dahlgren et al., 1985; Linder & Ferguson, 1985; Campbell & Kellogg, 1987; Linder, 1987; Doyle et al., 1992; Duvall et al., 1993; Linder & Rudall, 1993; Chase et al., 1995; Davis, 1995; Kellogg & Linder, 1995; Linder & Kellogg, 1995; Stevenson & Loconte, 1995), were used as outgroups: Flagellaria L. (Flagellariaceae), Baloskion Raf. (Restionaceae), and Joinvillea Brongn. and Gris (Joinvilleaceae). The sampled representative of Restionaceae, an Australian species widely known as Restio tetraphyllus Labill., is recognized as belonging to a separate genus, Baloskion (B. Briggs, pers. comm.), but as the formal combination has not been made in that genus we refer to it either as Baloskion or as Restio tetraphyllus.

B. CHLOROPLAST DNA RESTRICTION SITE CHARACTERS

Methods for DNA extraction and restriction site analysis are the same as in earlier studies (Soreng et al., 1990; Davis & Soreng, 1993; Choo et al., 1994). Briefly, total DNA was extracted, then digested with 10 restriction enzymes: Bam HI, Bcl II, Bgl II, Cla I, Hind III, Kpn I, Pst I, Pvu II, Sal I, and Sma I, using methods described by the supplier (Gibco/Bethesda Research Lab [G/BRL]). After electrophoresis in agarose gels (0.6-1%), the DNA was transferred to nylon membranes (Zetaprobe-GT+; G/BRL). The membranes first were probed with a set of large probes (10-21 kb) cloned from the grasses Permisetum americanum (L.) Leeke (Thomas et al., 1984) and Triticum aestivum L. (Bowman et al., 1981). One additional large clone also was used, from Phaseolus vulgaris L., (Palmer & Thompson, 1981) to cover the small single copy (SSC) region. Up to 34 smaller probes (1-4 kb) from Nicotiana tabacum L. (Sugiura et al., 1986) were used as needed to assess the homology and map the restriction sites that fell within and between the larger probe regions. This overall combination of probes covers more than 99% of the chloroplast genome. Mapping was facilitated by reference to the 271 restriction sites for the above enzymes in the Oryza sativa cpDNA sequence (Hiratsuka et al., 1989; Sugiura, 1989). The distribution of these sites by region within the chloroplast genome of Oryza sativa is as follows: 172 in the large single copy (LSC) region, 62 in the inverted repeat (IR)[TABULAR DATA FOR TABLE I OMITTED] region, and 37 in the small single copy (SSC) region. Restriction site maps were encoded as a data matrix in which each of 616 differentiable sites was scored as a binary character, with each site specified by restriction enzyme and map location, the latter keyed to the enumeration of nucleotides in the cpDNA sequence of Oryza sativa; cladistically informative characters of the matrix are published here (Appendix 1).

C. STRUCTURAL CHARACTERS

Forty-two characters representing anatomy and gross morphology, physiology, and chloroplast genome structural mutations also were scored. These characters constituted the "structural" character set. The principal reference for most structural characters is Watson and Dallwitz's (1992) Grass Genera of the World. With the goal of minimizing the number of polymorphisms, questionable homologies, and missing data, we consulted the primary literature and specimens of the exemplar species. Primary literature sources included surveys of particular structural attributes of grasses (e.g., lodicules, embryos) and taxonomic monographs; secondary sources include floras and prior reviews of structural variation and taxonomy of the grasses. Personal observations were made of living material, DNA vouchers, and other specimens. The structural characters are described in Appendix 2, which includes comments on their scoring when the scores used in the analysis are absent in Watson and Dallwitz (1992) or differ from those provided there. The structural data are presented in Appendix 3.

The following rules were adopted for the reconciliation of polymorphisms, ambiguity, and contradictory statements in the literature:

1. If a species consistently has a structure in fully developed organs, but not in vestigial ones, it is scored as having that structure.

2. If a monotypic genus is reported to have one character state by Watson and Dallwitz (1992), and another state in a primary source, the taxon is scored as polymorphic unless ancillary evidence suggests that one or the other source is in error.

3. If all reports for a polytypic genus specify the same state, but there is no report for the exemplar species in the analysis, the exemplar species is scored as having the state reported for the genus.

4. If a polytypic genus is reported to have one state by Watson and Dallwitz, and this is contradicted by a primary source, the primary source is followed if it includes an observation of the exemplar species; if the primary source does not mention the exemplar species, the taxon is scored as polymorphic.

5. If a polytypic genus is reported to be polymorphic by Watson and Dallwitz, and if a primary or secondary source reports an observation of the exemplar species as having a single state, or (for multistate characters) a subset of the states reported for the genus by Watson and Dallwitz, the species is scored as in the primary or secondary source. Thus, some polymorphisms in the data set represent generic polymorphisms where the state has not been observed for the exemplar species, and others represent observed infraspecific variation.

6. If the exemplar species has a character state that is intermediate between two recognized states, it is scored as ambiguous.

7. An additional situation pertains in multistate characters where one or more states of a multistate character are determined to be absent, but the observed attribute is not assigned to one of the remaining states. In this situation, termed "subset ambiguity," the taxon is scored as polymorphic for all states that have not been determined to be absent.

8. A situation that combines aspects of subset polymorphism and subset ambiguity also is recognized. If some individuals of an exemplar species have one state, and other individuals have a state that is intermediate between that and a second state, the taxon is scored as "polymorphic" for the two states. Similarly, polymorphism coding might be used if the state of an exemplar species is unknown, and one or more representatives of a genus have one state, while one or more others are intermediate between that state and another.

The 42 structural characters fall into several groups: morphology of culm and leaf (6 characters), morphology of spikelet (7), morphology of flower (9), morphology of fruit and embryo (9), morphology of seedling (1), anatomy of vegetative structures (5), biochemistry (2), and structure of the chloroplast genome (3). The character selection reflects consideration of ease of scoring, minimal ambiguity in the determination of states, and historical use in grass classification. All multistate structural characters were treated as nonadditive.

D. DATA ANALYSIS

Cladistically uninformative characters were removed from the data matrices prior to analysis, with the mop command of Dada vet. 1.1.4 (Nixon, 1997). However, when there are multistate characters in a matrix (as in the structural character set of the present analysis) removal of uninformative characters may leave autapomorphies. An informative multistate character can include one or more uninformative states, as it would, for example, if states 0 and 1 each occurred in several terminals and state 2 occurred in one terminal. In this case, state 2 would be an uninformative state (an autapomorphy) of an informative character if the character were treated as nonadditive (note, however, that if the character were additive, state 2 would be informative, though present in only one taxon). In the structural character set of the present analysis, with all characters treated as nonadditive, any state that occurs in only one terminal is autapomorphic, and as such it contributes one step to any possible tree and does not influence tree structure. Recoding of such a state as missing removes the autapomorphic step from all trees without altering the results of cladistic analysis. Note, however, that results obtained from analytical methods such as neighbor joining and maximum likelihood estimation are influenced by cladistically uninformative data, including uninformative states of informative characters.

The present structural matrix includes two informative characters with uninformative states (char. 23, state 2 unique to Baloskion; char. 38, state 2 unique to Aristida; Appendix 3). These data are included in the matrix published here, but all analyses (except character optimization on trees for the purpose of determining synapomorphies of clades and autapomorphies of terminals) were conducted with these two states recoded as missing. Thus, tree lengths obtained with the structural character matrix (Appendix 3) are two steps longer than those reported below, as are those obtained by simultaneous analysis of both matrices. Consistency indices (CI; Kluge & Farris, 1969), reported below, also were calculated using matrices with uninformative characters and uninformative states of informative multistate characters removed.

Cladistic analysis, with characters equally weighted, was conducted on the two character sets (restriction sites and structural characters) separately, and a simultaneous analysis was conducted of the combined matrix of all characters. The separate analysis of the structural data set did not resolve Poaceae as monophyletic (see results), and a second analysis of this character set was conducted with monophyly of Poaceae constrained, to explore implications of this data set under that condition. All four cladistic analyses were conducted with Nona ver. 1.16 (Goloboff, 1993b), using the default settings amb- (clades resolved only if they have unambiguous support) and poly = (polytomies allowed). In each analysis 3500 subsearches were conducted, each consisting of tree-construction using a random taxon entry sequence, followed by tree bisection-reconnection (tbr) swapping with up to 20 most-parsimonious trees retained (hold/20 mult*3500); shortest trees retained from the subsearches then were tbr-swapped to completion ([max.sup.*]), except for the two analyses of the structural character set. Neither of the analyses based on just the structural character set (unconstrained and with monophyly of grasses constrained) identified a small number of most-parsimonious trees, and in each case swapping was aborted after 10,000 trees had been accumulated and tbr-swapping had been conducted on all of those trees.

One additional cladistic analysis was conducted, based on the combined matrix of both character sets and using implied weighting (Goloboff, 1993a), as implemented in Piwe ver. 2.15 (Goloboff, 1993c), using the default weighting function (concavity = 3), and the same ambiguity and polytomy settings, plus the same search procedures as used in the unweighted analyses conducted with Nona (Goloboff, 1993b).

Bootstrap support (Felsenstein, 1985) was assessed with a "strict-consensus" bootstrap analysis. In this form of the bootstrap a clade is recognized as having been resolved by a particular bootstrap replicate only if it occurs in all most-parsimonious trees (i.e., if it occurs in the strict-consensus tree) obtained in that replicate. Thus, every potential clade receives a score of either 0 or 1 for each bootstrap replicate. In contrast, in the "frequency-within-replicates" bootstrap, as implemented in most software packages that conduct bootstrap analysis, each clade is assigned a score in each replicate in proportion to its frequency of occurrence among most-parsimonious trees. If a group occurs in all most-parsimonious trees in a particular replicate, it receives a score of 1.0 under both forms of the bootstrap; but if it occurs in some but not all trees, it receives a score of 0 in the strict-consensus bootstrap and a score [greater than]0 in the frequency-within-replicates bootstrap. Thus, if equally effective tree searches are conducted, observed strict-consensus bootstrap scores must be less than or equal to frequency-within-replicates bootstrap scores, and in empirical studies with various data sets the strict-consensus bootstrap frequency for a clade that occurs in the strict-consensus tree for a given data set is usually 0-10% lower than the frequency-within-replicates bootstrap frequency for the same clade (pers. ohs.). In the present analysis two strict-consensus bootstrap analyses were conducted, one on the restriction site data set, the other on the combined data set, using Clados vet. 1.4.98 (Nixon, 1993) running Nona (Goloboff, 1993b) as a daughter process for tree searches. In each analysis 1000 bootstrap replicates were conducted with the same ambiguity and polytomy settings as in the basic analyses, and with each replicate comprising 20 random taxon entry sequences with tbr-swapping to up to four trees, followed by tbr-swapping of all shortest trees obtained from the 20 subanalyses, with a total of up to 80 shortest trees saved (hold/4 hold80 mult*20 [max.sup.*]).

Minimum character removal (MCR) scores were determined by successive character removal analysis (Davis, 1993; Davis et al., 1993) conducted on the combined matrix. With 406 informative characters in the combined matrix it was not feasible to conduct more than one round of analysis (i.e., 406 analyses, each with a different one of the 406 characters removed from the matrix). The analyses were conducted with Nona (Goloboff, 1993b), using the same ambiguity and polytomy settings as in the basic analyses, and with analysis comprising 100 random taxon entry sequences with tbr-swapping to up to five trees, followed by tbr-swapping of all shortest trees obtained from the 100 subanalyses, with a total of up to 500 shortest trees saved (hold/5 hold500 mult*100 [max.sup.*]).

Unambiguous structural character apomorphies of terminals and clades were determined by optimizing the structural data matrix on most-parsimonious trees obtained by analysis of the combined data sets using the apo- command of Nona (Goloboff, 1993b). Character changes of ambiguous placement and other patterns of character variation were examined with Clados (Nixon, 1993).

The Mickevich/Farris incongruence index ([I.sub.MP]; Mickevich & Farris, 1981; Kluge, 1989) and lengths of various data subsets as optimized on trees obtained from the combined data or from the other data set (related to the Miyamoto incongruence index; Kluge, 1989) were calculated from tree lengths of the various character sets on trees obtained in the various analyses and from minimum possible tree lengths for data sets obtained using the fit command of Nona (Goloboff, 1993b). All analyses yielded more than one most-parsimonious tree, and for all calculations based on the number of extra steps required to map the various data sets on the various sets of trees, the minimum required length was used (Swofford, 1991).

IV. Results

A. COLINEARITY OF THE CHLOROPLAST GENOMES

Restriction site mapping indicates that all of the grass cpDNAs sampled here are colinear with respect to the 27 kb and 6.4 kb inversions (relative to the chloroplast genome of Nicotiana L.) previously documented among grasses (Quigley & Weil, 1985; Howe et al., 1988; Hiratsuka et al., 1989; Downie & Palmer, 1992; Doyle et al., 1992) and in Joinvillea (Doyle et al., 1992). Specifically, these mapping results are consistent with previously reported results obtained by polymerase chain reaction (PCR) assays using diagnostic primer combinations (Doyle et al., 1992; see char. 40 of the present analysis, Appendix 2). New PCR data demonstrate that Baloskion lacks the 6.4 kb inversion, like the two genera of Restionaceae previously tested (Doyle et al., 1992), whereas the restriction site data suggest that Baloskion, again like the other two genera, has the 27 kb inversion. Because Flagellaria is the only taxon in the present analysis that lacks the 27 kb inversion, that character is cladistically uninformative in this set of taxa and is not included in the structural character set. New PCR data for six of the exemplar grasses in this study (Anisopogon, Anomochloa, Brachyelytrum, Lygeum, Molinia Schrank, and Poa L.; see Table I for species names and voucher information) have the small trnT inversion that has been detected in all grasses sampled to date, and that is absent in all nongrasses (Doyle et al., 1992).

B. CHARACTERISTICS OF THE DATA MATRIX

The restriction site and structural character data in the present analysis represent 89.7% and 10.3% of the total data set, respectively. Of a total of 616 restriction sites mapped and scored as cladistic characters, 364 are cladistically informative (Appendix 1). In the restriction site data matrix, 11% of the cells (9% in Poaceae) are scored missing (i.e., unobserved or of ambiguous homology). In the structural character matrix of 42 characters scored for 75 taxa, 14% of all cells (12% in Poaceae) are scored missing (i.e., unobserved, of ambiguous homology, or inapplicable) and another 4% are scored as polymorphisms or subset ambiguities (Appendix 3).

A principal cause of missing data in the restriction site character set is attributable to difficulty in the differentiation of pairs of restriction sites that occur in close proximity to one another (primarily where sites are mapped within [less than]300 bp of each other). This difficulty arose in part because of the availability of precise locations for restriction sites that were detected by searching for the appropriate nucleotide sequences in the published cpDNA sequence of Oryza. When two sites for the same restriction enzyme occurred in close proximity to each other (e.g., at a distance of 300 bp or less from each other), and a site detected by restriction digest in another taxon was mapped near those two locations, it could not be determined in some cases (e.g., when the site was flanked by large restriction fragments) whether that taxon had one or the other or both of the sites observed in Oryza.

A second difficulty occurred in the alignment of restriction sites when slight size variation was apparent between adjacent fragments among exemplars. This difficulty was encountered most frequently in the alignment of maps of the three nongrass accessions with those of grasses, often in the LSC region, and particularly in areas of known inversion endpoints and in areas where length variation was detected among the grasses. Even within the grasses, ease of alignment between taxa tended to be correlated with proximity of phylogenetic relationship, and this resulted in a greater amount of missing data in phylogenetically isolated species and sparsely sampled large groups than occurred in Pooideae, where depth of sampling was greatest.

C. CLADISTIC ANALYSES

The provisional tribe and subfamily designations adopted for this study, and used in Table I and in figures, are those of Clayton and Renvoize (1986). However, the following changes, reflecting the results of the analyses presented here, are adopted in the text from this point onward: First, we recognize a monophyletic Pooideae that includes Diarrhena and Anisopogon, which are placed by Clayton and Renvoize (1986) in Bambusoideae and Arundinoideae, respectively. Second, Clayton and Renvoize place Ampelodesmos and Brachypodium within tribes Poeae and Triticeae (= Hordeeae), respectively, of subfam. Pooideae; although the results presented here favor the placement of these two genera within Pooideae, their phylogenetic positions are not within the tribes in which they are placed by Clayton and Renvoize (1986), and we recognize these genera as constituting separate tribes, Ampelodesmeae and Brachypodieae, respectively. Third, Brachyelytrum is removed from Bambusoideae, but because its placement is ambiguous it is left unassigned to subfamily. Fourth, five additional tribes are removed from subfam. Bambusoideae and placed in small subfamilies that represent early-diverging clades within the Poaceae: Ehrharteae and Oryzeae are recognized as tribes within Ehrhartoideae; Anomochloeae and Streptochaeteae as tribes within Anomochlooideae; and Phareae as a tribe within Pharoideae (Clark & Judziewicz, 1996). Further discussion of "early-diverging" lineages refers to Anomochlooideae, Pharoideae, Bambusoideae (including Bambuseae, Parianeae, and Olyreae), and Ehrhartoideae, four putative lineages that are often placed within a broadly defined Bambusoideae and that diverge earliest from the diverse lineage that includes Brachyelytrum, Pooideae, and the PACC clade. Finally, subfam. Arundinoideae is abandoned, and we refer to this group as the "Arundinoid" assemblage.

In the text that follows, and in Table II, where two or more genera of a tribe, or two or more tribes, are resolved as monophyletic in a consensus tree, we usually refer to them by the name of the tribe or subfamily that they represent (e.g., Olyreae for Lithachne + Olyra; Ehrhartoideae for Oryzeae + Ehrharteae) rather than repeatedly citing sets of genera. For monogeneric tribes we simply cite the genus (e.g., Anomochloa).

1. Chloroplast DNA Restriction Site Data

Cladistic analysis of the restriction site data recovered 10 most-parsimonious trees of length (L) 1277, consistency index (CI) 0.28, and retention index (RI; Farris, 1989) 0.70; 62 clades arc resolved in the strict consensus of these trees [ILLUSTRATION FOR FIGURE 1 OMITTED]. All polytomies in the consensus [TABULAR DATA FOR TABLE II OMITTED] tree represent lack of resolution below the tribal level, except for two multichotomies within the PACC clade. Joinvillea is placed as sister of Poaceae, and a lineage that consists of Anomochloa and Streptochaeta is placed as sister of all other grasses. Within the latter group (i.e., all grasses except Anomochloa and Streptochaeta) there is a pectinate arrangement among a series of lineages that have often been assigned to Bambusoideae, with Pharus, Ehrhartoideae, Eremitis, Bambuseae, and Olyreae diverging sequentially from the lineage that includes the PACC clade as sister to a clade in which Brachyelytrum is sister to Pooideae. One notable aspect of this structure, which is not universal among the relationships supported by the various analyses conducted in this study, is the sister-group relationship between Ehrhartoideae and all other grasses except Anomochloa, Streptochaeta, and Pharus; we refer to this relationship as the hypothesis of an "early-diverging" Ehrhartoideae. Another notable aspect of this structure, also inconstant among the various analyses, is the placement of Brachyelytrum as sister of Pooideae.

The PACC clade does not include Anisopogon, which is placed within the Pooideae. Within the PACC clade there is, in the consensus tree, a basal polytomy of all other individual Arundinoid exemplars (Aristida, Arundo L., Amphipogon, Danthonia DC., Molinia, and Phragmites Adans.) plus two subclades that together include all elements of Panicoideae, Centothecoideae, and Chloridoideae. The first of the subclades includes the three elements of Panicoideae, relationships among which are unresolved, and a monophyletic group consisting of the two elements of Centothecoideae; the second subclade consists of all elements of Chloridoideae. Five alternative sets of relationships underlie the overall lack of higher-level resolution in the PACC clade. The alternative relationships, which occur in various combinations among the 10 most-parsimonious trees are 1) Aristida sister of all other members of the PACC clade vs. the clade consisting of Panicoideae and Centothecoidea, sister to all other members of the PACC clade; 2) Panicoideae and Centothecoideae both monophyletic and sister to each other vs. Panicoideae paraphyletic with Centhothecoideae nested among its three elements; 3) Chloridoideae sister to a monophyletic set of all Arundinoids except Aristida vs. Chloridoideae nested within a paraphyletic set of all Arundinoids vs. Chloridoideae nested within a paraphyletic set of all Arundinoids except Aristida vs. Chloridoideae sister of Amphipogon, a structure that coincides with the placement of Aristida as sister of all other members of the PACC clade and with all remaining Arundinoids sister of the clade that consists of Centothecoideae plus Panicoideae.

Diarrhena and Anisopogon are situated within Pooideae, where each is provisionally recognized as the sole genus of a tribe. All tribes of Pooideae represented by more than one exemplar were resolved as monophyletic, except that Ampelodesmos does net fall within Poeae and Brachypodium does not fall within Hordeeae. Five separate pairwise combinations of tribes also are supported as monophyletic: Lygeum + Nardus; Ampelodesmos + Stipeae; Brachypodieae + Meliceae; Bromeae + Hordeeae; and Poeae + Agrostideae. The results support a pectinate structure within the Pooideae, with the following series of lineages diverging in sequence: Lygeum + Nardus; Anisopogon; Ampelodesmos + Stipeae; Brachypodieae + Meliceae; Diarrhena; and a clade in which Bromeae + Hordeeae is sister of Agrostideae + Poeae.

2. Structural Character Data

Unconstrained analysis of the structural character set was aborted after 10,000 most-parsimonious trees of length 158, CI = 0.32, and RI = 0.77 had been discovered, and each had been subjected to tbr-swapping. The consensus of these 10,000 trees (not illustrated), with 34 resolved clades, does not include a monophyletic Poaceae, for Baloskion is placed within a group that consists of all members of Pooideae and the PACC group, plus Brachyelytrum. In this tree Streptochaeta is the sister of all other grasses plus Baloskion, and within this group the next deepest branch is a multichotomy from which emerge three individual terminals (Anomochloa, Pharus, and Eremitis) and a clade that includes all other grasses plus Baloskion. There is a dichotomy in the latter group, with one of the two sister groups consisting of a monophyletic association of Lithachne, Olyra, and all sampled elements of Bambuseae, and the other consisting of Oryzeae as sister of a clade in which Ehrharta is sister of a group that includes Baloskion and all remaining grasses (i.e., Brachyelytrum, the PACC group, and Pooideae. Phylogenetic structure is highly unresolved within the group that is sister of Ehrharta; 25 individual terminals (including Baloskion, Brachyelytrum, and various elements of Chloridoideae, Arundinoideae, and Pooideae) and seven multitaxon clades emerge from a single multichotomy. The largest of the multitaxon groups consist of 1) a monophyletic Centothecoideae as sister of a monophyletic Panicoideae, 2) all elements of Chloridoideae except Uniola, 3) all elements of Meliceae, and 4) Vulpia as sister of all elements of Agrostideae.

Analysis of the structural data, with the grasses constrained to be monophyletic, also was aborted after 10,000 trees had been identified and each had experienced tbr-swapping. These trees are one step longer than those yielded by the unconstrained analysis (length 159 steps, CI = 0.32, RI = 0.76), and 45 clades are resolved in the strict-consensus tree (Fig. 2). In the consensus tree Streptochaeta is the sister of all other grasses, and within the latter group there is a multichotomy in which relationships are unresolved among three individual terminals (Anomochloa, Pharus, and Eremitis) and a clade that includes all remaining grasses. Within the latter group there is a sister-group relationship between two clades, the first including a monophyletic grouping of Lithachne and Olyra as sister to a monophyletic grouping of all elements of Bambuseae, the second including a monophyletic grouping of Oryza and Leersia as sister to a monophyletic grouping of all remaining grasses. In the latter there is a pectinate structure in which Ehrharta and Brachyelytrum diverge successively from a clade that includes all remaining grasses (Pooideae plus the PACC group), and within which there is a sister-group relationship between a clade that includes all elements of the PACC group and one that includes all elements of Pooideae. Within the PACC clade, Aristida is sister to a clade that includes all other elements of the group, and the remaining Arundinoid elements constitute a paraphyletic assemblage, within which is nested a monophyletic grouping that consists of a monophyletic Centothecoideae as sister of a monophyletic Panicoideae; the sister clade of this group includes Phragmites, Amphipogon, and all elements of Chloridoideae. Within Pooideae there is a basal multichotomy from which emerge 17 individual terminals and four multitaxon clades. These four clades are a grouping of Lygeum and Nardus; a group of four genera of Poeae (Sesleria Scop. as sister of a clade that includes Poa, Puccinellia Parl., and Catabrosa P. Beauv.); a group that includes all elements of Meliceae (with Schizachne sister of a clade that includes the two representatives of Glyceria and the two representatives of Melica); and a group in which Vulpia is sister of a clade that includes all elements of Agrostideae.

3. Simultaneous Analysis of Restriction Site and Structural Characters

Unweighted simultaneous analysis of the combined matrix of 364 restriction site characters and 42 structural characters resolved 12 most-parsimonious trees of length 1468, CI = 0.28, and RI = 0.70. The total number of steps in restriction site characters among these trees ranges from 1277 (the same number of steps as in all 10 trees resolved by the restriction sites alone) to 1284 (CI = 0.28 in all trees, RI = 0.69-0.70), and the number of steps in structural characters ranges from 184 to 191 (CI = 0.27-0.28, RI = 0.69-0.71). Of the 12 most-parsimonious trees for the combined matrix, two (i.e., the only two with 1277 steps in the restriction site data) are among the 10 resolved by restriction sites alone, while all 12 most-parsimonious trees for the combined matrix differ from those that are most-parsimonious for the structural characters alone.

