Amer. J. Bot. 74(10): 1577-1584. 1987.

A NUMERICAL ANALYSIS OF FLAVONOID VARIATION IN SUBGENUS AUSTROMONTANA ()'

STEVEN J. WOLF AND RICHARD WHITKUS Departmentof Biology, University of Missouri, Kansas City, Missouri 64110; and Departmentof Botany, Ohio State University, Columbus,Ohio 43210

ABSTRACT Species of Arnica subgenusAustromontana produce a total of 23 leaf flavonoids, including simple and methylatedflavone and flavonol glycosides as well as highly methylatedflavone aglyconesand a 6-hydroxylatedflavone. Most of the taxa exhibitconsiderable interpopulational variability,with the number of compounds per populationranging from 2 to 14. Analysis of flavonoid variation in 1 3 populationsrepresenting all 9 species of the subgenuswas carried out using cluster analysis, principal components analysis, and binary discriminantanalysis. Results indicate the flavonoid profileof the very rareA. viscosais the most distinctive in the subgenus.Although exhibiting considerable interpopulational variability, all populationsof A. gracilis, a hybridtaxon, form a very distinct and cohesive group,supporting its recognitionat the specificlevel. Additionally,chemical diversification from A. cordifoliahas takenplace largely in the Klamathregion of Oregonand . The rangeof variabilityexhibited by A. cordifolia is reflectedin these Klamathregion derivatives.

ARNICA L. subgenus Austromontana Maguire 1983, 1984b, c). In general,the rare Klamath consists of nine montane to alpine species re- region endemics are characterizedby reduced stricted primarily to western flavonoid profiles and/or more methylated (Wolf and Denford, 1984a). Members of the aglycones, but the wider rangingspecies have subgenus are extremely polymorphic, largely fewer methylated aglycones and more glyco- due to microspecies formation via apomixis sides (Wolfand Denford, 1984c). Additionally, (Wolf, 1980; Wolf and Denford, 1984a). Re- most of the taxa exhibit considerable inter- cent studies based on morphological,cytolog- populational variability, with the number of ical, geographical,and chemical analyses con- compounds per population rangingfrom 2 to cluded that the two widespread species A. 14. This distribution pattern is probably the cordifolia Hook. and its derivative A. latifolia result of a combination of factors including Bong., are ancestral species of the subgenus founder effect, genetic drift, and a general re- (Wolf and Denford, 1984b, c). From these two duction of flavonoid profile as a result of re- species, major evolutionary diversification productive isolation in both the rare taxa and within the subgenushas taken place largelyin apomicts (Wolf and Denford, 1983, 1984c). the Klamath region of southwestern Multivariatestatistical analysis increasingly and northwesternCalifornia (Wolf and Den- has become routine in systematic investiga- ford, 1984c). Four of the nine species are rel- tions. This is not surprising since the large atively rare and are restrictedto the Klamath number of computer programsavailable and region. Additionally,A. gracilis Rydbg., an al- their ease of use have made the handling of pine species of the Rocky Mountains and Pa- previously unwieldly data sets relatively sim- cific Northwest, is a hybrid between A. cor- ple. Numerical techniques are particularly difolia and A. latifoliaand has been recognized valuablefor analyzingvariability, both at pop- at the specificlevel (Wolfand Denford, 1984b). ulational and taxonomic levels, as well as de- Flavonoid chemistry has been especially limiting distinguishingcharacters and display- useful in suggestingevolutionary relationships ing relationships.However, the application of within Arnica subgenus Austromontana. numerical techniques in chemosystematic in- Species of the subgenuselaborate a total of 23 vestigationsof flavonoids is a relativelyrecent leaf flavonoids, including simple and meth- phenomenon, and as Bohm, Banek, and Maze ylated flavone and flavonol glycosides as well (1984) have noted, most such investigations as highly methylated flavone aglycones and a largely have ignored populational variation. 6-hydroxylated flavone (Wolf and Denford, Recent studies, such as those by Wolf and Den- ford (1983), Parker and Maze (1984), Mc- I Received for publication 15 October 1986; revision Dougal and Parks (1984), and Bohm et al. accepted 3 February 1987. (1984), have begun to addressthe significance

