chapter 8 developing classifications · 2020. 4. 1. · tion in the taxa of a classification and...

210 Plant Systematics

Systematics aims at developing classifica-tions based on different criteria and, often adistinct methodology is employed for theanalysis of data. Data handling to establishrelationships between the organisms oftenmakes use of one of the two methods: phe-netic methods and phylogenetic methods,often providing different types of classifica-tion. Distinction is sometimes also made be-tween phylogenetic and evolutionary clas-sification schemes. Phylogenetic methodsaim at developing a classification based onan analysis of phylogenetic data, and devel-oping a diagram termed cladogram or phy-logenetic tree, or more recently, simplytree, which depicts the genealogical descentof taxa. Biologists practicing this methodol-ogy are known as cladists, and the field ofstudy as cladistics. The term, however, isslowly being replaced by phylogenetic sys-tematics. The phylogenetic conceptspresent a huge diversity of variation, unfor-tunately often contradictory, leading to dif-ferent interpretations of similar results. Abrief understanding of these is, therefore,necessary before attempting to explore thiscomplex field. Before the development ofmodern methods of cladistics, the numeri-cal methods were largely used for drawingphylogenetic inferences from the data analy-sis. The modern Phylogenetic methods, how-ever, integrate the concepts and practices

of numerical taxonomy with cladistic meth-ods. It is, however, essential to understandthe concepts of each, and the final integra-tion in phylogeny reconstruction.

PHENETIC METHODSNumerical taxonomy received a great im-petus with the development and advance-ment of computers. This field of study is alsoknown as mathematical taxonomy (Jardineand Sibson, 1971), taxometrics (Mayr, 1966),taximetrics (Rogers, 1963), multivariatemorphometrics (Blackith and Reyment, 1971)and phenetics. The modern methods of nu-merical taxonomy had their beginning fromthe contributions of Sneath (1957),Michener and Sokal (1957), and Sokal andMichener (1958) which culminated in thepublication of Principles of Numerical Tax-onomy (Sokal and Sneath, 1963), with an ex-panded and updated version Numerical Tax-onomy (Sneath and Sokal, 1973). The latterauthors define Numerical taxonomy asgrouping by numerical methods of taxo-nomic units into taxa on the basis of theircharacter states. Before the developmentof modern methods of cladistics, thenumerical methods were also used fordrawing phylogenetic inferences from thedata analysis.

The last few decades have witnessed aforceful debate on the suitability of the

Chapter 8

Developing Classifications

Kamal

Rectangle

Developing Classifications 211

empirical approach or operational approachin systematic studies. Empirical taxonomyforms the classification on the basis of taxo-nomic judgment based on observation of dataand not assumptions. Operational taxonomy,on the other hand, is based on operationalmethods, experimentation to evaluate theobserved data, before a final classification.Numerical taxonomy finds a balance be-tween the two as it is both empirical andoperational (Figure 8.1).

It must be remembered that numericaltaxonomy does not produce new data or a newsystem of classification, but is rather a newmethod of organizing data that could help inbetter understanding of relationships. Spe-cial classifications are based either on oneor a few characters or on one set of data.Numerical taxonomy seeks to base classifi-cations on a greater number of charactersfrom many sets of data in an effort to pro-duce an entirely phenetic classification ofmaximum predictivity.

Principles of TaxometricsThe philosophy of modern methods of nu-merical taxonomy is based on ideas thatwere first proposed by the French naturalistMichel Adanson (1763). He rejected the ideaof giving more importance to certain char-acters, and believed that natural taxa arebased on the concept of similarity, which ismeasured by taking all the characters intoconsideration. The principles of modern nu-merical taxonomy developed by Sneath andSokal (1973) are based on the modern inter-pretation of the Adansonian principles andas such are termed neo-Adansonian prin-ciples. It would, however, be wrong to visu-alize Adanson as the founder of numericaltaxonomy, because he worked in a differentacademic environment from that of today,when tools of investigation were much dif-ferent. These principles of numerical tax-onomy are enumerated below.

1. The greater the content of informa-tion in the taxa of a classification andthe more characters it is based upon,

the better a given classification willbe.

2. A priori, every character is of equalweight in creating natural taxa.

3. Overall similarity between any two en-tities is a function of their individualsimilarities in each of the many char-acters in which they are being com-pared.

4. Distinct taxa can be recognized be-cause correlations of characters dif-fer in the groups of organisms understudy.

5. Phylogenetic inferences can be madefrom the taxonomic structures of agroup and also from character corre-lations, given certain assumptionsabout evolutionary pathways andmechanisms.

6. Taxonomy is viewed and practiced asan empirical science.

7. Classifications are based on pheneticsimilarity.

The methodology of numerical taxonomyinvolves the selection of operational units(populations, species, genera, etc., fromwhich the information is collected) and char-acters. The information from these isrecorded, and similarity (and/or distance)between units is determined using various

Figure 8.1 Relationship between empirical, op-erational and numerical taxonomy(after Sneath and Sokal, 1973).


statistical formulae. The ultimate analysisinvolves comparison of similarity data andconstructing diagrams or models, which pro-vide a summary of the data analysis. Thesediagrams or models are used for final syn-thesis and better understanding of the rela-tionships. The major advantages of numeri-cal taxonomy over conventional taxonomyinclude:

1. Numerical taxonomy has the power tointegrate data from a variety ofsources such as morphology, physiol-ogy, phytochemistry, embryology,anatomy, palynology, chromosomes,ultrastructure and micromorphology.This is very difficult to do by conven-tional taxonomy.

2. Considerable automation of the dataprocessing promotes efficiency andthe work can be handled by even lessskilled workers.

3. Data coded in numerical form can beintegrated with existing data-process-ing systems in various institutionsand used for the creation of descrip-tions, keys, catalogues, maps andother documents.

4. The methods, being quantitative, pro-vide greater discrimination along thespectrum of taxonomic differences,and can provide better classificationsand keys.

5. The creation of explicit data tables fornumerical taxonomy necessitates theuse of more and better described char-acters, which will necessarily improveconventional taxonomy as well.

6. The application of numerical taxonomyhas posed some fresh questions con-cerning classification and initiatedefforts for re-examination of classifi-cation systems.

7. A number of biological and evolution-ary concepts have been reinterpreted,thus introducing renewed interest inbiological research.

Numerical taxonomy aims at determin-ing phenetic relationships between organ-isms or taxa. Cain and Harrison (1960) de-

fined phenetic relationship as an arrange-ment by overall similarity, based on allavailable characters without any weight-ing. Sneath and Sokal (1973) define phe-netic relationship as similarity (resem-blance) based on a set of phenotypic char-acteristics and not phylogeny of organ-isms under study. It is distinct from a cla-distic relationship, which is an expressionof the recency of common ancestry and isrepresented by a branching network of an-cestor-descendant relationships. Whereasthe phenetic relationship is represented bya phenogram, the cladistic relationship isdepicted through a cladogram.

CLADISTIC METHODSAlthough phylogenetic diagrams (now appro-priately known as phylograms) have beenused by Bessey (1915), Hutchinson (1959,1973), and contemporary authors of classifi-cation systems to depict the relationshipsbetween taxa, the cladograms are distinctin the sense that they are developed usinga distinct methodology. This method was firstproposed by W. Hennig (1950, 1957), a Ger-man zoologist who founded the subject of phy-logenetic systematics. The term cladisticsfor this methodology was coined by Mayr(1969). An American Botanist, W. H. Wagner,working independently, developed a methodof constructing phylogenetic trees, called thegroundplan-divergence method, in 1948.Over the years, cladistics has developed intoa forceful methodology of developing phylo-genetic classifications.

Cladistics is a methodology that attemptsto analyse phylogenetic data objectively, ina manner parallel to taxometrics, whichanalyses phenetic data. Cladistic methodsare largely based on the principle of parsi-mony according to which, the most likelyevolutionary route is the shortest hypothet-ical pathway of changes that explains thepattern under observation. Taxa in a trulyphylogenetic system should be monophylet-ic. It has been found that symplesiomorphy(possession of primitive or plesiomorphic


character-state in common by two or moretaxa) does not necessarily indicate monophy-ly. Synapomorphy (possession of derived orapomorphic character-state in common bytwo or more taxa), on the other hand, is amore reliable indicative of monophyly. It isthus common to use homologous shared andderived character-states for cladistic stud-ies. Before analysing the methodology of han-dling data for phylogenetic analysis, it is im-portant to understand the major terms andconcepts used in Phylogenetic Systematics.

Phylogenetic TermsMany important terms have been repeatedlyused in discussions on the phylogeny of an-giosperms, with diverse interpretation,which has often resulted in different sets ofconclusions. A prominent case in point isMelville (1983), who regards the angiospermsas a monophyletic group. His justification—several ancestral forms of the single fossilgroup Glossopteridae gave rise to an-giosperms—renders his view as polyphyleticin the eyes of the greater majority of authorswho believe in the strict application of theconcept of monophyly. The involvement ofmore than one ancestor makes angiospermsa polyphyletic group, a view that has beenfirmly rejected. A uniform thorough evalua-tion of these concepts is necessary for properunderstanding of angiosperm phylogeny.

Plesiomorphic andApomorphic CharactersA central point to the determination of thephylogenetic position of a particular groupis the number of primitive (plesiomorphic)or advanced (apomorphic) characters (al-though the term character is often usedbroadly in literature, more appropriatelyprimitive or advanced and similarlyplesiomorphic and apomorphic refer to dif-ferent character-states of a character, andnot different characters) that the group con-tains. In the past, most conclusions on primi-tiveness were based on circular reasoning:

‘These families are primitive because theypossess primitive characters (or character-states) and primitive characters (or charac-ter-states) are those which are possessed bythese primitive families’. Over the recentyears, a better understanding of these con-cepts has become possible. It is generallyaccepted that evolution has proceeded at dif-ferent rates in different groups of plants sothat among the present-day organisms,some are more advanced than others. Thefirst step in the determination of relative ad-vancement of characters, is to ascertainwhich characters are plesiomorphic andwhich are apomorphic. Stebbins (1950) ar-gued that it is wrong to consider the charac-ters as separate entities, since it is throughthe summation of characters peculiar to anindividual, that natural selection operates.Sporne (1974) while agreeing with this, be-lieved that it is scarcely possible initially toavoid thinking in terms of separate charac-ters, which can be treated better statistically.Given insufficient fossil records of the ear-liest angiosperms, comparative morphologyhas been largely used to decide the relativeadvancement of characters. Many doctrineshave been proposed but unfortunately mostrely on circular reasoning. Some of theimportant doctrines are described below:

The Doctrine of conservative regionsholds that certain regions of plants havebeen less susceptible to environmental in-fluence than others and, therefore, exhibitprimitive features. Unfortunately, however,over the years, every part of the plant hasbeen claimed as conservative region. Also,the assumption that a flower is more con-servative than the vegetative parts is derivedfrom classifications which are based on thisassumption.

The doctrine of recapitulation holds thatearly phases in development are supposedto exhibit primitive features, i.e. ‘ontogenyrepeats phylogeny’. Gunderson (1939) usedthis theory to establish the followingevolutionary trends: polypetaly to gamopetaly(since the petal primordia are initiallyseparate, the tubular portion of the corollaarises later); polysepaly to gamosepaly;


actinomorphy to zygomorphy and apocarpy tosyncarpy. The concept originally applied toanimals does not always hold well in plantswhere ontogeny does not end with embryog-eny but continues throughout the adult life.Neoteny (persistence of juvenile features inmature organism) is an example wherein apersistent embryonic form represents an ad-vanced condition.

The doctrine of teratology was advocatedby Sahni (1925), who argued that when anormal equilibrium is upset, an adjustmentis often effected by falling back upon thesurer basis of past experience. Thus, tera-tology (abnormality) is seen as reminiscentof some remote ancestor. According toHeslop-Harrison (1952), some teratologicalphenomena are just likely to be progressiveor retrogressive, and each case must bejudged on its own merit.

The doctrine of sequences advocates thatif organisms are arranged in a series insuch a way as to show the gradation of a par-ticular organ or structure, then the two endsof the series represent apomorphy andplesiomorphy. The most crucial decision,however, is from which end should the se-ries be read.

The doctrine of association advocatesthat if one structure has evolved from an-other, then the primitive condition of thederived one will be similar to the generalcondition of the ancestral structure. Thus,if vessels have evolved from tracheids, thenthe vessels similar to tracheids (vessels withlonger elements, smaller diameter, greaterangularity, thinner walls and oblique endwalls) represent a more primitive conditionthan vessels with broader, shorter, more cir-cular elements with horizontal end walls.

The doctrine of common ground plan ad-vocates that characters common to all mem-bers of a group must have been possessedby the original ancestor and must, therefore,be primitive. The doctrine, however, cannotbe applied to angiosperms in which there isan exception for almost every character.

The doctrine of character correlation wasacknowledged during the second decade ofthe previous century when it was realized

that certain morphological characters arestatistically correlated and the fact can beused in the study of evolution. Sinnot andBailey (1914) demonstrated a positive cor-relation between trilacunar node andstipules. Frost (1930) believed that correla-tion between characters arises becauserates of their evolution have been correlated.Sporne (1974) has, however, argued that cor-relation can be shown to occur even thoughthe rates of evolution of characters are notthe same. Within any taxonomic group,primitive characters may be expected toshow positive correlation merely becausetheir distribution is not random. By defini-tion, primitive members of that group haveretained a relatively high proportion of an-cestral (plesiomorphic) characters, while ad-vanced members have dispensed with a rela-tively high proportion of these same char-acters—either by loss or replacement withdifferent (apomorphic) characters. It follows,therefore, that the distribution ofplesiomorphic characters is displaced to-wards primitive members, which have ahigher proportion of plesiomorphic charac-ters, than the average for the group as awhole. Departure from the random can bestatistically calculated in order to establishcorrelation among characters. Based onthese calculations, Sporne (1974) prepareda list of 24 characters in Dicotyledons and14 in Monocotyledons, which exhibit posi-tive correlation. These characters, becauseof their distribution, have been categorizedas magnoloid and amarylloid, respectively.Based on the distribution of these charac-ters, Sporne calculated an advancementindex for each family and projected theplacement of different families of an-giosperms in the form of a circular diagram,with the most primitive families near thecentre, and the most advanced along the pe-riphery. That the earliest members of an-giosperms are extinct is clear from the factthat none of the present-day families hasthe advancement index of zero. All livingfamilies have advanced in some respects.

The concept of apomorphic andplesiomorphic characters in understanding


the phylogeny of angiosperms has been con-siderably advanced with the recent develop-ment of cladistic methods. These employ adistinct methodology, somewhat similar totaxometric methods in certain steps in-volved, leading to the construction of cla-dograms depicting evolutionary relation-ships within a group. Certain groups of an-giosperms are reported to have a combina-tion of both plesiomorphic and apomorphiccharacters, a situation known asheterobathmy. Tetracentron has primitivevesselless wood but the pollen grains are ad-vanced, being tricolpate.

Homology and AnalogyDifferent organisms resemble one anotherin certain characters. Taxonomic groups ortaxa are constructed based on overall resem-blances. The resemblances due to homologyare real, whereas those due to analogy aregenerally superficial. A real understandingof these terms is, thus, necessary in orderto keep organisms with superficial resem-blance in separate groups. The two termsas such play a very important role in under-standing evolutionary biology.

These terms were first used and definedby Owen (1848). He defined Homology asthe occurrence of the same organ in dif-ferent animals under every variety offorms and functions. He defined Analogyas the occurrence of a part or an organin one animal which has the same func-tion as another part or organ in a differ-ent animal. If applied to plants, the rhizomeof ginger, the corm of colocasia, tuber ofpotato, and runner of lawn grass are allhomologous, as they all represent a stem.The tuber of potato and the tuber of sweetpotato, on the other hand, are analogous asthe latter represents a root.

Darwin (1959) was the first to apply theseterms to both animals and plants. He definedhomology as that relationship betweenparts which results from their develop-ment from corresponding embryonic parts.The parts of a flower in different plants arethus homologous and these, in turn, are

homologous with leaves because their de-velopment is identical.

During the latter half of the present cen-tury, phylogenetic interpretation has beenapplied to these terms. Simpson (1961) de-fined homology as the resemblance due toinheritance from a common ancestry. Anal-ogy, similarly, represents functional simi-larity and not due to inheritance from acommon ancestry. Mayr (1969) similarly de-fined homology as the occurrence of simi-lar features in two or more organisms,which can be traced to the same featurein the common ancestor of these organ-isms. It is, as such, imperative that homol-ogy between two organisms can result onlyfrom their having evolved from a commonancestor, and the ancestor must also con-tain the same feature or features for whichthe two organisms are homologous.

Wiley (1981) has provided a detailed in-terpretation of these terms. Homology mayeither be between two characters, two char-acter states, or between two organisms for a

Figure 8.2 Homology between characters (orcharacter states). In the first ex-ample, character A is plesiomorphicand B is apomorphic. In the secondexample, B is apomorphic in rela-tion to A but plesiomorphic in rela-tion to C as all three belong to anevolutionary transformation series.

particular character or character state. Twocharacters (or character-states) arehomologous if one is directly derived fromthe other. Such a series of characters iscalled an evolutionary transformationseries (also called morphoclines orphenoclines). The original, pre-existingcharacter (or character- state) is termedplesiomorphic and the derived one asapomorphic or evolutionary novelty.


