Recognition of Bromus Richardsonii and B. Ciliatus: Evidence from Morphology, Cytology, and DNA Fingerprinting (Poaceae: Bromeae)

Since our goal was to determine characteristic differences between Bromus richardsonii and B. ciliatus, a discriminate analysis (DA), principal components analysis (PCA), multidimensional scaling (MDS), bivariate analysis, and an amplified fragment length polymorphisms (AFLP) analysis were undertaken on 93 herbarium specimens and 31 field-collected populations. A cytological survey of B. ciliatus, B. richardsonii, and B. mucroglumis confirm previous reports that the first species is diploid (2n = 14) and the latter two are tetraploid (2n = 28). All taxa were correctly classified in the DA and important characters for each of the species were identified. Bromus richardsonii has lemmas with scattered hairs on the lower half between the midnerve and margins [glabrous in B. ciliatus], anthers (1.2) 1.6-2.7 (3.4) mm long [(0.9) 1-1.4 (1.6) mm long in B. ciliatus], second glumes (7.8) 8.9-11.3 (13.2) mm long [(6.2) 7.1-8.5 (9.5) in B. ciliatus]; and basal sheaths with dense, short to medium hairs [glabrous or with long hairs in B. ciliatus]. The PCA easily separated B. ciliatus and B. richardsonii into two well-defined groups and MDS mirrored the principal components analysis but displayed more overlap of individuals between the two groups. The AFLP-derived UPGMA dendrogram separated 154 individuals into two distinct clusters, one consisting entirely of B. ciliatus individuals and the other consisting of B. richardsonii individuals with six individuals of B. mucroglumis embedded within. Our study clearly indicates that there are distinctive morphological, cytological, and genetic differences to distinguish B. richardsonii and B. ciliatus as separate species.

and impo rta nt charac ters fo r each of the species were identified.Bromus rich ardson ii has lemmas with sca ttered hair s on the low er half between the mid nerve and margins [glabrous in B. cilia/US], anthers ( 1.2) 1.6-2.7 (3.4) mm lon g [(0.9) 1- 1.4 ( 1.6) mm long in B. cilia/us], second glumes (7.8) 8.9 -11.3 ( 13.2) mm lon g [(6.2) 7. 1-8.5 (9.5) in B. cilia /us ); and basal sheaths with den se, short to medium hairs [gl abr ou s or with lon g hairs in B. cilia /us].The PCA easily separated B. cilia/US and B. richa rdsonii into two we ll-de fined groups and MDS mirrored the principal components analysis but displ ayed mor e ove rlap of indivi dua ls between the two groups.The AFLP-derived UPGMA dendrogram se parated 154 ind ividua ls into two d istinct c lusters, one consisting entirely of B. cilia/us individuals and the other co ns isting of B. richa rdso n ii individuals with six individuals of B. mucroglumis embedded within .Our study c lear ly ind icates that there are distinctive morphological, cytological , and ge netic differen ces to distin gui sh B. richa rdso nii and B. ciliarus as separate species.

