Classification of the Leguminosae-Papilionoideae : A Numerical Re-assessment

The subdivision of the Leguminosae-Papilionoideae into taxa of lower rank was subject for major discrepancies between traditional classifications while more recent phylogenetic studies provided no decisive answer to this problem. As a contribution towards resolving this situation, 81 morphological characters were recorded comparatively for 226 species and infra-specific taxa belonging to 75 genera representing 21 of the 32 tribes currently recognized in this subfamily. The data matrix was subjected to cluster analysis using the Sørensen distance measure and Ward’s clustering method of the PC-ord version-5 package of programs for Windows. This combination was selected from among the 56 combinations available in this package because it produced the taxonomically most feasible arrangement of the genera and species. The 75 genera are divided into two main groups A and B, whose recognition requires little more than the re-alignment of a few genera to resemble tribes 1-18 (Sophoreae to Hedysareae) and tribes 19-32 (Loteae to Genisteae), respectively, in the currently accepted classification. Only six of the 21 tribes represented by two or more genera seem sufficiently robust as the genera representing each of them hold together in only one of the two major groups A and B. Of the 29 genera represented by more than one species each 17, 7 and 5 are taxonomically coherent, nearly coherent and incoherent, respectively. The currently accepted circumscription and inter-relationships among the disrupted tribes and genera are in need of much detailed investigation.


Introduction
The Papilionoideae DC. (nom.altern.Fabaceae Lindl.-Faboideae) is by far the largest of the three subfamilies of Leguminosae Adanson.It comprises 476 genera and 13855 species (APG 2012), distributed mainly in the temperate and subtropical parts of the world.Many of the species are of immense economic value globally as legume crops, including the innumerable cultivars of beans, faba beans, vetches (Vicia spp., Phaseolus spp., Vigna spp.), soybean (Glycine max), peas (Pisum sativum), chickpea (Cicer arietinum), lucerne or alfalfa (Medicago sativa), lentils (Lens esculentus), clovers (Trifolium spp.) and lupins (Lupinus spp.).Despite the huge range of variation in morphological features of their vegetative parts, members of the Papilionoideae are easily distinguishable from the other two subfamilies of the Leguminosae (Mimosoideae and Caesalpinioideae) by numerous morphological attributes including the papilionoid structure of the corolla, asymmetrical seeds, ovate-elliptical cotyledons, campylotropous ovules and the embryo axis is curved or a short spiral (Isely 1955 and1981;Watson and Dallwitz 1992;onwards, Kirkbride et al. 2003;APG 2012;El-Gazzar et al. 2012).This clear distinction of the Papilionoideae from the rest of the Leguminosae gained substantial support from phylogenetic studies (Käss andWink 1996-1997;Doyle et al. 1997;Ka-jita et al. 2001;Doyle and Luckow 2003;Wojciechowski et al. 2004;Champagne et al. 2007).
The subdivision of the Papilionoideae into taxa of lower rank was for many decades highly controversial (El-Gazzar and El-Fiki 1977, El-Gazzar 1979and 1981, Käss and Wink 1995, 1996and 1997, Doyle et al. 1997, Doyle and Luckow 2003, Wojciechowski et al. 2004, Champagne et al. 2007).This is clearly evident from the huge differences in the numbers and circumscription of tribes and sub-tribes recognized in the four major classifications of the Papilionoideae by De Candolle (1825), Bentham and Hooker (1865), Taubert (1894), Polhill and Raven (1981), who divided this subfamily into 6, 11, 10 and 32 tribes, respectively.Differences between the four systems in the number of subtribes are even greater.The apparently excessive fragmentation of the Papilionoideae in the system by Polhill and Raven (1981) into 32 tribes with 21 sub-tribes and 10 "groups" seems to have been universally accepted without being put to practical test.
The only common feature of all classificatory systems of the Papilionoideae to date is the recognition of tribes and sub-tribes on the basis of a limited range of floral characters with greater emphasis on petal morphology and stamen arrangement.Such few characters were used often singly to distinguish between chunky assemblages of genera.This seems to explain the unsettled disposition of (McCune, 1997).It consists of seven distance measures and eight clustering (sorting or linkage) methods thus offering 56 possible hierarchical arrangements of the species.Pre-requisites of the program necessitated the abbreviation of names of taxa into only eight digits.Full valid names of taxa with their author citations and abbreviations are presented in Appendix 1.
Tab. 1. List of the 81 characters and their character-states recorded comparatively for 226 species of the Leguminosae-Papilionoideae and used to build a system of classification of the subfamily most genera in the tribes and subtribes recognized in different classifications.The different arrangements of genera within the Papilionoideae are so great that a study based on a wider range of the plants' characters seemed urgently needed.
The present study was embarked upon in an attempt to answer the question: to what extent would the currently accepted classification of the Papilionoideae by Polhill and Raven (1981) withstand the test of numerical analyses of a comprehensive set of characters recorded comparatively from a cosmopolitan sample of genera and species using a number of computer programs with varying degrees of mathematical sophistication and taxonomic efficiency?