4. When the same relatively simple char-acter is found in a large number ofspecies, it is probably homologous inall the species. Sets of characters maysimilarly be homologous.

5. If two organisms share the charactersof sufficient complexity and judgedhomologous, other characters sharedby the organisms are also likely to behomologous.

Parallelism and convergenceUnlike homology, if the character shared bytwo organisms is not traced to a common an-cestor, the similarity may be the result ofhomoplasy (sometimes considered synonymof analogy). It can result in three differentways. One, the organisms have a commonancestor but the character-state was notpresent in their common ancestor (parallel-ism). It could also result from two differentcharacters in different ancestors evolvinginto identical character-states (conver-gence). Similarity could also arise from lossof a particular character (reversal), thusreverting to ancestral condition (loss of pe-rianth in some families). All the three situ-ations represent false synapomorphy

Three or more character-states may behomologous if they belong to the same evo-lutionary transformation series (ovary su-perior—>half-inferior—> inferior). The termsplesiomorphic and apomorphic are, however,relative. In an evolutionary transformationseries representing characters A, B and C(Figure 8.2), B is apomorphic in relation to Abut it is plesiomorphic in relation to C.

Two or more organisms may be homolo-gous for a particular character (or charac-ter-state) if their immediate common ances-tor also had this character. Such a charac-ter is called shared homologue. If the char-acter-state is present in the immediate com-mon ancestor, but not in the earlier ances-tor (Figure 8.3), i.e. the character-state is aderived one, the situation is known assynapomorphy. If the character-state ispresent in the immediate common ances-tor, as well as in the earlier ancestor, i.e. itis an original character-state, the situationis known as symplesiomorphy (note sym-).

The homology between different organ-isms is termed special homology, as repre-sented by different types of leaves in differ-ent species of plants. Different leaves in thesame plant such as foliage leaves, bracts,floral leaves would also be homologous, rep-resenting serial homology. The followingcriteria may be helpful in identifying homol-ogy in practice:

1. Morphological similarity with respectto topographic position, geometric po-sition, or position in relation to otherparts. A branch, for example, occursin the axil of a leaf, although it maybe modified in different ways.

2. Similar ontogeny.3. Continuation through intermediates,

as for example, the evolution of mam-malian year from gills of fishes, evo-lution of achene fruit from follicle inRanunculaceae. Similarly, vesselshaving evolved from tracheids, theprimitive forms of vessels are morelike tracheids, with elongated nar-rower elements with oblique endwalls.

Figure 8.3 Homology between two organismsB and C. In diagram I, similarity isdue to symplesiomorphy as thecharacter was unchanged in theprevious ancestor. In II, it is dueto synapomorphy as the previousancestor had a plesiomorphic char-acter and the two now share a de-rived character.

A B C A B C

I I I

A B C


Figure 8.4 Examples of convergence (I) and par-allelism (II) between organisms Aand B. In convergence, similarity isbetween organisms derived fromdifferent lineages. In parallelism,the ancestor is common but both Aand B have evolved an apomorphiccharacter independently. In bothcases, similarity represents falsesynapomorphy. Dissimilarity be-tween B and C in both diagrams isdue to divergence.

A B C A B C

III

because the similar character-state is de-rived and not traced to a common ancestor.

Simpson (1961) defined parallelism asthe independent occurrence of similarchanges in groups with a common ances-try, and because they had a common an-cestry. The two species Ranunculustripartitus and R. hederacea have a similaraquatic habit and dissected leaves and haveacquired these characters by parallel evolu-tion. The development of vessels in Gnetalesand dicotyledons also represents a case ofparallelism.

Convergence implies increasing similar-ity between two distinct phyletic lines, ei-ther with regard to individual organ orto the whole organism. The similar fea-tures in convergence arise separately in twoor more genetically diverse and not closelyrelated taxa or lineages. The similaritieshave arisen in spite of lack of affinity andhave probably been derived from different sys-tems of genes. Examples may be found inthe occurrence of pollinia in Asclepiadaceaeand Orchidaceae, and the ‘switch habit’ (cir-cular sheath at nodes) in Equisetum, Ephe-dra and Polygonum. The concepts of paral-lelism and convergence are illustrated inFigure 8.4.

Convergence is generally brought aboutby similar climates and habitats, similarmethods of pollination or dispersal. Once theconvergence has been identified betweentwo taxa, which have been grouped together,they are separated to make the groups natu-ral and monophyletic. The following criteriamay help in the identification of conver-gence:

1. Convergence commonly results fromadaptation to similar habitats. Wa-ter plants thus usually lack root hairsand root cap but contain air lacunae.Annuals are predominant in deserts,which also have a good number of suc-culent plants. The gross similaritybetween certain succulent species ofEuphorbiaceae and Cactaceae is avery striking example of convergence.

2. Convergence may also result fromsimilar modes of pollination such as

wind pollination in such unrelatedfamilies as Poaceae, Salicaceae andUrticaceae, pollinia in Asclepiadaceaeand Orchidaceae.

3. Convergence may also be due to simi-lar modes of dispersal, as seen inhairy seeds of Asteraceae,Asclepiadaceae and some Malvaceae.

4. Convergence commonly occurs be-tween relatively advanced members ofrespective groups. Arenaria andMinuartia form natural groups of spe-cies which were earlier placed withinthe same genus Arenaria. The two spe-cies Arenaria leptocladus and Minuartiahybrida show more similarity thanbetween any two species of these twogenera. If the similarity is patristic(result of common ancestry), then thetwo species would represent the mostprimitive members of respectivegroups (Figure 8.5-I) and it would havebeen advisable to place all of the spe-cies in the same genus Arenaria. The


studies have shown, however, thatthese two species are the most spe-cialized in each group (Figure 8.5-II)and thus show convergence. Separa-tion of the two genera is justified,because placing all the species withinthe same genus Arenaria would renderthe group polyphyletic, a situation thatevolutionary biologists avoid.

It is pertinent to mention that althoughthe concepts parallelism and convergenceseem to be distinct and theoretically sound,and often easy to apply when discussinghomoplasious (non-homologous) similarityin the case of closely related organisms (par-allelism), or distantly related organisms (con-vergence), the distinction is not always clear.In Figure 8.5-I, for example if we did not knowthe evolutionary history of the group beforelevel m, there was no way of telling whetherall the eight species had a common ances-tor or not. For practical reasons, it is alwayssafer to refer homoplasious situations to-gether. Some recent authors like Judd etal., (2002) treat parallelism and convergenceas same.

Reversal is a common evolutionary pro-cess, wherein loss of a particular charactermay lead to apparent similarity with ances-tral condition. The occurrence of reducedunisexual flowers without perianth or withreduced perianth in Amentiferae was onceconsidered to be primitive situation, but theevidence from wood anatomy, floral anatomyand palynology have shown that apparentsimplicity of these flowers is due to evolu-tionary reduction (reversal), and as such theassumed similarity to angiosperm ancestralcondition is representation of homoplasy, afalse similarity between an evolutionaryadvancement (secondary reduction) andancestral simple condition.

Monophyly, Paraphyly andPolyphylyThese terms have been commonly used intaxonomy and evolutionary literature withsuch varied interpretation that much con-fusion has arisen in their application.

Figure 8.5 Two possible reasons for similar-ity between species A and B. In (I),A (cf. Arenaria leptocladus) and B (cf.Minuartia hybrida) are the mostprimitive members of respective lin-eages FGHA (cf. Arenaria) and BCDE(cf. Minuartia). The two lineages havecommon ancestry and thus consti-tute a single monophyletic group(cf. Arenaria s. l.). In (II), A and Bhappen to be the most advancedmembers of the respective groups,the two lineages are distinct andas such similarity between A and Bis superficial due to convergence,justifying the independent recogni-tion of two lineages (cf. distinct gen-era Arenaria and Minuartia).

F G H A B C D E

F G H A B C D E

m

n

I

II


more lineages from one immediately an-cestral taxon of the same or lower rank.Such a definition would be true if, say, ge-nus B evolved from genus A through one spe-cies of the latter, since in that case, thegenus B would monophyletic at the samerank (genus) as well as at the lower (spe-cies) rank. On the other hand, if genus Bevolved from two species of genus A, it wouldbe monophyletic at the genus level but poly-phyletic at the lower rank.

Most authors, however, including Heslop-Harrison (1958) and Hennig (1966), adhereto a stricter interpretation of monophyly,namely the group should have evolved froma single immediately ancestral specieswhich, may be considered as belonging tothe group in question. There are thus twodifferent levels of monophyly: a minimummonophyly wherein one supraspecific taxonis derived from another of equal rank(Simpson’s definition), and a strict mono-phyly wherein one higher taxon is derivedfrom a single evolutionary species.

Mayr (1969) and Melville (1983) follow theconcept of minimum monophyly. Most au-thors, including Heslop-Harrison (1958),Hennig (1966), Ashlock (1971) and Wiley(1981), reject the idea of minimum mono-phyly. All supraspecific taxa are composedof individual lineages that evolve indepen-dent of each other and cannot be ancestralto one another. Only a species can be anancestor of a taxon. The supraspecific an-cestors and, for that matter, supraspecifictaxa are not biologically meaningful entitiesand are only evolutionary artifacts.

Hennig (1966) defined a monophyleticgroup as a group of species descended froma single (‘stem’) species, and which in-cludes all the descendants from this spe-cies. Briefly, a monophyletic group comprisesall the descendants that at one time be-longed to a single species. A useful analysisof Hennig’s concept of monophyly was madeby Ashlock (1971). He distinguished betweentwo types of monophyletic groups:

: those that

are holophyletic when all descendants ofthe most recent common ancestor are con-tained in the group (monophyletic sensu

Defined broadly, the terms monophyly (deri-vation from a single ancestor) and polyphyly(derivation from more than one ancestor)would have different meanings dependingupon how far back we are prepared to go inevolutionary history. If life arose only onceon Earth, all organisms (even if you placean animal species and a plant species in thesame group) are ultimately monophyletic inorigin. There is thus a need for a precisedefinition of these terms, to make themmeaningful in taxonomy.

Simpson (1961) defined monophyly asthe derivation of a taxon through one or

Figure 8.6 Concepts of monophyly, paraphylyand polyphyly. In (I) groups AB andCD are monophyletic as each hasa common ancestor at level m.Similarly. group ABCD is monophyl-etic as it has a common ancestorat level n. In (II) group ABC isparaphyletic as we are leaving outdescendant D of the common an-cestor at level n. In (III) group BCis polyphyletic as their respectiveancestors at level m do not belongto this group.

A B C D A B C D

m

n

I

III

A B C D

II

m

n


group, i.e., it represents more than onepiece of a branch.

Gerhard Haszprunar (1987) introduced theterm orthophyletic while discussing the phy-logeny of Gastropods. An orthophyletic groupis a stem group, i.e. a group that isparaphyletic because a single clade (thecrown group), has been excluded. The termhas not been followed in other groups, espe-cially in botanical systematics. Sosef (1997)compares the existent hierarchical modelsof classification. He argues that a phyloge-netic tree can be subdivided according to amonophyletic hierarchical model, in whichonly monophyletic units figure or, accordingto a ‘Linnaean’ hierarchical model, in whichboth mono- and paraphyletic units occur.Most present-day phylogeneticists try to fitthe monophyletic model within the set of no-menclatural conventions that fit the Lin-naean model. However, the two models areintrinsically incongruent. The monophyleticmodel requires a system of classification ofits own, at variance with currently acceptedconventions. Since, however, the mono-

Hennig) and those that are paraphyleticand do not contain all descendants of themost recent common ancestor of the group.A polyphyletic group, according to him, isone whose most recent ancestor is not cla-distically a member of that group. Theterms holophyletic and monophyletic arenow considered synonymous. Diagrammaticrepresentations of Ashlock’s concept ofpolyphyly, monophyly and paraphyly is pre-sented in Figure 8.6.

An excellent representation of mono-phyly, paraphyly and polyphyly is presentedby ‘cutting rules’, devised by Dahlgren andRasmusen (1983). The distinction is basedon how the group is separated from a repre-sentative evolutionary tree (Figure 8.7). Amonophyletic group is separated by asingle cut below the group, i.e. it repre-sents one complete branch. A paraphyleticgroup is separated by one cut below thegroup and one or more cuts higher up, i.e.it represents one piece of a branch. A poly-phyletic group, on the other hand, is sepa-rated by more than one cut below the

Figure 8.7 The application of cutting rules to distinguish between monophyly, paraphyly andpolyphyly. The group is represented by lighter portion of the tree. Monophyletic groupcan be separated by a single cut below the group, a paraphyletic group by one cutbelow the group and one or more higher up. A polyphyletic is separated by more thanone cut below the group. A monophyletic group represents one complete branch, aparaphyletic group one larger portion of the branch; whereas the polyphyletic grouprepresents more than one pieces of a branch (based on Dahlgren et al., 1985).

Monophyly Paraphyly Polyphyly


phyletic model is unable to cope with reticu-late evolutionary relationships; it isunsuited for the classification of nature. TheLinnaean model is to be preferred. Thisrenders the acceptance of paraphyleticsupraspecific taxa inevitable.

As is true for the distinction between par-allelism and convergence, similarly, the con-cepts of paraphyly and polyphyly (both of whichare rejected by modern phylogenetic system-atics while constructing classification), holdgood, when the former is applied to a group ofclosely related organisms and latter to dis-tantly related organisms. The conceptsbecome ambiguous when a small group oforganisms is considered. In Figure 8.6-III,taxa B and C— if brought together—would form

a polyphyletic group, because they are derivedfrom two separate ancestors at level m. If,however, A, B, and C are under one group, Band C would still now be components of aparaphyletic group, because one descendantof the common ancestor at level n is kept outof the group. A natural group would be one,which includes all descendents of the com-mon ancestor, or the group is monophyletic.

Phylogenetic DiagramsThe affinities between the various groupsof plants are commonly depicted with thehelp of diagrams, with several innovations.These diagrams also help in understandingthe classification of included taxa. An

Figure 8.8 A phylogenetic tree representing the evolutionary history of plants including angio-sperms. The vertical axis represents the geological time scale. Only extant (living)plants are shown reaching the top.


understanding of these terms is necessaryfor a correct interpretation of putative rela-tionships. These branching diagrams arebroadly known as dendrograms. Any dia-gram showing the evolutionary history of agroup in the form of branches arising fromone or more points has often been referredas a phylogenetic tree, but the use of termsis now becoming more precise, and more in-novative diagrams are being developed of-ten providing useful information about dif-ferent taxa mapped in the diagram.

The most common form of diagram is onewhere the length of branch indicates the de-gree of apomorphy. Such diagrams weresometimes classified as cladograms (Stace,1980), but the term has now been restrictedto diagrams constructed through the dis-tinct methodology of cladistics (Stace, 1989).Diagrams with vertical axis representing thedegree of apomorphy are now more appro-priately known as phylograms. The earli-est well-known example of such a phylogramis ‘Bessey’s cactus’ (see Fig 10.11). In suchdiagrams the most primitive groups endnear the base and the most advanced reachthe farthest distance.

Hutchinson (1959, 1973) presented hisphylogram in the form of a line diagram (fig-ure 10.13). The recent classifications ofTakhtajan (1966, 1980, 1987) and Cronquist(1981, 1988) have more innovativephylograms in which the groups are de-picted in the form of balloons or bubbleswhose size corresponds to the number ofspecies in the group (an approach also foundin Besseyan cactus). Such phylograms thusnot only depict phylogenetic relationshipsbetween the groups, they also show the de-gree of advancement as also the relativenumber of species in different groups. Suchdiagrams have been popularly known asbubble diagrams. The bubble diagram ofTakhtajan (Figure 10.16) is more detailedand shows the relationship of the orderswithin the ‘bubble’; as mentioned earlier,Woodland (1991) aptly described it as‘Takhtajan’s flower garden’.

The phylogenetic tree is a commonlyused diagram in relating the phylogenetic

history. The vertical axis in such a diagramrepresents the geological time scale. In sucha diagram, the origin of a group is depictedby the branch diverging from the main stockand its disappearance by the branch termi-nation. Branches representing the fossilgroups end in the geological time when thegroup became extinct, whereas the extantplant groups extend up to the top of the tree.As already mentioned, the relative advance-ment of the living groups is indicated by theirdistance from the centre, primitive groupsbeing near the centre, and advanced groupstowards the periphery. A phylogenetic treerepresenting possible relationships and theevolutionary history of seed plants is pre-sented in Figure 8.8.