1577 1578 AMERICAN JOURNAL OF BOTANY [Vol. 74

TABLE 1. Distributionofflavonoids in Arnica subgenusAustromontana

CE NE GR CO DI SP VE LA VI Compound (7)a (10) (11) (33) (1 1) (9) (4) (26) (2)

1.A 6-Me lOb 7 24 10 6 4 13 2 2. A 7-Me 1 4 1 4 2 3. A 6,7-Me 6 2 4. L 4'-Me 10 9 5. L 6-Me 6 6. L 6-OH, 4'-Me 2 7. L 6,4-Me 3 4 2 8. L 3',6,7-Me 2 9. Q 3-Me 7 2 10. Q 6-Me 7 3 11. Q 3',6-Me 2 12. A 7-O-glu 9 13. L 7-O-glu 11 15 3 14. L 6-Me, 7-O-glu 7 7 2 15. K3-O-glu 2 6 10 14 11 6 4 9 2 16. K 3-0-gal 11 17 17. K 6-Me, 3-O-glu 2 8 6 18 8 9 4 18. Q3-O-glu 5 10 6 18 9 3 12 2 19. Q 3-O-diglu 5 10 10 32 11 9 4 26 2 20. Q 3-0-gentiobioside 7 10 10 33 11 9 4 26 2 21. Q 6-Me, 3-O-glu 11 18 2 22. Flavone glycoside 2 23. Flavone glycoside 2 Total compounds 5 8 14 10 11 8 6 8 14 Abbreviations: A. cernua(CE), A. cordifolia(CO), A. discoidea(DI), A. gracilis(GR), A. latifolia(LA), A. nevadensis (NE), A. spathulata (SP), A. venosa (VE), A. viscosa (VI), A-apigenin, L-luteolin, Q-quercetin, K-kaempferol, glu-glucose, gal-galactose, Me-OCH3. a Number of populations surveyed. b Number of populations containing given compound.

of populational variation in flavonoid profiles. for the presence or absence of 23 flavonoid Although previous investigations of the fla- compounds (Table 1). Descriptions of popu- vonoid chemistry of Arnica subgenus Austro- lation sampling techniques, flavonoid extrac- have contributed greatly towards an tion and identification are given in Wolf and understanding of evolutionary relationships Denford (1983, 1984c). A copy of the data within the group, some previous hypotheses matrix is available upon request. were necessarily tentative due to small sample Variation in flavonoid profiles both within sizes for some taxa (Wolf and Denford, 1984c). and among species was first examined with Further analyses of a much larger number of cluster analysis of the OTU's to see whether populations has indicated a greater range of populations of the taxa formed cohesive groups. variation than previously noted and has al- Similarities among OTU's were calculated by lowed for more definitive hypotheses concern- the simple matching coefficient of Sokal and ing evolutionary relationships within the sub- Michener (1958): genus. The present study is based on a sampling A + D/A + B + C + D, of 1 3 populations representing the entire geo- graphical range and all nine species of Arnica where A is the number of characters whose subgenus Austromontana (Table 1). Due to the presence is shared by two OTU's, B and C the large number of populations sampled and the number of characters present in one OTU but considerable variability noted, the data were absent in the other, and D the number of char- examined using various clustering and ordi- acters whose absence is shared by two OTU's. nation techniques in an effort to both confirm This coefficient gives equal weight to the shared previous hypotheses and, perhaps, provide new presence and absence of characters. Because insights into evolutionary relationships within evolutionary diversification in the subgenus has the subgenus. been accompanied by a reduction in flavonoid variation (Wolf and Denford, 1984c), shared MATERIALSAND METHODS-To form the ba- absence of compounds is as important as shared sic data matrix, 113 populations (OTU's) of presence. Phenograms were produced with av- Arnica subgenus Austromontana were scored erage linkage clustering (UPGMA), which gen- October 1987] WOLF AND WHITKUS-FLAVONOIDS IN AUSTROMONTANA 1579