Material and methods
The sample of taxa The present study was based on a cosmopolitan sample of 226 species and infra-specific taxa representing 75 genera of the Papilionoideae.This sample is necessarily small relative to the enormous size of the subfamily.However, it is fairly representative of the subfamily because: (a) it covers 21 of the 32 tribes and most of the sub-tribes in the currently accepted classification, (b) the genera are proportionately represented so that larger genera are represented by relatively larger numbers of species except in some relatively large genera (e.g.Aspalathus, Eriosema, Mucuna and Ormosia), and (c) it is geographically balanced in that all of the chief centers of distribution are proportionately catered for.
The number of fresh and herbarium specimens of individual species ranged between one and 35 with most of the species being represented by five-ten specimens each.The specimens are collected from some of the local botanic gardens (Orman Botanic Gardens in Giza, Alexandria University and Aswan) and the four major local herbaria (ALEX, CAID, CAI, CAIM; acronyms are according to Holmgren et al. (1990).Identification of the specimens was double checked using appropriate floras (Andrews 1952;Heywood and Ball 1968;Davis 1969;Zohary 1987;Boulos 1999 and2009) and the online floras of China (Flora of China 2013), the Iberian Peninsula (Flora Iberica 2013) and Pakistan (Ali 2013).Identities were confirmed by matching with images of type and non-type material on various websites (e.g.www.aluka.org;www.tropicos.org, http://coldb.mnhn.fr;http://sweetgum.nybg.org).Nomenclature of all taxa was updated according to the two major online sources (www.tropicos.org;www.theplantlist.org).

Data analysis
A list of 81 characters recorded comparatively for 226 species belonging to 75 genera is given in Tab. 1.The data matrix was subjected to cluster analysis using the package of classificatory programs PC-ord version 5 for windows Each of the resulting dendrograms has a certain % of chaining which is an expression of its clustering intensity.Low values of % chaining indicate a high clustering intensity of the dendrogram, and vice versa: the higher the clustering intensity, the more discrete (and taxonomically reliable) the groups in the dendrogram are.Comparison between the resulting dendrograms was carried out manually to select those which were closest to each other and to the currently accepted classification by Polhill and Raven (1981).

Results
The 56 dendrograms were shortlisted into four with the least percentages of chaining.They were based on the following combinations of distance measures and clustering methods: Distance measure clustering method % chaining Sørensen Ward`s 0.70 Relative Sørensen Ward`s 0.63 Euclidean Ward`s 1.37 Relative Euclidean Ward`s 1.00 A 'stopping level' was selected to produce basic groups in each of the four dendrograms and the species composition of the eight basic groups in the first dendrogram singled it out and favored its selection for further discussion.The lower reaches of each of groups 1-8 in Fig. 1 are shown in Fig. 2.