Dahlgren (1975) presented the phyloge-netic tree (preferred to call it phylogenetic‘shrub’ in 1977) of flowering plants with allextant groups reaching the top, and thecross-section of the top of the phylogenetictree was shown as top plane of this diagram(Figure 8.9). In subsequent schemes ofDahlgren (1977, 1983, 1989), the branchingportion of the diagram was dropped and only

Figure 8.9 Phylogenetic tree of angiospermspresented by Dahlgren (1975) witha section of the top (subsequentlynamed phylogenetic ‘shrub’ byDahlgren, 1977).


the top plane (cross-section of the top) pre-sented as a two-dimensional diagram (Fig-ure 8.10), and this has been very useful inmapping the distribution of various charac-ters in different groups of angiosperms, andthe comparison of these provides a goodmeasure of correspondence of various char-acters in phylogeny. This diagram has beenpopularly known as ‘Dahlgrenogram’.

Thorne’s diagram (2000) is similarly thetop view of a phylogenetic shrub (Figure10.23), in which the centre representing theextinct primitive angiosperms, now absent,is empty.

A cladogram represents an evolutionarydiagram utilizing cladistic methodology,which attempts to find the shortest hypo-

thetical pathway of changes within a groupthat explains the present phenetic pattern,using the principle of parsimony. A cla-dogram is a representation of the inferredhistorical connections between the entitiesas evidenced by synapomorphies. The verti-cal axis of the cladogram is always animplied, but usually non-absolute time scale.Cladograms are ancestor-descendantsequences of populations. Each bifurcationof the cladogram represents a past specia-tion that resulted in two separate lineages.

It must be pointed out, however, thatconsiderable confusion still exists betweenapplication of the terms cladogram andphylogenetic tree. Wiley (1981) defines a cla-dogram as a branching diagram of entities

Figure 8.10 Mapping of pollen grain dispersal stage in different dicotyledons on a two-dimen-sional diagram (Dahlgrenogram) of Dahlgren, representing transverse section throughthe top of a phylogenetic shrub. Pollen grain dispersal in 2-celled stage (unshaded),3-celled stage (dotted), or mixed (hatched). (Courtesy Gertrud Dahlgren).


changes, depending on which species isancestral and being relegated lower down onthe vertical axis. In the case of higher taxa,the number of cladograms and phylogenetictrees could possibly be equal, because highertaxa cannot be ancestral to other higher taxasince they are not units of evolution buthistorical units composed of separatelyevolving species.

Over the recent years, it has been thusbecoming increasingly common to constructevolutionary diagrams using cladistic

where the branching is based on inferredhistorical connections between the enti-ties as evidenced by synapomorphies. Itis, thus, a phylogenetic or historical den-drogram. He defines a phylogenetic tree asa branching diagram portraying hypoth-esized genealogical ties and sequences ofhistorical events linking individual or-ganisms, populations, or taxa. At the spe-cies and population level, the number of pos-sible phylogenetic trees could be more thancladograms for particular character

Figure 8.11 Tree (cladogram) for different families of the order Alismatales. Support indicatedfor branches refers to bootstrap support, discussed in subsequent pages. (Repro-duced from APweb vesion 7 (June , 2008), with permission from Dr P. F. Stevens.)


methodology, assuming that these charac-ter-state changes (represented as evolution-ary scale or tree length) correspond to thegeological time scale, and call these evolu-tionary diagrams as evolutionary tree (Juddet al., 2008), phylogenetic tree, or simplytree (Stevens, 2008), synonymous with acladogram.

A Phenogram is a diagram constructedon the basis of numerical analysis of phe-netic data. Such a diagram is the result ofutilization of a large number of characters,usually from all available fields, and involvescalculating the similarity between taxa andconstructing a diagram through clusteranalysis. Such a diagram (Figure 8.20) isvery useful, firstly because it is based on alarge number of characters, and secondlybecause a hierarchical classification can beachieved by deciding upon the thresholdlevels of similarity between taxa assignedto various ranks.

It must be pointed out that the modernphylogenetic methods, which aim at con-structing phylogenetic trees, also some-times use large number of characters forcomparison, especially when dealing withmorphological data, and there seem to be alot of similarities in data handling and com-putation, but are unique in the utilizationof evolutionary markers and, consequently,produce slightly different results. With theincorporation of distance methods in the con-struction of trees, the classical differencebetween the terms is largely disappearing.Modern cladistic programs develop trees inwhich branch lengths are indicated, andplotting programs offer the choice to indicatebranch lengths (and often called phylograms)or not. In latter case branches may besquare (line running vertically and horizon-tally- and often called phenogram; Figure8:21) or V-shaped (cladogram; Figure 8.11).These may be presented as upright or ashorizontal trees (prostrate trees) . Moderntrees contain information about evolution-ary markers such as bootstrap support,branch length, and Bremer support, asdiscussed in subsequent pages.

Phylogeny and ClassificationThe construction of phylogenetic classifica-tion involves two distinct steps: determiningthe phylogeny or evolutionary history of agroup, and construction classification on thebasis of this history. Imagine a lineage (orclade- a group of individuals producing suc-cessively, similar and genetically related in-dividuals, generally represented by lines in acladogram) with woody habit, alternate leaves,cymose inflorescence, 5 red petals, 5 stamens,2 free carpels, and dry fruit with many seeds.Over a period of time, some population ac-quires herbaceous habit and the original lin-eage splits into two, one with woody habit andthe second with herbaceous habit (Figure8.12). In the lineage with woody habit, one lin-eage emerges with fused carpels, while theother loses one of the two carpels. The onewith fused carpels loses 3 of the five stamensin one or more populations, and that with asingle carpel doubles the number of stamensto ten in one or more populations. The herba-ceous lineage, similarly, splits into one withyellow petals and one with white petals, theformer developing fleshy fruits in one or morepopulations, and the latter having the num-ber of seeds reduced to one in some popula-tions. The present descendents of the origi-nal ancestor are thus represented by eight lin-eages, which have developed a fewapomorphic character-states, but also shareplesiomorphic character-states such as alter-nate leaves, 5 petals, and cymose inflores-cence. There must be hundreds of moreplesiomorphic states, but of little significancein classification, as the above three. Note thatthe ancestral species at level I, II (woody habit,2 free carpels), IA (herbaceous habit, red pet-als), III (herbaceous habit, yellow petals andfree carpels and dry fruit), and IV (herbaceoushabit, white petals, 5 stamens and manyseeds) and have disappeared, whereas thoseat level V, and VI are still represented (al-though with minor changes) in the form of Eand G, respectively. Also note that united car-pels have arisen twice independently. Thesame is also true for the loss of three stamens.


Having known the evolutionary history ofthe group, we could use synapomorphy andthe concept of monophyly. Assuming that alleight lineages (groups of populations) are suf-ficiently distinct to be recognized as distinctspecies, we would have eight species. Thesimplest way would be to group these eightspecies into four genera, each having a com-mon ancestor. Two of these common ances-tors have disappeared, but two are still liv-ing and would also be included in the respec-tive genera (it will be more appropriate toregard E as ancestral to F and G ancestral toH). These could be further assembled intotwo families of four species each (two gen-era each) having a common ancestor at levelIA and II, and these two families into oneorder with a common ancestor at level I.Please note that common ancestor at levelI, IA, II, III and IV are also no longer living.

The second option would be to includeABCD in one genus, and EFGH in another

genus, and include all 8 species (2 genera)in one family (and, of course, depending uponthe degree of diversity from related families,this could still be a constituent of a mono-typic order). The third option would be to havea single genus of eight species.

Note the importance of synapomorphy indetermining monophyly. Character-statesalternate leaves, cymose inflorescence and5 petals (character-states of different char-acters not same) have been passed un-changed in all the eight descendents (spe-cies), which as such are symplesiomorphicfor each of these, and this symplesiomorphywill be valid between any two (or more)species that you choose to combine into agenus, say, D and E, or C and F, or sayABCDE. On the other hand, if we consideronly synapomorphy, the monophyly is easilydeciphered. A and B are accordinglysynapomorphic for yellow petals, C and D forwhite petals, E and F for one carpel and

Figure 8.12 Evolutionary history of a hypothetical group of organisms which started with theancestral species with woody habit, alternate leaves, cymose inflorescence, 5 redpetals, 5 stamens, 2 free carpels and dry fruit with many seeds. Eleven characterstate transformations at different stages have resulted in 8 present day species.Note that two of the changes (carpel union and loss of three stamens) have occurredtwice, and as such only nine genetic switches are involved. The ancestral speciesat levels I , IA, II, III an IV have disappeared.

A B C D E F G H

I

IIIA

VIII IV VI

Herbaceous habit

Yellow petals White petals

Fleshy fruit Single seed 10 stamens

1 carpel Carpels united

2 stamens

A

2 stamensCarpels united


Figure 8.13 Diagrams based on evolutionary pattern depicted in Figure 10.11. A: Venn diagrambased on the assumption that there are 21 woody species of which, 13 are withunited carpels, and 8 with free carpels. Of the 13 with united carpels, 7 have 2stamens whereas 6 have more stamens. B: An unrooted tree based on Venn dia-gram. C: A possible rooted tree, if evolutionary history of the group was not known,15 possible rooted trees could be drawn. D. Rooted tree based on knowledge thatherbaceous habit arose from woody habit, and we know further evolution of woodyspecies. Other possibilities are discussed subsequently. E: Data matrix of above,where the number of species are not indicated.

Plants woody

Carpels united

Stamens 2

Plants herbaceous

B

Stamens 2

Stamens >2

Carpels united

Carpels free

Plants woody

Plants herbaceous

A

C

D

Stamens 2

Stamens >2

Carpels united

Carpels free

Plants herbaceous

Plants woody

Stamens 2

Stamens >2

Carpels united

Carpels free

Plants herbaceous

Plants woody

Plants

Plants

Plants

Plants

Habit

2UnitedWoody

> 2UnitedWoody

> 2FreeWoody

> 2FreeHerbaceous

Carpels

Stamens

E


G and H for united carpels. Similarly, A, B, Cand D are synapomorphic for herbaceoushabit. The development of fleshy fruit in A,single seed in C, 10 stamens in F representthe occurrence of a derived character statein a single taxon, and termed asautapomorphy. Such character states arenot helpful in cladogram construction, al-though indicative of divergence. Develop-ment of 2 stamens in D and H independentlyrepresents homoplasy, and may lead to ar-tificial grouping of these two, if history of thegroup was not properly known. It must, how-ever, be noted that symplesiomorphy maysometimes be helpful in detecting mono-phyly, especially where in some other taxa,it has evolved into another character-state.As such, out of the four plesiomorphic char-acter-states listed here, only the woody habithas changed to herbaceous habit, and assuch in the remaining taxa (E, F, G and H)symplesiomorphy of woody character-stateidentifies monophyly of the group ABCD. Itmust be remembered that synapomorphyand symplesiomorphy are a reflection of ho-mology for a particular character-state (ormore than one character-states, each be-longing to a different character).

All above options would render (genus, fam-ily or order) monophyletic groups, the ulti-mate goal of phylogenetic systematics. Anyother options won’t work. Keeping D and Ein one genus (or CDE or DEF) would make itpolyphyletic, promptly rejected once de-tected, because the group is derived frommore than one ancestor. Keeping ABC un-der one genus, or FGH under one genuswould make paraphyletic groups becausewe are not including all the descendents ofthe common ancestor (we are leaving out Din first genus and E in second). In the sameway, putting more than four species (but lessthan eight) under the same genus wouldmake it paraphyletic. Paraphyletic taxa arestrongly opposed by phylogenetic system-atists; the classical case is the demise oftraditional division of angiosperms intomonocots and dicots, over the last decade.

It must be noted that all eight species—whatever way you classify— share alternate

leaves, cymose inflorescence and five pet-als. The species at level IA and above addi-tionally share woody habit, VI and above ad-ditionally united carpels, V and above onecarpel (and not fused carpels). Similarly, spe-cies at level II and above share herbaceoushabit (instead of woody habit) in addition tothree common, at level III and above addi-tionally yellow petals, and at level IV andabove additionally (in place of yellow petals)white petals.

The situation depicted above can be moreeasily represented through the concept ofnested groups, more conveniently repre-sented as a set of ovals, a Venn diagram (Fig-ure 8.13A). The diagram drawn here is basedon the assumption that we have informa-tion from a large group in which there are21 woody species of which 13 are with unitedcarpels and 8 with free carpels. Of the 13with united carpels 7 have 2 stamens whereas 6 have more stamens.

The information is presented in the formof an unrooted network (unrooted tree) (Fig-ure 8.13B). The herbaceous species areshown towards the left of left double arrow,and the woody species towards the right.Similarly, the species towards the left of themiddle double arrow are with free carpelswhile those towards the right with fused car-pels. The species towards the left of the rightdouble arrow have more than 2 stamens andthose towards the right just 2 stamens.

It must be noted that in constructing theabove Venn diagram and the unrooted net-work, only three character-state transforma-tions are accounted for. We have completelyleft out grouping of herbaceous plants andthe woody plants with a single carpel. Inclu-sion of these would make the diagramsmuch more complicated, and present sev-eral alternatives. Also, the more meaning-ful trees have to be rooted (the most primi-tive end at the base), to reflect the phylog-eny. Even with the phylogenetic history ofthe group known, there could be severalvariations of the rooted tree, two simple onesbeing shown in the Figure 8.13C and 8.13D.If we did not know the evolutionary historyof the group, a number of variations would be


possible, depending upon which character-state is plesiomorphic, and which character(habit, carpel fusion or stamen number)forms the root, and what would be the se-quence of the character changes on the tree.

The character-states chosen for analysisshould necessarily be homologous (one de-rived from another) and non-overlapping. Theanalysis becomes more meaningful when itis established that the evolution of a par-ticular character-state has been the resultof a corresponding genetic change, and nota mere plastic environmental influence.This fact underlies the importance of theemerging field of molecular analysis in phy-logenetic systematics. It is believed thatthe recognition of molecular character-states(nucleotide sequences) is often easier andmore precise, although there are always ac-companying problems.

The problem with vascular plants, espe-cially the angiosperms, is that we know verylittle about their evolutionary history. Thefossil records, which generally give fair in-formation about evolution, are very scarce-ly represented. What we have available withus is a mixture of primitive, moderately ad-vanced and advanced groups. Almost eachgroup has some plesiomorphic and someapomorphic character-states, and relativeproportion of one or the another delimit therelative advancement of various groups. At-tempt to reconstruct the evolutionary histo-ry of the group involves comparative studyits living members, sorting out plesiomor-phic and apomorphic character states, anddistribution of these in various members.Once the evolutionary history of the grouphas been constructed, monophyletic groupsat various levels of inclusiveness are iden-tified, assigned ranks, and given appropri-ate names, to arrive at a working system ofclassification.

PHYLOGENETIC DATA ANALYSISThe methodology of cladistics with incorpo-ration of numerical methods involves a num-ber of steps.

Taxa-Operational UnitsThe first step in data analysis involves theselection of Taxa for data collection, oftencalled Operational Taxonomic Units (OTUs)in Taxometrics, Operational EvolutionaryUnits (OEUs) in cladistics, referring to thesample from which the data is collected. Al-though it would be ideal to select differentindividuals of a population, practical consid-erations make it necessary to select themembers of the next lower rank. Thus, forthe analysis of a species would need selec-tion of various populations, for the study of agenus they would be different species, andfor a family they would be different genera.It is not advisable, however, to use generaand higher ranks, as the majority of char-acters would show variation from one spe-cies to another and thus would not be suit-able for comparison. The practical solutionwould be to use one representative of eachtaxon. Thus, if a family is to be analysed andits genera to be compared, the data from onerepresentative species of each genus can beused for analysis. Once the taxa are selected,a list of such taxa is prepared. A unique fea-ture of cladistic studies, however, is that thelist of taxa generally includes a hypotheti-cal ancestor, the comparison with which re-veals crucial phylogenetic information, andis used for rooting of the tree. It is, increas-ingly being realized that only a species isthe valid evolutionary entity, and all taxa athigher ranks are artifacts, constructed forthe sake of convenience. A meaningfulanalysis would always be one derived fromdata from various species (taken from popu-lations) and not any higher rank directly.

CharactersA conventional definition of a taxonomiccharacter is a characteristic that distin-guishes one taxon from another. Thus,white flowers may distinguish one speciesfrom another with red flowers. Hence, thewhite flower is one character and red floweranother. A more practical definition espoused


information is not available (a large num-ber of plants in a population are not in fruit)or the information is irrelevant (trichometype if a large number of plants are withouttrichomes), or the characters which show amuch greater variation within the sametaxon. Such characters are omitted from thelist. This constitutes residual weighting ofcharacters. The characters (leaves, bracts,carpels) or character states (simple leaf, pal-mate compound leaf, pinnate compound leaf)chosen should also be homologous, in termsof sharing common ancestry or belongingto same evolutionary transformation series.The ‘petals’ of Anemone are modified sepalsand thus not homologous with the petals ofRanunculus and hence not comparable. Simi-larly, the tuber of sweet potato (a modifiedroot) cannot be compared with the tuber ofpotato (a modified stem).