The strict-consensus tree obtained by simultaneous analysis of restriction site and structural data resolves 64 clades (i.e., 2 more than the number resolved by restriction site characters alone, 30 more than resolved by unconstrained analysis of the structural characters, and 19 more than resolved by the constrained analysis). This consensus tree is similar to elements of trees resolved by the two data sets separately, but it includes clades that are resolved by neither of the other two. In this tree Joinvillea is sister of the grasses, and as in trees obtained by analysis of the restriction sites alone, the next divergence event is between. Anomochloa + Streptochaeta and all other grasses, and within the latter group the next divergence event is between Pharus and all other grasses. Within the latter group is a multichotomy from which five lineages diverge. One of the five groups is a clade that includes Brachyelytrum, Pooideae, and the PACC clade. This clade also occurs in most-parsimonious trees for the restriction site data alone, and in most-parsimonious trees for the structural characters alone when monophyly of grasses was constrained; the unconstrained analysis of the structural characters resolved a group that included these elements plus Baloskion. The four other elements within the unresolved clade that is sister to Pharus are Eremitis, Ehrhartoideae, Olyreae, and Bambuseae. Regardless of the various potential relationships among these four groups, the placement of Brachyelytrum with Pooideae and the PACC clade is sufficient to render Bambusoideae (as circumscribed to include Brachyelytrum) nonmonophyletic. Even the removal of Brachyelytrum from Bambusoideae leaves the remaining elements of an inclusive Bambusoideae nonmonophyletic, for the divergence of at least two lineages (Pharus and the clade that includes Anomochloa and Streptochaeta) from the larger lineage that includes the PACC clade, Pooideae, and other elements often assigned to Bambusoideae also is inconsistent with a monophyletic Bambusoideae.

There are five polytomies in the consensus tree [ILLUSTRATION FOR FIGURE 3 OMITTED], each representing conflict between two alternative sets of relationships:

1. Phragmites and Danthonia sister taxa, with Molinia sister of that clade, vs. Phragmites and Molinia sister taxa, with Danthonia sister of that clade;

2. Sesleria and Poa sister taxa, with that clade sister of all other Poeae, vs. Sesleria sister of all other Poeae, and within that group Poa sister of all remaining Poeae;

3. Aveninae sister of Alopecurinae, Phalaridinae sister of that clade, and Amphibromus sister of that clade, vs. Phalaridinae sister of Alopecurinae, Amphibromus sister of that clade, and Aveninae sister of that clade;

4. Brachyelytrum sister of Pooideae, and PACC clade sister of that clade, vs. PACC clade sister of Pooideae, and Brachyelytrum sister of that clade;

5. Olyreae sister of the clade that includes Brachyelytrum, Pooideae, and the PACC clade, with Bambuseae sister of that clade, Eremitis sister of that clade, and Ehrhartoideae sister of that clade, vs. Ehrhartoideae sister of the clade that includes Brachyelytrum, Pooideae, and the PACC clade, Olyreae sister of that clade, and a monophyletic grouping of Eremitis and Bambuseae sister of that clade.

Only the last two of these five polytomies involve alternative relationships above the level of the tribe. The alternative relationships underlying polytomies 4 and 5 are correlated in occurrence among the 12 most-parsimonious trees such that the first of the two alternatives described above for polytomy 4 occurs only with the first alternative listed for polytomy 5 (this combination occurs in six of the 12 trees), and the second alternative for polytomy 4 occurs only with the second alternative for polytomy 5 (this combination occurs in the other six trees). Thus the monophyletic grouping of Brachyelytrum plus Pooideae is associated in all cases with a pectinate structure in which the divergence of Ehrhartoideae immediately follows the divergence of Pharus, and is itself followed by divergence of the following taxa in sequence: Eremitis, Bambuseae, Olyreae, and the PACC clade from a monophyletic grouping of Brachyelytrum plus Pooideae [ILLUSTRATION FOR FIGURES 1, 4A OMITTED]. This set of relationships, which we will refer to as "Alternative Topology 1," conforms to those obtained by analysis of the restriction site data [ILLUSTRATION FOR FIGURE 1 OMITTED]; it was described as the hypothesis of an early-diverging Ehrhartoideae, and it also can be described as the hypothesis of a sister-group relationship between Brachyelytrum and Pooideae. The second of the two alternative structures, in which the PACC clade and Pooideae are sister groups, occurs in association with the placement of Eremitis as sister of Bambuseae, and with the placement of this clade as sister to a clade in which the following taxa diverge in sequence: Olyreae, Ehrhartoideae, and Brachyelytrum, from the monophyletic grouping of Pooideae and the PACC clade [ILLUSTRATION FOR FIGURES 4B, 5-7 OMITTED]. This set of relationships, which we will refer to as "Alternative Topology 2," can be called the hypothesis of a "late-diverging" Ehrhartoideae, or that of a sister-group relationship between Pooideae and the PACC clade.

Unambiguous character transformations associated with clades resolved by the combined data set are described in Table II. Clade numbers in Fig. 3 and character placements on most-parsimonious trees (e.g., [ILLUSTRATION FOR FIGURES 5-7 OMITTED], representing one of six most-parsimonious trees exhibiting Alternative Topology 2), are used as reference points in Table II.

4. Topological Similarities among Unweighted Analyses

The separate unweighted analyses of restriction sites and structural characters (Poaceae constrained), and the analysis of the combined data set ([ILLUSTRATION FOR FIGURES 1-3 OMITTED], Table III) resolve several major groupings in common. The PACC clade is resolved by all three of these analyses; in all cases it includes all exemplars recognized by Clayton and Renvoize (1986) as elements of Arundinoideae, Centothecoideae, Chloridoideae, and Panicoideae, except Anisopogon. Within the PACC clare, Arundinoideae as circumscribed by Clayton and Renvoize (again excluding Anisopogon) are paraphyletic and include the earliest diverging lineages in the PACC clade. Centothecoideae (i.e., the two included species of Chasmanthium) always are resolved as monophyletic, Panicoideae are resolved as monophyletic (except by restriction site characters alone, where Panicoideae are paraphyletic), and Panicoideae plus Centhothecoideae always constitute a monophyletic group. Chloridoideae are resolved as monophyletic by the restriction site data and by the combined character set, but not by the structural characters alone, which place Amphipogon (and no other taxa except representatives of Chloridoideae) in the smallest clade that includes all elements of Chloridoideae.

All three of these unweighted analyses resolve Pooideae as circumscribed by Clayton and Renvoize (1986), but expanded to include Anisopogon and Diarrhena, as monophyletic. Within Pooideae, three groupings at the level of tribe or above are resolved as monophyletic by all three analyses: Meliceae, Agrostideae, and the pairwise combination of Lygeum and Nardus.

As just noted, all three analyses resolve both the PACC clade and Pooideae as monophyletic, each with identical contents. All three analyses also resolve a larger monophyletic grouping that consists of these two clades plus Brachyelytrum, but two different relationships are detected among these three lineages. The restriction site data place Brachyelytrum as sister of Pooideae, with the PACC clade as sister of that pairwise grouping, while the structural data place the PACC clade as sister of Pooideae, with Brachyelytrum sister of that pairwise grouping. With the combined data set these two alternative relationships are equally parsimonious, and each occurs in half of the most-parsimonious trees. The constant resolution of this group of three lineages leaves a similarly constant paraphyletic assemblage of all remaining elements (i.e., a broadly defined Bambusoideae) as early-diverging lineages in the grass family. In all three consensus trees [ILLUSTRATION FOR FIGURES 1-3 OMITTED], Anomochloa, Streptochaeta, and Pharus are among the earliest lineages to diverge from the line that includes most other grasses. Analysis of the structural characters resolves Streptochaeta as sister of all other grasses, with Anomochloa, Pharus, and Eremitis diverging separately from the next node, while the other two analyses resolve a monophyletic grouping of Streptochaeta and Anomochloa as sister of all other grasses, and within that group, Pharus as sister of all remaining grasses. Bambuseae, Olyreae, and Oryzeae also are resolved by all three analyses, while Ehrhartoideae (i.e., Oryzeae plus Ehrharta) are resolved by restriction sites and by the combined data set, but not by the structural data set.

Although Alternative Topology 1 (of the combined analysis) corresponds to the overall structure supported by restriction sites alone (compare [ILLUSTRATION FOR FIGURES 1 AND 4A OMITTED]), Alternative Topology 2 of the combined analysis is consistent with some but not all aspects of the relationships resolved by structural characters alone (compare [ILLUSTRATION FOR FIGURES 2 AND 4B OMITTED]). Alternative Topology 2 resembles the relationships supported by structural characters in the placement of Brachyelytrum as sister of a clade that includes Pooideae and the PACC clade, but differs in the relationships resolved among the remaining taxa.

[TABULAR DATA FOR TABLE III OMITTED]

5. Incongruence among Unweighted Analyses

The amount of incongruence within the restriction site data set is 913 steps (1277 steps in most-parsimonious trees minus 364 steps "total variation," or minimum possible number of steps if all characters were congruent), and the amount of incongruence within the structural data set (monophyly of grasses not constrained) is 106 steps (158 - 52); thus, the total amount of within-data-set incongruence, [I.sub.w], is 1019 steps. The total amount of incongruence in the combined data set, [I.sub.T], is 1052 steps (1468 - [364 + 52]), and the amount of incongruence between the two data sets, [I.sub.B], is [I.sub.T] - [I.sub.w] = 1052 - 1019 = 33 steps. The Mickevich/Farris incongruence index, [I.sub.MF], is [I.sub.B]/[I.sub.T] = 33/1052 = 0.031. This amount of incongruence falls at the low end of the range of other published figures, such as 0.153 (Rodman et al., 1996), 0.114 (Kluge, 1989), 0.052 (Uhl et al., 1995), and a range of 0.0083-0.0497 (Davis et al., in press).

The structural character set, when optimized on the 10 shortest trees obtained by analysis of the restriction site data, has a range in lengths of 191-197, a range in CI of 0.26-0.27, and range in RI of 0.68-0.69. The restriction site character set, when optimized on the 10,000 trees retained after unconstrained analysis of the structural character set, has a range in lengths of 1552-1748 steps, a range in CI of 0.20-0.23, and a range in RI of 0.54-0.61. Because this set of trees is not a complete set of most-parsimonious trees for the structural data set, the actual ranges in length, CI, and RI of the restriction site dam on most-parsimonious trees for the structural data may be greater than the observed ranges; the ranges reported here, therefore, are minimum ranges.

The restriction site data require a minimum of 0 extra steps to be optimized on trees obtained by simultaneous analysis of the two data sets, relative to the number of steps on their own most-parsimonious trees (1277 in both cases); the restriction site data require a minimum of 22% more steps (1552 vs. 1277) to be optimized on trees obtained by analysis of the structural character set. The structural dam require a minimum of 16% more steps (184 vs. 158) to be optimized on trees obtained by simultaneous analysis of the two data sets, and a minimum of 21% more steps (191 vs. 158) to be optimized on trees obtained by analysis of the restriction site character set.

6. Implied Weighting

Simultaneous analysis of the combined matrix of 364 restriction site characters and 42 structural characters, using implied weighting, resolved one most-parsimonious tree of fit 2589.4 (rescaled fit 0.41). The length of this tree, when the data are optimized on it as unweighted characters, is 1486 steps, or 18 steps (1.2%) more than the 1468 steps of trees obtained by unweighted analysis. Of these 1486 steps, 1302 are in the restriction site dam (vs. minimum length of 1277 on shortest trees obtained by unweighted analysis of all data, an increase of 25 steps, or 1.9%, and versus 1277 steps on shortest trees obtained by unweighted analysis of just the restriction site data, for the same increase of 25 steps, or 1.9%). The remaining 184 steps are in the structural data; this represents no increase relative to the 184 steps on shortest trees obtained by unweighted analysis of all of the data, and an increase of 25 steps (15.8%) relative to the minimum length of 158 steps on shortest trees obtained by unconstrained and unweighted analysis of just the structural data.

The tree resolved by implied weighting (Fig. 8) is congruent in many respects with those obtained by unweighted simultaneous analysis of both character sets, but it also exhibits some unique elements. It resembles the unweighted analysis in resolving both Pooideae (including Anisopogon and Diarrhena) and the PACC clade as monophyletic and in resolving a monophyletic grouping of Brachyelytrum, Pooideae, and the PACC clade. It resembles Alternative Topology 2 in resolving Brachyelytrum as sister of the monophyletic grouping of Pooideae and the PACC clade within this group, and in placing Ehrhartoideae and Olyreae as successively less closely related lineages. Bambuseae are the next most closely related group, and the tree obtained by implied-weighting differs from other trees in uniting Pharus and Eremitis as sister taxa that together constitute a clade that is the next most closely related group. The two remaining grasses, Anomochloa and Streptochaeta, are again identified as sister taxa, and as in the results of all other analyses except that of the structural data, this clade is the sister of all other grasses, while Joinvillea is sister of the grasses.

Within the PACC clade, the tree obtained by implied weighting is congruent with that obtained by unweighted analysis in supporting a sister-group relationship between Aristida and a clade that includes all other elements in that group, and in resolving a monophyletic group in which a monophyletic Centhothecoideae is sister of a monophyletic Panicoideae. However, this tree differs in nesting Chloridoideae within a paraphyletic assemblage of all other Arundinoid elements (except Anisopogon, which is placed in Pooideae, in agreement with the unweighted analysis). This assemblage of Chloridoideae and Arundinoid elements is sister of the clade that comprises Centothecoideae and Panicoideae.

Within Pooideae, the tree obtained by implied weighting is congruent with the results obtained by equal weighting in resolving a clade consisting of Lygeum and Nardus as sister of a clade that includes all other members of Pooideae. However, the results of implied weighting differ substantially with regard to the rest of the Pooideae. The equal-weighted analysis resolved a monophyletic grouping of Bromeae, Hordeeae, Poeae, and Agrostideae, and a pectinate series of four diverging lineages between that group and the clade that consists of Lygeum and Nardus. The implied-weighting analysis resolves the same grouping of Bromeae, Hordeeae, Poeae, and Agrostideae, but the sister of this clade is a clade that includes all other taxa of Pooideae except the grouping of Lygeum and Nardus. Within this unique group there is a sister-group relationship between a clade that includes Anisopogon as sister of Ampelodesmos and Stipeae (Ampelodesmos is placed among elements of a paraphyletic Stipeae), and a second clade in which Diarrhena is sister of a clade that consists of Brachypodieae as sister of Meliceae.

V. Discussion

A. SCORING OF STRUCTURAL CHARACTERS AND USE OF EXEMPLAR TAXA

As noted above, Watson and Dallwitz (1992) was the principal source of the structural character data employed here. This book, derived from an exhaustive database of the grass literature, represents a monumental achievement and serves as a basic source for future studies. The same database was used previously for the development of character sets for cladistic analyses of the grass family (Kellogg & Campbell, 1987), and of subfamilies Pooideae, Bambusoideae, and tribe Andropogoneae (i.e., Saccharineae) (Kellogg & Watson, 1993). However, as Kellogg and Watson (1993) note, there are limitations to the use of these data directly for the purpose of phylogenetic analyses. Any such compendium of a vast literature must be used with caution. Indeed, we found a few contradictions, errors, and omissions, relative to the primary literature (Appendix 2). A different sort of problem is that the character definitions in the book were intended chiefly for the purpose of identification, rather than as homology assessments, as is appropriate for phylogenetic analysis. Thus, we evaluated character and state definitions with the latter purpose in mind.

Another problem encountered in the use of Watson and Dallwitz (1992) is that of nonmonophyletic genera. Few genera have been studied from a phylogenetic perspective and it is clear that many conventionally recognized genera of Poaceae and other families will be found to be nonmonophyletic as such studies are conducted. There is evidence that two genera in the present study are not monophyletic as circumscribed by Watson and Dallwitz. Bambusa Schreb. s.l., as recognized by Watson and Dallwitz (1992) and Clayton and Renvoize (1986), is divided here into Bambusa (Old World) and Guadua Kunth (New World), which were placed in separate subtribes by Soderstrom and Ellis (1987). That determination is further substantiated by the study of Clark et al. (1995), and by the present analysis, for neither of these analyses resolves Bambusa and Guadua as sister taxa [ILLUSTRATION FOR FIGURE 3 OMITTED]. Secondly, the circumscription of Festuca L. has recently been refined. Festuca subg. Schedonorus (P. Beauv.) Peterm., including the F. arundinacea complex, belongs in Lolium according to Darbyshire and Warwick (1992), and that relationship also is supported by the present analysis [ILLUSTRATION FOR FIGURE 3 OMITTED]. In addition to the previous two cases, Pariana Aubl., as recognized by Watson and Dallwitz (1992), may not be monophyletic. It is divided into Pariana and Eremitis by Clayton and Renvoize (1986) and by Hollowell (1987), and is represented here by Eremitis.

The use of multispecies taxa with multiple polymorphic characters as individual "composite terminals" in cladistic analyses influences the results in unpredictable ways (Nixon & Davis, 1991). Among the more pernicious effects is the possibility for most-parsimonious resolutions to imply character state combinations that have never been observed and that may never have existed; the determination that such character combinations exist or ever existed, and the secondary effects on tree structure implied by such combinations, are, of course, artifactual. For example, if a genus has a [01] polymorphism for one character and a [01] polymorphism for a second character, four potential combinations of individual states are implied, but it is possible that only two combinations exist today, or ever existed. If an unobserved combination is most-parsimonious in the context of the complete data set, which may include any number of additional polymorphisms in other characters and taxa, the effect of polymorphism scoring on the resulting tree structure may be dramatic. These effects can be manifested even if all scored polymorphisms reflect actual polymorphisms of the monophyletic terminals that are sampled (e.g., if an actually monophyletic Festuca were to be polymorphic for awn presence/absence). The potential problem is magnified if some of the composite terminals actually represent separate lineages that properly belong in different positions in a phylogenetic tree, as would have been the case with "Bambusa" and "Festuca" in the present analysis, had they been treated as individual terminals. Thus, the use of polymorphism coding for multispecies taxa that are used as terminals must be regarded as an expedient. In the present analysis we eliminated many generic polymorphisms, as reported by Watson and Dallwitz (1992), by checking specimens and searching the literature for data on the exemplar species in the study. A further advantage to scoring all characters as they occur in exemplar taxa is that the procedure improves the comparability of indexes of support and character homoplasy between structural and molecular data sets.

B. ANALYSES OF MOLECULAR DATA SETS

Cladistic analyses of a variety of molecular data have been used in previous studies to analyze the major internal phylogenetic structure of the Poaceae. Chloroplast DNA restriction sites have been used previously in studies of the phylogeny of grasses by Soreng et al. (1990), Davis and Soreng (1993), Yaneshita et al. (1993), and Duvall et al. (1994). Nucleotide sequences of a variety of genes of the chloroplast genome have been used as follows: rbcL by Doebley et al. (1990), Barker et al. (1995), and Duvall and Morton (1996); rpoC2 by Cummings et al. (1994); rps4 by Nadot et al. (1994); ndhF by Clark et al. (1995); and matK by Hongping Liang and Hilu (1996). Nucleotide sequences of genes of the nuclear genome have been used as follows: rRNA by Hamby and Zimmer (1988, 1992) and by Kellogg and Linder (1995:531, a reanalysis of Hamby and Zimmer's data, with the addition of unpublished data for Brachyelytrum provided by Issel and Zimmer); and Adh by Morton et al. (1996; also see Gaut et al., 1996). The analyses based on cpDNA gene sequences and restriction sites all estimate the history of the same nonrecombining, nonreticulating organelle lineage, while Adh is a low-copy-number nuclear-encoded gene that may experience recombination, and rRNA is a family of multiple-copy nuclear genes that are subject to concerted evolution, and for which evolutionary patterns may be complex (e.g., Wendel et al., 1995; Cronn et al., 1996).

Some general conclusions may be drawn from these molecular studies (note that here and in the following pages we mention genera and tribes that are not part of the present study but that provide a broader context for the discussion of phylogeny and character state distributions in the grasses). First, there is broad support for the monophyly of certain major groups, including Pooideae and the PACC clade, and within the latter, Panicoideae and Chloridoideae. In contrast, Bambusoideae, as recognized in classifications such as those of Clayton and Renvoize (1986), Watson and Dallwitz (1992), Soderstrom and Ellis (1987), and Tzvelev (1989), are paraphyletic, because the PACC clade and Pooideae arise from among elements assigned to Bambusoideae in those classifications (although different arrangements have been resolved by the various data sets). Some elements of a broadly defined Bambusoideae are resolved within or adjacent to the Pooideae or PACC clades, and these two groups can be expanded in membership to accommodate those elements. In particular, Diarrheneae and Phaenospermateae Renvoize & Clayton, two tribes included in Bambusoideae by Clayton and Renvoize (1986) or Watson and Dallwitz (1992), appear to belong in Pooideae, and one (Centotheceae) appears to belong in the PACC clade, while Brachyelytreae are resolved as closely related to the Pooideae. Among the remaining elements of a broadly defined Bambusoideae, the Bambuseae and Oryzeae (the most frequently sampled tribes of this group), plus the Anomochloeae, Streptochaeteae, Phareae, Olyreae, Parianeae, Ehrharteae, and Streptogyna P. Beauv. (the sole genus of Streptogyneae C. E. Calderon & Soderstr.), do not appear to belong within the PACC clade or in Pooideae, though relationships resolved among these elements vary among treatments, and some of these groups are in some cases resolved as closely related to Pooideae or the PACC clade.

C. PHYLOGENETIC STRUCTURE OF THE GRASSES

1. Support for Groups Resolved by the Present Analysis

One of the 12 most-parsimonious trees obtained by simultaneous analysis of the restriction site and structural character sets is illustrated with transformations in structural characters optimized on the internodes ([ILLUSTRATION FOR FIGURES 5-7 OMITTED]; also see consensus tree, [ILLUSTRATION FOR FIGURE 3 OMITTED]). This tree is one of the six with Alternative Topology 2 (late-diverging Ehrhartoideae), in which subfam. Ehrhartoideae is sister of the clade that comprises Brachyelytrum as sister of a subclade in which Pooideae and the PACC clade are sister groups.

The consensus tree derived by analysis of just the structural character set, constrained to resolve Poaceae as monophyletic, is poorly resolved within the Pooideae, and also includes a polytomy near the base of the family [ILLUSTRATION FOR FIGURE 2 OMITTED]. The consensus tree derived by analysis of just the restriction site character set is almost completely resolved except within the PACC clade (there is also a polytomy within the Bambuseae and another within the Agrostideae). As detailed above, the consensus obtained by simultaneous analysis of both character sets has five polytomies [ILLUSTRATION FOR FIGURE 3 OMITTED]; relationships within the PACC clade are more fully resolved by the combined character sets than by either of them separately (64 clades resolved by combined data sets, 62 by restriction sites, 44 by constrained analysis of structural characters), but polytomies 4 and 5 (see section III. C.3) represent a lesser degree of resolution than obtained with the restriction site data. Thus, the effect of combining the data is to add resolution in some areas and to decrease it elsewhere.

2. Comparison of Results of Simultaneous Analysis with Previous Classifications and Hypotheses of Phylogeny of the Grasses

The present results are congruent in many respects with modern classifications of the grasses. Of the two most comprehensive recent treatments of the family - namely, Clayton & Renvoize, 1986, and Watson & Dallwitz, 1992 - the present results are in greater agreement with the former. All of the taxa placed in Chloridoideae, Panicoideae, and Pooideae in those classifications, and sampled in the present study, were also placed in those groups by the present analysis. The Bambusoideae are circumscribed similarly in both of those classifications, except that Clayton and Renvoize (1986) exclude Centotheceae. The present results do not support a monophyletic Bambusoideae in the sense of either of these treatments, but among the various elements sometimes assigned to that group, the placement of Centothecoideae within the PACC clade, as sister of Panicoideae, is inconsistent with any circumscription of a monophyletic Bambusoideae that includes Centothecoideae and does not also include most or all elements Panicoideae, Arundinoideae, and Chloridoideae.

Other tribes of Bambusoideae recognized by Watson and Dallwitz (1992) and Clayton and Renvoize (1986), and represented by more than one exemplar in the present analysis, are resolved as monophyletic, but even after removal of Diarrheneae to Pooideae and of Centhotheceae to the PACC clade, the remaining elements of Bambusoideae do not constitute a monophyletic group.

Additional incompatibilities among the present results and classifications by Clayton and Renvoize (1986) and Watson and Dallwitz (1992) reflect the broad concept of Arundinoideae and the narrow concept of Pooideae adopted by Watson and Dallwitz (1992). Watson and Dallwitz do acknowledge that the Arundinoideae they circumscribe is likely to be polyphyletic, and the present results are consistent with that determination. Clayton and Renvoize (1986) attempted to circumscribe a natural Arundinoideae, in part by placing within Pooideae several tribes and genera (e.g., Ampelodesmos, Lygeeae, Nardeae, Stipeae, and Aveneae [i.e., Agrostideae] subtr. Duthieinae Potztal) that have characteristics thought to have been present among early-diverging elements of Pooideae. The present results, based on sampling of the first four of these taxa, support the placement of all of them (including Duthieinae by implication), plus Anisopogon, within Pooideae. Even with all of these taxa removed from Arundinoideae, however, the present results also indicate that the more strictly circumscribed Arundinoideae that remain are still paraphyletic.