CEI CE2 HM 4 I LA24

LA 16 3 4 Cal 2 4 CES CE6 6 4 6 SP2 33 Spa 4 SP9 7372 3 e 2 3 4 4 83 37 2 29224 72 2 2 82 2 2 4 _L{S C02~C0 2 4 I 2 2 2 4

3 6 CO3 C01 2 5 4 1 5

LAIS 2 5 5 CaLl 2 7 2 2 5~~~~~5 2 5 5 2 5 5 5 5

LAI3 Fig. 2. Projectionof 113 populations of Arnica sub- genusAustromontanaonto firsttwo principalcomponents. Axes I and II represent38.6% of the total variation. A. CasA LASS cernua(1), A. cordifolia(2), A. discoidea(3), A. gracilis(4), A. latifolia(5), A. nevadensis(6), A. spatulata(7), A. venosa 2 LAE (8), and A. viscosa(9).

Ca12 erally gives the least amount of distortion of a similaritymatrix (Rohlf, 1970; Sneathand So- kal, 1973). Ca02 Clustering methods tend to faithfully rep- resent the similarity matrix at lower levels of COA9 clustering,but distortthe matrixat higherclus- C032 teringlevels (Sneathand Sokal, 1973). To bet- ter representphenetic relationships at higher levels, ordination techniques are considered more appropriate(Sneath and Sokal, 1973). Phenetic relationshipsamong the taxa, there- fore, were explored with principalcomponent analysis (PCA) of the covariancematrix of the data. Results of such an analysis are identical to a principalcoordinate analysis of presence- absence data (Gower, 1966; Neff and Marcus,

CD 1 1980). However, an additional benefit accrues from a PCA in being able to interpretthe prin- cipal axes in terms of charactercombinations, a feature absent in principle coordinate anal- C0a8 COA ysis. Although a PCA indicates which char- actersexplain the majorityof the overall vari- ation among OTU's, it does not separate COAI between taxa vs. within taxa variation since no a priorigrouping of OTU's is done. To find those characters which maximally separate taxa, discriminant analysis is appropriateon pre-chosen groups of OTU's (i.e., taxa). Be- cause of various assumptions about the data,

Fig. 1. Phenogramfor 113 populationsof Arnicasub- 5.5 A.6 0.t 0.8 0.9 1.0 genusAustromontana. Taxa abbreviationsfollow Table 1. 1580 AMERICAN JOURNAL OF BOTANY [Vol. 74

II 4 2

38 2 1

7 I 9 Fig. 5. Projection of A. cordifoliaand its Klamath de- 8 7 1 rivatives onto the first two axes of binary discriminant 6 analysis. Numbers follow Fig. 2.

5 tions were performedon the IBM 3081 at Ohio State University. Fig. 3. Projectionof the nine species of Arnica sub- To explore different questions of relation- genus Austromontanaonto the first two axes of binary discriminantanalysis. Numbers follow Fig. 2. shipswithin the subgenus,three sets of analyses were performed.First, all taxa were analyzed simultaneously( 113 OTU's) to identify overall discriminant analysis is performed with con- patterns of relationshipswithin the subgenus. tinuous variables. However, binary discrimi- A second analysis explored the relationships nant analysis (BDA, Strahler,1978) is a meth- amongA. cordifoliaand its presumedKlamath od for findingcombinations ofbinary variables region derivatives (64 OTU's). Finally, an which are most important for discriminating analysis of A. cordifolia,A. latifolia and their groups. Thus, to identify those combinations hybrid derivative A. gracilis, was undertaken of compounds which maximally separate the (94 OTU's) to see if the chemical data support taxa, a BDA of the taxa was performedusing the recognitionof the latterat the specificlevel. the method of Strahler(1978). Clusteranalysis was performedwith the NT- RESULTS-Thephenogram for the analysis of SYS programpackage (Rohlf, Kishpaugh,and the subgenus(Fig. 1) indicates several clusters Kirk, 1972), PCA and BDA with the BMDP comprised of OTU's of a single taxon. Most program package (Dixon, 1981). All calcula- notable among these are A. gracilis and A. vis-