R ES UM EN
Puest o que nue stra meta era determinar diferencia s caracter fsticas ent re Br omus richardson ii y B. cilia /us.se emprendieron el ana lisis disc riminado (DA), el analisis de los co mpo ne ntes princ ipale s (PCA ), el escalamiento multidimen sion al (M DS), e l ana lisis bivariante, y un anal isis de los pol imorfism os de la lon gitud de fragrnent os am plificados (A FLP) en 93 espe cfm enes de herbario y 3 1 pobl aciones naturales.Una e ncues ta citol6gica de B. cilia/us, B. rich ardsonii.y B. mu croglumis co nfirma n pre vios informes que la primera es pecie es d iploide (2 n = 14) Y las ulrimas dos so n tetra ploides (2 n = 28 ).Todos taxo nes fuero n corr ecta rnente clasificados en el DA e irnport ant es ca racte res para ca da una de las espec ies fuero n ident ificados.Brom us richardson ii presenta lem as co n pel os d ispersos e n INTRODUCTION Bromus L. is a large genus of at least 150 species that occur throughout the world primarily in temperate regions (Clayton and Renvoize 1986 ;Pavlick 1995;Watson and Dallwitz 1992) .Along with Boissiera Hochst.ex Steud. and Littledalea Hemsl., Bromus has been placed in its own tribe, Bromeae, characterized by fused leaf sheath margins more than V<! their length (Clayton and Renvoize 1986;Soreng and Davis 1998).More recent classifications of the Poaceae align the Bromeae near the Triticeae (both tribes in subfamily Pooideae) by possessing simple endosperm starch grains and a hairy ovary with free style branches (Hsiao et al. 1999;Macfarlane and Watson 1982;Soreng and Davis 1998;Soreng et al. 2000).Bromus section Bromopsis Dumort., which includes Bromus ciliatus L. and Bromus richardsonii Link, is comprised of at least 26 species in North America (including Mexico ; NA), six species in Central America (all shared with NA), and 12 species indigenous to South America (14 in total, 2 shared with NA ) for a total of 38 species in the Western Hemisphere (Pavlick 1995;Planchuelo and Peterson 2000;Pohl and Davidse 1994;Soderstrom and Beaman 1968;Wagnon 1952 ).This section (Bromus sect.Bromopsis) contains mostly perennial (often short-lived) species that have I (3)-nerved lower glumes, 3(5)-nerved upper glumes, and spikelets that are terete (not laterally compressed) before anthesis (Pavlick 1995;Wagnon 1952).Our purpose is to clarify the species boundary between B. cilatus and B. richa rdsonii.
Bromus ciliatus is wide-ranging species often associated with wet meadows, stream banks, bogs, thickets, and occasionally talus slopes and roadsides.It occurs in the Aleutians Islands of Alaska and the Northwest Territories to Newfoundland, Canada south to Maryland, North Carolina, Tennessee, Illinois, Nebraska, Colorado to southern California (Pavlick 1995;Wagnon 1952).It is a diploid with a chromosome number of 2n = 14, as established by Wagnon (1952) who counted 67 individuals mostly collected from the Great Lakes-New England region where it is abundant.
Later reports from Canada and the United States confirmed the diploid status of B. ciliatus (Armstrong 1981(Armstrong , 1983;;Bowden 1960;Gervais 1979 ;Love and Love 1964 ;Mitchell and Wilton 1965;Schulz-Shaeffer 1956, 1960;Wilton 1965). Wilton (1965) also reported a diploid count from a Russian collection.Reports of tetraploid B. ciliatus from Wyoming (Reeder 1977 ) and the western United States (Barnett 1955;Elliot 1948Elliot , 1949) ) are probably in error and referable to B. richardsonii.A sterile triploid hybrid (2n = 21), apparently between B. ciliatus and B. richardsonii, with intermediate morphological characteristics was discovered in Alaska (Mitchell and Wilton 1965).Bro- mus ciliatus differs from B. richardsonii by possessing lemmas that are glabrous or scabrous on the back (although the margins are hairy), lacking a tuft of hairs at the summit of the sheath, and having anthers that are less than 2 mm long (Mitchell 1967;Mitchell and Wilton 1965;Pavlick 1995;Wagnon 1952).
Bromus richardsonii is found in open woods, canyons , exposed slopes, along creeks, trails, and roadsides in western North America from Southern Alaska (Little Susitna Valley) and Yukon, British Columbia south to western Texas, then west to Arizona, northern Mexico, and southern California (Mitchell 1967;Mitchell and Wilton 1965;Pavlick 1995;Wagnon 1952) .During field collecting in 1998 the second au-thor also discovered it in the Cypress Hills of southern Saskatchewan and adjacent Alberta.Unlike B. ciliatus, B. richardsonii is a confirmed tetraploid (2n = 28), as established by Armstrong (1981Armstrong ( , 1983)), MitcheIl and Wilton (1965), Ward and Spellenberg (1988) and Wagnon (1952), who surveyed 28 individuals from Alaska, Arizona, British Columbia, Colorado, New Mexico, and Wyoming .Bromus richardsonii can be separated from B. ciliatus by having lemmas with short appressed hairs on the back (in addition to hairy margins), a tuft of hairs at the summit of the sheath, and anthers that are generally more than 2 (2-3.5)mm long (Mitchell 1967;Mitchell and Wilton 1965;Pavlick 1995;Wagnon 1952) .
The autoecology of B. ciliatus and B. richardsonii also differs. Cayouette et al. (1997) found differences in plant height, winter survival, phenology (heading and maturity), clump diameter, and seed yield.Individuals of B. richardsonii had fewer and more erect branches, denser foliage, glaucous-green leaves (yellow-green leaves in B. ciliatusi, and greater rust resistance than individuals of B. ciliatus (Cayouette and Coulman unpubl.).
Bromus mucroglumis Wagnon, a species ranging from southwestern New Mexico to the San Francisco Mountains of Arizona south to Chihuahua and Sonora, Mexico, and then known from the Sierra de la Laguna (Baja California Sur, Mexico), was included in this study since it may be a possible introgressive hybrid with B. richardsonii (Wagnon 1950(Wagnon , 1952)).After counting seven individuals from Arizona and Baja California Sur, Wagnon (1952) concluded that B. mu- croglumis is also a tetraploid at 2n = 28 .Since its original description by Wagnon in 1950, B. mucroglu- mis has not generally been recognized as a distinct species until recently resurrected by Pavlick (1995), Scholz (1994), andKartesz (1999).We have only a cursory sampling of B. mucroglumis and expect to more fully investigate its relationship with other members of Bromus section Bromopsis in the future.
The present study was undertaken to clarify the distinctness of B. richardsonii and B. ciliatus by studying cytology, ecology, morphology, and molecular markers.The investigation involved: (I) determining chromosome numbers; (2) a numerical analysis based on morphological characters from a field study and herbarium specimens, and (3) a molecular study using AFLP markers to distinguish individuals and populations.