In comparing the present results to those obtained by other authors, it is necessary to consider the depth of sampling among studies. If the Pooideae and the PACC group represent two major monophyletic groups within the grass family, and if these two groups are closely related to each other and derived from among the various lineages of a broadly defined Bambusoideae, the deepest branches in the grass family are most likely to be resolved if the major sublineages of the Pooideae and the PACC clade are sampled, along with all major lineages assigned by various authors to Bambusoideae.

In the following discussion we consider the cladistic analyses of five molecular studies (Table IV) that have included members of at least six of the six to seven major subfamilies widely recognized in classifications: Barker et al. (1995; rbcL sequences); Duvall and Morton (1996; rbcL sequences); Clark et al. (1995; ndhF sequences); Hongping Liang and Hilu (1986; matK sequences); and the present cpDNA restriction site analysis (which supersedes Davis & Soreng, 1993). One cladistic analysis based on structural characters of grasses also is considered. Kellogg and Campbell (1987) presented a taxonomically comprehensive cladistic analysis wherein individual genera of Arundinoideae were used as terminals, and Bambusoideae, Chloridoideae, Panicoideae, and Pooideae were constrained, in large part, as putatively monophyletic composite taxa.

Despite a great deal of consensus among the results of DNA analyses, one area in which differences remain is that of the composition and relationships among three major groups or assemblages (i.e., Bambusoideae, Pooideae, and the PACC clade) that have been proposed in [TABULAR DATA FOR TABLE IV OMITTED] classifications and resolved, to varying degrees, in phylogenetic analyses. The present study and the analysis of ndhF sequences (Clark et al., 1995) sampled broadly in Bambusoideae s.l. (16 and 24 taxa, respectively, including Ehrhartoideae, Brachyelytrum, Diarrhena, and Phaenosperma). In the present analyses based on the restriction site characters and on both sets of characters, and in the analysis of ndhF sequences, Anomochloa and Streptochaeta are resolved as the solitary elements of a clade that is the sister group of all other grasses, and within the latter group Pharus is the sister of all remaining grasses [ILLUSTRATION FOR FIGURES 1, 3, 4 OMITTED]. The consensus tree of the present structural character analysis [ILLUSTRATION FOR FIGURE 2 OMITTED] resolves similar but not identical relationships at this juncture: Streptochaeta diverges first, followed by a polytomy of Anomochloa, Pharus, Eremitis, and a clade that includes all other grasses. Of these four genera or the tribes they represent, Duvall and Morton (1996) included Anomochloa, and it was resolved as the sister of all other grasses in their study. None of these genera is represented as an individual terminal in the other molecular studies. Various authorities had previously identified Anomochloa, Streptochaeta, and Pharus as among the "most primitive" extant grasses (e.g., Celakovsky, 1889; Arber, 1929; Roshevits, 1937; Stebbins, 1956; Tateoka, 1957; Soderstrom, 1981). Indeed, Nakai (1943) placed Anomochloa and Streptochaeta in their own families, and Clifford (1961) excluded Anomochloa from Poaceae. Two structural variants of the chloroplast lend additional support to the early divergence of these grasses: 1) a 15 bp insertion in the ndhF gene, present in other grasses sampled but absent in Anomochloa, Streptochaeta, Pharus, Leersia, Oryza, and nongrasses (Clark et al., 1995); and 2) a 740 bp intron in rpoC1, present in Anomochloa, Streptochaeta, and nongrasses, and absent in other grasses sampled (Clark et al., 1995; R. Wallace et al., unpubl. data).

If Anomochloa, Streptochaeta, and Pharus are recognized as representatives of the earliest lineages to diverge from the one that includes all other grasses, they must be excluded from the Bambusoideae if that subfamily is to be monophyletic. The first two of these genera can be accommodated in Anomochlooideae, and the latter in Pharoideae (Clark & Judziewicz, 1996). The present results also favor the exclusion from Bambusoideae of Diarrhena and Brachyelytrum, both of which were included in Bambusoideae by Clayton and Renvoize (1986) and Watson and Dallwitz (1992), who suggested affinities of these taxa with Oryzeae and Ehrharta. The present analysis, like that of Clark et al. (1995), places Diarrhena within Pooideae. The present analysis also resolves Brachyelytrum either as the sister group of Pooideae (a relationship previously detected by Clark et al. [1995] and Kellogg and Linder [1995]) or as the sister group of a clade that includes Pooideae plus the PACC clade. The first of these alternative positions for Brachyelytrum is supported by the restriction site data [ILLUSTRATION FOR FIGURES 1, 4 OMITTED], and the second by the structural character data [ILLUSTRATION FOR FIGURES 2, 4 OMITTED]. In the first of these positions Brachyelytrum could be accommodated taxonomically either as a member of Pooideae or as the sole genus of a separate subfamily. In the second of these positions it would be appropriate to recognize a separate subfamily for Brachyelytrum. Clark et al. (1995) also found a third genus sometimes assigned to Bambusoideae, Phaenosperma (not sampled in the present analysis), to be nested within Pooideae. Henceforth in this discussion we include Diarrhena and Phaenosperma in Pooideae and consider Brachyelytrum as an element of a clade that also includes Pooideae and the PACC clade ([ILLUSTRATION FOR FIGURE 3 OMITTED], node 13).

Some of the most-parsimonious trees resolved by the present analysis resemble previous results based on rbcL (Barker et al., 1995; Duvall & Morton, 1996) and matK (Hongping Liang & Hilu, 1996) in resolving Pooideae plus the PACC clade as a monophyletic grouping. The analysis of ndhF (Clark et al., 1995) resolves Pooideae and the PACC clade as lineages arising separately from among elements of a broadly defined Bambusoideae. After the initial divergences of Anomochloa plus Streptochaeta, and then Pharus, the PACC clade arises as the sister group of a clade that the authors designated the "BOP Clade," which comprises additional elements often assigned to Bambusoideae, Oryzoideae (i.e., Ehrhartoideae; and including Streptogyna, which was resolved as the sister of Oryzeae), and Pooideae (with Brachyelytrum sister of a clade that included all other elements of the subfamily), with relationships unresolved among these three groups. Clark et al. (1995) indicated that the trichotomy within the BOP clade in the consensus tree reflects two alternative equally parsimonious structures; in one tree Pooideae and Bambuseae were resolved as sister taxa, with Ehrhartoideae as the sister group of that clade, and in the other Bambuseae and Ehrhartoideae were resolved as sister taxa, with Pooideae as the sister group of that clade. The analysis by Duvall and Morton (1996) resolved one most-parsimonious tree, in which Anomochloa is the sister of all other grasses, and within the latter group, a clade that consists of Olyreae and Bambuseae is the sister of all remaining grasses. Within the latter group Ehrhartoideae were resolved as the sister group of a clade that consists of Pooideae plus the PACC clade. Other studies with more limited sampling among Bambusoideae s.1. favor Pooideae as sister group of a clade that includes Bambuseae, Oryzeae, and the PACC clade, or Oryzeae alone as sister group of the PACC clade (cpDNA restriction sites: Yaneshita et al., 1993; rRNA: Hamby & Zimmer, 1992; rps4: Nadot et al., 1994; rbcL: Doebley et al., 1990), or Pooideae (Morton et al., 1996). A consistent result of these molecular analyses is that both Pooideae and the PACC group are resolved as monophyletic. The five most comprehensive studies (Table IV) identify the Bambusoideae s.1. as paraphyletic, while the PACC clade and Pooideae (whether or not they are resolved as sisters of each other) are nested among the earliest diverging grass lineages.

Two narrower circumscriptions of Bambusoideae, Tzvelev's (1989) and Soderstrom and Ellis's (1987), also appear to define paraphyletic groups, as judged by the results of the ndhF analysis (Clark et al., 1995) and the present analysis. Tzvelev (1989) restricted the Bambusoideae by excluding not only Diarrheneae, Brachyelytreae, and Centotheceae but also Ehrharta, Oryzeae, Phyllorhachideae, and Phaenospermateae. Soderstrom and Ellis (1987) excluded all of these tribes except Oryzeae, and possibly Phyllorachideae, from their Bambusoideae. These treatments still circumscribe groups that would be recognized as paraphyletic, for Tzvelev included Anomochloeae, Streptochaeteae, and Phareae within the Bambusoideae, and Soderstrom and Ellis included these three tribes plus Oryzeae.

The placement of tribe Centotheceae is another area in which there is conflict between the results of phylogenetic studies and existing classifications (if it is intended that recognized taxa be monophyletic). Watson and Dallwitz (1992) place Centotheceae in Bambusoideae supertribe Oryzodae, while Clayton and Renvoize (1986) place the tribe its own subfamily, Centothecoideae, as a close relative of Chloridoideae, within a more inclusive group that is largely equivalent in composition to the PACC clade. The present results do place Centotheceae in the PACC clade, but rather than with Chloridoideae the tribe is resolved as sister group of Panicoideae (or possibly nested within Panicoideae), as previously resolved by Barker et al. (1995) and Clark et al. (1995). The present results also substantiate the conclusions of Yates (1966) that Chasmanthium and Uniola are not closest relatives.

Like the Bambusoideae s.l., many independent lines of evidence suggest that the Arundinoideae are not monophyletic. Four of the five major molecular analyses (Table IV) resolve a monophyletic PACC group that includes a nonmonophyletic Arundinoideae (minus a few disparate elements). The analysis of structural characters by Kellogg and Campbell (1987) also resolved a nonmonophyletic Arundinoideae, but all other grasses except Pooideae (i.e., the other elements of the PACC clade plus Bambusoideae s.1.) were nested among the Arundinoids. A complication in any discussion of these relationships is the broad circumscription of Arundinoideae by Watson and Dallwitz (1992). Even after several elements included among the Arundinoids by those authors are removed from consideration (i.e., Ampelodesmos, Anisopogon, Lygeum, Nardus, Stipeae, and probably a few other taxa; see below), the remaining taxa within the PACC clade still appear to be paraphyletic. The Arundinoideae, as more narrowly circumscribed by Clayton and Renvoize (1986), and with Anisopogon again removed, are similarly paraphyletic, in that Chloridoideae, Centothecoideae, and Panicoideae arise from among elements of that subfamily. In the trees obtained in the present study by simultaneous analysis, as in the results of the ndhF analysis, Aristida is the sister of all other members of the PACC clade, and the latter group includes two lineages, one of them including Chloridoideae, the other including Centothecoideae and Panicoideae, with a set of Arundinoids as early-diverging elements within each of the two lineages.

If attention is restricted to Arundinoid taxa that fall within the PACC clade, relationships still are complex, partly because different analyses have sampled different Arundinoid taxa, and partly because there is conflict among the results obtained by the various analyses. In the present study, Amphipogon is resolved as sister of a monophyletic Chloridoideae by simultaneous analysis of all characters, and as described above, some of the alternative relationships resolved by analysis of the restriction site data place various additional Arundinoid elements as close relatives of Chloridoideae; analysis of the structural data places Amphipogon among genera of Chloridoideae. Clark et al. (1995) included Micraria F. Muell. (and not Amphipogon) in their study, and resolved that genus as the sister group of a monophyletic Chloridoideae. The analysis of rbcL variation by Barker et al. (1995) includes a broader sampling of Arundinoid elements (14 taxa), and sheds additional light on this set of grass taxa. The sampling in that analysis was guided by the separate analysis of rpoC2 sequences of nearly all of the circa 40 genera of the Arundinoideae (Barker, 1995). With the rbcL data Barker et al. (1995) resolve Aristideae as the sister group of a subclade that consists of Danthonieae Zotov (a tribe subsumed in tribe Arundineae by Clayton and Renvoize, 1986), Centropodia Rchb., and Chloridoideae. With the rpoC2 data, however, Aristideae are resolved as one of three subclades within the PACC clade, with the second of the three subclades consisting of Danthonieae plus Chloridoideae, and the third including Arundineae (in a very restrictive sense), Panicoideae, and Centothecoideae (Barker, 1995); relationships among these three subclades of the PACC clade are unresolved.

In the present analysis four genera of Arundinoideae (Arundo, Phragmites, Molinia, and Danthonia; [ILLUSTRATION FOR FIGURE 3 OMITTED], node 23) are resolved as a monophyletic group that is sister of the clade that includes Centothecoideae and Panicoideae. A similar group was resolved by Clark et al. (1995; Danthonia not sampled), while Barker et al. (1995), who sampled Arundinoideae more deeply, resolved a somewhat different arrangement of Arundinoid taxa. They resolved a clade that includes Arundo, Monachather Steud., Phragmites, and Moliniopsis Hayata (included in Molinia by Clayton & Renvoize, 1986), as sister group to a clade that includes Gynerium Willd. ex P. Beauv. (tribe Arundineae sensu Clayton & Renvoize, 1986), Thysanolaena Nees (tribe Thysanolaeneae C. E. Hubb., Arundinoideae sensu Clayton & Renvoize, 1986, and Watson & Dallwitz, 1992), plus Centothecoideae and Panicoideae.

As noted earlier, the present analysis resolves a sister-group relationship between Centothecoideae and Panicoideae (or possibly a nesting of Centothecoideae among elements of Panicoideae). In analyses of nucleotide sequence variation of ndhF (Clark et al., 1995) and rbcL (Barker et al., 1995), and in a separate cpDNA restriction site analysis (Soreng & Davis, unpubl.data), Centothecoideae and Panicoideae are resolved in a clade that also includes Thysanolaena, and in the ndhF study the Panicoideae are polyphyletic. Clark et al. (1995) also resolved Zeugites P. Browne of Centotheceae in this clade, and demonstrated that Chasmanthium and Zeugites are closely related even if the tribe is questionably monophyletic. Thus, it appears from the small sampling to date that Centotheceae and Panicoideae either constitute the entirety of a monophyletic group or that the smallest monophyletic group that includes both of them also includes some Arundinoid elements.

There is a growing consensus that Pooideae are monophyletic if circumscribed to include a small number of "orphan" taxa in addition to the conventionally recognized elements. Analyses of cpDNA restriction site variation among 25 exemplar taxa of Pooideae (Soreng et al., 1990), 38 exemplars (the current study), and 72 exemplars (Soreng & Davis, unpubl. data), the latter representing 32% of the genera in the subfamily sensu Clayton & Renvoize (1986), have consistently resolved the subfamily as monophyletic under the broad circumscription proposed here. All four "core" tribes of Pooideae are resolved by the present analysis as individually monophyletic (if the additional tribe Seslerieae W. D. J. Koch is included in Poeae; see below), and the four tribes are united within a single clade in which Agrostideae and Poeae are sister taxa and together constitute the sister of a clade that includes Bromeae and Hordeeae as sister taxa [ILLUSTRATION FOR FIGURE 3 OMITTED]. This core group is included in Pooideae in all modem classifications, although its elements are grouped in different ways [e.g., compare the taxonomies of Clayton and Renvoize (1986) and Watson and Dallwitz (1992) with the present results]. As an apparently monophyletic assemblage, it would be reasonable to circumscribe the grouping of these four tribes as Pooideae, but several additional lineages are more closely related to this group than to the PACC clade or to any other major lineage, and they would have to be accommodated in some manner within a phylogenetic system, presumably as a set of additional subfamilies. Results of the present analyses suggest that Pooideae sensu Watson & Dallwitz (which includes the four "core" tribes plus Meliceae, Brachypodieae, and Seslerieae, the latter of which is nested within Poeae in the present results) remains monophyletic and nonoverlapping with the PACC clade if it is expanded to include Lygeum, Nardus, Anisopogon, Ampelodesmos, Stipeae, and Diarrhena [ILLUSTRATION FOR FIGURE 3 OMITTED]. If additional studies confirm a sister-group relationship between Brachyelytrum and the clade that includes the four core tribes of Pooideae and these additional elements, it will be reasonable to include Brachyelytrum in Pooideae as well. Phaenosperma (not sampled in the present study) also should continue to be considered as a potential element of Pooideae (Clark et al., 1995). An inclusive circumscription of Pooideae also is supported by the studies of Davis and Soreng (1993, see their Fig. 1 and commentary, 15 taxa in Pooideae), Nadot et al. (1994, 16 taxa), Cummings et al. (1994, 4 taxa), Clark et al. (1995, 6 taxa including Diarrhena and Phaenosperma), Barker et al. (1995, 6 taxa), and Kellogg and Linder (1995, 6 taxa). This broad circumscription of Pooideae is not contradicted by the analysis of structural characters in the present study (monophyly of grasses constrained), nor by the analysis of Kellogg and Campbell (1987).

All of the taxa in the broadly circumscribed Pooideae just described, except the four core tribes and the additional tribe Brachypodieae Harz [included in Triticeae (i.e., Hordeeae) by Clayton & Renvoize, 1986], have been included in the Arundinoideae or Bambusoideae (or in equivalent groups) by one or more contemporary authors (e.g., Tateoka, 1957; Clayton & Renvoize, 1986; Watson & Dallwitz, 1992). In other words, the only tribes within clade 29 in Figure 3 that have consistently been included in Pooideae are Agrostideae, Brachypodieae, Bromeae, Poeae, and Hordeeae. The other taxa have been excluded from Pooideae in various classifications because they exhibit "anomalous" features, such as "small" chromosomes of base numbers x = 8, 9, 10, 11, or 12, or in which the base number is indeterminate but it or chromosome size appears to depart from the "typical" states in core Pooideae of x = 7 and "large" chromosomes. The Diarrheneae include one or two genera and two to five species (if Neomolinia Honda & Sakis is included), and this tribe is placed in Bambusoideae by Clayton and Renvoize (1986) and Watson and Dallwitz (1992), but members of the group have some features typical of Pooideae, or are polymorphic for features characteristic of Pooideae and other taxa sometimes assigned to Bambusoideae (Appendix 2). Anisopogon is similar to Stipeae in characters in addition to those used here, such as features of the guard cells (Kellogg & Campbell, 1987), and amino-acid composition of caryopses (Hock-Hin & Watson, 1987). Phaenosperma (Phaenospermateae, Bambusoideae, sensu Clayton & Renvoize and Watson & Dallwitz) also is placed within Pooideae by Kellogg and Campbell's (1987) analysis of structural characters, and by Clark et al.'s (1995) analysis of ndhF. None of the genera resolved in the Pooideae in the present analysis has been placed outside the group by phylogenetic studies other than that of Baum (1987), which placed Stipeae among members of the PACC clade.

D. STRUCTURAL CHARACTER EVOLUTION IN THE GRASSES

The associated occurrence of several morphological specializations of grasses leave little doubt that the family is monophyletic and highly derived (e.g., Dahlgren et al., 1985; Campbell & Kellogg, 1987). Simultaneous analysis of the restriction site and structural character sets provides a hypothesis of the history of diversification of the grasses that is demonstrably different from that obtained by analyzing these data individually, and that represents a best-supported overall hypothesis of relationships (Kluge & Wolf, 1993; Nixon & Carpenter, 1996). Each set of characters helps to resolve topological differences that are weakly supported or unresolved by individual analyses (Eernisse & Kluge, 1993; and Kluge & Wolf, 1993). At the same time, the resolution of two distinct and alternative sets of relationships among the deeper branches within the family highlights areas of conflict. We discuss the following character state distributions because of their support for major clades of grasses or because of their apparent adaptive significance.

Many of the structural character state changes mapped in Figs. 5-7 can be optimized on several nodes on the most-parsimonious trees, especially in the nexus where the two major clades, Pooideae and the PACC clade, diverge from among the earlier-diverging lineages. This could indicate that taxon density still is too low for confidence in exact character placement, or that for certain branching events there is a weak phylogenetic signal in the assembled data or among the extant organisms. It is possible that additional studies will not yield much improved resolution, as representation in the present study is already fairly comprehensive at the tribal and generic level (many tribes and genera being monotypic) around this nexus. On the other hand, some of the multiple optimizations around nodes are the results of polymorphisms and missing data. Whatever the reasons are for locally unstable character placements, and topological differences among studies, it is of interest to examine structural character transitions obtained from the simultaneous analysis in a broader context, as these may shed light on some controversial aspects of grass evolution. Distributions of states in taxa not included in this study, but which are of interest for extrapolation from the results, were verified in Watson & Dallwitz, 1992, unless otherwise stated.

1. Spikelets

In grasses it is a matter of convenience and tradition to identify as a spikelet the combination of one or more florets (each comprising a lemma, palea, lodicules, stamens, and pistil, or in certain grasses one or more of these parts absent) with two proximally situated bracts, the glumes (sometimes one or both absent). In the present analysis, grass spikelets with one fertile floret (char. 9) are present among all early-diverging herbaceous lineages just beyond the point of divergence of Anomochloa and Streptochaeta ([ILLUSTRATION FOR FIGURE 3 OMITTED], nodes 4-13), and in Chusquea Kunth of the Bambuseae. Although the aggregations of flowers and bracts in Anomochloa and Streptochaeta cannot be identified readily as typical grasslike spikelets (see below), their flowers are solitary, or at least isolated by multiple bracts of questionable homology to "glumes," "lemmas," and "paleas." The first identifiable grass spikelets present after the divergence of these two genera have one fertile rioter with a lemma, a palea, no rachilla extension (a rachilla is a primary axis that joins the rioters and two glumes within a spikelet, and here it does not extend beyond the point of attachment of the fertile floret). The homology of "glumes" across the family is uncertain. There is no compelling reason to doubt the homology of glumes among Phareae, Parianeae, and Olyreae, within Pooideae, and within Arundinoids and Chloridoideae, and perhaps among these three major groups, and consequently there is no compelling reason not to call the terminal inflorescence units "true" grass spikelets in all of these groups. However, between this larger assemblage and Bambuseae, Ehrhartoideae, Centotheceae, and Panicoideae, the homology of glumes, and thus spikelets, is questionable (Clifford, 1987). In Oryzeae, glumes are rarely present. Although there is one fertile floret, there sometimes are one or two sterile florets below the fertile one that can be mistaken for glumes.

Spikelets with single fertile florets occur in early-diverging lines (Phareae) and in nearly all later-diverging herbaceous lines (Olyreae, Parianeae, Ehrharta, Oryzeae) before the point of divergence of Brachyelytrum, Pooideae, and the PACC clade, and this condition appears to be plesiomorphic in the Bambuseae. Spikelets of this sort appear to represent the plesiomorphic condition in genera and tribes around the nexus from which the earliest-diverging elements of the Pooideae and the PACC clades arise: Anisopogon, Nardus, Phaenosperma (the latter not in the present study, but placed here by the analysis of Clark et al., 1995), and Stipeae, among Pooideae; Aristideae, Amphipogon, and a few other genera not in the present study, among Arundinoids (see Kellogg & Campbell, 1987: figs. 28.2 & 28.3, for a similar result concerning the occurrence of a single floret and absence of rachilla extension around the nexus between Pooideae and Arundinoideae). Brachyelytrum, which also diverges somewhere in this nexus, also has a single floret, but is unusual among these taxa in possessing a rachilla extension. In the present analysis, spikelets with multiple fertile florets arise independently within tribe Bambuseae, Pooideae, and the PACC clade, and revert to the single fertile floret condition several times.

As difficult as it is to apply the term "spikelet" in a consistent way, it is even more difficult to define the "pseudospikelet." This term has been used to describe certain modified forms of the true grass spikelet, as well as various stages in the evolution of the proto-grass inflorescence, and has been applied to structures found in three groups of grasses. The first of these is the clade that consists of Anomochloa and Streptochaeta ([ILLUSTRATION FOR FIGURE 3 OMITTED], node 3). The other two occurrences of "pseudospikelets" are in the Bambuseae, where they are accounted for by independent origins in Bambusa and Guadua [ILLUSTRATION FOR FIGURE 3 OMITTED], or in larger clades that include these genera. Given the relationships resolved in this study and by Clark et al. (1995) - specifically, the resolution of Anomochloa and Streptochaeta as elements of the earliest lineage of grasses to diverge from the line that includes all others - the pseudospikelets present in these taxa can be regarded either as a synapomorphy of this lineage (i.e., Anomochlooideae) or as a plesiomorphy of the grasses that is retained only in these two genera, though in modified form in Streptochaeta (Judziewicz & Soderstrom, 1989).