D:i r:

2 2 2 2 2 3 2 2 22 2 2 2 3 2 3 2 3 2 2 1 32 3 2 2 4 2 3 2 2 4 2 2 7 2 4 2 2 4 2 4 4 -I 22 I 2 2 4 4 I 5 55 4 4 2 27 7 5 2 7

28 5 5 55 5 2 2 5 5 54 5 2~~~~~~~ 5 5 3

Fig. 4. Projection of 64 populations of A. cordifolia Fig. 6. Projection of 94 populations of A. cordifolia, and its Klamath derivatives onto the first two principal A. latifolia, and A. gracilis onto the first two principal components. Axes I and II represent46.1% of the total components. Axes I and II represent 47.7% of the total variation.Numbers follow Fig. 2. variation. Numbers follow Fig. 2. October 1987] WOLF AND WHITKUS-FLAVONOIDS IN AUSTROMONTANA 1581

GR 1 TABLE the GR2 2. Sortedfactorloadings greater than 70%for GR7 first threefactors of the BDA of Amica subgenusAus- 083 GR4 tromontana.Thefirst threefactors accountfor 76.12% 088 of the total variation GR8 OGR Character Factor I Factor 2 Factor 3 G R9 GR10 6 0.987 0.0 0.0 LA 1 LA3 8 0.987 0.0 0.0 LA2 11 0.987 0.0 0.0 LA12 LA9 22 0.987 0.0 0.0 LA 11 23 0.987 0.0 0.0 LA17 -LA8 14 0.0 0.987 0.0 -LA14 13 0.0 0.982 0.0 LA25 9 0.0 0.905 0.395 LA26 LA 19 12 0.0 0.835 0.499 LA23 LA4 16 0.0 0.263 0.936 LAI18 21 0.0 0.0 0.932 LA2 2 10 0.0 0.701 0.703 CLA7 LA 13 LA2 1 LA5 LA 15 cos which accounts for an additional 11.9%of the LA24 total variation. -C04 In the BDA (Fig. 3), A. viscosaseparates from LA2O C02 the remaining taxa along the first axis on the CO03 COG basis of compounds 6, 8, 11, 22 and 23 (Table rCO 1 2 2), all of which are unique to this taxon. The LCOI13 LAI10 remainingtaxa spreadout along axis 2, and to I LA 6 C0le a lesser extent on axis 3 (not shown). -CO22 Both the PCA (Fig. 4) and the phenogram C029 (not shown)for the analysisofA. cordifoliaand -CO215 018 its presumed Klamath region derivatives in- co 19 C024 dicate a lack of distinct clusters, indicating a CO017 shared overall variational pattern and a lack Cos C025 of unique flavonoid profilesfor any one taxon. C023 C020 Furtherevidence for this pattern is shown in C02 1 results of the BDA where all the EC028 the (Fig. 5) 0.7 0 0. 0 1C030 taxa are essentially distinct from one another. ~ ~~~ ~~~~ ~~~~~~~~~~~C03 1 LC032 Factor loadings (Table 3) indicate that only ~~~CO014 4 C027 compound 13 (presentin A. cordifoliaand A. C07 discoidea) can clearly discriminate along an C033 axis, while the remainingcharacters have load- ings on two or more axes. Colo AnalysisofA. latifolia,A. cordifoliaand their 0. 7 0.8 0.8 1.0 hybridderivative indicatesa unique patternin Fig. 7. Phenogramof 94 populationsof A. cordifolia, A. gracilis. In the PCA (Fig. 6), A. gracilis is A. latifolia and A. gracilis.Abbreviations follow Table 1. separatedfrom its parentson the firstaxis. The