Binary and multistatecharactersThe characters most suitable for computerhandling are two-state (binary or presence-absence) characters (habit woody or herba-ceous). However, all characters may not betwo-state. They may be qualitativemultistate (flowers white, red, blue) or quan-titative multistate (leaves two, three, four,five at each node). Such multistate charac-ters can be converted into two-state (flowerswhite or coloured; leaves four or more vsleaves less than four). Or else the charac-ters may be split (flowers white vs not white,red vs not red, blue vs not blue; leaves twovs not two, three vs not three and so on).Such a splitting may, however, give moreweightage to one original character (flowercolour or number of leaves). It is essentialthat different character states identified arediscrete or discontinuous from one another.Discreteness of character states can beevaluated by comparing the means, ranges,and standard deviations of each characterfor all taxa in analysis. Additionally t-testsand multivariate analysis may also be usedfor evaluating character state disreteness.

by numerical taxonomists defines charac-ter (Michener and Sokal, 1957) as a feature,which varies from one organism to an-other. By this second definition, flowercolour (and not white flower or red flower) isa character, and the white flower and redflower are its two character-states. Someauthors (Colless, 1967) use the term at-tribute for character-state but the two arenot always synonymous. When selecting acharacter for numerical analysis, it is im-portant to select a unit character, whichmay be defined as a taxonomic characterof two or more states, which within thestudy at hand cannot be subdivided logi-cally, except for the subdivision broughtabout by the method of coding. Thus, tri-chome type may be glandular or eglandular.A glandular trichome may be sessile orstalked. An eglandular trichome may, simi-larly, be unbranched or branched. In such acase, a glandular trichome may be recog-nized as a unit character and an eglandulartrichome as another unit character. On theother hand, if all glandular trichomes inOTUs are of the same type and all eglandulartrichomes are of the same type, the trichometype may be selected as a unit character.

The first step in the handling of charac-ters is to make a list of unit characters. Apreliminary step involves character compat-ibility study in which each character is ex-amined to determine the proper sequenceof character-state changes that take placeas the evolution progresses (morphoclinesor transformation series). The list shouldinclude all such characters concerningwhich information is available. A priori, allcharacters should be weighted equally (noweighting to be given to characters).Although some authors advocate that somecharacters should subsequently be assignedmore weightage than others (a posterioriweighting), such considerations generallyget nullified when a large number of char-acters is used. It is generally opined thatnumerical studies should involve not lessthan 60 characters, but more than 80 aredesirable. For practical consideration, theremay be some characters concerning which


Ordering of Character-statesA binary character will have single step orswitch (Figure 8.14-I) necessary for change.The minimum number of switches possible(Wagner parsimony) in a multistate char-acter will depend whether the characterstates are ordered or left unordered. In anunordered transformation series each char-acter state can evolve into every other char-acter state, and represents a single switch(Figure 8.14-II, III). A three-state characterwill have two switches or steps, and threepossible morphoclines (Figure 8.14-IV), four-state character three switches and severalmorphoclines. Whereas ordering of two-statecharacters is relatively easy, multi-statecharacters are often difficult to order, andchanges may often be reversible, and it isadvisable to leave them unordered, and iden-tify only one switch (Fitch parsimony). Themolecular characters are different DNA se-quences, that may differ in having one ofthe four bases (adenine, thymine, guanine

and cytosine) at a particular locus, and assuch present four character-states. As re-versals are common in these, these are al-ways left unordered.

Assigning PolarityIt is, however, necessary to determine therelative ancestry of the character-states, orthe assignment of polarity. The designationof polarity is often one of the more difficultand uncertain aspects of phylogenetic analy-sis. For this, the comparison may be madewithin the concerned group (in-group com-parison) or relatives outside the group (out-group comparison). The latter may often pro-vide useful information, especially when theout-group used is the sister-group of the con-cerned group. If two character-states of acharacter are found in a single monophyl-etic group, the state that is also found in asister-group is likely to be plesiomorphic andthat found only within the concerned mono-phyletic group is likely to be apomorphic.

A B A B

C DCA B

C A B

C

I

II III

A B

B A C

A C B

A B C

C B

A C

A

B

C AB

C B A

IV

VI

A B

B AV

Figure 8.14 Ordering and polarity of character states. I: Binary character with single possibleswitch. II: Unordered three-state character with single possible switch. III: Unor-dered Four-state character with single possible switch. IV: Ordered three-state char-acter with two possible switches and three possible morphoclines. V: Polarized bi-nary character with two possible morphoclines. VI: Ordered and Polarized three-state character with 6 possible morphoclines.


Ingroup comparison (also known as com-mon ground plan or commonality principle)is based on the presumption that in a givengroup (presumably monophyletic), the primi-tive structure would tend to be more com-mon. Thus all 8 species of cladogram in Fig-ure 8.12 share plesiomorphic characterstates: alternate leaves, cymose inflores-cence and five petals. Five species haveplesiomorphic 5 stamens, two derived 2 sta-mens and one with 10 derived stamens.Similarly four species have red petals’ andtwo each with white and yellow petals. It isassumed that the evolution of a derived con-dition will occur in only one of potentiallynumerous lineages of the group; thus theancestral condition will tend to be in the ma-jority. As is evident from Figure 8.14, thenumber of possible morphoclines increases

after the polarity criterion is included andthe selection of single appropriate mor-phocline representing the true sequenceeven more challenging.

Character Weighting andCodingThe coding of character states is done byassigning non-negative integer values. Bi-nary characters are conveniently assigned0 and 1 for two states. If possible to distin-guish, plesiomorphic state is assigned 0 andapomorphic state 1 code (Figure 8.15-IV). Itis often assumed that whereas the samecharacter-state may arise more than oncewithin a group between closely related spe-cies (parallelism), or between remotely re-lated species (convergence; the distinction

A B C

A

B

C

0 1 2

1 0 1

2 1 0

A B C D

A

B

C

D

0 1 2 3

1 0 1 2

2 1 0 1

3 2 1 0

A B C D

A

B

C

D

0 1 1 1

1 0 1 1

1 1 0 1

1 1 1 0

A B C D

A

B

C

D

0 1 5 5

1 0 5 5

5 5 0 1

5 5 1 0

III III

V

A

B

A B

0 1

1 0

IV

Figure 8.15 Data matrix of coded character states. I: Ordered three-state character. II: orderedfour-state character. III: Unordered character. IV: Binary character. V: Differentialweighting to character state changes; imagine A and B represent Purines (Adenineand Guanine), C and D Pyrimidines (Cytosine and Thymine), purine to purine orpyrimidine to pyrimidine change (transition) is given 1 step weight, but purine topyrimidine change or reverse (transversion) given 5 steps weight.


Characters �

OTUs (t) H

abit

0-w

oody

1-h

erba

ceou

s Fr

uit

0-fo

llicl

e 1-

ache

ne

Ova

ry

0-su

peri

or 1

-inf

erio

r L

eave

s 0-

sim

ple

1-c

ompo

und

Hab

itat

0-

terr

estr

ial

1-aq

uatic

Po

llen

1-

trip

orat

e 0-

mon

osul

cate

O

vule

1-

unite

gmic

0-b

iteg

mic

C

arpe

ls

0-fr

ee 1

-uni

ted

Plas

tids

1-P

I-ty

pe 0

-PII

-typ

e

1. 1 0 1 1 0 1 1 1 1 2. 1 0 1 1 1 1 0 0 1 3. 0 NC 0 1 0 0 1 1 1 4. 1 1 1 0 1 0 0 0 0 5. 1 0 1 1 0 1 1 0 1 6. 1 1 0 1 1 NC 0 1 0 7. 0 1 1 0 1 1 0 0 1 8. 0 0 0 1 0 0 1 1 1 9. 1 1 1 0 0 1 1 0 0 10. 0 0 0 1 1 0 1 1 1 11. 1 0 1 1 0 1 0 0 1 12. 1 1 0 0 1 1 0 0 1 13. 0 0 1 0 1 0 1 0 1 14. 1 0 1 0 1 0 1 0 1 15. 0 1 1 0 0 1 1 0 0

Table 8.1 A portion of the data matrix with hypothetical t OTUs and n characters. Binary codinginvolves for state a and 1 for state b. The NC code stands for characters not comparablefor that OTU. In this analysis a total of 100 characters were used but only nine arepictured here.

between parallelism and convergence issometimes omitted) for a simple character,it is highly unlikely for more complex char-acters. It is also assumed that whereas manygenes must change in order to create a mor-phological structure, one gene change isenough for its loss (reversal). This Dollo’slaw is taken into account when choosingtrees, gains of structures counted more thanlosses, a process known as Dollo parsimony.Such weighting of characters is often com-mon in phylogenetic analysis. In transfor-mation series leaf simple —> pinnately lobed—> pinnately compound, the development ofpinnate compound leaf from simple leaf oc-curred in two steps, and needs to be givenmore weightage. The coding may accordingly

be done as 0 for most primitive character-state (simple leaf), 1 for intermediate char-acter-state (pinnately lobed leaf) and 2 moremost advanced state (pinnately compoundleaf) (Figure 8.15-I). In molecular data,transversions (Purine to pyrimidine or py-rimidine to purine changes) are given moreweightage (Figure 8.15-V) over transitions(purine to purine or pyrimidine to pyrimi-dine), because the latter occur more fre-quently and are easy to reverse, whereas theformer is a less likely biochemical change.Restriction site gains may similarly beweighted over site losses. A complex char-acter, presumably controlled by many genes,may change less easily than a simple char-acter controlled by fewer genes. The former


is often given more weighting over a simplecharacter. It may be assumed that leafanatomy may not change easily but hairi-ness may change readily. The number ofsteps between two character states is con-veniently represented through characterstep matrix (Figure 8.15). One may, how-ever, be tempted to count leaf anatomy char-acter as equivalent to two changes in hairi-ness. This may often be the result of bias toobtain desired results. It is reasonable, how-ever, to adopt the approach of numerical tax-onomy to give equal weighting to all thecharacters in the preliminary analysis,identify those characters which show theleast homoplasy and give them moreweightage in the subsequent analysis, a pro-cess known as successive weighting. Thisavoids a bias towards a particular charac-ter, and as such enables rational treatmentof available data.

Residual weighting involves excluding acharacter from the list when information fora large number of taxa is not available, or isirrelevant. But in certain cases, informationmay be available for a particular characterfor large number of OTUs but not for a few.Alternately, the information may be irrel-evant for a few taxa (say, the number of spursin a taxon, which lacks spurs). Such char-acters are used in analysis but for the taxafor which information is not available or isirrelevant, an NC code (Not Comparable) isentered in the matrix. Whenever the NCcode is encountered, the program bypassesthat particular character for comparing theconcerned taxon. For data handling by com-puters, the NC code is assigned a particular(not 0 or 1) numeric value. Such residualweighting should, however, be avoided whenappreciable number of taxa are not compa-rable for a particular character. The codeddata may be entered in the form of a matrixwith t number of rows (OTUs) and n num-ber of columns (character-states) with thedimension of the matrix (and the number ofattributes) being t x n (Table 8.1).

Certain characters in plants evolvetogether. Occurrence of stipules andtrilacunar nodes is usually correlated.

Similarly sympetalous members tend tohave epipetaly and tenuinucellate ovules.Such correlated characters receive lesserweighting. If two characters are correlated,each gets 1/2 weighting, if three 1/3 weight-ing and so on.

It is always advisable to identify and in-clude the most ancestral taxon (outgroup) aslast taxon (or first taxon, as certain programschoose first taxon for rooting) in the list oftaxa. If it is possible to identify plesiomorphicand apomorphic character-states, 0 repre-sents plesiomorphic character-state and 1the apomorphic character-state of a particu-lar character. Outgroup taxon in the matrixgets 0 code for all character states (Table 8.4).For multistep changes, or unlikely eventsappropriate codes as indicated in Figure 8.15are transferred to the matrix. Outgroup taxonin the matrix is essential in final rooting ofthe most parsimonious cladogram (tree).

Measure of similarityOnce the data have been codified andentered in the form of a matrix, the next stepis to calculate the degree of resemblancebetween every pair of OTUs. A number offormulae have been proposed by variousauthors to calculate similarity or dissimi-larity (taxonomic distance) between theOTUs. If we are calculating the similarity(or dissimilarity) based on binary data codedas 1 and 0, the following combinations arepossible.

Number of matches m = a + dNumber of positive matches aNumber of mismatches u = b + cSample size n = a + b + c + d = m + uj and k are two OTUs under comparison


Yule coefficientThis coefficient has been less commonlyused in numerical taxonomy. It is calculatedas:

SY = ac – bcad – bc

Taxonomic distanceTaxonomic distance between the OTUs canbe easily calculated as a value 1 minussimilarity or 100 minus percentage similar-ity. It can also be directly calculated as Eu-clidean distance using formula proposed bySokal (1961):

1/22

1( – )

n

ij iki

X X�

� ��

jk

The average distance would be repre-sented as:

2

jkdn

��

jk

Other commonly used distance measuresinclude Mean character difference (M.C.D.)proposed by Cain and Harrison (1958),Manhattan metric distance coefficient(Lance and Williams, 1967) and Coefficientof divergence (Clark, 1952).

Once the similarity or distance betweenevery pair of taxa has been calculated, thedata are presented in a second matrix witht x t dimensions where both rows and col-umns represent taxa (Table 8.2; Table 8.3).It must be noted that diagonal t value in thematrix represents self-comparison of taxaand thus 100% similarity. These values areredundant as such. The values in thetriangle above this diagonal line would besimilar to the triangle below. The effectivenumber of similarity values as such wouldbe t x (t-1)/2. Thus if 15 OTUs are com-pared the number of values calculated wouldbe 15 x (15-1)/2 = 105.

A data matrix with coded character-statesfor each taxon can be used for calculating

Some of the common formulae are dis-cussed below:

Simple matching coefficientThis measure of similarity is convenientand highly suitable for data wherein 0 and1 represent two states of a character, and0 does not merely represent the absence ofa character-state. The coefficient was intro-duced by Sokal and Michener (1958). Thecoefficient is represented as:

SSM = Matches

Matches + Mismatches

orm

m u�

It is more convenient to record similarityin percentage (Table 8.2). In that case, theformula would read:

SSM

= m

m u� � 100

When comparing a pair of OTUs, a match isscored when both OTUs show 1 or 0 for a par-ticular character. On the other hand, if oneOTU shows 0 and another 1 for a particularcharacter, a mismatch is scored.

Jaccard Coefficient ofassociationThe coefficient was first developed by Jaccard(1908) and gives weightage to scores of 1only. This formula is thus suitable for datawhere absence-presence is coded and 1 rep-resents the presence of a particular char-acter-state, and 0 its absence. The formulais presented as:

SJ =

aa u�

where a stands for number of charactersthat are present (scored 1) in both OTUs .This can similarly be represented as apercentage similarity.


Table 8.2 Similarity matrix of the representative hypothetical taxa presented as percentage simple matching coefficient. OTUs

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

1 100

2 47.0 100

3 54.0 47.0 100

4 49.0 54.0 52.0 100

5 50.0 51.0 44.0 48.5 100

6 46.0 59.0 46.0 47.0 48.0 100

7 47.0 48.0 48.0 46.0 65.0 47.0 100

8 56.0 51.0 56.0 51.5 46.0 58.0 25.0 100

9 50.0 45.0 49.0 50.0 60.0 40.0 79.0 30.0 100

10 50.0 45.0 54.0 50.5 58.0 41.0 77.0 36.0 92.0 100

11 53.0 54.0 49.0 45.5 65.0 51.0 92.0 31.0 75.0 73.0 100

12 48.0 47.0 49.0 50.0 58.0 42.0 81.0 30.0 96.0 94.0 75.0 100

13 47.0 44.0 49.0 49.5 59.0 44.0 68.0 41.0 81.0 83.0 62.0 81.0 100

14 55.0 46.0 55.0 51.5 57.0 44.0 72.0 39.0 81.0 81.0 72.0 81.0 74.0 100

15 56.0 45.0 57.0 53.0 54.0 44.0 67.0 40.0 78.0 72.0 67.0 74.0 67.0 87.0 100

Table 8.3 Dissimilarity matrix of the representative hypothetical taxa based on the similarity matrix in Table 8.2. OTUs

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

1 0.0

2 53.0 0.0

3 46.0 53.0 0.0

4 51.0 46.0 48.0 0.0

5 50.0 49.0 56.0 51.5 0.0

6 54.0 41.0 54.0 53.0 52.0 0.0

7 53.0 52.0 52.0 54.0 35.0 53.0 0.0

8 44.0 49.0 44.0 48.5 54.0 42.0 75.0 0.0

9 50.0 55.0 51.0 50.0 40.0 60.0 21.0 70.0 0.0

10 50.0 55.0 46.0 49.5 42.0 59.0 23.0 64.0 8.0 0.0

11 47.0 46.0 51.0 54.5 35.0 49.0 8.0 69.0 25.0 27.0 0.0

12 52.0 53.0 51.0 50.0 42.0 58.0 19.0 70.0 4.0 6.0 25.0 0.0

13 53.0 56.0 51.0 50.5 41.0 56.0 32.0 59.0 19.0 17.0 38.0 19.0 0.0

14 45.0 54.0 45.0 48.5 43.0 56.0 28.0 61.0 19.0 19.0 28.0 19.0 26.0 0.0

15 44.0 55.0 43.0 47.0 46.0 56.0 33.0 60.0 22.0 28.0 33.0 26.0 33.0 13.0 0.0


the distance (and, consequently, the simi-larity) between every pair of taxa, includingthe hypothetical ancestor. The distance iscalculated as the total number of character-state differences between two concernedtaxa, the data presented as t x t matrix(Table 8.5).