The origin of the grass spikelet has long been debated. Radically different modern proposals were presented by Soderstrom (1981) and Clayton (1990). One general interpretation of the spikelet is that it is homologous in most grasses, the exceptions being Streptochaeta, Anomochloa, and some Bambuseae (Clayton, 1990; for alternative interpretations, see Schuster, 1910; Nunez, 1968; Clifford, 1987). We suggest that models of the origin of the spikelet should take into account that Streptochaeta and Anomochloa, which do not have typical grass-type spikelets, are the sole living representatives of the earliest-diverging line from the one that includes all other grasses, and that Joinvillea, the apparent sister group of grasses, has a typical monocot panicle. The implications of this situation are 1) that the pseudospikelets of Streptochaeta and Anomochloa are not homologous with those of the Bambuseae, and 2) that the typical grass spikelet may have arisen in the lineage that includes all grasses except Streptochaeta and Anomochloa. In both Soderstrom's (1981) and Clayton's (1990) interpretations, pseudospikelets were derived from typical grass or proto-grass inflorescences, respectively, by contraction of lateral branches into condensed compound floral axes. Pseudospikelets have been interpreted as plesiomorphic for the grasses (Soderstrom, 1981), or possibly synapomorphic only within Bambuseae, and between Anomochloa and Streptochaeta (Clayton & Renvoize, 1986; Clifford, 1987). Clayton (1990) interpreted the spikelet of Bambuseae with two or more fertile toters and two glumes as plesiomorphic among all extant grasses, except possibly Anomochloa and Streptochaeta, and spikelets with single florets, as well as the pseudospikelets of Bambuseae, as independently derived from this basic arrangement. Soderstrom (1981: 43) viewed pseudospikelets of Bambuseae as proto-spikelet structures for the family, and true grass spikelets as having arisen from these. He further postulated the apomorphic nature of spikelets of the monoecious, herbaceous, early-diverging lineages that have only one floret per spikelet (Buergersiochloeae S. T. Blake, Olyreae, Parianeae, and Phareae). The present results are in conflict with Soderstrom's hypothesis, but do not fully conform to Clayton's interpretation.

The grass floret, or anthoecium, is an identifiable structure that appears to be homologous in most grasses. The presence of a palea that subtends the flower and lodicules, these together subtended by a sheathlike bract called a lemma, is the rule in the grasses. Paleas are first clearly identifiable, among the present sampling of grasses, in the clade that comprises all grasses except Anomochloa and Streptochaeta ([ILLUSTRATION FOR FIGURE 3 OMITTED], node 4). In the inflorescences of Anomochloa and Streptochaeta, the bracts that directly subtend the flowers do not appear to be paleas, and the homology of these or other bracts of pseudospikelets with prophylls in these genera is ambiguous (Soderstrom, 1981; Judziewicz & Soderstrom, 1989). Paleas (or prophylls, if these are considered homologous with paleas; Stebbins [1974] doubts this interpretation, but Gould [1975], Linder [1987], and Clayton [1990] favor it) that directly subtend flowers are absent in the flowers of the three nongrass taxa in the present study and in closely related families of Poales sensu Dahlgren et al. (1985). Although Clayton and Renvoize (1986) disagree, we also note, as did Clark et al. (1995), that the uppermost bract enveloping a flower in the inflorescence of Anomochloa is nearly identical in anatomical structure to the lemma of the pistillate spikelet in Pharus. If this bract does represent a lemma, then Anomochloa does not have a palea above the lemma, and thus does not have a true grass floret. Because prophylls characteristically develop only in leaf axils, a palea that was homologous with a prophyll would be absent in a truly terminal flower, and we suggest that the "lemma and flower" structures of Anomochloa represent terminal bracteate flowers rather than subterminal ones.

Although the most parsimonious interpretation of the character variation pattern detected in this study is that a single-flowered spikelet is plesiomorphic for the family, the later de novo development of a rachilla extension and subsequent additional fertile florets also is problematic. If a palea is present directly below a flower (ignoring lodicules), and if paleas are homologous with prophylls and prophylls only develop on axillary branches, then the floret is not truly terminal. It is reasonable, then, to suggest that the first true grass spikelet may have been indeterminate, that it had several florets (Clayton, 1990), and that it was subsequently reduced to one floret as late as in the immediate ancestor of Pharus plus all other grasses except Anomochlooideae ([ILLUSTRATION FOR FIGURE 3 OMITTED], node 4). Later reversions to the multiple-flowered condition then might have occurred within Bambuseae, Pooideae, and the PACC clade. If Anomochloa retains a plesiomorphic presence of flowers that are truly terminal (see above), the indeterminate multiple-flowered grass spikelet may have evolved between nodes 2 and 4 on Fig. 3.

Thus, the trees and character state distributions in the present results lend some support to Clayton's (1990) interpretation of the evolution of the grass spikelet, with the caveat that all multiple-flowered spikelets present in living grasses, in Bambuseae, Pooideae, and the PACC clade, may represent reversions, while the plesiomorphic multiple-flowered spikelet no longer survives.

Topologies reported in two other cladistic studies (Kellogg & Campbell, 1987; Clark et al., 1995) are consistent with the interpretation that a spikelet with one fertile floret (rarely with one or more sterile "lemmas" below the fertile one as in Ehrhartoideae), and lacking a developed rachilla extension above the fertile floret, was present in most early grass lineages and that this is the plesiomorphic form of the spikelet in Pooideae and the PACC clade.

2. Lodicules

Two questions that arise when lodicules are considered are 1) the homology of lodicules and 2) the distribution and pattern of diversification of lodicule types. There are structures in Anomochloa and Streptochaeta that are sometimes called lodicules, but it is far from certain that they are homologous with the lodicules of most other grasses, and the relationship of lodicules (in Anomochloa and Streptochaeta, or in other grasses) to structures in nongrasses also is obscure. Because of this ambiguity regarding the homology of lodicules, which is reflected in the scoring of this character in the matrix of structural characters, the point of origin of lodicules (char. 14) can be fixed at any of a series of nodes in the phylogenetic structure presented here, but no later than in the last common ancestor of Pharus and all other grasses except Anomochlooideae ([ILLUSTRATION FOR FIGURE 3 OMITTED], node 4).

In Clifford's (1987) interpretation, the occurrence of lodicules in all true florets of the early grasses serves as evidence that most grasses have only subterminal flowers (see the previous section). Thus, the putative absence of lodicules in Anomochloa and Streptochaeta, neither of which has true florets, is perhaps best regarded as a plesiomorphic absence. The earliest recognizable lodicules, as interpreted from the present analysis (chars. 15-18), were three in number, unfused, apically membranous (unlobed, more or less lanceolate, and fleshy at the base), and heavily vascularized (except in Pharus, according to Judziewicz, 1987). Vascularization is reduced or lost within the PACC clade and in all Pooideae. In the PACC clade, above the point of divergence of Aristida from other members of the clade, the apical membranous portion is reduced or absent, giving the lodicule a truncate form and fleshy texture. Lodicules become fused in Meliceae, and are lost in a number of other taxa. Under Alternative Topology 1, lodicule number (char. 15) changes twice from three to two: once in the Ehrhartoideae and once in the clade that includes Brachyelytrum, Pooideae, and the PACC clade. Under Alternative Topology 2 (i.e., late-diverging Ehrhartoideae), subfam. Ehrhartoideae is sister of the clade that consists of Brachyelytrum, Pooideae, and the PACC clade, and the change from three to two lodicules is a synapomorphy of this set of four lineages [ILLUSTRATION FOR FIGURE 5 OMITTED]. Where the reduction from three to two lodicules has occurred it has typically been by loss of the posterior lodicule (Clifford, 1961). The posterior lodicule appears to have been regained and lost again in early-diverging elements of Pooideae ([ILLUSTRATION FOR FIGURE 3 OMITTED], in the region between nodes 31 and 36).

The homology of lodicules with other organs in nongrasses is ambiguous. They may represent tepals, staminodes, stipules, ligules, reduced bracts, leaves, branch systems, or novel structures (Clifford, 1987). That the posterior lodicule develops above the anterior lodicules, rather than below them, as would be the case if they were all part of the same whorl, and that the posterior lodicule is sometimes suppressed, favors an interpretation that they are derived from stipules, leaves, or branches (Clifford, 1987). Under the latter interpretation (Clifford, 1987) lodicules would be absent from truly terminal flowers (which would not have paleas either), and thus their presence suggests that the flower is subterminal and subtended by a vestigial branch system (i.e., the lodicules). This again points to Anomochloa (if it lacks lodicules) as having terminal flowers (see previous section). However, it is not clear what is represented by the fringe of hairs below the staminal whorl in Anomochloa, or the large, fleshy, lanceolate bracts in this position in Streptochaeta. The possibility that these represent vestigial tepals is consistent with the interpretation of flowers in these taxa as terminal.

Lodicules are involved in the opening of florets at the time of flowering (Clayton, 1990). However, some grasses that lack lodicules also open at anthesis (e.g., Nardus). The posterior lodicule may not have been essential to this operation at certain stages of grass evolution, and its development therefore may not have been maintained. The large, indurate florets characteristic of many elements of Stipeae and relatives thereof may be difficult to force open, and re-expression of the third lodicule therefore may have been favored in these groups.

3. Stamen Number

Judging by the positions of stamens in alternating whorls, Clifford (1961) suggested that stamen number (char. 19) is reduced to three in the grasses from a putatively ancestral number of six, by loss of the inner whorl. Consistent with that suggestion, the presence of six stamens appears to be plesiomorphic in the grasses. This state changed to three stamens prior to the divergence of the PACC clade and Pooideae. This event occurred either just after the divergence of Pharus (under Alternative Topology 2) or just after the divergence of Ehrhartoideae (under Alternative Topology 1). The inner whorl of stamens was regained in some Bambuseae in either case, and in Ehrhartoideae (under Alternative Topology 2). In early-diverging genera not sampled in the present analysis stamen number has increased further; Pariana of Parianeae has 6-40 stamens (Clayton & Renvoize, 1986; Watson & Dallwitz, 1992), and a few genera of Bambuseae may have as many as 120 stamens (Clayton & Renvoize, 1986).

4. Stigmatic Branching

A character that is particularly relevant to the evolution of wind pollination in grasses is the development of longer and higher orders of stigmatic branching (char. 22). Implications of this change in stigmatic structure for breeding system evolution are discussed below (section V.H). Here we discuss the character states and their distribution among taxa. Heslop-Harrison and Shivanna (1977) defined plumose stigmas as those with receptive cells dispersed on multiseriate branches. In their interpretation most stigmas in grasses are truly plumose, and those of Flagellaria, which have receptive cells concentrated in distinct ridges, zones or heads, are not. Stigmas of Joinvillea are similar to those of Flagellaria. Although Joinvilleaceae have been described as having "plumose" stigmas (Cronquist, 1981; Dahlgren et al., 1985), illustrations (Engler & Prantl, 1888; Dahlgren et al., 1985) show an absence of branches made up of series of receptive cells. Stigmas of Restionaceae, and the related families Anarthriaceae D. F. Cutler & Airy Shaw, Centrolepidaceae Endl., Ecdeiocoleaceae D. F. Cutler & Airy Shaw, and Eriocaulaceae Desv., generally are more narrow and elongate in their primary axis (as in grasses) than those of Flagellaria or Joinvillea (see illustrations in Engler & Prantl, 1888; Black, 1960; Dahlgren et al., 1985). Although stigmas in these five families do have short lateral branches in some cases, illustrations of their stigmas either show branches that are not made up of series of receptive cells, or do not clearly show that they are. An illustration of Baloskion (in Black, 1960) falls into the latter category. For this reason, we scored stigmas of Flagellaria and Joinvillea as simple, and those of Baloskion as ambiguous. We further restrict the definition of "plumose" in our analysis and discussion to stigmas that have elongate branches and thus a nonlinear appearance (i.e., states 2 and 3 of char. 22). Thus, with simple stigmas also present in Anomochloa, Streptochaeta, and Pharus, stigmas appear to have been simple at the point of origin of the grasses.

The occurrence of simple stigmas in certain early-diverging herbaceous lineages has been interpreted as a reversal from plumose stigmas (Soderstrom, 1981: 44). However, the present analysis suggests the opposite. Eremitis has simple stigmas, as do the three earliest diverging lineages of grasses. Under either of the Alternative Topologies, broadly plumose stigmas arose twice after the divergence of Phareae, and were plesiomorphically simple in Eremitis, or were lost once and then regained in Eremitis. A factor that merits consideration at this point is the presumed monophyly of Parianeae and the nature of stigmas in other members of that tribe. The combination of a spiciform inflorescence with a fragile rachis, bearing verticils of 4-6 spikelets, the spikelets with single unisexual flowers, the pistillate ones disarticulating below the glumes, along with a set of other unifying features, is good evidence that Parianeae represent a natural group (Clayton & Renvoize, 1986; Hollowell, 1987). However, Pariana has stigmas that are distinctly broadly plumose, and introrse (Hollowell, 1987), the sparsely arranged primary branches originate on the inner side of the main axes and are not rebranched, and the receptive cells express papillae only at the branch tips (pets. obs.). The condition in Pariana may represent a novel state or an intermediate state between simple stigmas and the more frequent plumose condition in grasses. As such, it remains unclear whether the simple stigma in Eremitis is plesiomorphic or apomorphic within Parianeae.

In the phylogenetic structures resolved by simultaneous analysis of the two data sets, stigmas with branches made up of elongate series of receptive cells arise one or more times in the remainder of the grasses ([ILLUSTRATION FOR FIGURE 3 OMITTED], around node 5). Secondary stigmatic branching is well developed in a few groups of grasses and is synapomorphic for tribe Meliceae ([ILLUSTRATION FOR FIGURE 3 OMITTED], node 40), where secondary branching is highly developed and tertiary branching occurs. Stigmas revert to the simple state in Nardus, Sesleria, and Sporobolus and in many genera of the PACC clade and Pooideae not sampled in this study.

5. Fruit Characteristics

Embryo and caryopsis characters tend to be phylogenetically stable in the grasses. Although such characteristic grass features as spikelets and lodicules may not have been present in the earliest grasses (see above), the grass-type fruit, with a lateral and well-developed embryo (char. 25), is resolved as a synapomorphy of the grasses ([ILLUSTRATION FOR FIGURE 3 OMITTED], node 2). The four characters (chars. 26-29) of the embryo formula of Reeder (1957, 1962) all provide synapomorphies for the PACC clade and Pooideae, and for subclades within these two major groups. However, they were of little use in determining relationships among these clades and the earlier-diverging lineages, at least with the present sampling, for they are either optimizable at different nodes around the nexus of these lineages in most-parsimonious trees or they are apomorphies of lineages within the PACC clade or Pooideae. Epiblast occurrence (char. 26) is quite homoplasious (8 steps, CI = 0.12, RI = 0.46). Epiblasts are present in Anomochloa and absent in Streptochaeta. They are absent in Ehrharta, Arundinoids, Uniola L., and Panicoideae (except Andropogon, which was not sampled), but are present in other early-diverging lines, Chasmanthium, many Chloridoideae, and nearly all Pooideae (they are absent in Bromus, and are polymorphic in Brachypodium and Hordeeae). The absence of a scutellar tail (char. 27, 2 steps, CI = 0.50, RI = 0.96) is a synapomorphy of Pooideae or of Brachyelytrum plus Pooideae. A scutellar tail is present in all early-diverging lineages that have been studied, except Pharus and some Oryzeae, and it is variable in occurrence in Brachyelytrum and Oryza. A scutellar tail has been found to be present in all but two genera of the PACC clade studied. Mesocotyl internodes elongate (char. 28, 1 step, CI = 1.0, RI = 1.0) is a synapomorphy of the PACC clade, where it occurs in all but a few genera of Panicoideae (not sampled). The internodes are short or negligible among Pooideae and among the early-diverging lines in which they have been observed (except Ehrharta in part). Under Alternative Topology 2, the occurrence of embryonic leaf margins that do not overlap (char. 29, 3 or 4 steps, CI = 0.25-0.33, RI = 0.82-0.88) is a synapomorphy of the PACC clade plus Pooideae. Embryonic leaf margins overlap in embryos of taxa in the clade that consists of Centotheceae plus Panicoideae and in all lineages that diverge earlier than the point of divergence of Pooideae and the PACC clade other than Anomochloa, and Potamophila R. Br. of Oryzeae (not sampled).

A linear elongate hilum (char. 24) is plesiomorphic in the grasses. It is characteristic of all lineages that diverge before the point of divergence of Pooideae and the PACC clade, but variable within Pooideae and the PACC clade (7-8 steps, CI = 0.12-0.14, RI = 0.68-0.72). Hilum shape changes from long and linear to short relative to the caryopsis length and broad or punctiform within the PACC clade ([ILLUSTRATION FOR FIGURE 3 OMITTED], node 15) and then is reversed among some Arundinoids. It is also narrow and elongate in some genera of Panicoideae not sampled in this study. It also reverses from short to long within Pooideae within tribes Poeae and Agrostideae.

6. First Seedling Leaf

Absence of a blade on the first seedling leaf(char. 32) may be plesiomorphic in the Poaceae. The character state is not known for Joinvillea or Anomochloa, and the blade is absent in Flagellaria and present in Baloskion and other Restionaceae (B. Briggs, pers. comm.; H. P. Linder, pers. comm.). Among other early-diverging lineages, a blade on the first true leaf is known to occur only in Pharus, Ehrharta, Streptogyna, and Zizania L. (the latter two genera not sampled in the present analysis). The blade is absent for all genera of Bambuseae and early-diverging herbaceous lineages surveyed by Hoshikawa (1969), Soderstrom (1981), and Judziewiez and Soderstrom (1989). In Alternative Topology l, gain of the blade represents a synapomorphy for the clade that includes Brachyelytrum, Pooideae, and the PACC clade, and it is present in all taxa in this group for which the character has been observed (i.e., above node 13 in [ILLUSTRATION FOR FIGURE 3 OMITTED]). Under Alternative Topology 2, in which subfam. Ehrhartoideae is the sister group of this clade, gain of the blade can be optimized parsimoniously as a synapomorphy of this clade, or as a synapomorphy of the larger clade that also includes Ehrhartoideae, in which case absence of the blade in Oryza and Leersia represents a loss. Further investigation of seedling leaf characters including presence/absence, angle of divergence, and width are needed (Kuwabara, 1961). Erect, narrow first seedling leaves may be a synapomorphy for Ehrhartoideae if that group is circumscribed to include Ehrharta, Oryzeae, and Streptogyna, or, as already noted, for this group plus Brachyelytrum, Pooideae, and the PACC clade.

7. Cross-Venation

As noted in Appendix 2, the potential for inconsistency in the scoring of this character is greater than with most others. If cross-venation is regarded as present only when it is frequent and regular in occurrence (i.e., tessellation), taxa with prominent but infrequent cross-venation must be scored as lacking cross-venation. On the other hand, if the rare to occasional presence of distinct cross-veins is regarded as sufficient for the recognition of cross-venation as present, additional taxa should be scored as having this feature. For these reasons, and because it appears that various authors have scored absence of the character somewhat inconsistently, the following statements should be regarded as tentative.

Presence of cross-venation of the leaf blade (char. 33) appears to be plesiomorphic in the grasses. Although we considered the character inapplicable for Baloskion and other Restionaceae, cross-venation is present in Flagellaria, Joinvillea, and the earliest-diverging grass lineages. It is characteristic of most other early-diverging lineages, but is absent from some Ehrhartoideae. Alternative Topology 1, with Brachyelytrum (which has cross-venation) sister of Pooideae, suggests that the loss of cross-venation arose independently in Pooideae and the PACC clade. Under Alternative Topology 2, loss of cross-venation can be interpreted as a synapomorphy of the clade that consists of Pooideae plus the PACC clade; however, the character still would be homoplasious by virtue of the occurrence of cross-venation within the PACC clade in five genera of Panicoideae, Thysanolaena, all genera of Centotheceae except Chasmanthium, and Melanocenchrus Nees of Chloridoideae, as well as within Pooideae, where it is known from Phaenosperma, Pseudobromus K. Schum. (neither of which was sampled), and a few other taxa. In this and in other respects, Pseudobromus, an African genus placed within Festuca by Clayton and Renvoize (1986) and Watson and Dallwitz (1992), is similar to Brachyelytrum (pets. obs.), which also has cross-venation, and it may be properly placed near that genus rather than in Festuca.

8. Microhairs

Multicellular microhairs (char. 34) can, but need not, be optimized as plesiomorphic in the Poaceae. They occur in Joinvilleaceae and are absent in Restionaceae, but the character is unobserved for Flagellariaceae (Appendix 2). Although multicellular microhairs have more than two cells in Joinvilleaceae and some early-diverging lineages (where they are bi- and multicellular), they are bicellular in most Poaceae. Bicellular microhairs are present in the Anomochlooideae and are widespread among other early-diverging lines but are absent in several of the tribes and genera in this region of the phylogeny: Pharus and Phareae [except Suddia Renvoize, which is excluded from the Phareae by Judziewicz (1987) and which was not sampled in the present study], Ehrharta, Streptogyna, and Porteresia (the latter two genera not sampled). In the PACC clade, they are absent in Spartina Schreb., Distichlis Raf., and Uniola, among taxa in the present sample, and sporadically elsewhere. In Alternative Topology 1 (Brachyelytrum sister of Pooideae) their absence is a synapomorphy within Pooideae for the clade that includes all members of the subfamily except Nardus and Lygeum ([ILLUSTRATION FOR FIGURE 3 OMITTED], node 31) or for the clade that includes Brachyelytrum and Pooideae if bicellular microhairs were lost initially and regained in the clade that consists of Nardus plus Lygeum. The absence of multicellular microhairs in Brachyelytrum is interpreted as an independent loss under Alternative Topology 2 (PACC clade sister of Pooideae), while under Alternative Topology 1 it can, but need not, be a synapomorphy of the clade that includes Brachyelytrum and Pooideae.

Bicellular microhairs with a short broad terminal cell relative to the basal cell, termed "chloridoid," occur in Lygeum, Amphipogon, and most Chloridoideae (except in Eragrostis Wolf in part, including the species sampled in this study). Those in Lygeum would appear to have arisen independently. The chloridoid microhairs in Amphipogon resemble those found in tribe Pappophoreae Kunth of Chloridoideae (Tateoka et al., 1959; Watson & Dallwitz, 1992), and chloridoid microhairs may be a synapomorphy of Amphipogon plus Chloridoideae.

9. Arm Cells

Arm cells (char. 36) are not known to occur among the nongrass taxa sampled in this analysis. Arm cells are present in all early-diverging lineages examined except for Streptochaeta, two genera of Phareae, a few of Olyreae, two genera of Bambuseae, most Ehrharta, and one genus of Oryzeae. This cell type is absent from Brachyelytrum, all Pooideae, and most members of the PACC clade (except Phragmites among the current sampling, and in a few other unsampled genera of Arundinoids, Chloridoideae, and Panicoideae). With the present sampling, loss of arm cells optimizes as a synapomorphy of the clade that includes Brachyelytrum, Pooideae, and the PACC clade ([ILLUSTRATION FOR FIGURE 3 OMITTED], node 13) under either of the alternative topologies, but more extensive sampling within the PACC clade might alter that result. Lynn Clark (pers. comm.) notes that highly invaginated arm cells apparently occur only in Olyreae, Bambuseae, and Parianeae, and suggests that they may be a synapomorphy of this set of tribes.

10. Fusoid Cells

Fusoid cells (char. 37) have been considered a uniting feature of Bambusoideae s.l. (Clayton & Renvoize, 1986; also see Kellogg & Campbell, 1987). However, this assemblage is not resolved as monophyletic in the present study, and there is evidence that these cells occur outside the grasses, in Joinvillea (Smithson, 1957), so the presence of fusoid cells may be a plesiomorphy of the grass family. Within the grasses, loss of fusoid cells in Alternative Topology 1 (early-diverging Ehrhartoideae) is a parallel occurrence in Ehrhartoideae and in the clade that includes Brachyelytrum, Pooideae, and the PACC clade, while in Alternative Topology 2 (late-diverging Ehrhartoideae), it is a synapomorphy of the clade that includes all four of these groups. Among early-diverging lineages, fusoid cells are absent in Phyllorhachideae (not sampled) and in two genera of Bambuseae (not sampled), and are variable in occurrence in Ehrharta and Oryzeae, though absent in all taxa sampled here. Fusoid cells are absent in Brachyelytrum and Pooideae but infrequently present in the PACC clade (in Hakonechloa Makino ex Honda, Viguierella A. Camus, Homolepis Chase, and Streptostachys Desv., none of which are sampled in the present study). Fusoid cells in some members of Panicoideae may be structurally different from those in other lineages (L. Clark, pers. comm.) and further investigation is warranted.

11. Photosynthetic Pathway

[C.sub.4] photosynthesis (chars. 38, 39) in the grasses is known only in members of the PACC clade. The results of the present analysis suggest that [C.sub.4] photosynthesis has originated three times, once each in Aristida, Chloridoideae, and Panicoideae. Some form of [C.sub.4] photosynthesis occurs in Aristida, a few other Arundinoids, all hloridoideae reported in the literature (except Eragrostis walteri Pilg.), and in most Panicoideae. Among Arundinoids, it is reported in Aristida and Stipagrostis Nees of tribe Aristideae, and in Centropodia and Alloeochaete C. E. Hubb. of tribe Arundineae sensu Clayton and Renvoize (1986). Barker et al. (1995) resolved Centropodia as sister group of tribe Pappophoreae of Chloridoideae, and on the basis of this placement Centropodia could be placed within Chloridoideae. Tzvelev (1989), without mentioning references, cites a few additional Arundinoid genera as [C.sub.4], but H. P. Linder, R. P. Ellis, and G. Davidse (pers. comm.), have confirmed that these genera are [C.sub.3]. In Panicoideae, Ca is constant within Saccharineae (i.e., Andropogoneae) and Eriachneae (Owhi) EckBoorsb., but is variable in occurrence in Paniceae and Arundinelleae Stapf and is absent in Hubbardieae C. E. Hubb. and Isachneae Benth. (Clayton & Renvoize, 1986; Hattersley & Watson, 1992). Thus it may have evolved more than once in Panicoideae. [C.sub.4] subtypes were of little significance as characters in the present study, in which sampling of [C.sub.4] groups is limited. However, the results are consistent with the hypothesis of Hattersley and Watson (1992) that the NAD-ME system originated before the PCK subtype in the Chloridoideae ([ILLUSTRATION FOR FIGURE 3 OMITTED], nodes 17 and 20). Other cladistic studies (Kellogg & Campbell, 1987; Barker et al., 1995; Clark et al., 1995) also indicate that Aristideae arose independently from the Paniceae and Chloridoideae. The structure resolved here supports suggestions that [C.sub.4] photosynthesis arose a minimum of three or possibly four times or more within the grass family (Brown, 1977; Clayton & Renvoize, 1986; Hattersley & Watson, 1992; Sinha & Kellogg, 1996).