TABLE 3. Sortedfactorloadings greater than 70%for the cosa which are two ofthe most distinct clusters. first threefactors of the BDA of A. cordifoliaand its Other evident clusters are those comprised of Klamathderivatives. Thefirst threefactors account for A. latifolia and A. nevadensis.Shared flavonoid 90.99%of the total variation profiles in the remaining OTU's results in their Character Factor I Factor 2 Factor 3 intermixing throughout the remainder of the 1 0.928 0.334 0.0 phenogram. 15 0.811 -0.464 0.333 The overall pattern seen in the phenogram 17 0.763 -0.525 0.0 is also exhibited by the PCA (Fig. 2). Again, 2 0.741 -0.379 0.0 A. gracilis is the only taxon which forms a 13 0.0 0.978 0.0 distinct group on the first two axes. Although 7 0.368 -0.758 0.0 A. viscosa does not separate out on the first 18 -0.299 0.0 0.940 two axes, it does so on the third (not shown), 3 0.470 -0.253 0.842 1582 AMERICAN JOURNAL OF BOTANY [Vol. 74

TABLE 4. Sortedfactorloadings greater than 70%for the in the subgenus(Fig. 2), and it takes a central twofactors of the BDA of A. cordifolia, A. latifolia, position in the discriminant analysis (Fig. 3). and A. gracilis. These twofactors accountfor all the It previously has been suggested that A. cor- variationamong the three taxa difolia gave rise to A. discoideaBenth. via dip-

Character Factor I Factor 2 loid populations in the Klamath region (Wolf and Denford, 1984c). Not only is this hypoth- 14 0.994 0.0 morphologicaland cytolog- 15 0.977 0.0 esis supportedby 9 0.959 0.285 ical evidence, but also by chemical evidence. 13 0.952 -0.306 Three relictual Klamath region diploid pop- 12 0.907 0.421 ulations of A. discoidea (D12, D13 and D14), 4 0.907 0.421 largelyan apomictic polyploid complex, clus- 10 0.754 0.656 ter with six northwestern populations of A. 21 0.0 0.982 mostly on the basis of com- 16 0.0 0.975 cordifolia(Fig. 1), 17 0.633 -0.774 pounds 13 and 14. Additionally, several other populations of these two species form distinct clusters in the phenogram. Furtherevolutionary diversification from A. compounds which contributehigh loadings to discoideahas takenplace in the Klamathregion the firstaxis are 16 and 2 1 (sharedby A. gracilis giving rise to two rare Klamath endemics, A. and A. latifolia), 4 and 12 (unique to A. gracilis) spathulata Greene and A. venosa H. M. Hall and 13 (shared by A. gracilis and A. cordifolia). (Wolf and Denford, 1984a, c). All four pop- and A. cordifolia separate along ulations of A. venosa cluster with A. discoidea axis II which has compounds 13 and 17 (found and most of the populations of A. spathulata in A. cordifolia),and 16 and 21 (found in A. cluster with the latter (Fig. 1). The three taxa latifolia) as those which contributehigh load- also are in close proximity in the BDA of the ings.Interestingly, A. gracilisfalls between these subgenus (Fig. 3) and of the Klamath region two taxa on axis II, a result of its additive taxa (Fig. 5). Since A. cordifoliais ancestralto flavonoid profilein these compounds. Like the all three species, it is not surprisingthat it clus- phenogramof Fig. 1, Fig. 7 shows A. gracilis ters with additionalpopulations of each in Fig. distinct from the parental species. 1 and the range of variation in A. cordifolia In the BDA of A. cordifolia, A. latifolia, and encompasses all these Klamath derivatives in A. gracilis(ordination not shown), compounds the PCA (Fig. 4). of the PCA are seen as a subset of compounds Based on morphological, cytological, geo- which separatethe three taxa (Table 4). As in graphical,and flavonoidanalyses, it previously the PCA, the three taxa are spread along the has been suggestedthat A. nevadensisA. Gray first axis with A. gracilis more displaced than has been derived from A. cordifolia(Maguire, the others. Seven compounds have loadings 1943; Wolf and Denford, 1984a, c). However, greater than 75% on the first axis (Table 4), the previous flavonoid data, based on only two with compounds 4, 12, and 13 also used in the population samples, was inconclusive (Wolf firstaxis of the PCA. The second axis separates and Denford, 1984c). Further populational A. cordifolia and A. latifolia with compounds samplinghas confirmedthe close phenetic re- 16 and 21 (found in A. latifolia) and 17 (found lationship between these two species. All ten in A. cordifolia).Again, these compounds con- populations of A. nevadensisform a very co- tribute high loadings for the PCA. hesive group and cluster with populations of A. cordifolia(Fig. 1). DIscussION-Results of the present study Previous studies have demonstratedthat A. support previous hypotheses concerningevo- gracilis is a hybrid between A. cordifoliaand lutionaryrelationships within Arnicasubgenus A. latifolia (Wolf and Denford, 1984b). How- Austromontana. Even though most of the ever, it is morphologically, ecologically, and species of the subgenus exhibit considerable reproductively distinct and has been recog- variation with respect to flavonoid profiles, nized at the specific level (Wolf and Denford, most of this variation is systematicallysignif- 1984b). Results of the present investigation icant. Previous investigations have suggested also demonstratethat this species is chemically that A. cordifolia is probably the ancestral distinct as well. In both phenograms (Fig. 1, species of the subgenus (Maguire, 1943; Wolf Fig. 7) all populationsofA. gracilisform a very and Denford, 1984a, c). It is thereforenot sur- cohesive group. Additionally, in both PCA's prising that this species forms several clusters (Fig. 2, Fig. 6) as well as the BDA (not shown), with more than one species (Fig. 1), its range A. gracilis is quite distinct from both its pre- of variation encompasses most of the species sumed parents.These results supportthe pre- October 1987] WOLF AND WHITKUS-FLAVONOIDS IN AUSTROMONTANA 1583