Plant Material
Herbarium vouchers and seeds were collected from 31 local ities for B.

Cytology
Cytological observations were carried out on living material from seeds collected in the field between 1993 and 1999 and grown in the Department of Botany greenhouse at the Museum Support Center (see Appendix I).One specimen per population was used for cytological study.Young root tips were pretreated in cold water and kept at 4°C for at least 18 h to condense the chromosomes (Darbyshire et al. 1992) before being fixed in Farmer's fixative (glacial acetic acid :absolute ethanol, I:3) for 12-24 h and then stored in 70% ethanol.Staining was in alcoholic hydrochloric acid -carmine for 6 days at room temperature (Snow 1963).Excised root tips were squashed in 45% glacial acetic acid mixed with a few drops of glycerin.Representative cells were counted and photographed with an Olympus BH-2 using Kodak TMAX film.Chromosome counts were made on 1-6 cells from each collection.

Morphological Data and Analyses
Data Matrix: Data were available for 20 morphological characters (Table I, characters 2-21) and a single ecological character (elevation, character I in Table I).For the multivariate analyses onJy characters 5-18 were used; the other characters exhibited high levels of missing data and would have affected those analyses too strongly.The other morphological characters (2-4, J9-21), as well as elevation, were investigated separately in bivariate scatter plots with other characters, where missing data have no effect other than not to be plotted.A complete set of 93 operational taxonomic units X 21 characters is avai lable from the first author upon request.Three forms of multivariate analysis were applied to the morphological data for characters 5-18; a Discriminant Analysis (DA), a Principal Components Analysis (PCA), and a Non-Metric Multidimensional Scaling (MDS).First, a discriminant analysis was performed using SYSTAT (Version 9, 1999, SPSS, Inc .)and this included specimens from all three taxa.Bro- mus mu croglumis was not included in the subsequent analyses because it was well se parated from the other taxa in the results of the DA, and because the number of specimens available for this study was small compared with the number available for the other two taxa.Instead, the general ordinations were run to identify variation patterns for these two well-represented species .The PCA was performed using the program PC-ORO (Version 4, McCune and Mefford 1999) using a correlation matrix of standardized data for the vari-ables, and three axes were interpreted as having variation relevant to this study.The choice of using the correlation matrix based on standardized data was to give each character the sa me contribution to the overall variance in the analysis.The mixture of pre senceabsence and measurement data is undesirable for either DA or PCA (Davis 1986) and this led to the application of MDS, also using PC-ORO, based on a matrix of Euclidean distances between the specimens.For most data, MDS has been demonstrated as the optimum technique for finding the optimum ordination in two or three dimensions (Clarke and Warwick 1994;McCune and Mefford 1999 ).
Final evaluation of the two general ordinations was done by calculating a cophe netic correlation coefficient (Sneath and Sokal 1973) and plotting the final, inferred distances between the specimens against the to the restriction fragments in order to create primary templates for pre-amplification; (2) pre-amplification of the primary templates with AFLP primers with an additional sing le nucleotide at the 3' end; (3) selective amplification was performed with -y33P-Iabeled EcoR I primers having three selective nucleotides at the 3 ' end (AGG, ACG, AAG) in combination with Mse I primers having three se lective nucleotides at the 3' end (CTC, CTG, CGC, CAC); and ( 4) the amplification products were separated on 5% polyacrylamide gels for 2.30 h at 80 W. The gels were transferred to Whatman paper and dried on a gel dryer for 2 h at 80 aC.Gels were exposed to Kodak BIOMAX film at -80 aC for 1-7 days depending on the signal intensity.
AFLP products were scored as present (I) or absent (0) on autographs to create a binary matrix.Euclidean distance matrices between individuals and locations were computed by using NTSYS-PC 2.0 I (Rohlf 1997).Dendrograms were constructed using unweighted pair-group method of arithmetic averages (UPG-MA) cluster analysis using the Euclidean distance matrices.The goodness of fit of each of the dendrograms to the relevant Euclidean distance matrices were assessed using cophenetic correlation coefficients (Sneath and Sokal 1973)  starting Euclidean distance matrix, although this method was also used in an exploratory manner to compare Cytology the results of the MDS and PCA.