12. Chromosome Numbers

Chromosome number diversity is extensive in Poales and in Poaceae. Chromosome number was not used as a character in the analysis because as a result of polyploidy, aneuploidy, dysploidy, and possible hybrid combinations, base chromosome numbers are ambiguous for many taxa (de Wet, 1954). In several early-diverging genera of the grasses, and in Pooideae, all or most species are polyploid and the scoring of base numbers would be unduly speculative. Although chromosome numbers were scored and were included in some provisional character analyses, the state of this character still was ambiguous at many nodes. For example, the plesiomorphic base number for Pooideae might be x = 7, 8, 9, 10, 11, 12, or 13, and that of Poaceae might be x = 11 or 12. The outgroups provide ambiguous evidence in this regard: The chromosome number of Flagellaria is n = 19, that of Joinvillea is n = 18, and those of Restionaceae s.l. have been interpreted as consistent with x = 7, 8, 9, 11, 12, and 13 (Dahlgren et al., 1985). Joinvillea and Anomochloa share the gametic chromosome number of n = 18 (Hunziker, 1989), and Flagellaria, Diarrhena, and Miscanthus Andersson share the gametic chromosome number n = 19, but it is not clear if these are base numbers. When chromosome numbers are overlaid on most-parsimonious trees, Olyreae are united by x = 11; Ehrhartoideae by x = 12; Chasmanthium and Arundinoids have x = 12 [except Aristida (x = 11), Danthonia (x = 97), Molinia (x = 9), and possibly Amphipogon (x unknown)]; Chloridoideae are plesiomorphically x = 10; and within Pooideae, Brachypodieae and the "core" Pooideae are each united by x = 7.

E. SYNAPOMORPHIES OF VARIOUS GROUPINGS OF EHRHARTOIDEAE, POOIDEAE, BRACHYELYTRUM, AND THE PACC CLADE

In the consensus tree that summarizes most-parsimonious trees obtained by simultaneous analysis of both character sets, the polytomy at node 13 [ILLUSTRATION FOR FIGURE 3 OMITTED] results from conflict between Alternative Topologies 1 and 2. In the former, the PACC clade is the sister group to Pooideae plus Brachyelytrum; in the latter, Brachyelytrum is the sister group of Pooideae plus the PACC clade, and Ehrhartoideae is the sister group of the clade that includes all three of those taxa. Resolution of any group that includes Pooideae and the PACC clade but does not include Ehrhartoideae (or other early-diverging lineages, including Bambuseae) is in conflict with the resolution of a group that includes Pooideae and Ehrhartoideae (i.e., Oryzoideae) with Bambuseae, Olyreae, or Parianeae (i.e., the "BOP" clade resolved by Clark et al., 1995) but excludes the PACC clade. The structural synapomorphies of clades that are resolved by the present analysis, and that impinge on this question, are detailed in the following paragraphs. Character state transformations that are optimizable to other internodes are listed after unambiguous synapomorphies of the various groups.

Unambiguous synapomorphies of the clade that includes Brachyelytrum, Pooideae, and the PACC clade in all six trees with alternative topology 1 are the following: loss of pseudopetioles (char. 6), gain of awn or mucro (char. 10), loss of stylar fusion (char. 20), gain of lamina of first seedling leaf (char. 32), loss of arm cells (char. 36), and loss of fusoid cells (char. 37). Three additional characters may be synapomorphies of this group, but can be optimized equally parsimoniously elsewhere: presence of two lodicules (char. 15), embryonic leaf margins meeting rather than overlapping (char. 29), and loss of cross-venation in leaf blades (char. 33). There is only one unambiguous synapomorphy of this clade in the six trees with Alternative Topology 2, and it is one of the six that are unambiguous synapomorphies under Alternative Topology 1: loss of arm cells (char. 36). A potential synapomorphy of this group is gain of lamina of first seedling leaf (char. 32).

Unambiguous synapomorphies of the clade that includes Pooideae and the PACC clade (present only in Alternative Topology 2) are embryonic leaf margins meeting rather than overlapping (char. 29), and loss of cross-venation in leaf blades (char. 33); both of these character transformations are potential synapomorphies, in Alternative Topology 1, of the larger clade that also includes Brachyelytrum.

There are five unambiguous synapomorphies of the clade that includes Ehrhartoideae, Brachyelytrum, Pooideae, and the PACC clade (present only in Alternative Topology 2): loss of pseudopetiole (char. 6), gain of awn (char. 10), change in lodicule number from 3 to 2 (char. 15), loss of stylar fusion (char. 20), and loss of fusoid cells (char. 37). Two additional characters can, be but need not, be optimized on this internode: gain of disarticulation above the glumes (char. 12) and gain of lamina of first seedling leaf (char. 32).

Under Alternative Topology 1, in which Ehrhartoideae is not resolved as a close relative of Brachyelytrum, Pooideae, and the PACC clade, there are six structural characters that arise in parallel between these groups, either as unambiguous synapomorphies or as potential synapomorphies: loss of pseudopetiole (char. 6), gain of awn (char. 10), change in lodicule number from 3 to 2 (char. 15), loss of stylar fusion (char. 20), gain of lamina of first seedling leaf (char. 32), and loss of fusoid cells (char. 37). These six characters are elements of the syndrome of structural features that support a close relationship between Ehrhartoideae, Brachyelytrum, Pooideae, and the PACC clade (i.e., Alternative Topology 2), to the exclusion of Bambuseae and other early-diverging lines, and that are in conflict with Alternative Topology 1 as well as with the resolution of a BOP clade. Analyses that resolve relationships in conflict with these characters should account for parallelism in these characters.

F. PLACEMENT OF FIVE PROBLEMATIC TAXA WITHIN POOIDEAE

The distribution of character states in the present analysis, along with subsidiary character state data, provide evidence for the placement of Anisopogon in Pooideae, along with four additional orphan genera not sampled in this study. We suggest that Anisopogon, placed in Arundinoideae by Clayton and Renvoize (1986) and Watson and Dallwitz (1992), and Danthoniastrum, Duthiea, Metcalfia, and Pseudoclanthonia, placed in Arundinoideae by Watson and Dallwitz and in Pooideae tribe Aveneae (i.e., Agrostideae) subtribe Duthieinae by Clayton and Renvoize, have characteristics generally diagnostic for Pooideae and not Arundinoideae. In all five of these genera, the species for which character scores are available have membranous, nonciliate ligules; absence of bicellular microhairs (except possibly Duthiea; see Watson & Dallwitz, 1992); two or three lodicules that are lanceolate, unlobed, and non- or weakly vascularized (including Duthiea, pers. obs., contra Clayton & Renvoize, 1986, and Watson & Dallwitz, 1992); long-linear hila; hairy ovaries; and small embryos. The first of these characters, membranous ligules without ciliate margins, is rare among Arundinoids (occurring only in Arundo L., Dichaetaria Nees ex Steud., Elytrophorus P. Beauv., Hakonechloa, Monachather, and Thysanolaena). Bicellular microhairs are present in most Arundinoideae but occur in Pooideae only in Nardus and Lygeum. If these five genera are excluded from the Arundinoids, taxa with three lodicules, or Pooideae-type lodicules (except in Aristideae), are absent in the PACC clade. Lanceolate, membranous-tipped, non- or weakly vascularized lodicules are both plesiomorphic and common in Pooideae, and the three-lodicule state is present among early-diverging lineages of the Pooideae (Stipeae and Ampelodesmos), while these lodicule characters (except the infrequent occurrence of weak vascularization) are rare or absent among Arundinoids. Long-linear hila are plesiomorphic and common in Pooideae, including the early-diverging members, and uncommon among Arundinoids. Hairy ovaries are frequent in Pooideae but rare among Arundinoids (present only in Dregeochloa Conert and Pentameris P. Beauv.). As in Pooideae (except Cyathopsis Stapf), embryos are small in Anisopogon, Danthoniastrum, Duthiea, Metcalfia, and Pseudodanthonia (pers. obs.), whereas embryos among Arundinoids are mostly large. Embryo formula characters, which are of major importance in distinguishing the PACC clade from the Pooideae, are known only for Metcalfia. That genus has formula F + FP (i.e., embryo mesocotyl internode negligible, epiblast present, scutellar tail absent, embryonic leaf margins overlapping), a pattern typical of Oryzeae, not unknown in Pooideae, and unrecorded within the PACC clade. A further piece of evidence favoring placement in Pooideae is the occurrence of large chromosomes of base chromosome number x = 7 in Duthiea, a combination of size and base number otherwise unknown from the PACC clade. Although most of the above features are homoplasious in both subfamilies, the presence of this suite of synapomorphic or plesiomorphic characteristics of the Pooideae, and absence of any characters diagnostic for the PACC clade, suggests that Anisopogon, Danthoniastrum, Duthiea, Metcalfia, and Pseudoclanthonia should be recognized as early-diverging elements of Pooideae.

G. TAXONOMIC CONCLUSIONS

The analyses presented here suggest that among extant grasses, a clade consisting of Anomochloa and Streptochaeta diverged first from the lineage that includes all other grasses, and that Pharus was the next to diverge; further diversification resulted in one of two possible phylogenetic structures: 1) an "early-diverging" Ehrhartoideae, followed by Parianeae, Bambuseae, Olyreae, the PACC clade, and then Brachyelytrum from Pooideae; or 2) a clade consisting of Parianeae plus Bambuseae, followed by Olyreae, a "late-diverging" Ehrhartoideae, Brachyelytrum, and then the PACC clade from Pooideae. These alternatives are equally well supported by the combined data set. The second structure, however, is more strongly supported by the analysis of complex character structural data, as opposed to simple restriction site changes. Given the general correspondence between the phylogeny resolved by restriction sites and by simultaneous analysis of both data sets, and those resolved by other molecular studies, we propose the following phylogenetic structure for the grasses:

Anomochlooideae (Pharoideae ((Parianeae Bambuseae Olyreae) (Ehrhartoideae (Brachyelytreae Pooideae (Aristidoideae (Chloridoideae [including some Arundinoids] (Arundineae s.str. (Thysanolaeneae Centotheceae Panicoideae)))))))).

The present results thus are consistent with the recognition of Anomochlooideae and Pharoideae (Clark & Judziewicz, 1996). In light of the incongruence between the present analysis and that of ndh F concerning relationships among other major clades, it is reasonable to retain a Bambusoideae that is provisionally circumscribed to include Parianeae, Olyreae, and Bambuseae, until additional phylogenetic evidence becomes available. Clark et al. (1995) accepted subfam. Oryzoideae, and the present results are consistent with the recognition of that group with the name corrected to Ehrhartoideae. Acceptance of subfam. Ehrhartoideae would not affect the monophyly of other subfamilies if Streptogyna was also removed from Bambusoideae and placed in Ehrhartoideae (Clark et al., 1995). Brachyelytrum may be submerged in Pooideae (if future studies confirm its placement as sister of an already broadly circumscribed Pooideae) or recognized as a separate subfamily if confirmed to be the sister group of Pooideae plus the PACC clade. Either of these taxonomic placements would be compatible with results of the present analysis, but the former is consistent with the analyses of Kellogg and Linder (1995) and Clark et al. (1995), while the latter is not. Phaenosperma (according to Clark et al., 1995) and Diarrhena (present results) both appear to belong within Pooideae. Further, the present results agree with Clayton and Renvoize (1986) in the placement of certain orphan taxa such as Ampelodesmos, Anisopogon, Lygeum, Nardus, Stipeae, and Duthieinae in Pooideae.

The entire PACC clade (i.e., Aristidoideae Cato and all the names that follow it in the preceding list) could well be subsumed within a broadly defined Panicoideae, as proposed by Clark et al. (1995). However, an alternative approach, which would involve a less dramatic change to current classifications, would be to recognize Aristidoideae, to assign Centotheceae to Panicoideae, to apportion some of the remaining Arundinoids to Chloridoideae and Panicoideae, and to recognize one or more additional subfamilies that accommodate any monophyletic groups of Arundinoids that do not fall within the more broadly circumscribed Panicoideae and Chloridoideae. This proposal is consistent with the rbcL analysis of Barker et al. (1995) and with the ndhF analysis of Clark et al. (1995), which suggests that Panicoideae sensu Clayton and Renvoize (1986), Tzvelev (1989), or Watson and Dallwitz (1992), are polyphyletic unless they include, at minimum, Centotheceae and Thysanolaena, and that Micraria and Amphipogon can be accommodated within the Chloridoideae (Clark et al., 1995; and the present analysis, respectively). Despite the incongruous placement of Danthonia in the present study, we suggest that the comprehensive analyses of Arundinoids by Verboom et al. (1994) and Barker et al. (1995) be followed (provisionally), in placing the Danthonioid taxa in the Chloridoideae.

H. BREEDING SYSTEMS IN GRASSES

1. Evolution of Wind Pollination in Grasses

The locations in the present phylogenetic trees of transformations in the degree of stigmatic branching, from simple in Anomochloa, Streptochaeta, Phareae, and Parianeae to plumose, are suggestive of corresponding changes among early grasses from insect to wind pollination. Plumose stigmas function more efficiently as traps for airborne pollen than do simple stigmas (Niklas, 1985), and they are clearly advantageous for wind pollination. Stigmatic surfaces are simple and are distributed on short styles in Joinvillea and Flagellaria, the state is ambiguous in Baloskion, and the surfaces are simple or plumose in the Poaceae (see section V.D.4; Appendix 2, char. 22). The relatively simple style surface found in the earliest-diverging grasses in the present analysis is correlated with other features that suggest that the species possessing them are dependent, at least in part, upon insect pollination. Moreover, there is evidence that some degree of entomophily is present in some early-diverging lineages with plumose stigmas.

Insect pollination is documented in some early-diverging herbaceous lineages (Olyra L. and Piresia Swallen, of Olyreae, and Pariana and Eremitis, of Parianeae; Soderstrom & Calderon, 1971, 1979; Salgado-Labouriau & Rinaldi, 1990; Salgado-Labouriau et al., 1992) and is implicated for some species of Chusquea, Bambusa (Janzen, 1976), Dendrocalamus Nees, and Schizostachyum Nees, of Bambuseae (Chapman, 1990). Observations of insect pollination in bamboos have tended to be anecdotal, and further study is warranted. Although representatives of most of the early-diverging grass lineages appear to be wind pollinated (Soderstrom & Calderon, 1971), little is known of pollination modes in Anomochloa, Streptochaeta, or Pharus. However, the occurrence of simple stigmas and stamen filaments fused and exserted terminally from among the surrounding bracts in these three genera suggest insect rather than wind pollination (Soderstrom, 1981: 35).

Page (1978) found a correlation among grasses between degree of pollen exine sculpturing and the taxonomic distribution of states of this character. That correlation is still apparent when evaluated using the phylogeny of grasses resolved here. Nearly all of the earlydiverging lineages compared in Page's study had grouped granules, and half of these had high surface relief. Although only two had separate granules, and none had separate granules and low relief, one-third of the members of the PACC clade and Pooideae had separate granules and two-thirds had low relief. Species known or suspected to be insect pollinated on other grounds had grouped granules (Anomochloa, Streptochaeta, Leptaspis R. Br. [Phareae], Dendrocalamus, Olyra) or high exine reliefs (Olyra, Pariana, Piresia). Species of Pooideae and the PACC clade tend to have widely spaced granules of low relief. Thus, a functional connection may exist between insect pollination and pollen with fused granules or high exine relief, and between wind pollination and pollen with low exine relief.

Joinvillea and Flagellaria have perfect flowers, occur in mesic tropical forests often in the shade, and probably are insect pollinated to some degree (Newell, 1969). Cronquist (1981) indicated that Flagellaria is apparently anemophilous, but did not comment on the pollination mode of Joinvillea. Newell (1969) reported ants as possible pollinators of Flagellaria, and suggested that bees and/or wind may pollinate Joinvillea. Species of Restionaceae are mostly monoecious or dioecious (including Baloskion), and they often inhabit wetlands and are wind pollinated (Black, 1960; Cronquist, 1981; Dahlgren et al., 1985). Thus, there is no compelling evidence from pollination mode in the closest relatives of grasses that wind pollination was fixed in the ancestor of the grasses, and some evidence to suggest that entomophily was plesiomorphic and possibly supplemented by wind pollination.

Relationships of taxa and taxonomic distribution of simple stigmas suggested by the present analysis, supplemented with data on floral emergence, insect visitation, and pollen sculpturing, suggest that the earliest grasses were at least partially entomophilous. It can further be suggested that grasses only became dependent on wind pollination some time after the divergence of Anomochloa, Streptochaeta, Phareae, and possibly Parianeae.

2. Evolution of Self-Incompatibility in Grasses

The distribution of monoecy, protogyny, and self-incompatibility in grasses, when compared with the phylogenetic structures resolved here, are suggestive of the possibility that self-incompatibility evolved only once or twice in the grasses. Furthermore, an initial incompatibility system may have arisen in the ancestor of the clade that comprises Brachyelytrum, Pooideae, and the PACC clade, and separately in Ehrhartoideae; or, if these two clades are sister groups, it may have evolved in their most recent common ancestor.

It is possible that both protogyny and monoecy preceded the evolution of self-incompatibility. Protogyny is documented in several early-diverging herbaceous lineages: Anomochloa and Streptochaeta (Judziewicz & Soderstrom, 1989), Pharus and Leptaspis (Phareae; Judziewicz, 1987), and Zizania (Oryzeae; Conner, 1987). Monoecy is ubiquitous in species of Phareae, Olyreae, Parianeae, Phyllorachideae, and Puelia Franch. (Bambuseae, or given separate tribal status; Soderstrom & Ellis, 1987), and occurs in several genera of Oryzeae (Clayton, 1990).

The complex "S, Z incompatibility" system, called "the grass system," has been demonstrated only sporadically, and then only in genera of core tribes of the Pooideae (see review of pertinent literature in Chapman, 1990). A complex diallelic system such as this may be broken down easily, but it is unlikely to have evolved multiple times. Although the nature of the incompatibility system has not been determined in other tribes and subfamilies, self-incompatibility is widely documented in grasses. Outside of the core Pooideae, self-incompatibility is known in species of Meliceae and Stipeae of Pooideae, and in the PACC clade in Arundineae, Paniceae, Sacchareae (i.e., Andropogoneae), Chlorideae (i.e., Cynodonteae), and Eragrostideae (Conner, 1979, 1981). Given this wide distribution in the clade that comprises Brachyelytrum, Pooideae, and the PACC clade, it is reasonable to suggest that their most recent common ancestor was self-incompatible.

There is limited evidence regarding self-compatibility among early-diverging lines. Selfincompatibility is known in Oryza and Ehrharta (Conner, 1979), and suggested for Phyllostachys Siebold & Zucc. of Bambuseae (Chapman, 1990, and references therein). The suggestion of self-incompatibility in Phyllostachys may not be reliable, however, as many Bambuseae take up to three years from initiation of flowering to setting of mature fruit (L. Clark, pers. comm.). Self-compatibility is apparent in Anomochloa, Streptochaeta, and Pharus (L. Clark, pers. comm., from observations of seed set in greenhouse situations), Eremitis (Hollowell, 1987), Zizania, Oryza, and Ehrharta (Conner, 1979). Coincident with known occurrences of self-compatibility among early-diverging herbaceous lineages, given the relationships resolved here, it appears that several of these groups (Phareae, Parianeae, Olyreae, and possibly Ehrhartoideae) share the occurrence of unisexual florets as a synapomorphy. That the occurrence of unisexual flowers is negatively correlated with the occurrence of self-incompatibility (Conner, 1979; Chapman, 1996) suggests that most of these taxa may be self-compatible. The hypothesis that best fits the data on self-incompatibility in the context of the phylogeny resolved here is that self-incompatibility evolved late in the diversification of grasses, just prior to the divergence of Ehrhartoideae from the clade that consists of Brachyelytrum, Pooideae, and PACC clade, if Ehrhartoideae are the sister group of that clade, or independently in each lineage if these are not sister groups.

I. BIOGEOGRAPHY

Grasses are widely believed to have evolved in Gondwana during the late Cretaceous (Clayton, 1975; Stebbins, 1981, 1987; Linder, 1987), and Clayton (1975) suggested that they had diversified to the tribal level by the end of the Cretaceous. However, as there are no confirmed grass fossils from the Cretaceous, a minimum age for the origin of grasses is Paleocene. According to Muller (1981), pollen of Poaceae has not been confirmed from earlier than the Paleocene, but by the end of this epoch grass pollen occurred in Brazil, Africa, and Australia [see comments by Linder (1987) and Crepet and Feldman (1991) on pollen of grasses from the Cretaceous]. Grass pollen was widespread by the end of the Eocene but not abundant until the Oligocene (Muller, 1981). Only a few confirmed grass macrofossils date from the early Tertiary, in the Paleocene, Eocene, and Oligocene (Litke, 1968; Galbreath, 1974; Muller, 1981; Thomasson, 1987; Crepet & Feldman, 1991). Miocene and younger fossils have been found in abundance and often are referable to modern genera [Muller, 1981; Thomasson, 1987; Poinar & Columbus, 1992 (a corrected date for the latter authors' fossil was established by Iturralde-Vinent & MacPhee, 1996)].

Modern representatives of the earliest-diverging grass lineages detected in the present phylogenetic analysis are Anomochloa and Streptochaeta. If grasses did arise in Gondwana in the Cretaceous, representatives of these or earlier grass lineages have since died out on the African continent (or have yet to be discovered). If this is the case, the occurrence of the above and other early-diverging groups in northeastern South America may simply represent survival in a refugium and not the center of origin of the family (Soderstrom & Calderon, 1974; Soderstrom, 1981).

However, as these two genera are known only from the New World, and there are no confirmed Cretaceous grass fossils, it could also be postulated that grasses first arose in South America in the Paleocene. In any case, grasses could have been dispersed between South America and Antarctica/Australia via the land bridge that persisted between those continents well into the early Tertiary (Clayton, 1975; Stebbins, 1981). If grasses did originate in South America during the Paleocene, they may have been dispersed from South America to Eurasia and Africa via a North American land bridge, as has been postulated in the reverse direction for Leguminosae Juss. (Lavin & Luckow, 1993).

The earliest confirmed grass fossil was collected from North American strata assigned to the Paleocene/Eocene boundary (Crepet & Feldman, 1991), and more recently assigned to the early Eocene (S. Wing, pers. comm.). This fossil was reported to have racemose inflorescence fragments, spikelets with two glumes that are keeled, two toters (the bracts are presumed to be lemmas and paleas but are not well preserved), and a total of six exserted dorsifixed stamens. No evidence of gynoecia was found. Although the fossils could not be definitively assigned to any tribe of grasses, they were said to be "suggestive of Aveneae [i.e., Agrostideae, but the reference is to the two-flowered members of that tribe]. . . or Arundineae" (Crepet & Feldman, 1991: 1013). Given the ambiguity of the floret bract identification, and absence of evidence of gynoecia, these spikelets might have contained one floret with six stamens or two florets with three stamens each, and the floret or florets may have been bisexual or unisexual. We note that single-flowered, unisexual spikelets with six stamens are common among earlydiverging herbaceous grass lineages, and that this combination of characteristics is not known in the Pooideae or the PACC clade. The associated impression fossils of slender roots, shoots, and base of the plant suggest that the plant was a herbaceous perennial (Crepet & Feldman, 1991: [ILLUSTRATION FOR FIGURE 4 OMITTED]). These vegetative parts suggest that the grass was a member of Streptogyneae, Olyreae, Ehrhartoideae, the PACC clade, Pooideae, or an immediate ancestor of these, and that it was not a member of any earlier-diverging extant lineage.

Paleontologists now believe that at the Cretaceous/Tertiary boundary, ca. 65 million years ago, many lineages went extinct as a result of a widespread inhospitable climate attributable to an impact event and/or volcanic activity (Alvarez et al., 1980, 1982; Hallam, 1987; Srivastara, 1994). The collapse of complex ecosystems may have set the stage for a rapid and widespread establishment of nonspecialist organisms with short generation times. Tropical and subtropical forest ecosystems did rebound in the Paleocene, but by the Eocene (or somewhat later; S. Wing, pers. comm.) more arid and cooler climates were spreading, and these changes favored the emergence of cold- and drought-tolerant vegetation, including savanna (Wolfe, 1975, 1977, 1978; Stebbins, 1981). These vegetational shifts are correlated with the rise in South America of mammals with hypsodont teeth adapted to grazing (Stebbins, 1981). S. Wing (pers. comm.) notes that the timing of the rise of hypsodonts and of savanna in South America is more in line with that of North America than previously was thought, and thus was upper Oligocene to lower Miocene rather than lower or middle Eocene as reported by Stebbins. The available data are consistent with the hypothesis that the first grasses appeared in the earliest Tertiary, after forest landscapes were cleared by Cretaceous/Tertiary boundary events, survived and diversified in seasonally or aseasonally dry tropical and subtropical forests and wetland openings, and then emerged as dominants in the Oligocene and Miocene with the development of climatic conditions favorable to the evolution of savanna habitats.