vious recognition of A. gracilis at the specific , which clusters with several level (Wolf and Denford, 1984b). of the species at various levels, appears to be Arnicaviscosa A. Grayis the rarestand most the ancestral species of the subgenus as pre- morphologicallydistinctive species in Arnica, viously hypothesized (Wolf and Denford, and probablythe most recentlyevolved species 1984a). The resultsalso supportthe hypothesis in subgenus Austromontana(Wolf and Den- that this species probablyhas given rise to A. ford, 1984a, c). It is known from only a few nevadensis,A. discoidea (in the Klamath re- populationson very recentvolcanic substrates, gion), and A. gracilis (via hybridization with at high elevations, largely in the Klamath re- A. latifolia).Arnica discoideain turn probably gion. In addition to its unique morphology, has given rise to the two Klamath region en- this species also has a very distinctive flavo- demics A. spathulataand A. venosa. The pres- noid profileas evidencedby the clusteranalysis ent study also refutes Straley'shypothesis that (Fig. 1), PCA (Fig. 2, third axis not shown), A. venosaand A. viscosaare closely related. In and BDA ofthe subgenus(Fig 3). Straley(1980) addition to morphologicaldifferences, the lat- erected a new subgenusin which he placed A. ter's flavonoid profile is the most distinctive viscosaand A. venosa.However, based on mor- within the subgenus. The remaining, nonsys- phological, ecological, and chemical analyses, tematically significant variation in flavonoid as well as Straley's (1980) own hybridization profilesand resultingclusters are likely the re- results,this treatmentwas rejectedby Wolf and sult of a combination of factors, including Denford(1 984a, c). Resultsofthe presentstudy founder effect, genetic drift, and a general re- also indicate the artificialityof a new subgenus duction of flavonoid profile as a result of re- for these two species. All analyses of the sub- productive isolation in both the rare taxa and genus demonstratethe close phenetic relation- apomicts. ship between A. venosa and A. discoidea, which contrasts considerablywith the placement of A. viscosa. LITERATURE CITED As previously noted, some taxa, as well as BOHM,B., H. M. BANEK, AND J. R. MAzE. 1984. Fla- some populations of Arnica subgenusAustro- vonoid variation in North American Menziesia (Er- montana exhibit very reduced or depleted fla- icaceae). Syst. Bot. 9: 324-345. vonoid profiles (Wolf and Denford, 1983, DIXON, W. J. [ED.] 1981. BMDP statisticalsoftware. Uni- 1984c). For example, the flavonoid profile of versity of California Press, Berkeley. GOWER, J. C. 1966. Some distance properties of latent the rare Klamath endemic A. cernua Howell, root and vector methods for systematics. Biometrika which consists of only five compounds, is a 53: 325-338. small subset of its presumed parental species MAGUIRE, B. 1943. A monograph of the genus Arnica. A. cordifolia(Wolf and Denford, 1984a). How- Brittonia 4: 386-5 10. ever, since there are as few as two compounds MCDOUGAL, K. M., AND C. R. PARKS. 