This method is closer to taxometric meth-ods, because both plesiomorphic andapomorphic character-states are givenequal weightage, but the inclusion of hypo-thetical ancestor is always crucial for thestudy.

Another method of calculating distanceinvolves calculation of the number ofapomorphic character-states common be-tween the pairs of concerned taxa, ignoringthe possession of plesiomorphic character-states in common (Table 8.6). Since onlysynapomorphy is likely to define monophyl-etic groups, this method is closer to theoriginal cladistic concept.

Construction of TreesDifferent methods are available for the finalanalysis of cladistic information. Three ofthese commonly used in phylogenetic analy-sis include Parsimony-based methods, Dis-tance methods and Maximum likelihoodmethod.

Parsimony-based methodsThe methods are largely based on the bio-logical principle that mutations are rareevents. The methods attempt to minimisethe number of mutations that a phylogenetictree must invoke for all taxa under consid-eration. A tree that invokes minimum num-ber of mutations (changes) is considered tobe the tree of maximum parsimony. Theevolutionary polarity of taxa is decided forconstruction of such trees. The Wagnergroundplan divergence method, anexample of this, was first developed by H. W.Wagner in 1948 as a technique for deter-mining the phylogenetic relationshipsamong organisms that he hoped would

replace intuition with analysis. The methodwas based on determining the apomorphiccharacter-states present within a taxon andthen linking the subtaxa based on relativedegree of apomorphy. Interestingly, whereasthe method found little favour with zoologists,it has been used in many botanical studies.Kluge and Farris (1969) and Farris (1970)developed a comprehensive methodology forthe development of Wagner trees, based onthe principle of parsimony. The method isthe basis of many phylogeny computer algo-rithms currently in use. A given datasetmay, however, yield many possible equallyparsimonious trees due to homoplasy, asmore than one character-state change mayoccur during the evolutionary process of aparticular group of organisms.

The following steps are involved in theanalysis:

1. Determine which of the various char-acters (or character-states) in a seriesof character transformations areapomorphic.

2. Assign the score of 0 to theplesiomorphic character and 1 to theapomorphic character in each trans-formation series. If the transformationseries contains more than two homo-logues, then these ‘intermediateapomorphies’ may be scaled between0 and 1. Thus, a transformation se-ries of three characters may be scoredas 0, 0.5 and 1 (or 0, 1 and 2 depend-ing on the weighting assigned).

3. Construct a table of taxa (EUs) andcoded characters (or character-states:see Table 8.3).

4. Determine the divergence index foreach taxon by totalling up the values.Since apomorphic character-statesare coded 1, the divergence index ineffect represents the number ofapomorphies (character-states) in ataxon, except in cases of weighted cod-ing. For the data matrix in Table 8.4,the divergence index for 15 taxa wouldbe calculated as:


Characters (n)�

Taxa (t)

Hab

it

0-w

oody

1-h

erba

ceou

s

Fru

it 0-

foll

icle

1-a

chen

e

Ova

ry

0-su

peri

or 1

-inf

erio

r

Lea

ves

0-si

mpl

e 1-

lobe

d 2

-com

poun

d

Hab

itat

0-

terr

estr

ial

1-aq

uati

c

Pol

len

1-

trip

orat

e 0

-mon

osul

cate

Ovu

le

1-un

iteg

mic

0-b

iteg

mic

Car

pels

0-

free

1-u

nite

d

Pla

stid

s 1-

PI-

type

0-P

II-t

ype

1. 1 0 1 1 0 1 1 1 1 2. 1 0 1 1 1 1 0 0 1 3. 0 1 0 1 0 0 1 1 1 4. 1 1 1 0 1 0 0 0 0 5. 1 0 1 2 0 1 1 0 1 6. 1 1 0 1 1 1 0 1 0 7. 0 1 1 0 1 1 0 0 1 8. 0 0 0 1 0 0 1 1 1 9. 1 1 1 0 0 1 1 0 0 10. 0 0 0 1 1 0 1 1 1 11. 1 0 1 2 0 1 0 0 1 12. 1 1 0 0 1 1 0 0 1 13. 0 0 1 0 1 0 1 0 1 14. 1 0 1 0 1 0 1 0 1 15. 0 0 0 0 0 0 0 0 0

Table 8.4 Data matrix of t taxa and n characters scored as 0 (plesiomorphic) and 1 (apomorphic)character-states. Multistate character is assigned 0 for ancestral state, 1 for interme-diate and 2 for most advanced state.The matrix is similar to Table 9.1 but only 9 char-acters pictured are used for calculations. Also the last taxon included is the hypotheti-cal ancestor in which all character-states are scored as 0 (plesiomorphic), as it ispresumed that the ancestor would possess all characters in a plesiomorphic state.

Taxon Divergence index1 72 63 54 45 76 67 58 49 510 511 612 513 414 515 0

Note that the hypothetical ances-tral taxon 15 has an index of 0.

5. Plot the taxa on a graph, placing eachtaxon on a concentric semicircle that

equals its divergence index. The linesconnecting the taxa are determinedby shared synapomorphies (see Table8.6). The cladogram (Wagner tree) ispresented in Figure 8.16.

Not all cladistic methods apply the prin-ciple of parsimony. The methods of compat-ibility analysis or clique analysis utilize theconcept of character compatibility. Suchmethods can detect and thus omit ho-moplasy. They can be carried out manuallyor using a computer program, and can gen-erate both rooted as well as unrooted trees.Groups of mutually compatible charactersare termed cliques. Let us consider twocharacters, A and B, with two character-states each. Four character-state combina-tions are possible:

Assuming the evolution has proceededfrom A1 to A2 and from B1 to B2. If all the


T ab le 8 .5 t x t m a trix p resen ting d istan ce be tw een taxa exp ressed a s the n um b er o f ch arac ter- s ta te d iffe rences b etw een p airs o f taxa.

E u s 1 2 3 4 5 6 7 8 9 10 1 1 1 2 1 3 1 4 15 __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ ___ _ 1 0 2 3 0 3 4 7 0 4 7 4 8 0 5 2 3 6 7 0 6 5 5 6 4 7 0 7 6 3 7 3 6 6 0 8 3 6 1 8 5 6 7 0 9 4 5 7 3 4 6 4 7 0 10 4 5 2 7 6 5 6 1 8 0 11 3 2 7 6 1 6 5 6 5 7 0 12 6 3 7 3 6 4 2 7 4 6 5 0 13 5 4 5 4 5 8 3 4 5 3 6 5 0 14 4 3 6 3 4 7 4 5 4 4 5 4 1 0 15 7 6 5 4 7 6 5 4 5 5 6 5 4 5 0 T ab le 8 .6 t x t m a trix p resen ting d istan ce be tw een taxa exp ressed a s n um b er o f d erived (ap o m orp h ic) cha racte r-sta tes co m m on b etw een p airs o f taxa.

E U s 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

__ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __

1 X 2 5 X 3 4 2 X 4 2 3 0 X 5 5 4 2 2 X 6 3 3 2 2 2 X 7 3 4 1 3 3 2 X 8 4 2 4 0 2 2 1 X 9 4 3 1 3 4 2 3 1 X 10 4 3 4 1 2 3 2 4 1 X 11 4 4 1 2 5 2 3 1 3 1 X 12 3 4 1 3 3 3 4 1 3 2 3 X 13 3 3 2 2 3 1 3 2 2 3 2 2 X 14 4 4 2 3 4 2 3 2 3 3 3 3 4 X 15 0 0 0 0 0 0 0 0 0 0 0 0 0 0 X ____________________________________________________


four combinations are met in nature thenobviously there must have been at least onereversal (A2 to A1) or parallelism (A1 to A2occurring twice), and as such A and B areincompatible. On the other hand, if only twoor three of the combinations occur, then Aand B are compatible. Cliques are formed bycomparing all pairs of characters and find-ing mutually compatible sets. The largestclique is selected from the data to produce acladogram. Finally, a rooted tree or networkis obtained according to whether or not a hy-pothetical ancestor was included in theanalysis.

Multiple TreesThe unrooted tree constructed in Figure8.13-B represented only a small portion ofthe evolutionary sequence. Extension of thistree would make it more complicated andpresent a lot of possibilities. Let us add a

small portion of the herbaceous lineage withyellow petals, again assuming that there area total of 15 herbaceous species of which sixare with red petals and 9 with yellow petals.Of these nine 4 are with united carpels and5 with free carpels. The additional Venn dia-

Fig. 8.16 General representation of a Wagnertree.

Figure 8.17 A: The Venn diagram for woody species, the same as Figure 10.12A. B: The Venndiagram for a small portion of the herbaceous lineage of assumed 15 species ofwhich 6 are with red petals and 9 with yellow petals, latter with 4 species havingunited carpels and 5 free carpels. C: Extension of the unrooted tree of Figure 10.12Bto include the species depicted in the Venn diagram B here. There are 5 actualcharacter state changes but with 4 switches as united carpels have arisen twice.

Plants herbaceous

Petals yellow

Carpels united

Plants woody

Stam

ens 2

Stam

ens >2

Carpels united

Carpels free

Plants w

oody

Plants herbaceous

C

Plants woody

Carpels united

Plants herbaceous

Stamens 2

Carpels free

Carpels united

Petals red

Petals yellow

B A

R


gram for the herbaceous species and theextended unrooted tree is presented in theFigure 8.17-C. It must be noted that herewe know the evolutionary history of thegroup—which normally is never known—andthe aim of phylogenetic analysis is to recon-struct and depict this evolutionary historythrough trees. The unrooted tree here hasfive character-state changes (actualchanges, tree length) involved. The changefrom free to united carpels has occurredtwice, and as such there are only four ge-netic switches involved. If we did not havethe knowledge about the evolutionary his-tory of the group, we would try a number ofvariations. One possible variation of theunrooted tree would be to link 4 herbaceousspecies with united carpels to the woody spe-cies with united carpels, thus presenting asingle change of free to united carpels. Butthis brings in further changes. Now, changefrom woody habit has occurred twice, changefrom red to yellow petals has occurred twice,and more significantly the number of actualchanges (tree length) has increased to six(Figure 8.18), with same four geneticswitches involved. With more descendents

being included in the tree, the number ofoptions would increase. Also we have to con-vert each unrooted tree into a rooted treeso that the most primitive basal end of thetree is known, and different lineages pre-sented as the more advanced branches. Thisbrings in many more options, as indicatedearlier. In our example, where we know thehistory of the tree, the tree can be rooted atR, as indicated by an arrow (Figure 8.17-C),but in a large majority of cases, it is a com-plicated process, and a lot of hypotheses,strategies and algorithms come into play.

A number of sophisticated computer algo-rithms are available which compare treesand calculate their lengths. The widely usedones include NONA, PAUP, and PHYLIP.These programs determine the number ofpossible trees, and then sort out the short-est of all these. If we are dealing with threespecies, three rooted trees are possible [A(B, C)], [B (A, C)] and [C (A, B)] (Figure 8.19),if 4 taxa are mapped the 15 trees are pos-sible, 5 then 105, for 10 taxa 34,459,425 treesand so on.

The number of possible rooted trees for nnumber of taxa can be calculated as:

Figure 8.18 Possible variation of the unrooted tree presented n figure 10.14C, if we did not haveany idea about the evolutionary history of the group. Note that tree length hasincreased to six, and habit has changed twice from woody to herbaceous and fromred to yellow petals. Such homoplasious situations are uncommon.

Stamens 2

Stamens >2

Carpels united

Carpels free

Plants woody

Plants herbaceous

Petals red

Petals yellow

Plants woody

Plants herbaceous

Petals red

Petals yellow


Nr = (2n-3)! / [(2 n –2 ) X (n-2)! ]

It can also be calculated as:

Nr = P (2 i - 1)

where P represents the product of all factors(2 i -1) from i = 1 to i = n - 1.

A simpler way to calculate the possiblenumber of rooted trees is as follows:

Nr = (2(n + 1) –5) X number oftrees for (n –1) taxa

As noted above, the number of possiblerooted trees is much more than number ofpossible unrooted trees. Latter can be cal-culated as:

Nu = (2n - 5)! / [(2 n –3 ) X (n-3)! ]

or more simply as:

Nu = number of rooted trees for (n – 1) taxa

Thus, for 3 taxa, 3 rooted trees and 1unrooted trees are possible (Figure 8.19), for4 taxa 15 rooted trees and 3 unrooted treesand for 5 taxa, 105 rooted trees are possiblebut only 15 unrooted trees. For our 8 spe-cies in Figure 8.12, if evolutionary history

was not known we should expect 135135rooted trees and 10395 unrooted trees. Thefigures also highlight the enormous chal-lenges in reconstructing the evolutionaryhistory of any group.

A large number of trees generated aresorted and, ones presenting the shortest evo-lutionary path, in agreement with the prin-ciple of parsimony, are shortlisted.

Distance methodsDistance methods were originally developedfor handling phenetic information and con-struction of phenograms, some of these havenow been incorporated in cladistic method-ology. Cluster analysis is the most commonlyused method of constructing trees.

Cluster analysisData presented in OTUs x OTUs (t x t)matrix are too exhaustive to provide anymeaningful picture and need to be furthercondensed to enable a comparison of units.Cluster analysis is one such method inwhich OTUs are arranged in the in the

A B A B C B A C C B A B

C

AI II III IV

V

A B C D

VI

A C B D A B C D A C B D

VII VIII IX

Figure 8.19 Possible number of rooted and unrooted trees. I: Single rooted tree for two taxa.II-IV: Three possible rooted trees for three taxa. V: One possible unrooted tree forthree taxa. VI-IX: Some of the possible 15 rooted trees for four taxa.


order of decreasing similarity. The earliermethods of cluster analysis were cumber-some and involved shifting of cells with simi-lar values in the matrix so that OTUs withclosely similar similarity values werebrought together as clusters. Today, with theadvancement of computer technology, pro-grams are available which can perform anefficient cluster analysis and help in the con-struction of cluster diagrams or pheno-grams. The various clustering proceduresare classified under two categories.

Agglomerative methodsAgglomerative methods start with t clustersequal to the number of OTUs. These are suc-cessively merged until a single cluster hasfinally been formed. The most commonlyused clustering method in biology is the Se-quential Agglomerative Hierarchic Non-overlapping clustering method (SAHN). Themethod is useful for achieving hierarchicalclassifications. The procedure starts withthe assumption that only those OTUs wouldbe merged which show 100% similarity. Asno two OTUs would show 100% similarity,we start with t number of clusters. Let usnow lower the criterion for merger as 99%similarity; still no OTUs would be merged asin our example the highest similarity re-corded is 96.0%. The best logical solutionwould be to pick up the highest similarityvalue (here 96.0) and merge the two con-cerned OTUs (here 9 and 12). By inference,if our criterion for merger is 96.0 we willhave t-1 clusters. Subsequently the nextlower similarity value is picked up and thenumber of clusters reduced to t-2. The pro-cedure is continued until we are left with asingle cluster at the lowest significant simi-larity value. Since at various steps of clus-tering a candidate OTU for merger wouldcluster with a group of OTUs, it is importantto decide the value that would link the clus-ters horizontally in a cluster diagram. Anumber of strategies are used for thepurpose.

In the commonly used single linkageclustering method (nearest neighbour

technique or minimum method), the can-didate OTU for admission to a cluster hassimilarity to that cluster equal to the simi-larity to the closest member within the clus-ter. The connections between OTUs and clus-ters and between two clusters are estab-lished by single links between pairs of OTUs.This procedure frequently leads to long strag-gly clusters in comparison with other SAHNcluster methods. The phenogram for our datausing this strategy is shown in Figure 8.20.

The highest similarity value in our ma-trix (see Table 8.2) is 96.0 between OTUs 9and 12, and as such they are linked at thatlevel. The next similarity value of 94.0 isbetween OTUs 10 and 12, but since 12 hasalready been clustered with 9, 10 will jointhis cluster linked at 94.0. The process isrepeated till all OTUs have been agglomer-ated into single cluster at similarity valueof 53.0.

In the complete linkage clusteringmethod (farthest neighbour or maximummethod) the candidate OTU for admissionto a cluster has similarity to that clusterequal to its similarity to the farthest mem-ber within the cluster. This method will gen-erally lead to tight discrete clusters that joinothers only with difficulty and at relativelylow overall similarity values.