VI. Acknowledgments

We thank Victoria Hollowell for assistance in the scoring of morphological characters of Eremitis and H. Peter Linder and Barbara Briggs for similar assistance with Baloskion/Restio Rottb. (though we take responsibility for the character scoring); Kelly Allred, Anne Bruneau, Nigel Barker, Lynn Clark, Keith Clay, Stephen Darbyshire, John LaDuke, H. Peter Linder, Paul Peterson, Steve Renvoize, Mary Sangrey, Hildemar Scholz, J. K. Wipff, and the curators of Fairchild Tropical Botanical Gardens, Kew Gardens, and the U.S.D.A. Plant Introduction Station at Pullman, Washington, for providing material for DNA extractions; Kevin Nixon for technical assistance and discussions of data analysis; Jeff Doyle, Jane Doyle, and Viviana Tharp for technical assistance; Jim Rawson and Michael Clegg, Catherine Bowman, and Jeff Palmer for providing clones of Permisetum Rich., Triticum L., and Nicotiana, respectively; Doug Goldman, Meng Keong Choo, and especially Karen Hansen for laboratory assistance; Scott Wing for discussion of Tertiary ecological events; Jim Reveal for updates of suprageneric nomenclature; and Lynn Clark for valuable comments on the contents of this paper. This study was supported by NSF grants BSR-8696101 and BSR-9006660 to JID, by a U.S.D.A. Plant Exploration Grant to JID and RJS, and by the L. H. Bailey Hortorium.

VII. Literature Cited

Alvarez, L. W., W. Alvarez, F. Asaro & H. V. Michel. 1980. Extraterrestrial cause for the Cretaceous-Tertiary extinction. Science 208: 1095-1108.

-----, -----, ----- & -----. 1982. Current status of the impact theory for the terminal Cretaceous extinction. Geol. Soc. Amer. Spec. Pap. 190: 305-315.

Arber, A. 1929. Studies in the Gramineae. VI. 1. Streptochaeta. 2. Anomochloa. 3. Ichnanthus. Ann. Bot. (London) 43: 35-53.

Barker, N. P. 1995. A molecular phylogeny of the subfamily Arundinoideae (Poaceae). Ph.D. dissertation, University of Cape Town, South Africa.

-----., H. P. Linder & E. H. Harley. 1995. Polyphyly in the Arundinoideae (Poaceae): Evidence from rbcL. Syst. Bot. 20: 423-435.

Baum, B. R. 1987. Numerical taxonomic analyses of the Poaceae. Pages 334-342 in T. R. Soderstrom et al. (eds.), Grass systematics and evolution. Smithsonian Institution Press, Washington, DC.

Bessey, E. A. 1917. The phylogeny of grasses. Rep. Michigan Acad. Sci. 19: 239-245.

Bews, J. W. 1929. The world's grasses: Their differentiation, distribution, economics, and ecology. Longmaas Green, London.

Black, J. M. 1960. Flora of South Australia. Part 1. Ed. 2. W. L. Hawes, Adelaide.

Bor, N. L. 1968. Flora of Iraq. Vol. 9, Gramineae. Ministry of Agriculture, Baghdad.

Bowman, C. M., B. Koller, H. Delius & T. A. Dyer. 1981. A physical map of wheat chloroplast DNA showing the location of the structural genes for the ribosomal RNAs and the large subunit of ribulose 1, 5-bisphosphate carboxylase. Molec. Gen. Genet. 183: 93-101.

Brown, W. V. 1977. The Kranz syndrome and its subtypes in grass systematics. Mem. Torrey Bot. Club 23: 1-97.

-----, W. F. Harris & J. D. Graham. 1959. Grass morphology and systematics: I. The internode. Southw. Naturalist 4: 115-125.

Butzin, F. 1965. Neue Untersuchungen uber die Blute der Gramineae. Ph.D. dissertation, Freie Universitat Berlin.

Campbell, C. S. & E. A. Kellogg. 1987. Sister group relationships of the Poaceae. Pages 217-224 in T. R. Soderstrom et al. (eds.), Grass systematics and evolution. Smithsonian Institution Press, Washington, DC.

-----, P. E. Garwood & L. P. Specht. 1986. Bambusoid affinities of the North American temperate genus Brachyelytrum (Gramineae). Bull. Torrey Bot. Club 113: 135-141.

Camus, E. G. 1913. Les Bambusees. Paul Lechevalier, Paris.

Caro, J. A. 1982. Sinopsis taxonomica de las Gramineas Argentinas. Dominguesia 4: 1-51.

Celakovsky, L. 1889. Uber den Ahrchenbau der brasilianische Grasgattung Streptochaeta Schrader. Sitzungsber. Konigl. Bohm. Ges. Wiss. Prag., Math.-Naturwiss. Cl. 3: 14-42.

Chapman, G. P. 1990. The widening perspective: Reproductive biology of bamboos, some dryland grasses and cereals. Pages 240-257 in G. P. Chapman (ed.), Reproductive versatility in the grasses. Cambridge University Press, Cambridge. -----. 1996. The biology of grasses. CAB International, Wallingford, UK.

Chase, M. W., D. W. Stevenson, P. Wilkin & P. J. Rudall. 1995. Monocot systematics: A combined analysis. Pages 685-730 in P. J. Rudall et al. (eds.), Monocotyledons: Systematics and evolution. Vol. 2. Royal Botanic Gardens, Kew.

Choo, Meng Keong, R. J. Soreng & J. I. Davis. 1994. Phylogenetic relationships among Puccinellia and allied genera of Poaceae and inferred from chloroplast DNA restriction site variation. Amer. J. Bot. 81: 119-126.

Clark, L. G. & E. J. Judziewicz. 1996. The grass subfamilies Anomochlooideae and Pharoideae (Poaceae). Taxon 45: 641-645.

----- & X. Londono. 1991. Miscellaneous new taxa of bamboo (Poaceae: Bambuseae) from Colombia, Ecuador and Mexico. Nordic J. Bot. 11: 323-331.

-----, Weiping Zhang & J. F. Wendel. 1995. A phylogeny of the grass family (Poaceae) based on ndhF sequence data. Syst. Bot. 20: 436-460.

Clayton, W. D. 1975. Chorology of the genera of Gramineae. Kew Bull. 30: 111-132.

-----. 1990. The spikelet. Pages 32-51 in G. P. Chapman (ed.), Reproductive versatility in the grasses. Cambridge University Press, Cambridge.

----- & S. A. Renvoize. 1986. Genera Graminum, grasses of the world. Kew Bull. XIII.

Clifford, H. T. 1961. Floral evolution in the family Gramineae. Evolution 15: 455-460.

-----. 1965. The classification of Poaceae: A statistical study. Papers, Dept. of Biology, University of Queensland 4: 243-253.

-----. 1987. Spikelet and floral morphology. Pages 21-30 in T. R. Soderstrom et al. (eds.), Grass systematics and evolution. Smithsonian Institution Press, Washington, DC.

-----, W. T. Williams & G. N. Lance. 1969. A further numerical contribution to the classification of the Poaceae. Austral. J. Bot. 17: 119-131.

Conner, H. E. 1979. Breeding systems in the grasses: A survey. New Zealand J. Bot. 17: 547-574.

-----. 1981. Evolution of reproductive systems in the Gramineae. Ann. Missouri Bot. Gard. 68: 48-74.

-----. 1987. Reproductive biology in grasses. Pages 117-132 in T. R. Soderstrom et al. (eds.), Grass systematics and evolution. Smithsonian Institution Press, Washington, DC.

Crepet, W. L. & G. D. Feldman. 1991. The earliest remains of grasses in the fossil record. Amer. J. Bot. 78: 1010-1014.

Cronn, R. C., X. Zhao, A. H. Paterson & J. H. Wendel. 1996. Polymorphism and concerted evolution in a tandemly repeated gene family: 5S ribosomal DNA in diploid and allopolyploid cottons, J. Molee. Evol. 42: 685-705.

Cronquist, A. 1981. An integrated system of classification of flowering plants. Columbia University Press, New York.

Cummings, M. P., L. M. King & E. A. Kellogg. 1994. Slipped strand mispairing in a plastid gene: rpoC2 in grasses (Poaceae). Molec. Biol. Evol. 11: 1-8.

Dahlgren, R. M. T., H. T. Clifford & P. F. Yeo. 1985. The families of the monocotyledons. Springer-Verlag, New York.

Darbyshire, S. J. & S. I. Warwick. 1992. Phylogeny of North American Festuca (Poaceae) and related genera using chloroplast DNA restriction site variation. Canad. J. Bot. 70: 2415-2429.

Davidse, G., M. Sousa S. & A. O. Charter. 1994. Flora Mesoamericana. Vol. 6, Alismataceae a Cyperaceae. Universidad National Autonoma de Mexico, D.F., Mexico.

Davis, J. I. 1993. Character removal as a means for assessing stability of clades. Cladistics 9: 201-210.

-----. 1995. A phylogenetic structure of the monocotyledons, as inferred from chloroplast DNA restriction site variation, and a comparison of measures of clade support. Syst. Bot. 20: 503-527.

----- & R. J. Soreng. 1993. Phylogenetic structure in the grass family (Poaceae) as inferred from chloroplast DNA restriction site variation. Amer. J. Bot. 81: 1444-1454.

-----, M. W. Frohlich & R. W. Soreng. 1993. Cladistic characters and cladogram stability. Syst. Bot. 18: 188-196.

-----, M. P. Simmons, D. W. Stevenson & J. F. Wendel. In press. Data decisiveness, data quality, and incongruence in phylogenetic analysis: An example from the monocotyledons using mitochondrial atpA sequences. Syst. Biol.

Davis, P. H. 1985. Flora of Turkey. Vol. 9. University Press, Edinburgh.

Decker, H. F. 1964. An anatomic-systematic study of the classical tribe Festuceae (Gramineae). Amer. J. Bot. 51: 453-463.

de Wet, J. M. J. 1954. The genus Danthonia in grass phylogeny. Amer. J. Bot. 41: 204-211.

Doebley, J., M. Durbin, D. M. Golenberg, M. T. Clegg & Din Pow Ma. 1990. Evolutionary analysis of the large subunit of carboxylase (rbcL) nucleotide sequence among the grasses (Gramineae). Evolution 44: 1097-1108.

Downie, S. R. & J. D. Palmer. 1992. Use of chloroplast DNA rearrangements in reconstructing plant phylogeny. Pages 14-35 in P. S. Soltis et al. (eds.), Molecular systematics of plants. Chapman & Hall, New York.

Doyle, J. D., J. I. Davis, R. J. Soreng, D. Garvin & M. J. Anderson. 1992. Chloroplast DNA inversions and the origin of the grass family (Poaceae). Proc. Natl. Acad. U.S.A. 89: 7722-7726.

Duvall, M. R. & B. R. Morton. 1996. Molecular phylogenetics of Poaceae: An expanded analysis of rbcL sequence data. Molec. Phylogenet. & Evol. 5: 352-358.

-----, M. T. Clegg, M. W. Chase, W. D. Clark, W. J. Kress, H. G. Hills, L. E. Eguiarte, J. F. Smith, B. S. Gaut, E. A. Zimmer & G. H. Learn Jr. 1993. Phylogenetic hypotheses for the monocotyledons constructed from rbcL sequence data. Ann. Missouri Bot. Gard. 80: 607-619.

-----, P. M. Peterson & A. H. Christensen. 1994. Alliances of Muhlenbergia (Poaceae) within New World Eragrostideae are identified by phylogenetic analysis of mapped restriction sites from plastid DNAs. Amer. J. Bot. 81: 622-629.

Ebinger, J. E. & J. L. Carlen. 1975. Culm morphology and grass systematics. Trans. Illinois State Acad. Sci. 68: 87-101.

Eernisse, D. J. & A. G. Kluge. 1993. Taxonomic congruence versus total evidence, and amniote phylogeny inferred from fossils, molecules, and morphology. Molec. Biol. Evol. 10: 1170-1195.

Engler, A. & K. Prantl. 1888. Die naturlichen Pflanzenfamilien. Vol. 2, subvol. 4. Engelmann, Leipzig.

Farris, J. S. 1976. Phylogenetic classification of fossils with recent species. Syst. Zool. 25: 271-282.

-----. 1980. The information content of the phylogenetic system. Syst. Zool. 28: 483-519.

-----. 1989. The retention index and the rescaled consistency index. Cladistics 5: 417-419.

-----, M. Kallersjo, A. G. Kluge & C. Bult. 1994. Testing significance of incongruence. Cladistics 10: 315-319.

Felsenstein, J. 1985. Confidence limits on phylogenies: An approach using the bootstrap. Evolution 39: 783-791.

Fernald, M. L. 1950. Gray's manual of botany. Ed. 8. Dioscorides Press, Portland, Oregon.

Galbreath, E. C. 1974. Stipid grass "seeds" from the Oligocene and Miocene deposits of northeastern Colorado. Trans. Illinois State Acad. Sci. 67: 366-368.

Gaut, B. S., B. R. Morton, B. C. McCaig & M. T. Clegg. 1996. Substitution rate comparisons between grasses and palms: Synonymous rate differences at the nuclear gene Adh parallel rate differences at the plastid gene rbcL. Proc. Natl. Acad. Sci. U.S.A. 93: 10274-10279.

Goloboff, P. 1993a. Estimating character weights during tree search. Cladistics 9: 83-91.

-----. 1993b. Nona, version 1.16 (computer software and manual). Distributed by the author.

-----. 1993c. Piwe, version 2.15 (computer software and manual). Distributed by the author.

Gould, F. W. 1975. The grasses of Texas. Texas A&M University Press, College Station.

Great Plains Flora Association. 1986. Flora of the Great Plains. University Press of Kansas, Lawrence.

Hallam, A. 1987. End-Cretaceous mass extinction event: Argument for terrestrial causation. Science 238: 1237-1242.

Hamby, R. K. & E. A. Zimmer. 1988. Ribosomal RNA sequences for inferring phylogeny within the grass family Poaceae. Pl. Syst. Evol. 160: 29-38.

----- & -----. 1992. Ribosomal RNA as a phylogenetic tool in plant systematics. Pages 50-91 in P. S. Soltis et al. (eds.), Molecular systematics of plants. Chapman & Hall, New York.

Hattersley, P. W. 1987. Variations in photosynthetic pathway. Pages 49-64 in T. R. Soderstrom et al. (eds.), Grass systematics and evolution. Smithsonian Institution Press, Washington, DC.

----- & L. Watson. 1992. Diversification of photosynthesis. Pages 38-116 in G. P. Chapman (ed.), Grass evolution and domestication. Cambridge University Press, Cambridge.

Heslop-Harrison, Y. & K. R. Shivanna. 1977. The receptive surface of the angiosperm stigma. Ann. Rev. Bot. 41: 1233-1258.

Hilu, K. W. & K. Wright. 1982. Systematics of Gramineae: A cluster analysis study. Taxon 31: 9-36.

Hiratsuka, J., H. Shimada, R. Whittier, T. Ishibashi, M. Sakamoto, M. Mori, C. Kondo, Y. Honji, Sun Chong-Rong, Meng Bin-Yuan, Li Yu-Qing, A. Kanno, Y. Nishizawa, A. Hirai, K. Shinozaki & M. Sugiura. 1989. The complete sequence of the rice (Oryza sativa) chloroplast genome: Intermolecular recombination between distinct tRNA genes accounts for a major plastid DNA inversion during the evolution of the cereals. Molec. Gen. Genet. 217: 185-194.

Hitchcock, A. S. 1951. Manual of the grasses of the United States. Ed. 2, revised by A. Chase. U.S.D.A. Misc. Pub. 200, Washington, DC.

Hitchcock, C. L. 1969. Gramineae. Pages 384-725 in C. L. Hitchcock et al. (eds.), Vascular plants of the Pacific Northwest. Part 1. University of Washington Press, Seattle.

Hock-Hin, Y. & L. Watson. 1987. Taxonomic patterns in protein amino acid profiles of grass leaves and caryopses. Pages 88-96 in T. R. Soderstrom et al. (eds.), Grass systematics and evolution. Smithsonian Institution Press, Washington, DC.

Hollowell, V. C. 1987. Systematics of the subtribe Parianinae (Poaceae: Bambuseae: Olyreae). Ph.D. dissertation, University of South Carolina, Columbia.

Holmgren, A. H. & N. H. Holmgren. 1977. Poaceae. Pages 175-464 in A. Cronquist et al. (eds.), Inter- mountain Flora. Vol. 6. Columbia University Press, New York.

Holmgren, P. K., N. H. Holmgren & L. C. Barnett. 1990. Index Herbariorum, part I: The herbaria of the world. Ed. 8. New York Botanical Garden, Bronx.

Holub, J. 1980. Glyceria. Pages 179-181 in T. G. Tutin et al. (eds.), Flora Europaea.. Vol. 5. Cambridge University Press, Cambridge.

Hongping Liang & K. W. Hilu. 1996. Application of the mark gene sequences to grass systematics. Ca- nad. J. Bot. 74: 125-134.

Hoshikawa, K. 1969. Underground organs of the seedlings and the systematics of Gramineae. Bot. Gaz. 130: 192-203.

Howe, C. J., R. F. Bowman & C. M. Dyer. 1988. Common features of three inversions in wheat chloro- plast DNA. Curr. Genet. 13: 339-349.

Hunziker, J. H. 1989. Chromosome studies on Anomochloa and other Bambusoideae (Gramineae). Darwiniana 29: 41-45.

Iturralde-Vinent, M. A. & R. D. E. MacPhee. 1996. Age and palcogeographical origin of Dominican amber. Science 273: 1850-1852.

Jacques-Felix, H. 1988. Les Liliopsida (ex Monocotyledones) n'ont pas de cotyledon. II. La pre feuille de la plantule: Ses rapports avec celles des axes feuilles. Adansonia 3: 275-333.

Janzen, D. H. 1976. Why bamboos wait so long to flower. Ann. Rev. Ecol. Syst. 7: 347-391.

Jirasek, V. 1969. Morphologic der Schuppchen (Lodiculae) von Grasern und ihre Terminologie ein witerer Beitrag zur Kenntnis des Baues der Lodiculae. Acta Univ. Carol. Biol. 1968: 321-344.

-----. 1970. Beitrag zur Kenntnis zweizellinger Haare bei Grasern mit enutzunt von Pfeifengrasern - Molinia caerulea (L.) Moench s.l. Acta Univ. Carol. Biol. 1969: 383-402.

----- & M. Jozifova. 1968. Morphology of lodicules, their variability and importance in the taxonomy of the Poaceae family. Bol. Soc. Argent. Bot. 12: 324-349.

Johnston, C. R. & L. Watson. 1976. Microhairs: A universal characteristic of non-festucoid grass genera? Phytomorphology 26: 297-301.

----- & -----. 1981. Germination flaps in grass lemmas. Phytomorphology 31: 78-85.

Judziewicz, E. J. 1987. Taxonomy and morphology of the tribe Phareae (Poaceae: Bambusoideae). Ph.D. dissertation, University of Wisconsin, Madison.

-----. 1990. Flora of the Guianas, fasc. 8, family 187. Poaceae (Gramineae). Koeltz, Koenigstein.

----- & T. R. Soderstrom. 1989. Morphological, anatomical and taxonomic studies in Anomochloa and Streptochaeta (Poaceae: Bambusoideae). Smithsonian Contr. Bot. 68: 1-52.

Kellogg, E. A. & C. S. Campbell. 1987. Phylogenetic analyses of the Gramineae. Pages 310-322 in T. R. Soderstrom et al. (eds.), Grass systematics and evolution. Smithsonian Institution Press, Washington, DC.

----- & H. P. Linder. 1995. Phylogeny of the Poales. Pages 511-542 in P. J. Rudall et al. (eds.), Monocotyledons: Systematics and evolution. Vol. 2. Royal Botanic Gardens, Kew.

----- & L. Watson. 1993. Phylogenetic studies of a large data set. I. Bambusoideae, Andropogonodae, and Pooideae (Gramineae). Bot. Rev. (Lancaster) 59: 273-343.

Kennedy, P. B. 1899. The structure of the caryopsis of grasses with reference to their morphology and classification. U.S.D.A. Div. Agrostol. Bull. 19: 1-44.

Kim, K. J. & R. K. Jansen. 1994. Comparisons of phylogenetic hypotheses among different data sets in dwarf dandelions (Krigia, Asteraceae): Additional information from internal transcribed spacer sequences of nuclear ribosomal DNA. Pl. Syst. Evol. 190: 157-185.

Kinges, H. 1961. Merkmale des Gramineenembryos, ein Beitrag zur Systematik der Graser. Bot. Jahrb. Syst. 81: 50-93.

Kluge, A. G. 1989. A concern for evidence and a phylogenetic hypothesis of relationships among Epicrates (Boidae, Serpentes). Syst. Zool. 38: 7-25.

----- & J. S. Farris. 1969. Quantitative phyletics and the evolution of anurans. Syst. Zool. 18: 1-32.

----- & A. J. Wolf. 1993. Cladistics: What's in a word? Cladistics 9: 183-199.

Kunth, C. S. 1835. Enumeratio plantarum. I: Agrostographia synoptica. Sumptibus J. G. Collae, Stuttgart.

Kuwabara, Y. 1961. On the shape and direction of leaves of grass seedlings J. Japan. Bot. 36: 368-373.

Lavin, M. & M. Luckow. 1993. Origins and relationships of tropical North America in the context of the boreotropics hypothesis. Amer. J. Bot. 80: 1-14.

Linder, H. P. 1987. The evolutionary history oft he Poales-Restionales: A hypothesis. Kew Bull. 42: \297-318.

----- & I. K. Ferguson. 1985. On the pollen morphology and phylogeny of the Restionales and Poales. Grana 24: 65-76.

----- & E. A. Kellogg. 1995. Phylogenetic patterns in the Commelinid clade. Pages 473-496 in P. J. Rudall et al. (eds.), Monocotyledons: Systematics and evolution. Vol. 2. Royal Botanic Gardens, Kew.

----- & P. J. Rudall. 1993. The megagametophyte in Anarthria (Anarthriaceae, Poales) and its implications for the phylogeny of the Poales. Amer. J. Bot. 80: 1455-1464.

Litke, R. 1968. Uber den Nachweis tertiarer Gremineen. Monatsberichte der Deutschen Akademia der Wissenschaften zu Berlin 19: 462-471.

McClure, F. A. 1973. Genera of bamboos native to the New World (Gramineae: Bambusoideae). Smithsonian Contr. Bot. 9: 1-148.

Macfarlane, T. D. & L. Watson. 1980. The circumscription of Poaceae subfamily Pooideae, with notes on some controversial genera. Taxon 29: 645-666.

Martinovsky, J. O. 1980. Stipa. Pages 247-252 in T. G. Turin et al. (eds.), Flora Europaea. Vol. 5. Cam- bridge University Press, Cambridge.

Metcalfe, C. R. 1960. Anatomy of the monocotyledons, I. Gramineae. Clarendon Press, Oxford.

Mickevich, M. F. & J. S. Farris. 1981. The implications of congruence in Menidia. Syst. Zool. 30: 351-370.

Mlada, J. 1977. The histological structure of the grass embryo and its significance for the taxonomy of the family Poaceae. Acta Univ. Carol. Biol. 1974: 51-156.

Morton, B. R., B. S. Gant & M. T. Clegg. 1996. Evolution of alcohol-dehydrogenase genes in the palm and grass families. Proc. Natl. Acad. U.S.A. 93: 11735-11739.

Muller, J. 1981. Fossil pollen records of extant angiosperms. Bot. Rev. (Lancaster) 47: 1-142.

Nadot, S., R. Bajon & B. Lejeune. 1994. The chloroplast gene rps4 as a tool for the study of Poaceae phylogeny. Pl. Syst. Evol. 191: 27-38.

Nakai, T. 1943. Ordines, familiae, tribi, genera, sectiones, species, varietates, formae et combinationones novae a Prof. Nakai, Takenoshin adhuc ut novis edita. Festschrift, Tokyo.

Nees ab Essenbeck, T. F. L. 1843. Genera plantarum florae Germanieae, iconibus et deseriptionibus illustrata, plantarum monocotyledonearum. Vol. 1, Gramineae. Sumtibus Henry & Cohen, Bonn.

Newell, T. K. 1969. A study of the genus Joinvillea (Flagellariaceae). J. Arnold Arbor. 50: 527-555.

Niklas, K. J. 1985. The aerodynamics of wind-pollination. Bot. Rev. (Lancaster) 51: 328-386.

Nixon, K. C. 1993. Clados, version 1.4.98 (computer software and manual). Distributed by the author.

-----. 1996. Palcobotany in cladistics and cladistics in palcobotany: Enlightenment and uncertainty. Rev. Palaeobot. & Palynol. 90: 361-373.

-----. 1997. Dada, version 1.1.4. (computer software and manual). Distributed by the author.

----- & J. M. Carpenter. 1994. On outgroups. Cladistics 9: 413-426.

----- & -----. 1996. On simultaneous analysis. Cladistics 12: 221-241.

----- & J. I. Davis. 1991. Polymorphic taxa, missing values and cladistic analysis. Cladistics 7: 233-241.