DNA Fingerprinting
For the AFLP analysis seeds from the 31 localities were grown in the Department of Botany greenhouse at the Museum Support Center and harvested when young (at two months of age).Four to six leaf blades from four to six individuals of each accession were harvested and lyophilized prior to DNA extraction.From each sample two leaves were placed in a 2 ml microcentrifuge tube with two 3 mm glass beads.The tubes were placed on a horizontal shaker until the leaf tissue was ground to a fine powder and the DNA was extracted using DNeasy@l Plant Mini Kit (QIAGEN lnc.) according to the manufacturer's directions.Total genomic DNA was quantified by fluorimetry using Hoechst 33258 stain (Sigma Chemical Co, St. Louis, USA).
Four EcoR I:Mse I primer combinations were used in this study.The AFLP method was performed as described by Vos et al. (1995) using the AFLp@l Analysis System I provided by Life Technologies.This general protocol included four main steps : (I) restriction digestion of 250 ng of genomic DNA with EcoR I and Mse I restriction enzymes and ligation of adapters Chromosome counts are given for 36 specimens (Appendix I).Thirty-one came from our vouchers and the rest from voucher collections of previous studies (Cayouette, Wojtas, and Fillion unpubl. data;Wagnon 1952).All twenty-two counts for B. ciliatus were diploid 2n = 14, whereas all 12 counts of B. richardsonii and both counts of B. mucroglumis were tetraploid 2n = 28 (Fig. 2).These mitotic counts were made at late prophase, metaphase, and/or anaphase.Satellites were observed for both diploid and tetraploid cytotypes.

Discriminant Analysis
Despite some characters being binary and therefore not optimal for DA (Davis 1986), the results obtained were excellent in that all specimens (93 specimens, see Appendix I for locations) from all three taxa were correctly classified using characters 5 through 18. DA axis I easily discriminated between the two main taxa, B. ciliatus (47 specimens) and B. richardsonii (42 specimens), and DA axis II separated B. mucroglumis (4 specimens) from the other two species (Fig. 3; Table 2).The three most important characters and their states (presence for binary characters, magnitude for lengths) that served to separate each taxon from the others are B. ciliatus 2n = 14 B. richardsonii 2n = 28 as follows: for B. ciliatus, abaxially hairy top culm blade (9), hairy top culm node (5), and hairy top culm sheath (6); B. mucroglumis, glume I length with a mean of 6.60 mm (12), lemma 2 with hair s not restricted to the margin area ( 16), and adaxially hairy top culm blade (10); B. richardson ii, lemma 2 with hairs not restricted to margin area (16), hairy top sheath margin (8), and glume 2 length with a mean of 10.10 mm (13).
The mean and standard deviation for all qu antitat ive characters are given in Table 3. Bromus