In the average linkage clusteringmethod, an average of similarity is calcu-lated between a candidate OTU and a clus-ter or between two clusters. Several varia-tions of this average method are used. Theunweighted pair-group method using arith-metic averages (UPGMA) computes the av-erage similarity or dissimilarity of a candi-date OTU to a cluster, weighting each OTUin the cluster equally, regardless of its struc-tural subdivision. The method originallydeveloped for the procedures of numericaltaxonomy has been applied in phylogeneticanalysis with relevant modifications, andused for the construction of trees. UPGMAmethod procedure begins with as manyclusters as the number of taxa. The two taxawith minimum distance merged to reducethe number of clusters by one. In the next


step average distance between new clusterand remaining taxa are determined by tak-ing the average distance between these twomembers and all other remaining taxa,weighting each taxon in the cluster equallyregardless of its structural subdivision, andmerging the taxon with smallest distanceto the first cluster. The process is repeatedwith this new cluster of three taxa, and theprocedure continues till all the taxa aremerged, the most distant taxon joining lastof all. From measure of similarity or dissimi-larity of taxa (OEUs) as presented in Table8.5 and 8.6, a network presenting minimumdissimilarity is constructed. Analysis of datafrom first six taxa of table 8.4 is presentedin Figure 8.21. The procedure begins by unit-ing nearest taxa A and E (with minimumdistance of 2). Next matrix in now con-structed in which distance between (AE) andrest of the taxa is recalculated. The lowestvalue in this matrix (step 1 matrix) is be-tween (AE) and E, which are next united atdistance level 3 into (AE)B. The distance be-tween this cluster and rest of the taxa is

now recalculated as presented in step 2matrix. The lowest distance in this matrixis 4 between D and F which are united intoone cluster. With this merger the distancebetween taxa/clusters is recalculated andpresented in step 3 matrix. The lowestdistance 5.5 is now between clusters (AE)Band DF, which are next united. Finally thedistance between this enlarged cluster andC is recalculated as presented in step 4matrix. Finally the two clusters (((AE)B)(DF))and C are united at distance of 6.5 to formfinal cluster ((((AE)B)(DF))C). The resultingphenogram and reconstructed phylogenetictree constructed from the analysis arepresented in Figure 8.21.

The distance matrix can similarly be gen-erated from single nucleotide differencesbetween homologous DNA sequencesderived from different species. Six such hy-pothetical sequences from different speciesare presented in Figure 8.22. The distancematrix is based on number nucleotide dif-ferences between different sequences. A andF with lowest distance are merged, followed

Figure 8.20 Cluster diagram of 15 OTUs based on similarity matrix in Table 8.2 using singlelinkage strategy.


Taxa A B C D E F

A 0

B 3 0

C 4 7 0

D 7 4 8 0

E 2 3 6 7 0

F 5 5 6 4 7 0

Taxa AE B C D F

AE 0

B 3 0

C 5 7 0

D 7 4 8 0

F 6 5 6 4 0

Taxa (AE)B C D F

(AE)B 0

C 6 0

D 5.5 8 0

F 5.5 6 4 0

Taxa (AE)B FD C

(AE)B 0

FD 5.5 0

C 6 7 0

Taxadistance

7 6

5 4

3 2 1

0

A E B D CF

A E B D F C

Distance MatrixStep 1 matrix

Step 2 matrix

Step 3 matrix

PhenogramPhylogenetic tree

Taxa ((AE)B)(FD) C

((AE)B)(FD) 0

C 6.5 0

Figure 8.21 Construction of phenogram and phylogenetic tree (cladogram) based on distancematrix concerning first six taxa in Table 8.4 using UPGMA clustering method. Thetaxa with minimum distance are united and treated as single cluster in next matrix,and distance values recalculated as average of distance from either of united taxa.The procedure repeated till all taxa are united. The phylogenetic tree is constructedbased on sequence of clustering of taxa.

by recalculation of new matrix in which Aand F form one cluster and value of eachtaxon is calculated as average distance fromA and F. Now lowest value is shared by B andD which form second cluster. The values arerecalculated similarly , and successively Cjoins AF cluster, and then E joins (AF)Ccluster. The two clusters are finally mergedto enable construction of phylogenetic treeeither as phenogram or as cladogram. Sometypes of genetic polymorphism data such asRAPD are best handled when sharing of 0code by two taxa in the matrix is ignoredwhen both taxa lack a given polymorphicband in gel electrophoresis. Jaccard coeffi-cient is best suited for handing such data.

Figure 8.23 presents results of RAPD analy-sis of 8 taxa, where only polymorphic bandsare shown, monomorphic bands being omit-ted. The distance matrix based on similar-ity matrix was processed using PHYLIP, asshown in Figure 8.24. Distance methods aresuitable for handling both morphological andmolecular data, or a combination of both.These methods use all data with usuallyequal importance, whereas the parsimonymethods use only informative moleculardata. In general for a site to be informative,when handling sequence data, irrespectiveof how many sequences are aligned, it hasto have at least two different nucleotides,and each of these nucleotides has to be


10 20 30 40 50

TaxaA 0 B 13 0C 11 17 0D 13 10 24 0E 13 12 12 18 0F 9 16 12 12 18 0

A B C D E F Taxa AF B C D

B 14.5 0 C 11.5 17 0D 12.5 10 24 0E 15.5 12 12 18 0

E Taxa AF C EBD

E 15.5 15 12 0

BD 13.5 0

Taxa (AF)C E BD

E 13.5 15 0

AF 0 AF 0

C 11.5 20.5 0

(AF)C 0BD 16.75 0

Taxa ((AF)C)E BD

BD 15.875 0 ((AF)C)E 0

A F C E B D

Distance matrix Matrix step 1

Matrix step 2

Matrix step 3

Final Matrix Cladogram

A GCCAACGTCG ATGCCACGTT GTTTAGCACC GGTTCTTGTC CGATCACAGA TGT

B GCCAACAATG ATACCACGCC GTCCAGCACC GATTCTCGTC CGAGTACCGA TGT

C GGTAACGTCA ATGGGACGTT GTCCAGCACC GGTTCATGTC CAAGCAGAGA TGT

D GCCAACATTG ATACCACGCC GTTTAGCTGC GACACTCGTC CGATCACCAA TGT

E GCTAACGACA ATACCAGGCT GTCCAGCTCC GGTTTACGTC CGAGCACAGA TGT

F GCCAACATCG ATGGGACGTT GTTTAGCTCC GATTCATGTC CAATCACCAA TGT

0

5

10

15

20

A F C E B D

Phenogram

((((A,F:9),C:11.5),E:13.5),:15.875 (B,D:10));

Figure 8.22 Construction of phylogenetic tree based on single nucleotide differences in hypo-thetical DNA sequences of six species. Distance matrix is constructed based onthe number of nucleotide differences between each pair of DNA sequences andpresented in distance matrix. Further analysis proceeds as detailed in Figure 8.21,and also in the text on these pages.

present at least twice. Thus in the sequencedata presented in Figure 8.22, out of 28 sitesshowing nucleotide differences in six se-quences, there are only 12 informative siteswhich can be used in parsimony analysis.Procedures based on UPGMA method, how-ever, don’t account for different rates of evo-

lution occurring in different lineages. Somedistance methods such as transformed dis-tance method and neighbour-joiningmethod, although more complex are capableof incorporating different rates of evolutionwithin the lineages.


R A B C D E F G H

12

3

45

6

7

89

1011

12

13

11111100

10101011

00000101

10110111

01001011

10011101

01110110

11101010

10101011

01010110

10110101

01101110

11010011

A B C D E F G H

1

2

3

4

5

6

7

8

9

10

11

12

13

1

1

1

0

1

0

1

0

1

0

1

0

1

0

1

1

0

1

1

1

0

1

1

0

1

0

0

1

1

1

0

1

0

1

0

0

1

0

1

1

11010101

10001011

11010110

10011101

11000010

10111110

01011 011

01110101A

B

C

D

E

F

G

H

1 2 3 4 5 6 7 8 9 10 11 12 13

II

I

III

1.00.250.600.500.420.330.420.60

1.00.360.400.400.360.600.15

1.00.360.500.450.540.33

1.00.160.660.330.36

1.00.250.450.36

1.00.300.33

1.00.46

1.0A

B

C

D

E

F

G

H

A B C D E F G H

IV

Figure 8.23 Phylogenetic analysis of data concerning polymorphic bands from gel electrophore-sis from DNA of 8 taxa. I: Polymorphic bands of DNA from 8 taxa (A-H), R represent-ing reference bands; II: Binary coded matrix of the polymorphic bands; III: The samematrix presented in conventional format; IV: Lower triangular matrix of similaritymatrix using Jaccard coefficient, wherein sharing of 0 state (absence of bands) isignored. Further handling of data using UPGMA program of PHYLIP is presented inFigure 8.24.

Divisive methodsDivisive methods as opposed to agglo-merative methods, start with all t OTUs asa single set, subdividing this into one ormore subsets; this is continued untilfurther subdivision is not necessary. Thecommonly used divisive method is associa-

tion analysis (William, Lambert and Lance,1966). The method has been mostly used inecological data employing two state charac-ters. It builds a dendrogram from the topdownwards as opposed to cluster analysis,which builds a diagram from the bottom up.The first step in the analysis involves


8

Taxona 0.00 0.54 0.67 0.64 0.64 0.67 0.85 0.40

Taxonb 0.54 0.00 0.70 0.55 0.67 0.46 0.40 0.58

Taxonc 0.67 0.70 0.00 0.75 0.34 0.55 0.64 0.67

Taxond 0.64 0.55 0.75 0.00 0.84 0.50 0.60 0.58

Taxone 0.64 0.67 0.34 0.84 0.00 0.64 0.60 0.50

Taxonf 0.67 0.46 0.55 0.50 0.64 0.00 0.64 0.40

Taxong 0.85 0.40 0.64 0.60 0.60 0.64 0.00 0.75

Taxonh 0.40 0.58 0.67 0.58 0.50 0.40 0.75 0.00

(((Taxona:0.20000,Taxonh:0.20000):0.11000,(Taxonc:0.17000,

Taxone:0.17000):0.14000):0.01500,((Taxonb:0.20000,Taxong:0.20000):0.08125,

(Taxond:0.25000,Taxonf:0.25000):0.03125):0.04375);

I

II

IVIII

Figure 8.24 Construction of phylogenetic tree based on polymorphic bands from gel electrophoresisfrom DNA of 8 taxa using UPGMA program of PHYLIP; I: Square distance generatedfrom Figure 8.23-IV, each value calculated 1-similarity value. II: Outtree file gener-ated by UPGMA option of NEIGHBOUR program; III: Upright square tree (Phenogram)plotted through DRAWGRAM program; IV: Cladogram, but with branch lengthsomitted.

calculating chi square value between everypair of characters using the formula:

22 (ad – bc)

[(a + b)(a + c)(b + d)(c + d)]hin

X �

where i stand for the character being com-pared and h for any character other than i.

For each character the sum of chi-square iscomputed and the character showing maxi-mum chi square value is chosen as the firstdifferentiating character. The whole set ofOTUs is divided into two clusters, one con-taining the OTUs which show the charac-ter-state a and another containing OTUswhich show the character-state b. Within


each cluster, again, the character with thenext value of the sum of chi square is se-lected and the cluster subdivided into twoclusters as before. The process is repeatedtill further subdivision is not significant.

Hierarchical classificationsThe phenogram constructed using any tech-nique or strategy can be used for attempt-ing hierarchical classification, by decidingabout certain threshold levels for differentranks. One may tentatively decide 85 percent similarity as the threshold for the spe-cies, 65 for genera and 45 for families andrecognize these ranks on the basis of num-ber of clusters established at that thresh-old. Whereas such an assumption can helpin hierarchical classification, the point ofconflict would always be the threshold levelfor a particular rank. Some may argue—andare justified in doing so—to suggest 80 percent (or any other value) as the thresholdfor species. It is more common, therefore,to use terms 85 per cent phenon line, 65per cent phenon line, and 45 per centphenon lines. These terms may conve-niently be used till such time that sufficientdata are available to assign them formal taxo-nomic ranks to the various phenon lines.

The results of cluster analysis are com-monly presented as dendrograms known asphenograms. They can also be presented ascontour diagrams (Figure 8.25), originallydeveloped under the name Wroclaw dia-gram by Polish phytosociologists. The con-tour diagram may also incorporate thelevels at which clustering has taken place.

OrdinationOrdination is a technique which determinesthe placement of OTUs in two-dimensionalor three-dimensional space. The results oftwo-dimensional ordination are conve-niently represented with the help of a scat-ter diagram and those of three-dimensionalordination with the help of a three-dimen-sional model. The procedure works on dis-tance values calculated directly from thecoded data or indirectly from the already cal-

culated similarity values as 100 minus simi-larity (if similarity values are in percent-age) or 1 minus similarity (if similarity val-ues range between 0 to 1). A dissimilaritymatrix based on Table 8.2 is presented inTable 8.3.

The first step in the ordination starts withconstruction of the x-axis (horizontal axis).In the commonly used method of polar ordi-nation, the two most distant OTUs are se-lected as the end points (A and B) on x-axis.In our example, these are OTU 8 and 7 witha distance (dissimilarity value) of 75. Theposition of all other OTUs on this axis canbe plotted one by one. OTU 10 has a distanceof 64 from A (OTU 8) and a distance of 23from B (OTU 7). A compass with a radius of64 units is swung from A and a compass witha radius of 23 units is swung from B, form-ing two arcs. A line joining the intersectionof two arcs forms a perpendicular on the x-axis, and the point at which the line crossesthe x-axis is the position of the OTU. Thedistance between the x-axis and the point ofintersection of arcs is the poorness of fit ofthe concerned OTU. The location of OTU onthe axis from the left (point A) can also becalculated directly instead of plotting:

x = 2 2 2L AC – BC

2L

d d�

where x is the distance from the left end, Lis the dissimilarity value between A and B(length of x-axis), dAC is dissimilarity be-tween A and the OTU under considerationand dBC as the dissimilarity between B andthe OTU under consideration. The poornessof fit (e) of this OTU can be calculated as:

2 2AC –e d x�

After the position of all OTUs has been de-termined and the poorness of fit calculated,a second axis (vertical axis or y-axis) has tobe calculated. For this, the OTU with thehighest poorness of fit (most poorly fitted tox-axis) is selected and this forms the firstreference OTU of y-axis. The second refer-ence OTU is selected as that one with the


highest dissimilarity to the first referenceOTU of y-axis, but within 10 per cent (of thelength of x-axis) distance on x-axis. The po-sition of all other OTUs on the y-axis andtheir poorness of fit is determined as ear-lier. By using the values of poorness of fit toy-axis, a z-axis can be similarly generatedand the position of all OTUs on z-axis deter-mined similarly. The values can be used for

constructing a scatter diagram or a three-dimensional model.

A commonly used ordination techniqueknown as principal component analysis alsocalculates values for a two-dimensional scat-ter diagram. In this method, however, the val-ues on the horizontal as well as the verticalaxis are non-zero, ranging from -1 to 1 (calcu-lated as eigenvalues) and as such the

Figure 8.25 Contour diagram based on the phenogram shown alongside.


scatter diagram is presented along four axes:positive horizontal, negative horizontal, posi-tive vertical and negative vertical (Fig 8.26).The technique is based on the assumptionthat if a straight line represented a singlecharacter, all the OTUs could be placed alongthe line according to their value for that char-acter. If two characters were used, a two-di-mensional graph would suffice to locate allOTUs. With n characters, n-dimensionalspace is required to locate all OTUs as pointsin space.

Principal component analysis determinesthe line through the cloud of points that ac-counts for the greatest amount of variation.This is the first principal component axis. Asecond axis, produced perpendicular to thefirst, accounts for the next greatest amountof variation. The procedure ultimately pro-duces axes one less than the number ofOTUs. The first two axes are generally plot-ted to produce a scatter diagram. The proce-dure also calculates eigenvectors, whichindicate the importance of a character to aparticular axis. The larger the eigenvectorin absolute value, the more important isthat particular character.

A related method of ordination is princi-pal co-ordinate analysis developed by Gower(1966). This technique enables computationof principal components of any Euclideandistance matrix without being in possessionof original data matrix. The method is alsoapplicable to non-Euclidean distance and as-sociation coefficients as long as the matrixhas no large negative eigenvalues. Princi-pal co-ordinate analysis also seems to be lessdisturbed by NC entries than principal com-ponents.

Maximum Likelihood methodThe method is similar to distant method inthat all data is taken into consideration. Inthis method, similarly character-state trans-formations are compared, and the probabil-ity of changes determined. These probabili-ties are used to calculate the likelihood thata given tree would lead to the particular dataset observed, and the tree with maximum

likelihood is selected. The method is espe-cially suited to molecular data, where theprobability of genetic changes can be mod-eled more easily. With this approach, theprobabilities are considered for every indi-vidual nucleotide substitution in a set ofsequence alignments. It is commonly un-derstood that transitions occur three timesmore frequently as compared totransversions. Thus if C, T, and A occur inone column (representing one site), the se-quences with C and T (pyrimidines) are morelikely to be closely related than sequencewith A (Purine). Using objective criteriaprobability for each site and every possibletree that describes the relationship of se-quences. The tree with highest aggregateprobability is selected as representation ofa true phylogenetic tree.