Nunez, O. 1968. El problema de la palea de Oryza L. Bol. Soc. Argent. Bot. 12: 57-97.

Page, J. S. 1978. A scanning electron microscope survey of grass pollen. Kew Bull. 32: 313-319.

Page, V. M. 1947. Leaf anatomy of Streptochaeta and the relation of this genus to the bamboos. Bull. Torrey Bot. Club 74:232-239

Palmer, J. & W. F. Thompson. 1981. Rearrangements in the chloroplast genomes of mung bean and pea. Proc. Natl. Acad. Sci. U.S.A. 78: 5533-5537.

Poinar, G. O. & J. T. Columbus. 1992. Adhesive grass spikelet with ammalian hair in Dominican amher: First fossil evidence of epizoochory. Experientia 48: 906-908.

Prat, H. 1960. Vers une classification naturelle des Graminees. Bull. Soc. Bot. Fra. 107: 32-79.

Quigley, F. & J. H. Well. 1985. Organization and sequence of five tRNA genes and of an unidentified reading frame in the wheat chloroplast genome: Evidence for gene rearrangements during the evolution of chloroplast genomes. Curr. Genet. 9: 495-503.

Reeder, J. 1957. The embryo in grass systematics. Amer. J. Bot. 44: 756-768.

-----. 1962. The Bambusoid embryo: A reappraisal. Amer. J. Bot. 49: 639-641.

Renvoize, S. A. 1985a. A survey of leaf-blade anatomy in grasses: V. The bamboo allies. Kew Bull. 40: 509-535.

-----. 1985b. A survey of leaf-blade anatomy in grasses: VI. Stipeae. Kew Bull. 40: 731-736.

Rieppel, O. 1993. The role of paleontological data in testing homology by congruence. Acta Palaeontol, Polon. 38: 295-302.

Rodman, J. E., K. G. Karol, R. A. Price & K. J. Sytsma. 1996. Molecules, morphology, and Dahlgren's expanded order Capparales. Syst. Bot. 21: 289-307.

Rosengurtt, B., A. Laguardia & B. R. Arrillaga de Maffei. 1972. El character lipido del endosperma central en especies de Gramineas. Bol. Fac. Agron. Univ. Montevideo 124: 1-43.

Roshevits, R. Yu. 1937. Zlaki. Vvedenie v izuchenie kormovykhi i khlebnyky zlakov. [Grasses. An introduction to the study of fodder and cereal grasses. English translation for the Smithsonian Institution and the National Science Foundation. 1980. Indian National Scientific Documentation Center, New Delhi.]

Salgado-Labouriau, M. L. & M. Rinaldi. 1990. Palynology of Gramineae of the Venezuelan mountains. Grana 29: 119-128.

-----, S. Nilsson & M. Rinaldi. 1992. Exine sculpture in Pariana pollen (Gramineae). Grana 32: 243-249.

Scholz, H. 1982. Uber Mikro- un Makrohaare einiger Piptatherum und Stipa-arten (Stipeae, Gramineae). Willdenowia 12: 235-240.

Schuster, J. 1910. Uber die Morphologic der Grasblute. Flora 100: 213-266.

Seberg, O., S. Frederiksen, C. Baden & I. Linde-Laursen. 1991. Peridictyon, a new genus from the Balkan peninsula, and its relationship with Festucopsis (Poaceae). Willdenowia 21: 87-104.

Sharma, M. L. 1979. Some considerations on the phylogeny and chromosomal evolution in grasses. Cytologia 44: 679-685.

Sinha, N. R. & E. A. Kellogg. 1996. Parallelism and diversity in multiple origins of [C.sub.4] photosynthesis in the grass family. Amer. J. Bot. 83: 1458-1470.

Smith, A. B. & D. T. J. Littlewood. 1994. Paleontological data and molecular phylogenetic analysis. Paleobiology 20: 259-273.

Smithson, E. 1957. The comparative anatomy of the Flagellariaceae. Kew Bull. 11: 491-501.

Soderstrom, T. R. 1980. A new species of Lithachne (Poaceae: Bambusoideae). Brittonia 32: 495-501.

-----. 1981. Some evolutionary trends in the Bambusoideae (Poaceae). Ann. Missouri Bot. Gard. 68: 15-47.

----- & C. E. Calderon. 1971. Insect-pollination in tropical rain forest grasses. Biotropica 3: 1-16.

----- & -----. 1974. Primitive forest grasses and evolution of the Bambusoideae. Biotropica 6: 141-153.

----- & -----. 1978. Chusquea and Swallenochloa (Poaceae: Bambusoideae): Generic relationships and new species. Brittonia 30: 279-312.

----- & -----. 1979. A commentary on the Bamboos (Poaceae: Bambusoideae). Biotropica 11: 161-172.

----- & R. P. Ellis. 1987. The position of bamboo genera and allies in a system of grass classification. Pages 225-238 in T. R. Soderstrom et al. (eds.), Grass systematics and evolution. Smithsonian Institution Press, Washington, DC.

----- & F. O. Zuloaga. 1989. A revision of the genus Olyra and the new segregate genus Parodiolyra (Poaceae: Bambusoideae: Olyreae). Smithsonian Contr. Bot. 69: i-iv, 1-79.

Soreng, R. J., J. I. Davis & J. J. Doyle. 1990. Phylogenetic analysis of chloroplast DNA restriction site variation in Poaceae subfam. Pooideae. Pl. Syst. Evol. 171: 83-97.

Srivastava, S. K. 1994. Palynology of the Cretaceous-Tertiary boundary in the Scollard Formation of Alberta, Canada, and global KTB events. Rev. Palaeobot. & Palynol. 83: 137-158.

Stace, C. A. 1980. Vulpia. Pages 154-156 in T. G. Tutin et al. (eds.), Flora Europaea. Vol. 5. Cambridge University Press, Cambridge.

Stebbins, G. L. 1956. Cytogenetics and the evolution of the grass family. Amer. J. Bot. 43: 890-905.

-----. 1974. Flowering plants: Evolution above the species level. Belknap Press, Harvard University, Cambridge.

-----. 1981. Coevolution of grasses and herbivores. Ann. Missouri Bot. Gard. 68: 75-86.

-----. 1987. Grass systematics and evolution: Past, present, and future. Pages 359-367 in T. R. Soderstrom et al. (eds.), Grass systematics and evolution. Smithsonian Institution Press, Washington, DC.

----- & B. Crampton. 1961. A suggested revision of the grass genera of Temperate North America. Pages 133-145 in Recent Advances in Botany, Vol. 1. University of Toronto Press, Toronto.

Stevenson, D. W. & H. Loconte. 1995. Cladistic analysis of monocot families. Pages 543-578 in P. J. Rudall et al. (eds.), Monocotyledons: Systematics and evolution. Vol. 2. Royal Botanic Gardens, Kew.

Sugiura, M. 1989. Oryza sativa chloroplast DNA 134,525 bp. Nagoya University, Center for Gene Research, Nagoya, Japan.

-----, K. Shinozaki, N. Zaita, M. Kusuda & M. Kumano. 1986. Clone bank of the tobacco (Nicotiana tabacum) chloroplast genome as a set of overlapping restriction endonuclease fragments: Mapping of eleven ribosomal protein genes. Pl. Sci. 44: 211-216.

Swofford, D. L. 1991. When are phylogeny estimates from molecular and morphological data incongruent? Pages 295-333 in M. M. Miyamoto & J. Cracraft (eds.), Phylogenetic analysis of DNA sequences. Oxford University Press, New York.

Tateoka, T. 1957. Miscellaneous papers on the phylogeny of Poaceae, X: Proposition of a new phylogenetic system of Poaceae. J. Japan. Bot. 32: 275-287.

-----. 1962. Starch grains of endosperm in grass systematics. Bot. Mag. (Tokyo) 75: 377-383.

-----. 1964. Notes on some grasses. XVI. Embryo structure of the genus Oryza in relation to systematics. Amer. J. Bot. 51: 539-543.

-----, S. Inoue & S. Kawano. 1959. Notes on some grasses. IX. Systematic significance of bicellular microhairs of leaf epidermis. Bot. Gaz. 121: 80-91.

Terrell, E. E. 1971. Survey of occurrences of liquid or soft endosperm in grass genera. Bull. Torrey Bot. Club 98: 264-268.

Thomas, K. M., B. J. Wood, C. L. Bassett & J. R. Y. Rawson. 1984. A restriction endonuclease map of the chloroplast genome of pearl millet. Curr. Genet. 8: 291-297.

Thomasson, J. R. 1987. Fossil grasses: 1820-1986 and beyond. Pages 159-167 in T. R. Soderstrom et al. (eds.), Grass systematics and evolution. Smithsonian Institution Press, Washington, DC.

Tillich, H.-J. 1996. Seeds and seedlings in Hanguanaceae and Flagellariaceae (Monocotyledons). Sendtnera 3: 187-197.

-----. 1980. Ampelodesmos, Anthoxanthum, and Triticeae tribal description. Pages 190, 229-230, 252 in T. G. Tutin et al. (eds.), Flora Europaea. Vol. 5. Cambridge University Press, Cambridge.

Tzvelev, N. N. 1977. Zlaki S.S.S.R. [Grasses of the Soviet Union. English translation for the Smithsonian Institution. 1983. Amerind Publishing Co., New Delhi.]

-----. 1989. The system of grasses (Poaceae) and their evolution. Bot. Rev. (Lancaster) 55:141-204.

Uhl, N. W., J. Dransfield, J. I. Davis, M. A. Luckow, K. S. Hansen & J. J. Doyle. 1995. Phylogenetic relationships among palms: Cladistic analyses of morphological and chloroplast DNA restriction site variation. Pages 623-661 in P. J. Rudall et al. (eds.), Monocotyledons: Systematics and evolution. Vol. 2. Royal Botanic Gardens, Kew.

Verboom, G. A., H. P. Linder & N. P. Barker. 1994. Haustorial synergids: An important character in the systematics of Danthonioid grasses (Arundinoideae: Poaceae). Amer. J. Bot. 81: 1601-1610.

Warming, E. 1895. A handbook of systematic botany. Macmillan, New York

Watrous, L. E. & Q. D. Wheeler. 1981. The out-group comparison method of character analysis. Syst. Zool. 30: 1-11.

Watson, L. & M. J. Dallwitz. 1992. The grass genera of the world. CAB International, Wallingford, UK.

-----, H. T. Clifford & M. J. Dallwitz. 1985. The classification of Poaceae: Subfamilies and supertribes. Austral. J. Bot. 33: 433-484.

Wendel, J. F., A. Schnabel & T. Seelanan. 1995. An unusual ribosomal DNA sequence from Gossypium gossypioides reveals ancient, cryptic, intergenomic introgression. Molec. Phyl. Evol. 4: 298-313.

Wheeler, W. C., P. Cartwright & C. Y. Hayashi. 1993. Arthropod phylogeny: A combined approach. Cladistics 9: 1-39.

Wiley, E. O. 1981. Phylogenetics, the theory and practice of phylogenetic systematics. John Wiley, New York.

Williams, S. E., V. A. Albert & M. W. Chase. 1994. Relationships of Droseraceae: A cladistic analysis of rbcL sequence and morphological data. Amer. J. Bot. 81: 1027-1037.

Wolfe, J. A. 1975. Some aspects of the plant geography of the Northern Hemisphere during the Late Cretaceous and Tertiary. Ann. Missouri Bot. Gard. 62: 264-279.

-----. 1977. Paleogene floras from the Gulf of Alaska region. Geol. Surv. Prof. Paper no. 997. 108 pp. + 30 pls.

-----. 1978. A palcobotanical interpretation of Tertiary climates in the Northern Hemisphere. Amer. Scientist 6: 694-703.

Yakovlev, M. S. 1950. Struktura endosperma i zarodysha zlakov kak sistematicheskiy priznak. Trudy Bot. Inst. Komarova, Ser. 7, Morfol. Anat. Rast. 1: 121-128. [Structure of the endosperm and embryo of grasses as a systematic criterion. Unpublished English translation. 1970. Indian National Scientific Documentation Center, New Delhi.]

Yaneshita, M., T. Sasakuma & Y. Ogihara. 1993. Phylogenetic relationships of turfgrasses as revealed by restriction fragment analysis of chloroplast DNA. Theor. Appl. Genet. 87: 129-135.

Yates, H. O. 1966. Morphology and cytology of Uniola (Gramineae). Southw. Naturalist 11: 145-189.

VIII. Appendix 1: Restriction Site Character Data

Data matrix depicting occurrence of 364 mapped restriction sites among 75 exemplar taxa of Poaceae and related families (Table I). The total number of sites scored "missing" (i.e., not scored 0 or 1) is indicated for each taxon; "?" = site scored missing because unobserved; "\" = site scored missing because of uncertainty regarding homology (see text). The three delimited groups of characters for each restriction enzyme are those of the Large Single Copy (LSC) region, Inverted Repeat (IR) region, and Small Single Copy (SSC) region, respectively. The position of each site is keyed to the numbered nucleotides of the complete cpDNA genome sequence of Oryza sativa (Hiratsuka et al., 1989; Sugiura, 1989) and is provided, in thousands of base pairs, as a vertically oriented number with one decimal position (e.g., the first Bam HI site in the LSC, IR, and SSC regions are at locations 6.0, 99.9, and 101.8, respectively.

[TABULAR DATA FOR APPENDIX 1 OMITTED]

IX. Appendix 2: Structural Characters and States

Forty-two structural characters were scored for 75 exemplar taxa of Poaceae and related families (Table I), and notes on scoring are given where different from that of Watson and Dallwitz (1992) or where Watson and Dallwitz provide no report; where scoring differs from that of Watson and Dallwitz, their scoring is noted prior to that used in the analysis. Abbreviations: "C&R" = Clayton & Renvoize, 1986; "DCY" = Dahlgren et al., 1985; "H&W" = Hattersley & Watson, 1992; "W&D" = Watson & Dallwitz, 1992; "nr" = no report in Watson & Dallwitz, 1992 (implicit for taxa in families other than Poaceae); "pers. obs." = personal observation of specimens at the L. H. Bailey Hortorium; "-" = inapplicable; "?" = unobserved; "\" = homology ambiguous or observed state intermediate between recognized states. Multiple states [in brackets] signify polymorphism or subset ambiguity in multistate characters, where one or more states are determined to be absent (e.g., [01] for a character with states 0, 1, and 2, signifies polymorphism of states 0 and 1 OR indeterminant homology where state 2 is determined to be absent, but discrimination between states 0 and 1 has not been made). Guadua is included in Bambusa, and Eremitis in Pariana, by Watson and Dallwitz. Information on Baloskion is to be found in the primary literature under Restio, except when provided by H. P. Linder or B. Briggs. All multistate characters are scored as unordered. Four individual numbers (or ranges) following each character description represent results of optimization of the character on 12 most-parsimonious trees obtained by simultaneous analysis of restriction site and structural data (see text); the four numbers are number of steps on most parsimonious trees, number of extra steps beyond minimum required if character were consistent with trees, consistency index, and retention index, respectively. Additional steps implied by polymorphism within terminals not included in calculation of steps or CI.

CULM

1: Perennating and branching woody culms: 0 = absent; 1 = present. NOTE: See Soderstrom (1981: 16) for a strict definition of "woody" versus "herbaceous" in Bambusoid grasses. 2, 1, 0.50, 0.80.

Flagellaria = \. Stems perennate, but whether vegetative stems branch or are lignified is uncertain (pers. obs.).

Baloskion (Restio = [01]; DCY). Restio tetraphyllus = 0 (Black, 1960).

Joinvillea = 0 (Newell, 1969).

Eremitis (W&D 1). Eremitis = 0, persisting alive for one year, or two at maximum, non-branching (V. Hollowell, pers. comm.).

2: Culm internodes: 0 = solid; 1 = hollow. NOTE: Occurrence of a small pore is considered intermediate. 11, 10, 0.09, 0.09.

Flagellaria = 0 (DCY; pers. obs.).

Baloskion = 1 (B. Briggs, pers. comm.).

Joinvillea = 1 (DCY; pers. obs.).

Aristida (W&D [01]). Aristida purpurea = 0 (Ebinger & Carlen, 1975; Brown et al., 1959).

Danthonia (nr). Danthonia californica = 1 (pers. obs.).

Eragrostis (W&D [01]). Eragrostis curvula = \, small pores (Brown et al., 1959).

Eremitis (nr). Eremits = 1 (V. Hollowell, pers. comm.).

Leersia (W&D [01]). Leersia virginica = 1 (Ebinger & Carlen, 1975).

Nassella (nr). Nassella viridula = 1, included in Stipa in Hitchcock, 1969.

Panicum (W&D [01]). Panicum virgatum = 1 (Ebinger & Carlen, 1975). Permisetum (W&D [01]).

Pennisetum alopecuroides = 0 (Ebinger & Carlen, 1975). Spartina (W&D [01]).

Spartina pectinata = 1 (Ebinger & Carlen, 1975). Sporobolus (W&D [01]).

Sporobolus giganteus = 0 (Holmgren & Holmgren, 1977). Uniola (nr).

Uniola paniculata = \, culms solid or with a small pore (Brown et al., 1959), or hollow (Ebinger & Carlen, 1975).

LEAF

NOTE: Characters 4 through 6 are inapplicable for Baloskion (B. Briggs, pers. comm.).

3: Leaf sheath margins: 0 = free; 1 = fused at least 1/4 of length. 5, 4, 0.20, 0.60.

Flagellaria indica = 1 (pers. obs.), but the genus is polymorphic (pers. obs., contra DCY).

Baloskion = 0 (pers. obs.).

Joinvillea = 0 (DCY; pers. obs.).

4: Adaxial ligule type: 0 = membrane (with or without fringe of hairs); 1 = fringe of hairs only. 4, 3, 0.25, 0.62.

Flagellaria = -, no ligules (pers. obs.).

Joinvillea = 0, when present (DCY; pers. obs.).

Aristida (W&D [01]). Aristida purpurascens = [01], either a minute fringed collar or a fringed membrane (Gould, 1975).

Ehrharta (W&D [01]). Ehrharta calycina = 0, membranous, not ciliate (pers. obs.).

Panicum (W&D [01]). Panicum virgatum = 0, a fringed membrane (Gould, 1975). Pennisetum (W&D [01]).

Pennisetum alopecuroides = 1, ciliate with a very short collar (pers. obs.).

Sporobolus (W&D [01]). Sporobolus giganteus = [01], either a minute fringed collar or a fringed membrane (Gould, 1975).

Zoysia (W&D [01]). Zoysia japonica = [01], densely ciliate on the margin, but the membranous portion is variable in occurrence (pers. obs.).

5: Abaxial (contra-) ligule: 0 = absent; 1 = present. 1, 0, 1.00, 1.00.

Flagellaria = 0 (pers. obs.).

Joinvillea = 0 (pers. obs.).

Bambusa (nr). Bambusa = 1 (Davidse et al., 1994), Bambusa multiplex = 1 (Judziewicz, 1990; pers. obs.).

Chusquea (nr). Chusquea = 1 (Soderstrom & Calderon, 1978: fig. 2).

Eremitis (nr). Eremitis = 0 (V. Hollowell, pers. comm.). Guadua (nr).

Guadua = 1 (Judziewicz, 1990). Olyra (nr).

Olyra latifolia = 0 (pers. obs.).

Oryza (nr). Oryza sativa = 0 (pers. obs.).

Pharus (nr). Pharus latifolius = 0 (pers. obs.).

Stipa (W&D [01]). Stipa barbata = 0 (Bor, 1968; Martinovsky, 1980).

6: Pseudopetiole: 0 = absent; 1 = present. 1-2, 0-1, 0.50-1.00, 0.90-1.00.

Flagellaria = [01] (pers. obs.).

Joinvillea = 1 (pers. obs.).

SPIKELET

NOTE: Spikelet characters 7-13 are inapplicable or ambiguous except when grass-type spikelets or florets are present. The three nongrass genera do not have grass-type spikelets or florets. Pseudospikelets in Bambusa and Guadua are possibly homologous to each other, and grass-type florets are evident. However, pseudospikelets in Anomochloa and Streptochaeta are neither alike nor of the Bambuseae form, and the homology of any or all of their floral bracts with those of typical grass spikelets or florets is ambiguous (C&R; Judziewicz & Soderstrom, 1989).

7: Pedicel of spikelet: 0 = absent; 1 = present. 5, 4, 0.20, 0.31.

Amphipogon (W&D equivocal, not truly sessile). Amphipogon strictus = 1, spikelets subsessile (Black, 1960).

Chasmanthium (W&D [01]). Chasmanthium latifolium = 1, pedicels present, Chasmanthium nitidum = 1, spikelets subsessile (Hitchcock, 1951).

Eremitis (W&D refer to a false spike). Eremitis = 1, never sessile (V. Hollowell, pers. comm.).

8: Proximal female-sterile florets in female-fertile spikelets: 0 = absent; 1 = present. 7, 6, 0.14, 0.45.

Leersia and Oryza = \, homology of the vestigial bracts below the single fertile floret is ambiguous, as these may represent sterile florets or glumes (Nunez, 1968; Clifford, 1987).

Ehrharta = 1.

9: Number of female-fertile florets per female-fertile spikelet: 0 = two or more; 1 = one. 13, 12, 0.07, 0.57.

Melica (W&D [01]). Melica cupanii = 1, Melica altissima = [01] (Davis, 1985).

10: Awn or mucro on fertile or sterile lemma: 0 = absent; 1 = present. 13-14, 12-13, 0.07-0.08, 0.45-0.50.

Ehrharta (W&D [01]). Ehrharta calycina = 1 (pers. obs.).

Panicum (W&D [01]). Panicum virgatum = 0 (Gould, 1975).

11: Awn attachment: 0 = terminal / subterminal; 1 = from a sinus; 2 = dorsal. NOTE: Character 11 is inapplicable for taxa scored 0 for character 10, and is considered applicable for taxa polymorphic for character 10. 6, 4, 0.33, 0.69.

Ampelodesmos (W&D 1). Ampelodesmos mauritanicus = [01], awn terminal or from a very short sinus (Tutin, 1980; pers. obs.).

Amphipogon (description by W&D does not clearly translate to our states). Amphipogon strictus = [01], ambiguous, but not dorsal (pers. obs.).

Anisopogon (W&D 1). Anisopogon avenaceus = 1, three awns, the central from a distinct sinus in our interpretation (Black, 1960).

Arundo (W&D [012]). Arundo donax = [01], apical or subapical from a minute to short sinus (Gould, 1975).

Bromus (W&D [012]). Bromus inermis = 0, awnless or with a short, almost terminal awn (i.e., with a minute flange of tissue between the apex and awn; pers. obs.).

Ehrharta (W&D [01]). Ehrharta calycina = 0, terminal (pers. obs.).

Guadua (W&D Guadua mucronate). Guadua = 0, apiculate or in one species with a 3-5 mm awn (Davidse et al., 1994).

Schizachne (W&D awned from just below a sinus). Schizachne = [ 12], variable (Hitchcock, 1969).

Stipa, Nassella, and Piptatherum (W&D 0). The awn in Stipeae is interpreted as from a sinus when lateral lobes are present. Stipa barbata = [01] (Bor, 1968), and Nassella viridula = [01], lobes to 1.5 mm present or absent (Hitchcock, 1969), Piptatherum miliaceum = 0, no lobes evident (pers. obs.).

12: Disarticulation above glumes: 0 = absent; 1 = present. 8, 7, 0.12, 0.30.

13: Germination flap in lemma: 0 = absent; 1 = present. 5, 4, 0.20, 0.33. Calamagrostis (nr). Calamagrostis = 0 (Johnston & Watson, 1981). Eremitis (nr). Eremitis = 0 (V. Hollowell, pers. comm.). Oryza (nr). Oryza = 1 (Johnston & Watson, 1981). Phragmites (nr). Phragmites = 0 (Johnston & Watson, 1981).

FLOWER

NOTE: Characters 14-18 are scored as ambiguous for taxa that have tepals rather than lodicules, and characters 15-18 are scored as inapplicable for taxa that are scored 0 for character 14. Character 14 is scored as ambiguous in Streptochaeta and Anomochloa, as the homology of structures sometimes called lodicules in those taxa is ambiguous. Reporting and application of lodicule characters 17 and 18 by W&D is inconsistent with the literature and available illustrations for our taxa (e.g., lodicules in Stipeae reported as fleshy or membranous, heavily vascularized). We observed lodicules of as many taxa as possible from specimens and original literature (Jirasek & Jozifova, 1968; Tzvelev, 1977).

14: Lodicules: 0 = absent; 1 = present. 3, 2, 0.33, 0.50.

Aristida (W&D [01]). Aristida purpurea = 1 (Gould, 1975; Holmgren & Holmgren, 1977).

15: Lodicule number: 2 = two; 3 = three; 4 = four; 5 = five; 6 = six. 4-5, 3-4, 0.20-0.25, 0.63-0.72.

Nassella (W&D 2 "in material seen"). Nassella viridula = 2 (Hitchcock, 1969, where it is included in Stipa).

Stipa (W&D 3 or rarely 2). Stipa = [23], three or very rarely two lodicules (Bor, 1968; Tzvelev, 1977).

16: Fusion of anterior pair of lodicules: 0 = free; 1 = fused. 1, 0, 1.00, 1.00.