Comparison with the accepted classification
Although the traditional classifications of the Papilionoideae have the advantage of being comprehensive and cover the entire generic content of the subfamily known at the time of their establishment, the majority of groupings recognized in them are based on single characters of the flowers and pods.On the other hand, the present study is based on a much smaller sample of genera and species but with a much wider range of the variation in vegetative and floral morphology.In view of this basic difference, it is to be expected that the groupings in the present study would diverge to some extent from those in any of the traditional classifications.This is clearly evident from Tab. 2 where the agreements and disagreements between groups A and B and their eight sub-groups in Figs. 1 and 2 and Tab. 2 and the classification proposed by Polhill and Raven (1981) may be summed up in the following: 1.The sequence of tribes 1-32 in Polhill and Raven's (1981) classification seems taxonomically meaningful in terms of the two major groups A and B in Tab. 2. Thus, the great majority of genera belonging to tribes 2-18 (Sophoreae to Hedysareae) fall in group A, whereas most of the genera representing tribes 19-32 (Loteae to Genisteae; shaded in Tab. 2) appear in group B. It seems that the recognition of groups A and B requires only the re-allocation of some genera to coincide with the sequential arrangement of traditional tribes and that a major division within the Papilionoideae is worthy of further investigation.Tab. 2. Comparison between the distribution of 75 genera of the Papilionoideae among groups 1-8 in the arrangements presented in Figs. 1 and 2, and the tribes recognized by Polhill and Raven (1981).Numbers of species are given in parentheses.Genera belonging to Group B are shaded.Genera treated by Polhill and Raven (1981)  Out of these 29 genera, 17 seem taxonomically robust, followed by seven almost equally coherent genera, whereas the two species representing each of Onobrychis and Hedysarum were disbanded in groups A.C.1 and A.C.2.Representatives of Tephrosia, Macrotyloma and Argyrolobium together with some species of Vicia, Lathyrus and Trifolium crossed the boundary between the two major groups A and B in Fig. 1 and Tab. 2. In the category of seven nearly coherent genera, the separation of some splinter species from the main aggregation of their relatives in the same genus emphasizes the need to re-evaluate the intraspecific relationships of such genera by more detailed studies based on larger numbers of their species.This is best exemplified by the isolation of Astragalus kahiricus in group A.D.4 and A. fresenii and A. penduliflora in group A.C.1 instead of group A.C.2 where all other 26 Astragalus species are placed together.Generally, this categorization of genera seems to indicate a relatively sound generic concept in the Papilionoideae, especially in the case of such megagenera as Astragalus which occupies a prominent position among the largest angiosperm genera ( Jer-Ming Hu et al., 2013).
It is worth noting that Tab. 2 includes six genera treated by Polhill and Raven (1981) as sections or sub-sections of larger genera but recognized as distinct genera comprising species with currently accepted names in "www.theplant list.org"and "www.tropicos.org".Four of these segregate genera (Spiesia as ~Oxytropis; Dorycnium as ~Lotus; Orobus as ~Vicia; Teline as ~Genista), appear inseparable from their larger relatives, thus supporting the treatment of Polhill and Raven (1981) and providing fresh evidence favoring re-uniting them with their close allies.
In view of the foregoing remarks it seems advisable to subject larger samples of genera and species of the Leguminosae-Papilionoideae to much detailed studies based on the comparative recording of the widest possible range of variation in the plants' characters and subjecting the output to a number of numerical analyses with variable combinations of similarity assessment algorithms and clustering procedures, in order to select the classification which best imposes a clear pattern on the distribution of the recorded characters among the plants.Special emphasis should be laid on such genera as Tephrosia, Argyrolobium, Macrotyloma, Hedysarum and Onobrychis as well as such relatively large tribes as the Genisteae and Loteae whose representative species failed to emerge together in only one of groups 1-8 or were dispersed between the two major Groups A and B.
Based on data of sequencing nuclear ribosomal DNA, Allan and Porter (2000) proposed submerging the Coronilleae into the Loteae and a indicated a bio-geographic distinction within the combined tribe between the New Word Lotus s.l clade and the Old World Lotus s.l.clade.Although the sample of species representing these two tribes in the present study belongs predominantly to the Old World, the dispersal of their species among the three basal groups of group A.C. and three of the basal groups of Group B (B.E.5, B.E.6 and B.F.7) in Tab. 2 does not corroborate the results of Allan and Porter (2000).

Fig. 1 .
Fig. 1.Dendrogram based on the 81 characters listed in Tab. 1, recorded comparatively for 226 species of the Leguminosae-Papilionoideae, and analyzed by the Sørensen distance measure and Ward`s clustering method.The % chaining is 0.7

2 .
Only six of the 32 tribes represented by two or more genera in the present study seem coherent as their genera hold together in either group A or group B. They are the Sophoreae (3 genera), Robinieae (2 genera), Desmodieae (2 genera) and Aeschynomeneae (2 genera) in group A, and the Crotalarieae (2 genera) and Trifolieae (5 genera) in group B. The nearly coherent tribes include the Phaseoleae (10 genera), Tephrosieae (5 genera), Vicieae (5 genera) and Galegeae (8 genera) in group A, and the Coronilleae

Fig. 2 .
Fig. 2. The genera and species of groups 1-8 in Fig. 1.Names of taxa are abbreviated according to the list in the Appendix as sections of other genera are asterisked A.C.3 and B.E.6, and Onobrychis and Hedysarum with 1 species each in each of groups A.C.1 and A.C.2.