Using any one of the methods, a largedataset commonly used, and which includesmany homoplasies, large number of short-est trees may be generated by these auto-mated algorithms. These short listed treeshave to be further compared.

The Consensus TreeThe use of automated methods based on par-simony, even after applying relevant strate-

Figure 8.26 Plot of the results of the princi-pal component analysis of 18hypothetical taxa.


gies yield several trees, all presenting short-est pathways, based on parsimony but withdifferent linkages among the taxa (OEUs),and often presenting different evolutionaryhistory. Molecular studies of Clusiaceae byGustafsson et al., (2002) for example, includ-ing 1409 nucleotides of chloroplast gene rbcLpositions using PAUP*4.0b8a parsimonyanalysis method, yielded 8473 most parsi-monious trees for the 26 species compared.Interestingly, the number of trees generatedwas so large that search for trees 3 stepslonger than most parsimonious trees wasaborted. More significantly different datasets (molecular, morphology) may yield dif-ferent trees. While selecting the consensustree, the commonest approach is to identifythe groups, which are found in all the shortlisted trees, and build a consensus tree. Thiscould be achieved in different ways.

Strict consensus treeA more conservative approach in building aconsensus tree involves including onlymonophyletic groups that are common to allthe trees. The tree developed this way isknown as strict consensus tree. Considerthe two most parsimonious trees (althoughthere could often be numerous trees of sameshortest length available for comparison) asshown in Figure 8.27-I and 8.27-II.

Imagine that all groups A to J are mono-phyletic. Tree I shows that A and B are veryclosely related, and so are H and I. C, D, E,and F are shown arising successively andare related in that sequence. Tree II showsa similar relationship between H and I, andbetween A and B (but group J is shown re-lated to these two). The tree also shows thatC and D are closely related. As relationshipsbetween E, F, and G are ambiguous, they areshown arising from the same point in evo-lutionary history. The consensus tree IIIwould thus omit taxon J (which is absentfrom tree I), show A and B, as also H and I asin the two trees I and II. The other taxa C,D, E, F, and G are shown arising from thesame point.

Majority-rule consensus treeMajority-rule consensus tree shows all thegroups which appear in a majority of trees,say, more than 50 per cent of the trees. It isuseful to indicate for each group on the con-sensus tree the percentage number of themost parsimonious trees in which the groupappeared. Such a consensus tree, however,provides a partial summary of the phyloge-netic analyses, and may be inconsistent withthe trees from which it is derived.

Semi-strict consensus treeA semi-strict consensus tree is useful whencomparing trees from different data sets, orwith different terminal taxa. The consensustree developed indicates all the relationshipssupported by both type of trees or any one ofthese, but not contradicted by any. Thus, inFigure 8.27, tree II does not give us any in-formation about the time of origin of E, F andG, the tree I indicates that they originatedsuccessively. Similarly, tree I does not indi-cate any close relationship between C andD, whereas the tree II does. The semi-strictconsensus tree IV as such presents such in-formation, not contradicted by either tree.

Evaluating consensus treeDeveloping a consensus tree involves theuse of intuition, making guesses and devel-oping hypothesis. A number of evaluationstrategies are used to test the soundness ofthe tree and measuring support for eitherthe tree as a whole or for its individualbranches. These values are generally pub-lished along with the tree, to allow the fairassessment of the final results for compari-son of trees based on different datasets.

Consistency IndexThe principle of Parsimony is based on a ba-sic rule of science known as Ockham’s ra-zor, which says ‘do not generate a hypoth-esis any more complex than is demandedby the data’. Some information in the datamay be representing homoplasy (reversals,


parallelisms). Dollo parsimony (as indicatedabove) minimizes the use of homoplasiouscharacters. The commonest measure of ho-moplasy is the Consistency Index (CI),which is calculated by dividing the numberof genetic switches by actual geneticchanges on the tree.

Consistency Index CI = Min /L

Min stands for the minimum possible treelength or genetic switches, and L for the ac-tual tree length or actual number of genetic

changes. In the tree shown in Figure 8.13B,there are three character-state changes,each involving one switch, and, as such, theconsistency index would measure 3/3 = 1.The tree shown in Figure 8.17C has fiveactual character-state changes (tree lengthis 5), but it involves only four geneticswitches. As carpel fusion has occurredtwice, the consistency index would accord-ingly be 4/5 = 0.8. In the tree shown inFigure 8.18, the number of genetic switchesremains the same as four but the treelength has increased to six due to two

Figure 8.27 Two most parsimonious trees for a particular group of organisms, with monophyletictaxa A to J. I: showing C, D, E and F arising successively. II: E, F, G are shownarising at the same time from a common point, C and D being closely related.III: Strict consensus tree of trees I and II. IV: Semi-strict consensus tree of treesI and II.

A B C D E F G H I A B J C D E F G H I

A B C D E F G H I A B C D E F G H I

I II

III IV


parallel (or convergent) evolutions; the CIwould be calculated as 4/6 = 0.66.

Consistency Index may also be calculatedfor individual characters. In Figure 8.13B assuch CI for all characters is one, while in8.17-C, it is 0.5 for carpel fusion (minimumnumber of changes possible—one for binarycharacter divided by actual number ofchanges—here 2 since the character haschanged twice) and 1 for rest. In Figure 8.18,CI is 0.5 for habit and petal colour, and 1 forstamen number and carpel fusion. The char-acters that lower the CI of a tree (or whichhave lower CI) are considered to behomoplasious. The inclusion of a largernumber of homoplasious characters in theanalysis lowers CI for the tree and contra-dicts phylogeny. There may also be a char-acter, which changes only in one (or a veryfew) species, and may be of no relevance inothers. Suppose one species develops spinyfruits. The length or number of spines wouldnot be of any relevance in rest of the spe-cies without spines. Such a situation (asingle species having a particular charac-ter) is known as autapomorphy. Since sucha character has changed only once, it givesCI of 1, and as such the inclusion of manysuch characters would increase the consis-tency index of the tree, and provide falsesupport. Such uninformative characters areas such omitted before calculating CI.

The Consistency index values are oftendependent on the number of taxa analyzed.Any increase in number of taxa lowers CIvalues, and this is true for data from differ-ent sources, morphological or molecular.

Retention IndexAlthough theoretically the value of CI couldrange between 0 and 1, it rarely goes below0.5. For a character that, has changed fivetimes on a tree (this is a remote possibil-ity), CI will be 0.2. More so, the value of CIfor a tree, very rarely may go below 0.5, andthe values thus range between 0.5 and 1.The Retention Index (RI) corrects this nar-row range of CI by comparing maximum (andnot minimum as in CI) possible number of

changes in the character with actual num-ber of changes in the character. RI is com-puted by first calculating the maximum pos-sible tree length, if the apomorphic charac-ter-state originated independently in everytaxon that it appears in, or say, the taxa areunrelated for the said character-state. Thevalue of RI is calculated as:

Retention Index RI = (Max – L)/(Max – Min)

Max stands for the maximum tree lengthpossible, L the actual tree length and Minthe minimum tree length possible. The treein Figure 8.17-C thus has a maximum pos-sible tree length of 9 (minimum length of 4and actual length of 5 as we already know)and the RI would be (9-5)/(9-4) = 0.8. Higherthe RI, sounder is the tree.

Bremer Support (Decay Index)The principle of parsimony, followed in phy-logenetic analyses, aims at selecting theshortest tree. Some parts of the tree maybeore reliable than others. This is commonlyevaluated by comparing the shortest treewith those one or more steps longer. Decayindex or Bremer Support is the measure ofhow many extra steps are needed before theoriginal clade (group) is not retained. Thusif an internode has decay index of 3, thenthe clade (monophyletic group) arising fromit is maintained even in the cladogram 3steps longer than the shortest tree (see Fig-ure 8.29). Certain branches of the tree whichappear in the shortest tree, but disappear,or ‘collapse’ in the tree one step longer, arenot drawn in the strict consensus tree.Greater the decay index value, more robustis that internode of the cladogram.

Branches of the tree may also be testedby comparing the number of genetic changesleading up to a particular group, and theCI of individual characters involved. Doyleet al., (1994) on the basis of morphologicaldata, developed a tree having 18 characterchanges leading to angiosperms. Of these18 characters 11 had CI of 1, thus


supporting the view that angiosperms forma unique group of plants.

Bootstrap AnalysisAny realistic analysis requires that the dataused is randomized. Many techniques areavailable for randomizing the data. Bootstrapanalysis is the commonly used method de-veloped by Bradley Efron (1979). Its use inphylogeny estimation was introduced byFelsenstein (1985). Matrix in the Figure8.28-A contains information on the basis ofwhich the unrooted tree in Figure 8.17-C isconstructed. Without touching the rows, anycolumn is chosen at random to become thefirst column; similarly any other as secondand the process is repeated till the numberof columns in the new matrix is the same asin the original matrix. As the columns arepicked up from the original matrix, the newmatrix may contain some characters repre-sented several times (the same column mayhave been picked up at random more thanonce), while others may have been omitted(the columns were not picked up at all). Themethod is known as random sampling withreplacement. The resultant matrix B showsthat character carpel fusion was picked uptwice, whereas the random selection processmissed the stamen number.

Repeating the method of random selection,multiple such matrices (usually more than100) are constructed, and for each matrixthe most parsimonious tree/trees found.The consensus tree is developed from thesemost parsimonious trees. In this consensustree, the percentage number of trees (gen-erated by bootstrap analysis) that containthat clade is indicated as bootstrap supportvalue of that clade. Bootstrap analysis basedon the assumption that differential weight-ing by resampling of the original data willtend to produce same clades if the data aregood, and reflect actual phylogeny and verylittle of homoplasy. A bootstrap value of 70per cent or more is generally considered asgood support to the clade.

Several variations of bootstrap analysisare available. The partial bootstrapping in-

volves sampling fewer than the full numberof characters. The user is asked for the frac-tion of characters to be sampled. Block-bootstrapping is useful for handling corre-lated characters. When this is thought tohave occurred, we can correct for it by sam-pling, not individual characters, but blocksof adjacent characters. Block bootstrap andwas introduced by Künsch (1989). If the cor-relations are believed to extend over somenumber of characters, you choose a blocksize, B, that is larger than this, and chooseN/B blocks of size B. In its implementationhere the block bootstrap “wraps around” atthe end of the characters (so that if a blockstarts in the last B-1 characters, it contin-ues by wrapping around to the first charac-ter after it reaches the last character). Notealso that if you have a DNA sequence dataset of an exon of a coding region, you canensure that equal numbers of first, second,and third coding positions are sampled byusing the block bootstrap with B = 3. Partialblock-bootstrapping is similar to partialbootstrapping except sampling blocks ratherthan single characters.

Jackknife analysis (Jackknifing) issimilar to bootstrap analysis but differs inthat each randomly selected character maybe resampled only once, and not multipletimes, and the resultant resampled datamatrix is smaller than the original. Delete-half-jackknifing involves sampling a ran-dom half of the characters, and includingthem in the data but dropping the others.The resulting data sets are half the size ofthe original, and no characters are dupli-cated. The random variation from doing thisshould be very similar to that obtained fromthe bootstrap. The method is advocated byWu (1986). Delete-fraction jackknifing wasadvocated by Farris et. al. (1996) and involvesdeleting a fraction 1/e (1/2.71828). Thisretains too many characters and will lead tooverconfidence in the resulting groups whenthere are conflicting characters. This andthe preceding options form a part of theSEQBOOT program of Phylip software, andthe user is asked to supply the fraction of


Figure 8.28 A: Matrix based on the tree 9:16A. B: One possible matrix after procedure of randomsampling with replacement.

Plants

Plants

Plants

Plants

Habit

Yellow> 2FreeHerbaceous

Yellow> 2UnitedHerbaceous

2

> 2

> 2

> 2

RedUnitedWoody

RedUnitedWoody

RedFreeWoody

RedFreeHerbaceous

Carpels

Stamens

Plants

Plants

Petals

Plants

Plants

Plants

Plants

Habit

United

United

Free

Free

Free

United

Red

Red

Red

Red

Yellow

Yellow

FreeHerbaceous

UnitedHerbaceous

UnitedWoody

UnitedWoody

FreeWoody

FreeHerbaceousC

arpels

Plants

Plants

Petals

Carpels

A B

characters that are to be retained. The pro-gram also offers permuting method, withfollowing alternatives. Permuting specieswithin characters involves permuting thecolumns of the data matrix separately. Thisproduces data matrices that have the samenumber and kinds of characters but no taxo-nomic structure. It is used for different pur-poses than the bootstrap, as it tests not thevariation around an estimated tree but thehypothesis that there is no taxonomic struc-ture in the data: if a statistic such as num-ber of steps is significantly smaller in theactual data than it is in replicates that arepermuted, then we can argue that there issome taxonomic structure in the data(though perhaps it might be just the pres-ence of a pair of sibling species). Permutingcharacters simply permutes the order of thecharacters, the same reordering being ap-plied to all species. It is included as a pos-sible step in carrying out a permutation testof homogeneity of characters (such as theIncongruence Length Difference test). Per-muting characters separately for each spe-cies permute data so as to destroy all phylo-genetic structure, while keeping the basecomposition of each species the same as be-

fore. It shuffles the character order sepa-rately for each species.

It is a common practice, and conse-quently more informative, to indicate thebranch length (number of steps needed toreach that clade), bootstrap or jackknife sup-port and Bremer support (decay index) foreach clade in the consensus tree (Figure8.29).

Effect of Different OutgroupsAn important component of procedures gen-erating rooted trees is the incorporation ofan outgroup in the analysis. In morphologi-cal data, the outgroup choice can influencephylogenetic inference. In molecular data,one specific concern is the levels of se-quence divergence between outgroups andingroups and the subsequent possibility ofspurious long-branch attraction (Albert et al.,1994). The robustness of tree can be testedby using randomly-generated outgroup se-quences, excluding all outgroups, and usingoutgroups selectively. Sytsma and Baum(1996), investigating the molecular phylog-enies of angiosperms, found that removal ofall outgroups generated 27 shortest unrootedtrees. Using Ginkgo only as outgroup yielded


Figure 8.29 Tree developed from the study of 16 species of paleoherbs and 2 outgroup taxa, using58 morphological and ontogenetic characters. The cladogram requires 214 steps andhas CI = 0.51 and RI = 0.65. Bootstrap values are underlined and indicated below abranch. Decay Index is indicated above the branch. Ranunculus repens and Aquilegiaformosa were chosen as outgroup taxa. (Drawn from Tucker and Douglas, 1996).

Illicium floridanum

Magnolia denudata

Chloranthus spicatus

Saruma henryii

Lactoris fernadeziana

Cabomba caroliniana

Scilla violacea

Lilium tigrinum

Acorus calamus

Anemopsis californica

Houttuynia cordata

Gymnotheca chinensis

Saururus cernuus

Peperomia metallica

Piper marginata

Piper amalago

Aquilegia formosa

Ranunculus repens

74+1

28

+176

94+3

80 *

60-G +1

90+4

29**

28**

-G

53+3

16 ***-G

96>5

-A 22+181

+2


lineages identical with baseline study (whichincluded all outgroups); when only coniferswere used as outgroup, the consensus treewas less resolved and many nodes collapsed.Use of Gnetales us outgroup increased thenumber of steps needed to yield baseline to-pologies, and interestingly, Ceratophyllum isshown as sister to all angiosperms excepteudicots.

Effect of Lineage RemovalLineage removal strategy highlights theproblems of lineage extinction, which oftenleads to a particular group (especially criti-cal in angiosperms where fossil record ismeager) not being sampled in analysis, thusgiving distorted phylogenies. The same mayalso be true for extant taxa, for which verylittle data is available. The removal of all ma-jor lineages, one at a time (Sytsma andBaum, 1994), provided useful information.The removal of Ceratophyllum, paleoherbs IIb(Chloranthaceae, and Magnoliales) had noeffect on the remaining angiosperm topol-ogy, whereas the removal of paleoherbs I(Aristolochiales and Illiciales), Laurales andeudicots showed substantial changes.

Effect of ExemplarsThe large computational load in handling alarge data is often reduced by using place-holders or exemplars. These are often usedto represent large lineages. The use of ex-emplars can warn about the possible arti-facts when sparsely-sampled lineages ap-pear in basal positions. In such cases, moretaxa can be added to the data set for furtheranalyses. But in the case of basal cladewhere a large number of taxa are extinct,the results could be ambiguous. The resultsfrom angiosperms have shown that cladesshift around with ease when the number oftaxa sampled for each lineage is reduced,and the use of exemplars at times could givemisleading results.

Automated TreesA number sophisticated computer programsare available to construct phylogenetic trees.