Glyceria (W&D 1 for whole genus). Glyceria = [01], variable; G. declinata = [01], free or fused in G. fluitans, a close relative (Jirasek, 1969; Jirasek & Jozifova, 1968); G. striata = [01], free or possibly fused only near the base (Hitchcock, 1969).

Schizachne (W&D 1). Schizachne = [01], free or fused (Hitchcock, 1969).

17: Distally membranous portion of lodicule: 0 = absent; 1 = present. 2, 1, 0.50, 0.93.

Eremitis (nr). Eremitis = 1 (V. Hollowell, pers. comm.). Nassella (W&D fleshy).

Nassella viridula = 1 (Hitchcock, 1969).

Panicum (W&D 0). Panicum virgatum = [01], ambiguous, rudimentary (pers. obs.).

Stipa (W&D fleshy or membranous). Stipa barbata = 1 (Bor, 1968).

18: Lodicule vascularization: 0 = very faint to absent; 1 = prominent. 2, 1, 0.50, 0.95.

Anisopogon (nr). Anisopogon avenaceus = 0 (pers. obs.).

Brachyelytrum (W&D 0). Brachyelytrum = 1 (Macfarlane & Watson, 1980; Campbell et al, 1986; pers. obs.).

Danthonia (nr). Danthonia californica = 1 (pers. obs.). Eremitis (nr).

Eremitis = 1 (V. Hollowell, pers. comm.).

Panicum (nr). Panicum virgatum = 1 (pers. obs.).

Piptatherum (W&D 1). Piptatherum miliaceum = 0 (pers. obs. for this species; Tzvelev, 1977, for other species).

Stipa (nr). Stipa barbata = 0 (Bor, 1968).

19: Stamen number: 1 = one; 2 = two; 3 = three; 4 = four; 5 = five; 6 = six. 10-11, 6-7, 0.36-0.40, 0.36-0.45.

Flagellaria = 6 (DCY).

Baloskion (Restio = [123]; DCY). Restio tetraphyllus = 3 (H. P. Linder & B. Briggs, pers. comm.).

Joinvillea = 6 (DCY).

Anthoxanthum (W&D [23]). Anthoxanthum odoratum = 2 (Tzvelev, 1977; Turin, 1980).

Diarrhena (W&D [123]). Diarrhena obovata = [23] (Great Plains Flora Association, 1986).

Ehrharta (W&D [346]). Ehrharta calycina = 6 (Black, 1960).

Eremitis (W&D = 10-40). Eremitis = 2 (V. Hollowell, pers. comm.; C&R), and Pariana = 6-30 (C&R).

Glyceria (W&D [23]). Glyceria striata = 2 (Hitchcock, 1969). G. declinata = 3, (Holub, 1980; Tzvelev, 1977).

Leersia (W&D [1236]). Leersia virginica = [12] (Fernald, 1950).

Vulpia (W&D [123]). Vulpia alopecuros = 3 (Stace, 1980).

20: Styles fused at least at base: 0 = absent; 1 = present. NOTE: The solitary style in some taxa (see character 21) can be interpreted as the product of fusion or reduction in number, and those taxa are scored as ambiguous for character 20. 8-9, 7-8, 0.11-0.12, 0.50-0.56.

Flagellaria = 0 (DCY).

Baloskion = 1 (B. Briggs, pers. comm.).

Joinvillea = [01], variable within the species, sometimes fused for a short distance (DCY).

Anomochloa (nr). Anomochloa = \ (Clifford, 1987: 27; Judziewicz & Soderstrom, 1989).

Eremitis (W&D [01]). Eremitis = \ (V. Hollowell, pers. comm.).

Nardus and Lygeum (W&D 1). Both = \ (Clifford, 1987).

21: Number of stigmas: 1 = one; 2 = two; 3 = three; 4 = four. 7, 5, 0.28, 0.44.

Flagellaria = 3 (DCY).

Baloskion (Restio = [123]; DCY). Resrio tetraphyllus = 2 (Black, 1960).

Joinvillea = 3 (DCY).

Eremitis (W&D 2). Eremitis = 1 (V. Hollowell, pers. comm.).

Lithachne (W&D 1). Lithachne humilis = 2 (Judziewicz, 1990; pers. obs.).

Peridictyon (nr). Peridictyon inferred to be = 2, no reports to the contrary found in Elymus (in which it was formerly included), or for the tribe (C&R; Tutin, 1980).

22: Highest order of stigmatic branching present: 1 = simple (unbranched, or with branches composed of single elongate papillate receptive cells, or with very short branches composed of a few papillate receptive cells, but in the latter case the stigmas linear in outline); 2 = primary (branches well developed, composed of series of dispersed papillate receptive cells, secondary branches absent or minimally developed, stigmas lanceolate or broader); 3 = secondary (secondary to tertiary branches well developed, branches composed of series of dispersed papillate receptive cells). NOTE: In the present study only stigmas with state 2 or 3 are called plumose. This character is not recorded by W&D. Major general references are Kunth, 1835; Nees, 1843; Tzvelev, 1977; Heslop-Harrison & Shivanna, 1977. Specific references for Bambusoids are Camus, 1913; Clark & Londono, 1991; Judziewicz, 1987; Judziewicz & Soderstrom, 1989; McClure, 1973; Soderstrom, 1980; Soderstrom & Calderon, 1978; Soderstrom & Zuloaga, 1989. 5, 3, 0.40, 0.72.

Flagellaria = 1 (DCY; Engler & Prantl, 1888; Heslop-Harrison & Shivanna, 1977).

Baloskion (Restio = [12]; DCY). Restio tetraphyllus has minimal primary branching (Black, 1960), and B. Briggs (pers. comm.) refers to it as plumose. We recognize it as ambiguous with regard to states 1 and 2, and regard state 3 as absent.

Joinvillea = 1 (DCY; Engler & Prantl, 1888). Both Cronquist (1981) and DCY report stigmas of Joinvillea as plumose, but the stigmas are simple, without well-developed primary branches evident in illustrations (DCY; Engler & Prantl, 1888; see also Heslop-Harrison & Shivanna, 1977).

Eremitis (nr). Eremitis = 1, (V. Hollowell, pers. comm.).

FRUIT AND EMBRYO

NOTES: Four embryo characters that refer to parts of the typical grass-type embryo (26-29) are inapplicable for the three nongrass genera, which lack the grass-type embryo (i.e., have state 0 of character 25). Because W&D indicate that data provided for Piptatherum for embryo characters may refer to Oryzopsis Michx., Piptatherum was left unscored for characters 26-31 unless data were specifically available under that name.

23: Number of ovules per pistil: 1 = one; 2 = two; 3 = three. 2, 0, 1.00, 1.00.

Flagellaria = 3, one ovule per locule and 3 locules (DCY).

Baloskion = 2, two locules, one ovule per locule (Black, 1960).

Joinvillea = 3, one ovule per locule and 3 locules (DCY).

24: Hilum: 0 = linear-long, [greater than]1/3 length of grain; 1 = nonlinear, [less than]1/3 length of grain, elliptical or broader to punctiform. NOTE: linear-short, [less than]1/3 length of grain, is considered ambiguous. 7-8, 6-7, 0.12-0.14, 0.68-0.72. Amphipogon (W&D short).

Amphipogon = 1, punctiform (Black, 1960).

Aristida (W&D long-linear or short). Aristida purpurea = 0, long-linear (pers. obs.).

Arundo (W&D short). Arundo donax = 1, short, broadly obovate, 1/5 length of grain, subbasal (pers. obs.).

Calamagrostis (W&D short). Calamagrostis = 0, long-linear (Nees, 1843; pers. obs.).

Chasmanthium (W&D short). Chasmanthium latifolium = 1, narrowly elliptical, 1/5 length of grain, basal to sub-basal (pers. obs.).

Danthonia (W&D usually long-linear). Danthonia californica = 0, long-linear (pers. obs.).

Distichlis (W&D short). Distichlis stricta = 1, punctiform, sub-basal or basal (pers. obs.).

Eremitis (nr). Eremitis = 0 (V. Hollowell, pers. comm.).

Pennisetum (W&D short). Pennisetum alopecuroides = 1, elliptical small, 1/5 length of grain (pers. obs.).

Phragmites (W&D short). Phragmites = \, 1/3 length, narrowly elliptical (Bor, 1968; Tzvelev, 1977).

25: Embryo position and structure: 0 = embedded, simple; 1 = lateral, grass-type. NOTE: The grass-type embryo is lateral, peripheral to the endosperm, and differentiated in fruit (Reeder, 1957; Cronquist, 1981; DCY). Although presence of this type of embryo is not explicitly recorded by W&D, state 1 is inferred for all grasses for which they record embryo characters. 1, 0, 1.00, 1.00.

Flagellaria, Joinvillea, and Baloskion = 0, central, undifferentiated in fruit (DCY).

26: Embryo epiblast: 0 = absent; 1 = present. 8, 7, 0.12, 0.46.

Bambusa (nr). Bambusa = 1 (Kinges, 1961; Yakovlev, 1950: fig. 63),

Brachypodium (W&D [01]). Brachypodium sylvaticum = 0 (Mlada, 1977: pl. I); B. pinnatum = 1 (Mlada, 1977: pl. II); B. distachyon = [01] (scored as in W&D).

Elymus (W&D [01]). Elymus trachycaulus = 1 (Mlada, 1977: pl. V).

Guadua (nr). Guadua = 1 (Reeder, 1962),

Lygeum (nr). Lygeum = 1 (Kennedy, 1899: pls. 25 & 26).

Vulpia (nr). Vulpia hispanica = 1 (Nardurus [Bluff, Nees, Schauer] Rchb. in Decker, 1964).

27: Embryo scutellar tail: 0 = absent; 1 = present. 2, 1, 0.50, 0.96.

Anthoxanthum (W&D 0). Anthoxanthum odoratum = [01], polymorphic (Kennedy, 1899: figs. 29 & 51; Mlada, 1977: pl. IV).

Avena (W&D 0). Arena sativa = [01], a distinct notch present (Yakovlev, 1950) or absent (Mlada, 1977: pl. VII).

Brachyelytrum (W&D 0). Brachyelytrum erectum = [01], present (Kennedy 1899: fig. 43; Campbell et al., 1986) or absent (Reeder, 1957).

Diarrhena (W&D 1). Diarrhena = [01], present (Macfarlane & Watson, 1980) or absent (Reeder, 1957).

Oryza (W&D 1). Oryza sativa = 0 (Yakovlev, 1950), and Oryza = 0 (Mlada, 1977: pl. X; Kennedy, 1899: fig. 27; Reeder, 1957). Tateoka (1964) reports tail absent in O. sativa and other species of sect. Sativa but present in some other sections. C&R (p. 7) describe the Oryza-type embryo as lacking a scutellar tail.

Vulpia (nr). Vulpia hispanica = 0 (Nardurus in Decker, 1964).

28: Embryo mesocotyl internode: 0 = negligible; 1 = elongate. 1, 0, 1.00, 1.00.

Vulpia (nr). Vulpia hispanica = 0 (Nardurus in Decker, 1964).

29: Embryonic leaf margins: 0 = meeting; 1 = overlapping. 3-4, 2-3, 0.25-0.33, 0.82-0.88.

Avena (W&D 0). Avena = \, A. sativa and A. byzantina C. Koch illustrated as somewhat overlapping (Mlada, 1977: pl. VII).

Diarrhena (W&D 1). Diarrhena = [01], meeting (Reeder, 1957) or overlapping (Macfarlane & Watson, 1980; W&D).

Puccinellia (W&D 0). Puccinellia distans = [01], P. distans subsp. sevangeana (Grossh.) Tzvelev = 1 (Mlada, 1977: pl. II).

Triticum (W&D 0). Triticum = \, T. durum Desf., T. monococcum L., and T. polonicum L. illustrated as somewhat overlapping (Mlada, 1977: pls. V & VI).

Vulpia (nr). Vulpia hispanica = 0 (Nardurus in Decker, 1964).

30: Endosperm lipid: 0 = absent; 1 = present. NOTE: Liquid and semi-liquid endosperm are indicative of the presence of lipid, but "semi-solid" and solid states do not imply absence of lipid (Terrell, 1971; Rosengurtt et al., 1972). 3, 2, 0.33, 0.83.

Baloskion = 0 (B. Briggs, pers. comm.).

Ampelodesmos (nr). Ampelodesmos = 0, lipid absent (Rosengurtt et al., 1972).

Brachyelytrum (W&D 1 "liquid"). Brachyelytrum = ?. We regard the report of "liquid" endosperm to be erroneous; specimens at BH have hard endosperm. Although Macfarlane and Watson (1980) report the endosperm as "not hard," they do not say "liquid." "Not hard" may be interpreted as semi-solid, so there is no positive evidence of endosperm lipid in this genus.

Diarrhena (nr). Diarrhena = 0, lipid absent (Rosengurtt el al., 1972).

Lithachne (nr). Lithachne = 0, lipid absent (Rosengurtt el al., 1972).

Lygeum (nr). Lygeum = 0, lipid absent (Rosengurtt el al., 1972).

Nassella (nr). Nassella = 0, lipid absent (Rosengurtt el al., 1972).

31: Endosperm starch grain syndromes: 0 = Triticum-type (simple only, dimorphic in size, round or lenticular, free); 1 = Festuca-type (highly compound grains present, with or without simple grains); 2 = Andropogon-type (simple and compound together, the latter consisting of few granules); 3 = Panicum-type (simple only, uniform in size, small to medium, angular or sometimes smooth walled, densely packed); 4 = Brachyelytrum-type (simple only, large). NOTE: W&D report starch grains as "simple only" (= [034]) or "compound" (= [12]). Scoring here follows Tateoka's (1962) classification with one exception. He scored Brachyelytrum as having simple Panicum-type grains, but emphasized a major size difference. We recognize Brachyelytrum type as a separate state (see also Campbell et al., 1986). Additional reports come from Pammel, 1898; Yakovlev, 1950; and Tzvelev, 1977. 8, 4, 0.57, 0.72.

Flagellaria = 1 (Tillich, 1996).

Baloskion = 1 (Yakovlev, 1950: 160).

Anisopogon (W&D simple only). Anisopogon = 1 (Tateoka, 1962). There is only one species, and we follow the original report. Brachyelytrum (W&D simple only).

Brachyelytrum = 4. Chasmanthium (nr).

Chasmanthium = 3 (Tateoka, 1962).

Danthonia (W&D compound). Danthonia = [12], mostly Festuca-type, but a few species exhibit Andropogon-type (Tateoka, 1962).

Distichlis (W&D simple only). Distichlis = [034], from the report by W&D states 1 and 2 can be eliminated.

Eremitis (nr). Eremitis = 1 (V. Hollowell, pers. comm.).

Festuca (W&D 1). Festuca ovina = 1. Festuca generally has Festuca-type (Tateoka, 1962), but F. paradoxa Desv. has Triticum-type (Pammel, 1898). The latter is a member of an unusual North American section of the genus (though clearly Festuca s.str.) that Tateoka apparently did not sample.

Miscanthus (W&D compound). Miscanthus = 2 (Tateoka, 1962).

Nassella (W&D compound). Nassella viridula = [12], from the report by W&D states 0, 3, and 4 can be eliminated.

Panicum (W&D [13]). Panicum = [13], Tateoka (1962) reported 30 species with Panicum-type, two species with Festuca-type. Pennisetum (W&D simple). Permisetum = 3 (Tateoka, 1962).

Pharus (nr). Pharus = 1, compound only (Yakovlev, 1950).

Piptatherum (nr). Piptatherum = 1, included in Oryzopsis by Tateoka (1962). Pseudososa (nr).

Pseudososa = 3 (Tateoka, 1962). It should be noted that Arundinaria Michx. has Festuca-type (Pammel, 1898).

Sporobolus (W&D compound). Sporobolus = [13], Panicum-type in S. cryptandrus (Pammel, 1898: pl. xix, fig. vii), or Festuca-type (Tateoka, 1962).

Stipa (W&D report as simple only, or compound). Stipa = [0134]. Among Stipeae, Tateoka (1962) scored eight species of Stipa, and eight species of other genera, Oryzopsis, Piptochaetium, Achnatherum P. Beauv., and Orthoraphium Nees, as Festuca-type. However, given W&D's report, only the Andropogon-type can be eliminated for Stipa.

Zoysia (W&D compound). Zoysia = [12]. According to Tateoka (1962) this is compound, but the granules are few in number as in Danthonia. Tateoka scored Zoysia as Festuca-type but stressed intermediacy to Andropogon-type.

SEEDLING

32: Lamina of first seedling leaf: 0 = absent; 1 = present. 4, 3, 0.25, 0.57.

Flagellaria = 0, lacks a first blade (Jacques-Felix, 1988: fig. 3; Tillich, 1996).

Baloskion = I (H. P. Linder & B. Briggs, pers. comm.).

Pharus (nr). Pharus = 1, "the first blade is expanded, large, ovate" (Soderstrom, 1981: 23, 24, 26).

Bambusa (nr). Bambusa = 0 (Hoshikawa, 1969).

Eremitis (nr). Eremitis = 0 (V. Hollowell, pers. comm.). Miscanthus (nr).

Miscanthus = 1 (Hoshikawa, 1969).

VEGETATIVE ANATOMY

NOTE: Vegetative leaf anatomy characters 33, 36, and 37 are inapplicable for Baloskion (B. Briggs, pers. comm.).

33: Cross-venation of leaf blades: 0 = absent or obscure, or if distinct, then rare or occasional; 1 = present, distinct, more frequent than occasional. NOTE: The two states recognized for this character are intended to discriminate the regular occurrence of crossvenation from rare or accidental occurrences. In four cases we have assigned state 1 to species of genera recognized by W&D as lacking cross-venation (Ehrharta, Lithachne) or being variable for this character (Bambusa, Brachyelytrum). It is possible that these differences represent variation among observers in the interpretation of frequency and prominence of cross-veins, and reevaluation of other taxa may be warranted. 2-3, 1-2, 0.33-0.50, 0.81-0.90.

Flagellaria = 1 (pers. obs.).

Joinvillea = 1 (pers. obs.).

Bambusa (W&D [01]). Bambusa multiplex = 1 (pers. obs.).

Brachyelytrum (W&D [01]). Brachyelytrum erectum = 1 (Campbell et al., 1986; pers. obs.).

Ehrharta (W&D 0). Ehrharta calycina = 1 (Renvoize, 1985a); E. erecta also has cross-veins in mature leaves (pers. obs.).

Lithachne (W&D 0). Lithachne humilis = 1 (pers. obs.).

34: Multicellular microhairs: 0 = absent; 1 = present. 4, 3, 0.25, 0.90.

Baloskion = 0 (H. P. Linder & B. Briggs, pers. comm.).

Joinvillea = 1, often several-celled (Smithson, 1957).

Diarrhena (W&D [01] for microhairs). Diarrhena = 0. Bicellular microhairs lacking; unicellular microhairs are reported as possibly present, but rare, in one species (Scholz, 1982; Renvoize, 1985b).

Stipa and Piptatherum (W&D [01] for microhairs). All Stipeae = 0 for bicellular microhairs. Unicellular microhairs are reported as present, but rare, in a few species (Scholz, 1982; Renvoize, 1985b).

Molinia (W&D 0). Molinia caerulea = 1, (Jirasek, 1970).

35: Occurrence of a broad, short terminal cell, often with a longer basal cell in microhair, the walls of the terminal and basal cells similar in thickness: 0 = absence; 1 = presence. NOTE: Tateoka et al. (1959) distingished hairs of this form as "Chloridoid-type" as they are mainly restricted to subfamily Chloridoideae. This type of hair is contrasted with "Panicoid-type" microhairs, which have relatively longer and thin-walled terminal cells and are widespread among non-Chloridoid grasses. This distinction has been recognized in subsequent studies (Johnston & Watson, 1976; C&R; W&D, p. 28, #309). This character is inapplicable for taxa scored 0 for character 34.2, 1,0.50, 0.83.

36: Arm cells: 0 = absent; 1 = present. 4, 3, 0.25, 0.72.

Flagellaria = 0 (Smithson, 1957).

Joinvillea = \, slight lobing possibly present (Smithson, 1957).

Diarrhena (W&D 1). Diarrhena = 0. Soderstrom and Ellis (1987) report arm cells as absent, Metcalfe (1960) does not mention them (voucher, W. G. Dore 646, is D. americana P. Beauv. or D. obovata), and Macfarlane and Watson (1980) report them as absent. We have not found a source for the report by W&D, and consider it erroneous.

Streptochaeta (W&D, p. 880, "without arm cells, or at least without conspicuous arm cells, in the material seen"). Streptochaeta = 0. The genus has been examined in at least three studies and found wanting arm cells. Arm cells are not mentioned by Judziewicz and Soderstrom (1989), Soderstrom (1981), or Metealfe (1960) or Metcalfe's discussion of Page's (1947) work on the genus.

37: Fusoid cells: 0 = absent; 1 = present. 2-3, 1-2, 0.33-0.50, 0.81-0.90.

Flagellaria = 0 (Smithson, 1957).

Joinvillea = 1 (Smithson, 1957).

BIOCHEMISTRY

NOTE: Our primary source for photosynthesis subtype data was H&W.

38: Carbon fixation pathway: 0 = [C.sub.3]; 1 = [C.sub.4] NADP-ME classical-type; 2 = [C.sub.4] NADP-ME Aristida-type; 3 = [C.sub.4] NAD-ME. 4, 1, 0.75, 0.85.

Baloskion = 0 (H. P. Linder & B. Briggs, pers. comm.).

Aristida (W&D 1). Aristida = 2 (H&W).

Distichlis (nr). Distichlis = 3 (H&W).

Eragrostis (W&D [03]). Eragrostis curvula = 3 (Hattersley, 1987).

Panicum (W&D [0123]). Panicum virgatum = 3 (Hattersley, 1987).

Spartina (W&D 3). Spartina = 3 (H&W). Sporobolus (W&D 3).

Sporobolus = 3 (H&W).

Uniola (nr). Uniola = 3, "NAD-ME or PCK" (H&W).

Zoysia (W&D 3). Zoysia = 3 (H&W).

39: Carbon fixation PCK: 0 = absent; 1 = present. NOTE: PCK-type carbon fixation is recognized as a subtype of NAD - ME by H&W, who indicate that taxa with the PCK subtype exhibit both PCK and NAD-ME pathways, while other taxa exhibit only the NAD-ME pathway. Character 39 therefore is applicable only to taxa with state 3 for character 38.1, 0, 1.00, 1.00.

Distichlis (nr). Distichlis = 0 (H&W).

Eragrostis (W&D 0). Eragrostis curvula = 0 (Hattersley, 1987). Eragrostis species can be [C.sub.3], NAD-ME, or NAD-ME and PCK (H&W). PCK occurs among the annual species (H&W).

Panicum (W&D [01]). Panicum virgatum = 0 (Hattersley, 1987).

Spartina (W&D 1). Spartina = 1 (H&W).

Sporobolus (W&D [01]). Sporobolus = [01] (H&W).

Uniola (nr). Uniola = \, indeterminate, "NAD-ME or PCK" (H&W).

Zoysia (W&D 1). Zoysia = 1 (H&W).

CHLOROPLAST GENOME STRUCTURE

40: 6.4 kb inversion in the chloroplast genome relative to the arrangement in Nicotiana: 0 = absent; 1 = present. NOTE: Most scores are from Doyle et al. (1992), who used the polymerase chain reaction (PCR) to test for the occurrence of this inversion among a variety of exemplar monocots. The present restriction site mapping confirms presence of this inversion in additional grass species and in Joinvillea. 1, 0, 1.00, 1.00.

Baloskion = 0, determined by PCR (Jane Doyle, pers. comm.).

41: trnT inversion in the chloroplast genome relative to the arrangement in Nicotiana: 0 = absent; 1 = present. NOTE: Most scores are from Doyle et al. (1992); new scores are listed below. 1, 0, 1.00, 1.00.

Anisopogon, Anomochloa, Brachyelytrum, Lygeum, Molinia Schrank, and Poa L. = 1, determined by PCR (V. Tharp & J. Davis, pers. obs.).

42: 15 bp ndhF insertion in the chloroplast genome at Oryza sativa position 101951: 0 = absent; 1 = present. NOTE: Scores are from Clark et al. (1995). 2, 1, 0.50, 0.83.

X. Appendix 3: Structural Character Data

Data matrix (see facing page) for 42 structural characters (Appendix 2) as scored for 75 exemplar taxa of Poaceae and related families (Table I). Polymorphisms and subset ambiguities (see text) are signified as follow: A = [01 ]; B = [02]; C = [03]; D = [12]; E = [13]; F = [23]; G = [34]; H = [034]; I = [123]; J = [234]; K = [0134]. "Y-" = character inapplicable for a taxon; "?" = character state unobserved for a taxon; "\" = observed state is intermediate between recognized states, or there is uncertainty regarding homology of the observed state (see text and Appendix 2).

[TABULAR DATA FOR APPENDIX 3 OMITTED]
COPYRIGHT 1998 New York Botanical Garden
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1998 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Soreng, Robert J.; Davis, Jerrold I.
Publication:The Botanical Review
Date:Jan 1, 1998
Words:34366
Previous Article:Survey of Leguminosae - a study in adaptive radiation.
Next Article:Cecropia schreberiana in the Luquillo Mountains of Puerto Rico.
Topics:

Terms of use | Privacy policy | Copyright © 2018 Farlex, Inc. | Feedback | For webmasters