1984. Elevational perpopulation in this species,little affinitywith variation in foliar flavonoids of Quercus rubra L. (Fa- the latteris reflectedin the numericalanalyses. gaceae). Amer. J. Bot. 71: 301-308. NEFF, N. A., AND L. F. MARCUS. 1980. A survey of In general,most of the clustersin Fig. 1 above multivariate methods for systematics. New York: pri- the large A. latifolia cluster consist of popu- vately published. lations which have very few compounds and PARKER, W. H., AND J. M. MAZE. 1984. Intraspecific to an extent, cluster on the basis of shared variation in Abies lasiocarpa from absences of compounds. However, the major- and . Amer. J. Bot. 71: 1051-1059. ity of these clusters do in fact ROHLF, F. J. 1970. Adaptive hierarchial clustering represent the schemes. Syst. Zool. 19: 58-82. same taxa or closely relatedtaxa. For example, , J. KISHPAUGH, AND D. KIRK. 1972. NT-SYS. most of the populations of A. cordifolia, A. Numerical system of multivariate statis- cernua, and A. latifolia cluster together. Like- tical procedures. State University of New York, Stony wise, the threepopulations ofA. spatulatawhich Brook. cluster at the top of Fig. 1 have very reduced SNEATH, P. H. A., AND R. R. SOKAL. 1973. Numerical flavonoidprofiles (i.e., threecompounds each), taxonomy. W. H. Freeman, San Francisco. SOKAL, R. R., AND C. D. MICHENER. 1958. A statistical and therefore they do not cluster below with method for evaluating systematic relationships. Univ. the remaining populations from the Klamath Kansas Sci. Bull. 38: 1409-1438. region. STRAHLER, A. H. 1978. Binary discriminant analysis: a Even though the species of Arnica subgenus new method for investigating species-environment re- Austromontanaexhibit considerable popula- lationships. Ecology 59: 108-116. STRALEY, G. B. 1980. Systematics of Arnica, subgenus tional variation with respect to flavonoid pro- Austromontana and a new subgenus Calarnica (As- files, most of this variation is systematically teraceae: Senecioneae). Ph.D. dissertation, University significantand confirms previous hypotheses of British Columbia, Vancouver. concerningrelationships within the subgenus. WOLF, S. J. 1980. Cytogeographical studies in the genus 1584 AMERICAN JOURNAL OF BOTANY [Vol. 74

Arnica (Compositae:Senecioneae). I. Amer. J. Bot. ,AND . 1984b. Arnica gracilis (Composi- 67: 300-308. tae), a naturalhybrid between A. latifoliaand A. cor- , AND K. E. DENFORD. 1983. Flavonoid variation difolia. Syst. Bot. 9: 12-16. in Arnicacordifolia: an apomicticpolyploid complex. - AND . 1984c. Flavonoid diversity and Biochem. Syst. Ecol. 11: 111-1 14. endemism in Arnica subgenusAustromontana. Bio- - AND . 1984a. Taxonomyof Arnica(Com- chem. Syst. Ecol. 12: 183-188. positae)subgenus Austromontana. Rhodora 86: 239- 309.