These programs are basically similar tothose designed for development ofphenograms, but differing essentially in therequirement to select one taxon for rootingin most programs. PHYLIP (Phylogeny In-ference package), is a commonly used set ofprograms for inferring phylogenies (evolu-tionary trees) by parsimony, compatibility,distance matrix methods, and likelihood. Itcan also compute consensus trees, computedistances between trees, draw trees,resample data sets by bootstrapping orjacknifing, edit trees, and compute distancematrices. It can handle data that are nucle-otide sequences, protein sequences, genefrequencies, restriction sites, restrictionfragments, distances, discrete characters,and continuous characters. Distance matrixcan be generated using programs such asDNADIST (which handles nucleotide se-quence data; it gives you choice to setweightage for transversions/transitions),PRODIST (which works with protein se-quences) and RESTDIST (which works withrestriction site data). The most commonlyused programs of PHYLIP for handling dis-tance matrix data include FITCH, KITSCH,and NEIGHBOR These deal with data whichcomes in the form of a matrix of pairwisedistances between all pairs of taxa NEIGH-BOR offers UPGMA option in which no taxonneeds to be selected for rooting, whereasneighbor-joining option of this program, aswell as FITCH and KITSCH need one taxonto be selected for rooting, otherwise by de-fault first taxon is used for rooting. Theouttree generated by these programs can beplotted using DRAWGRAM or DRAWTREE,latter plotting only unrooted trees.DRAWGRAM provides a variety of options tochoose from. The trees can be drawn hori-zontal or vertical, branches square(phenogram), v-shaped (cladogram), curvedor circular. The branch lengths may be de-picted (phylogram) on the tree. The DNA se-quence data presented in Figure 8.22 wasanalysed using PHYLIP programs. Outputsare presented in Figure 8.30. DNA sequencedata can also be handled by DNAPARS pro-gram which performs Parsimony analysis


Figure 8.30 Analysis of the DNA sequence data presented in Figure 8.22 using PHYLIP. I: Infile,first line indicating number of taxa and number of nucleotides in each sequence;II: Square distance matrix (outfile) generated by DNADIST program; III: Outtree filegenerated by NEIGHBOR program using UPGMA option; IV: DNA sequence of the7th hypothetical taxon (taxono) used for rooting; V-VI: Square tree (Phenogram) andV-shaped tree (cladogram); VII: Unrooted tree of same; VIII-XVI: Diagrams based on7-taxa sequences; VIII: Phenogram, UPGMA option; IX-XI: Phylogram, Phenogramand Cladogram based on neighbour-joining option of NEIGHBOR; XII: Phenogrambased on DNAML program; XIII: Phenogram based on FITCH program; XIV: Phenogrambased on KITSCH program; XV: Tree (Phenogram) generated based on DNAPARSprogram; XVI: Majority-rule consensus tree based on CONSENSE program, usingouttree files of above six programs. (All trees except VII (plotted using DRAWTREE)plotted using different options of DRAWGRAM program).

6 53

Taxona GC C A A C G TC G A TG C CA C G TT G TT TA GC A C C G G TTC TTG TC C GA TC A C A GA TG T

Taxonb GC C A A C A A TG A TA C C A C G CC G TC C A GC A C C GA TTC TC G TC C GA G TA C C GA TG T

Taxon c G G TA A C G TC A A TG G GA C G TT G TC C A G C A C C G G TTC A TG TC C A A GC A GA G A TG T

Taxond GC C A A C A TTG A TA C C A C GC C G TTTA GC TGC GA C A C TC G TC C G A TC A C CA A TG T

Taxon e GC TA A C GA C A A TA C C A G GC T G TC C A G C TC C G G TT TA C G TC C GA GC A C A G A TG T

Taxon f GC C A A C A TC G A TG G GA C G TT G TTTA GC TC C GA TTC A TG TC C A A TC A C C A A TG T

6

Taxona 0 .000000 0.297888 0.253844 0.301576 0.306645 0.199728

Taxonb 0 .297888 0.000000 0.470560 0.229212 0.276609 0.401604

Taxon c 0 .253844 0.470560 0.000000 0.736034 0.286112 0.276199

Taxond 0 .301576 0.229212 0.736034 0.000000 0.475079 0.278527

Taxon e 0 .306645 0.276609 0.286112 0.475079 0.000000 0.460436

Taxon f 0 .199728 0.401604 0.276199 0.278527 0.460436 0.000000

((((Taxona:0 .09986,Ta xonf:0 .09986):0.03265,Taxon c:0 .13251):0.04302,

Taxon e:0 .17553):0.02684,(Ta xonb:0 .11461,Taxond:0 .11461):0 .08776);

I

IIIII

Taxon o G GC A A C GA C G A TA C C A C G TT G TT TA GC TC C G G TTC TC G TC C C A G CA GC C A TG TIV


15 9

Taxon1 101001111

Taxon2 101011001

Taxon3 010000111

Taxon4 111010000

Taxon5 101101101

Taxon6 110011010

Taxon7 011011001

Taxon8 000000111

Taxon9 111001100

Taxon10 000010111

Taxon11 101101001

Taxon12 110011001

Taxon13 001010101

Taxon14 101010101

Taxon15 000000000

B C

D

A

Figure 8.31 Construction of trees using MIX program of PHYLIP based on matrix in the Table8.4. A: Input file with fourth character converted into binary (simple and compoundleaves) character. Out of the 34 parsimonious trees generated by Mix, Consensustree generated by CONSENSE program presented as Phenogram (B), Cladogram(C) and Phylogram (D).

and selects the best tree. It gives you choiceto select the number of trees to be saved,10000 being the default. The program di-rectly yields the outtree file. PROTPARS,similarly performs Parsimony analysis ofProtein sequences. The protein sequencesare given by the one-letter code used by thelate Margaret Dayhoff’s group in the Atlas ofProtein Sequences, and consistent with theIUB standard abbreviations. DNAMOVEwhich handles data similar to DNAPARS,

allows the user to choose an initial tree, anddisplays this tree on the screen. The usercan look at different sites and the way thenucleotide states are distributed on thattree, given the most parsimonious recon-struction of state changes for that particu-lar tree. The user then can specify how thetree is to be rearranged, rerooted or writtenout to a file. By looking at different rearrange-ments of the tree the user can manuallysearch for the most parsimonious tree, and


can get a feel for how different sites are af-fected by changes in the tree topology.DNAML program carries out analysis of DNAsequences using Maximum LikelihoodMethod. The program uses both informativeand non-informative sites and yields theouttree file directly. RESTML similarlyhandles restriction site data using maxi-mum likelihood method. Binary data codedas 0 (ancestral state) and 1 (advanced state)is handled by MIX, which performs parsimonyanalysis and generates outtree which canbe plotted using DRAWGRAM. Input datafrom Table 8.4 and most parsimonious treegenerated using MIX program is presentedin Figure 8.31. For this analysis fourthmultistate character was converted into bi-nary character (simple and compoundleaves). Using Wagner parsimony the pro-gram was able to generate 34 trees. Taxon15 was used for rooting. CONSENSE programwas used to select the majority rule consen-sus tree. MOVE handles binary data and isan interactive program which allows theuser to choose an initial tree, and displaysthis tree on the screen. The user can lookat different characters and the way theirstates are distributed on that tree, given themost parsimonious reconstruction of statechanges for that particular tree. The userthen can specify how the tree is to berearraranged, rerooted or written out to afile. By looking at different rearrangementsof the tree the user can manually searchfor the most parsimonious tree, and can geta feel for how different characters are af-fected by changes in the tree topology.

Multistate data can similarly be handledby PARS, and can be converted into binarydata by FACTOR program. Data from Genefrequencies and continuous characters ishandled by CONTML (constructs maximumlikelihood estimates of the phylogeny;handles both types of data), GENDIST (com-putes genetic distances for use in the dis-tance matrix programs; handles data fromgene frequencies) and CONTRAST (exam-ines correlation of traits as they evolve alonga given phylogeny; handles continuous char-acters data). The data matrix for gene fre-

quencies contains number of species (orpopulations) and number of loci , where asthe second line contains number of allelesfor each locus. the default number of datafor each species (A-all) contains one alleleless for each locus. thus for three loci with2, 3 and 2 alleles respectively there wouldbe four values. Without A option, thereshould be 7 values. The values in datasetare preceded and followed by blanks. The datafrom continuous characters does not containthe second line, the data would include num-ber of species and the number of charactersin the first line (only line above speciesdata).

PHYLIP also offers programs to yield con-sensus tree (CONSENSE), Bootstrapping(SEQBOOT) and a host of related programs.The following information may be useful inhandling DNA sequence data.

Prepare infile of DNA sequences in whichtaxon name takes 10 characters followed bysequences in groups of 10 (separated by aspace), last three nucleotides being termi-nating codon. First taxon should be one in-tended as one used for rooting. Number oftaxa (sequences used) and number of nucle-otides in each sequence forms first line offile. Longer sequences can be interleaved(giving first part of sequences of all taxa andthen next part of all taxa) or aligned (finish-ing one sequence and then going to second).Save this file in text format in notepad (ANSIcode should be used; is default in notepad).Distance matrix can be prepared usingdnadist.exe program. When program asks forinfile, type above file name along with .txtending. Choose the ratio of transitions andtransversions, so that program can handleit accordingly. You can also choose distancemodel such as F84, Kimura, Jukes-Cantorand Logdet. Give name of your output file(preferably in txt format so that you can openand see it in notepad). The file can be savedas square matrix or only lower triangle. Theabove file can be used for generating clus-ters through Fitch, Kitsch, and Neighbor pro-grams. Each program searches for the short-est tree. When program asks for infile, typethe name of above output file. Give name of


Figure 8.32 Attempts towards construction of monophyletic groups. I: Strict consensus tree aspresented in the Figure 8.27-III. With poorly resolved phylogenies, the separation ofH and I in a group distinct and of the same rank as group CDEFG would create aparaphyly, as HI are left out of the descendents of common ancestor o. II: A consen-sus tree (hypothetical) with better resolved phylogenies. Both groups CDEFG and HIare monophyletic and, in turn could be assembled into more inclusive group withcommon ancestor at level o, now containing all descendents of the common ances-tor. This group (CDEFG, HI) and AB (also monophyletic) could be assembled into onemost inclusive monophyletic group, containing all descendents of the common an-cestor at p.

A B C D E F G H I

I

A B C D E F G H I

II

mn

o

p

mn

q

o

p

output file (of this program) or simply ask forreplacement if program reports that file isalready present. Neighbour provides a choicebetween Neighbour-joining (in which onetaxon is to be chosen for rooting) and UPGMA(in which no taxon for rooting has be se-lected). After selection of choice press Y. Pro-gram will generate outtree file, if alreadypresent replace it. You can read this file ifsaved in txt format. The above outtree filecan be used for plotting trees usingDRAWGRAM or DRAWTREE programs. Drawprogram asks for intree file. Type in thename of above outtree file. It will next askfor name of font file. Type font1 or any otherwithin the Phylip folder. The program pro-vides you number of choices includingphenogram and cladogram. It also provideschoice between indicating branch lengths(construction of conventional phylogram) ornot (conventional phenograms and cla-dograms where taxa end at same height).On typing Y tree preview will appear. PressPrint screen on keyboard and paste on new

paint file to save as image file in Paint. Youcan change options by clicking File->changeparameters in tree preview and go back todrawgram to generate other types of trees.

Binary data can similarly be input in infilewith just replacing nucleotide alphabets withbinary 0 and 1 data as presented in Table8.31 and handled by various programs men-tioned earlier.

Gene Trees and Species TreesTraditional the phylogenetically trees areconstructed using data from multiple char-acters, and if genetic data is used, fromanalysis of multiple genes. Such trees, ap-propriately known as species trees reflectthe evolutionary history of related groups ofspecies, and consequently a single species.A phylogenetic tree based on the divergenceobserved within a single homologous geneis most appropriately called a gene tree. Thegenes commonly used for the constructionof gene trees have been described in


chapter 7. Although they have been broadlyused in recent years in the construction ofphylogenies, a single gene may not alwaysreflect relationships between species, be-cause divergence within genes, especiallythe sequence polymorphism occurs beforethe splitting of populations that give rise tonew species.

Developing ClassificationOnce the phylogeny of a group has been de-veloped, the evolutionary process within thegroup can be reconstructed, the morphologi-cal, physiological and genetic changes canbe described, and the resultant informationused in the classification of the group. Phy-logenetic classifications are based on therecognition of monophyletic groups and avoidincluding paraphyletic and often completelyreject paraphyletic groups. Such classifica-tions are superior over classifications basedon overall similarity in several respects:

1. Such a classification reflects the ge-nealogical history of the group muchmore accurately.

2. The classification based on mono-phyletic groups is more predictive andof greater value than classificationbased on some characteristics.

3. Phylogenetic classification is of ma-jor help in understanding distributionpatterns, plant interactions, pollenbiology, dispersal of seeds and fruits.

4. The classification can direct thesearch for genes, biocontrol agentsand potential crop species.

5. The classification can be of consider-able help in conservation strategies.

The evolutionary history of the of the groupof 8 living species shown in Figure 8.12 wasknown with precise point of character trans-formations, and the construction of mono-phyletic groups, assembled into successivelymore inclusive groups, did not pose muchproblem. But it is often not the case. Even,most resolved consensus trees are oftenambiguous in several respects.

Consider the strict consensus tree rep-resented in the Figure 8.27-III. This tree isreproduced in the Figure 8.32-I. As notedearlier, the phylogenetic relationships be-tween taxa (these could be different species,genera, etc.) C, D, E, F, and G are poorly re-solved, and as such they are shown arisingfrom the common point, and consequentlycommon ancestor as level o. Although H andI form a distinct group with a common an-cestor as m, but leaving these two out of thegroup including CDEFG would render latteras paraphyletic (cf. traditional separation ofdicots and monocots). The safest situationwould be to include all the seven taxa intoone group, which may be regarded as belong-ing to the same rank as the group includingA and B. All the nine taxa may next be in-cluded into the single most inclusive groupwith common ancestry at level p. We are thusable to construct groups at two ranks only.

Now supposing the phylogenies of the taxawere better resolved and we had obtained aconsensus tree as shown in Figure 8.32-II.Now taxa C,D,E,F and G belong to a lineagewhich diverged from the main lineage, suc-cessive to the divergence of the lineageformed by A and B. Placement of H and I intoone group HI would not create any problemas both this group as well as the groupCDEFG are monophyletic with separate com-mon ancestors at level m and q, respectively.The groups CEDFG and HI could next be as-sembled into group CDEFGHI with commonancestor at o. Note that the group AB cannext be merged with CDEFGHI to form singlemost inclusive group ABCDEFGHI. Now wehave been able to construct taxa at threeranks instead of two from tree I.

Supposing the taxa A to I included in thetree, are different species. From tree II, thuswe are able to recognize three genera AB,CDEFG and HI. The last two are next as-sembled into family CDEFGHI and theformer a monotypic family AB. The other al-ternative was to place A and B in two sepa-rate monotypic genera (depending on the de-gree of morphological and genetic divergenceobtained) which are then assembled into


family AB. The two families may next be as-sembled into order ABEDEFGHI. There couldbe other possibilities also. The second rankcould be a subfamily and the third a family.Similarly, a third rank could be a suborderinstead of an order. These final decisionsare often made, based on the size of thegroup, degree of divergence, and the reliabil-ity of characters. All the groups recognizedabove would be monophyletic at the respec-tive ranks.

Next, let us look at the tree shown in theFigure 8.29, a study on paleoherbs. Ranun-culus and Aquilegia were used as outgrouprepresenting family Ranunculaceae; theirisolated position from paleoherbs is clearlydepicted in the tree. Paleoherbs constitutea group of taxa of uncertain affinities, whichhave been placed differently in various clas-sification schemes, but a few points seemto have been resolved. Piper, and Peperomia(both belong to Piperaceae) form a distinctgroup, and so do Saururus, Gymnotheca,Houttuynia and Anemopsis (all four belongingto Saururaceae), and the two families a wellsupported (bootstrap support of 90 per cent).This was confirmed by comparison of sevenpublished trees of paleoherbs. Cabomba,Lactoris and Saruma have least resolved

affinities with very poor support, with highlyunstable position.

The final decisions on the recognition ofgroups are, however, often based on personalinterpretation of phylogenies. Chloranthus,in this tree as well in several others, iscloser to Magnolia (Magnoliales) and Laurus(Laurales), but often finds different treat-ment. APG II places Chloranthaceae afterAmborellaceae at the start of Angiosperms.Judd et al., had earlier (1999) placedChloranthaceae under order Laurales ofMagnoliid complex, but have now (2008)shifted the family among basal ANITA Gradewith uncertain position. APweb of Stevens(2008), which places the family under orderChloranthales. Thorne had earlier (1999,2000) placed Chloranthaceae inM a g n o l i i d a e — > M a g n o l i a n a e —>Magnoliales—> Chloranthineae (other sub-orders within the order being Magnoliineae,and Laurineae), but subsequently (2003) in-cluded the family after Amborellaceae un-der order Chloranthales, the first order ofMagnoliidae, finally (2006, 2007) separatedunder subclass Chloranthidae, a placementsomewhat similar to APG II. Further discus-sion on angiosperm affinities will beresumed in the next chapter.

chapter 8 developing classifications · 2020. 4. 1. · tion in the taxa of a classification and...

Documents