Principal Co mpo ne nts Analysis
Using ch aracters 5 through 18, the first three PC's accounted for 60 .4% of the total variation in the data.The eigenvector information and general PC data are gi ven in Table 4.The first axis is by far the largest, accounting for just under 40 % of the total variance.It has positive loadings for all lengths and negative loadings for most binary characters denoting hairiness (c haracte rs 5, 6, and 9).The position of each specimen   1944 :!:: 157 (4 ) 6.60 :!:: 0.65 (4) 8. 15 :!: 0.99 (4) 1.70 :!: 0.32 (4) 7 .35± 1.34 (2) along PCl, therefore, is a general indication of its hairiness (more hairy to the negative side, less hairy to the positive side) and size (larger specimens to the positive side, smaller one s to the negative .The plot of PCl vs .PCl l (Fig. 4 ) shows that specimens belonging to B. richardsonii have larger glumes (both I & 2) than those belonging to B. cillatus, but the latter tend to be much more hairy, providing a definite separation between the taxa.The characters that have negative high loadings, and thus are c har acteristic of the hairy and shorter-glumed taxon B. ciliatus (characters 5, 6, and 9) show hairs that occu r on the top c ulm node, top culm sheath, and the top culm blade abaxial surface ; se en also in the DA .The other two PCs reflect variation with B. richardsonii, with pcn (Fig. 4 , 5) loading highly for top sheath margin (8), glume I vestiture (12), glume 2 vestiture ( 15) and palea 2 vestiture pattern (18 ) and PC III (Fig. 6) loading negatively for top blade collar (7) , top she ath margin (8) and glume 2 apex (14) and mo st highl y positive for lemma 2 central nerve basal half (17).The general geometry of the vari ance of the three dimensions, as shown (Fig .4-6) , is for a small sphere of specimens in negative space for PCI with near-zero values for the other two axes for B. ciliatus, and a terete-ring for B. richardsonii in the positive space of PCI and showing large variation in PCs Il and III, with relatively few near-zero values for those axes ; a small sphere of B. ciliatus in front of a terete-ring of B. richardsonii.These data argue for a detailed investigation of variation within the species B. richardsonii.

Multidimensional Scaling Analysis
The mo st robust ordination method available for these data is MDS (Fig. 7 The overlap between the two species is greater than in the PCA plots, which might suggest the method was less successful, but the cophenetic correlation for the MDS (r = 0 .98) is very much better than that for the PCA (r = 0 .76), even allowing one more dimension to the PCA.Bivariate plots of the inferred versus starting distances show the MDS to be much more representative of the true relationships inferred by the Euclidean distance matrix.Overall, the differences between the taxa are still quite distinct, although a few specimens of B. richardsonii do overlap within the Table 4. Eigenvector loadings for c haracte rs 5-18 o n the first three principal co mpo ne nts (PC ).Relative eigen values, percent of vari ance, and c umulative percent of var ian ce are listed below.Further investigation is possible for those variables represented by continuous data (the lengths and elevation) through bivariate comparisons.This especially helped in examining trends in those variables not used in the multivariate analyses; specimens with missing data were simply excluded from the graphics and statistics.These bivariate graphs each contain four comparisons (Fig. 8-15).The data are shown for each taxon using different symbols and correlation coefficients were calculated for each pair of variables for B. ciliatus and B. richardsonii.Bromus mucroglumis was represented by too few specimens to be analyzed.
The effect of elevation on glume length (Fig. 8, 9) clearly shows that B. richardsonii has larger glumes (higher values for the y-axes) and is typically found at higher elevation (to the right in the graphs), when compared with B. ciliatus, reinforcing the results of the summary statistics and multivariate analyses.Neither taxon shows a significant correlation of glume length with elevation.However, if all Bromus data are combined, there is an apparent, highly significant correlation (P ~0.001) in both cases, a theme that will be seen in most of the bivariate graphs.This suggests potential utility in looking at data in a more combined form for an analysis of allometry in the genus by including additional species.However, with only two species with any significant sample sizes, it also suggests extreme caution in that the apparent, highly significant correlations may simply be an artifact of the positions of two uncorrelated distributions of individual taxa rather than a trend for the genus.
The most straight-forward morphometric comparison is that of the two glume lengths (Fig. 10) which show highly correlated (P ~0.00 I) trends in both individual taxa, and as combined data.The slopes of the reduced major axes are sub-parallel (approximately 1.1 for B. ciliatus and 1.06 for B. richardsoniiy showing that glume 2 increases in length slightly quicker than glume 1 length in both species, also reflected in the higher values for glume 2 length.The combined analyses show a slightly higher slope of 1.135.The next step was to analyze variation in two continuous measurements (anther length and caryopsis length) which could not be used in the multivariate analyses as a result of missing data.First comparison of these two variables (Fig. II) with each other shows the expected smaller sample sizes for all taxa, the expected larger size of B. richardsonii, and non-significant correlations for both species.Their combined data again show the highly significant correlation (P ~ Fig. 7. Mu ltidi me nsio na l sca ling of morphological data (charac ters 5-18) for Brom us cilia/us ("') and B. richardsonii (+) .Th e sco res of a ll indiv idua ls are proj ected onto the two-dimensional space defin ed by axis 1 and axis II (r = 0.98).

Oth er Cha rac te rs
Basal sheath ves titure (3) appears to be a very goo d character since all 44 speci mens of B. ciliat us were g labro us or had long hairs wh ereas 35 out of 38 specimen s (92%) of B. richa rdsonii had den se, short or medium hairs.All 40 spec ime ns of B. ciliatus had g labrou s or long basal hairy shea th c row ns ( 4) whe reas 8/34 (24%) spec ime ns of B. richardsonii had hairs that we re longer and crow de d .Finall y, cary opsis co lor (2 1) is some wha t subjective since color can be very hard to qu anti fy, particularl y whe n see n by more than on e per son .However, 31/34 (9 1%) specimens of B. cilia tus had Iight to amber bro wn frui ts and 14/19 (74 %) of speci me ns of B. richardsonii had dark amber to purple fru its.

A FLP Analy sis
In total fo ur primer co mbina tions produced 41 3 band s (Table 5).In each primer combination the number of band s ranged from 77 to 136.Of all the band s fro m each primer combination only 15 were mon omorphic.Th e percentage of polymorphic band s for each primer combination ranged from 92% to 98 %.From eac h of the four primer combina tio ns used , an

C
• -:« .. - average of 20 polymorphic bands were used in the analysis .These bands were selected on the basi s of robustness of the band, and ea se of scoring w ithout any ambiguity.There were many other polymorph ic bands that were not used in this analysi s because of diffi culty in scoring.
Th e UPGMA dendrogram divides all 154 individuals into two distinct clusters, one consisting entirely of B. ciliatus individuals and a second consisting of B. richardsonii and B. mucroglumis individuals (Fig. 16).The cophenetic correlat ion coefficient for this plot is very robust at r = 0.9 8. Within thi s second clu ster, all six individuals of B. mucroglumis form a cluster embedded in the B. richardsonii group.There is no overl ap of indi viduals between species.Within each species, individuals from one location usually clu stered with each other.
A second UPGMA dendrogram using a matri x of 31 locations als o shows two distinct clusters corresponding to B. ciliatus and B. mucroglumis/B.richardsonii gro ups (Fig. 17

DISCUSSION
Our chromosome data confirmed that B. ciliatus is diploid and B. richardson ii and B. mucroglumis are tetr aploid (Armstrong 1981(Armstrong , 1983 ; ;Mitchell and Wilton 1965;Wagnon 1952;Cayouette et al. 1997;Cayouette, Wojt as, and Fillion unpubl. data;Ward and Spellenberg 1988 ).Triploid cytotypes (2n = 2 1) were not encounte red .The pre sen ce of small pin -head satell ites on some chromosomes of B. ciliatus and B. richardsonii has been previously documented (Armstro ng 1981, 198 3;Gerv ais 1979;Schulz-Schaeffer 1956, 1960;Wilton 1965 ), but their number s vary according to the cytotypes and to the interpretations of the authors.Satellite observation is sometimes ob scured by various techniques used to shorten chromosomes (Armstrong 1983 ;Gervai s 1979 ).Arm strong (1981) con sidered their presence and qu antity importa nt when asses sing  B. richardsonii into two well-defined groups, and the MDS also mirrored the PCA but displayed more overlap of individuals between the two groups.The latter technique (MDS) appears to be much better at representing the Euclidean distance matrix since the cophenetic correlation was high (r = 0.98) as compared to the PCA (r = 0.76).In general, the bivariate analyses show the new characters acting much like the continuous data used in the multivariate analyses; i.e., the two glume lengths.Each shows a strong size-based discrimination between B. cilia/us and B. richardsonii, although comparisons amongst these variables shows weaker or no correlations between the variables compared with the two glume lengths.
Clearly more intense sampling of B. mucroglumis is needed to ascertain whether or not this entity is indeed a distinct species.As mentioned in the introduction, Wagnon (1950Wagnon ( , 1952) ) indicated that "most of the variation [of B. mucroglumis] is apparently in the direction of B. richardsonii, thus suggesting introgression with this species."During our most recent collecting trip (September/October of 2000) the first two authors observed B. mucroglumis from the Chiricahua Mountains (Arizona) as having fairly consistent morphological features in which to separate it from B. richard- sonii, also from the same locality.These were hairy, first (character 12) and second (15) glurnes, hairy, top culm blades both upper (10) and lower surfaces (9) , hairy top culm nodes (5), and long hairy top sheath margins (8).Hairy first glumes and upper top culm blades were identified earlier in our DA as being important for B. mucroglumis.However, when we collected and observed other specimens of B. mucroglumis (apparently) and B. richardsonii from Sonora, Chihuahua, and Durango, Mexico, hairiness of the top culm blade, nodes, and top sheath margins became less reliable features to differentiate between these taxa.We hope to clarify the relationship of these two species in future studies.
The high percentage of polymorphic AFLP bands (92 %-98%) and low number of monomorphic bands observed in this study are indicative of a highly outcrossing species that are widely distributed geographically.Both of these latter characteristics are found in B. cilia/us and B. richardsonii.In a diversity study of wheat (self-pollinating) cultivars, the AFLP marker polymorphism rate varied from 1.2% to 26 .1% (Barrett and Kidwell 1998).Whereas, in maize (cross-pollinating but intensely selected) the range was 43% to 63% (Marsan et al. 1998).There were two types of polymorphic markers observed in this study: (I) AFLP markers that were common to all three species but present in different frequencies and (2) AFLP markers that were specific to a species.This combination of common and species-specific markers was previously observed in a study conducted on cultivated bromegrass species, B. inermis Leyss.and B. riparius Rehmann (Ferdinandez 1999).
The ability to distinguish closely related groups by AFLP markers, are well documented in other genera (Fuentes et al. 1999;Powell et al. 1996) and proved decisive in our study.The large numbers of polymorphic bands yields large amounts of information to assess genetic diversity between individuals and populations of B. cilia/us and B. richardsonii.The dendrogram was able to cluster the 154 individuals into two distinct groups.The genetic relationships determined by the AFLP markers are consistent with the chromosomal differences and morphological conclusions.The clustering of B. mucroglumis within the B. richardsonii group indicates a close genetic relationship between the two species.Similar results were observed in a phylogenetic study conducted on tall fescue using RFLPs (Xu and SIeper 1994).In this study genotypes from the same species with similar chromosome num-
Despite overlap of some individuals between locations the majority of the individuals from a particular location clustered together.It may be possible to differentiate closely related individuals with greater resolution by increasing the number of polymorphic markers.However, it is also possible that clustering of individuals from different locations is a result of genomic similarity rather than the inability of the AFLP markers to determine the differences.If the former is true, increasing the number of markers will only heighten the detectable differences and will not alter the fact that certain individuals from one location are genetically more similar to individuals from another location.In the present study, it appears that it is possible to differentiate closely related individuals collected from the same location and of the same species.

Taxonomic Treatment
Since it appears that B. ciliatus and B. richardsonii are distinct species, the following key using morphological features is given to separate these taxa.The characters used in this key are listed in descending order for ease of use.
) .All the B. ciliatus individuals form a tight cluster in the upper left of the plot and these are somew hat surrounded by four individuals of B. richardsonii.One individual outlier of B. ciliatus near the upper center of the plot is depicted within the more open cluster of B. richardsonii individuals.

.Fig. 4 -
Principal components analysis of morphological data (characters 5-18) for Bromus cilia/US (A) and B. richardsonii (+).The scores of all individuals are projected onto the two-dimensional space (r = 0.76).--4.Axis I (accounts for 39.7% of the variation) and axis II (accounts for 10.7% of the variation).-5.Axis II and axis III (accounts for 10.0% of the variation).-6.Axis I and axis III.
).Because all individuals of a population are combined one would expect a lower cophenetic corre lation coefficient for this plot ( r = 0.8 9).The sin gle population of B. mucroglumis from the Chiricahua Mountains again fell within the B. ri- chardsonii group.Within species, different collections from similar geographic areas (province or area of country) tend to clu ster together.However, there we re som e e xceptions, for example an Alb erta co llection of B. cil iatus (LAM-c-AB) clustered with a Qu ebec (GRA I-c-Qc) collection.
Fig. 17.Genetic relation sh ips of 31 popul ation s of Bro mus ciliarus (21 total) , B. mucroglumis ( I total), and B. richardson ii (9 total) as depi cted by a UPG MA dendrogram (r = 0.89).Geog raphica l ide ntifiers (see Append ix I) follow ed by state abbreviati ons are give n for each popul at ion .

Table 2 .
Summary of the canonical discriminant function s (O F) used to generate Figure 3.