Biological Classification: Toward a

Synthesis of Opposing MethodologiesErnst Mayr

For nearly a century after the publica-tion of Darwin's Origin (1) no well-de-fined schools of classifiers were recog-nizable. There were no competing meth-odologies. Taxonomists were unanimousin their endeavor to establish classifica-tions that would reflect "degree of rela-tionship." What differences there wereamong competing classifications con-cerned the number and kinds of charac-ters that were used, whether or not anauthor accepted the principle of reca-pitulation, whether he attempted to"base his classification on phylogeny,"and to what extent he used the fossilrecord (2). As a result of a lack of

or unspoken starting point of virtually allsystems of classification. Any classifica-tion incorporating the method of group-ing taxa by similarity is, to that extent,phenetic.

In the 1950's to 1960's several investi-gators went one step further and suggest-ed that classifications be based exclu-sively on "overall similarity." They alsoproposed, in order to make the methodmore objective, that every character begiven equal weight, even though thiswould require the use of large numbersof characters (preferably well over ahundred). In order to reduce the valuesof so many characters to a single mea-

Summary. Currently a controversy is raging as to which of three competingmethodologies of biological classification is the best: phenetics, cladistics, or evolu-tionary classification. The merits and seeming deficiencies of the three approachesare analyzed. Since classifying is a multiple-step procedure, it is suggested that thebest components of the three methods be used at each step. By such a syntheticapproach, classifications can be constructed that are equally suited as the basis ofgeneralizations and as an index to information storage and retrieval systems.

methodology, radically different classifi-cations were sometimes proposed for thesame group of organisms; also new clas-sifications were introduced without anyadequate justification except for theclaim that they were "better." Dissatis-faction with such arbitrariness and seem-ing absence of any carefully thought outmethodology, led in the 1950's and1960's to the establishment of two newschools of taxonomy, numerical phenet-ics and cladistics, and to a more explicitarticulation of Darwin's methodology,now referred to as evolutionary classifi-cation.

The Major Schools of Taxonomy

Numerical phenetics. From the earli-est preliterary days, organisms weregrouped into classes by their outwardappearance, into grasses, birds, butter-flies, snails, and others. Such grouping"by inspection" is the expressly stated

sure of "overall similarity," each char-acter is to be recorded in numericalform. Finally, the clustering of speciesand their taxonomic distance from eachother is to be calculated by the use ofalgorithms that operationally manipulatecharacters in certain ways, usually withthe help of computers. The resultingdiagram of relationship is called a pheno-gram. The calculated phenetic distancescan be converted directly into a classifi-cation.The fullest statement of this method-

ology and its underlying conceptualiza-tion was provided by Sokal and Sneath(3). They called their approach "numeri-cal taxonomy," a somewhat misleadingdesignation, since numerical methods,including numerical weighting, can beand have been applied to entirely differ-ent approaches to classification. Theterm numerical phenetics is now usuallyapplied to this school. This has intro-duced some ambiguity since some au-thors have used the term phenetic broad-

0036-8075/81/1030-0510$01.00/0 Copyright 1981 AAAS

ly, applying it to any approach makinguse of the "similarity" of species andother taxa, while to the strict numericalpheneticists the term phenetic means the"theory-free" use of unweighted charac-ters.

Cladistics (or cladism). This methodof classification (4), the first comprehen-sive statement of which was published in1950 by Hennig (5), bases classificationsexclusively on genealogy, that is, on thebranching pattern of phylogeny. For thecladist phylogeny consists of a sequenceof dichotomies (6), each representing thesplitting of a parental species into twodaughter species; the ancestral speciesceases to exist at the time of the dichoto-my; sister groups must be given the samecategorical rank; and the ancestral spe-cies together with all of its descendantsmust be included in a single "holophy-letic" taxon.

Evolutionary classification. Pheneticsand cladistics were proposed in the en-deavor to replace the methodology ofclassification that had prevailed eversince Darwin and that was variouslydesignated as the "traditional" or the"evolutionary" school, which bases itsclassifications on observed similaritiesand differences among groups of organ-isms, evaluated in the light of their in-ferred evolutionary history (7). The evo-lutionary school includes in the analysisall available attributes of these organ-isms, their correlations, ecological sta-tions, and patterns of distributions andattempts to reflect both of the majorevolutionary processes, branching andthe subsequent diverging of the branches(clades). This school follows Darwin(and agrees in this point with the cla-dists) that classification must be basedon genealogy and also agrees with Dar-win (in contrast to the cladists) "thatgenealogy by itself does not give classifi-cation" (8).The results of the evolutionary analy-

sis are incorporated in a diagram, calleda phylogram, which records both thebranching points and the degrees of sub-sequent divergence. The method of in-ferring genealogical relationship with thehelp of taxonomic characters, as it wasfirst carried out by Darwin, is an applica-tion of the hypothetico-deductive ap-proach. Presumed relationships have tobe tested again and again with the help ofnew characters, and the new evidencefrequently leads to a revision of theinferences on relationship. This methodis not circular (9) as has sometimes beensuggested.

The author is Alexander Agassiz Professor Emeri-tus, Museum of Comparative Zoology, HarvardUniversity, Cambridge, Massachusetts 02138.

SCIENCE, VOL. 214, 30 OCTOBER 1981510

on

Nov

embe

r 1,

200

8 w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

Is There a Best Way to Classify?

Each of the three approaches to classi-fication-phenetics, cladistics, and evo-lutionary classification-has virtues andweaknesses. The ideal classificationwould be one that would meet best asmany as possible of the generally ac-knowledged objectives of a classifica-tion.A biological classification, like any

other, must serve as the basis of a conve-nient information storage and retrievalsystem. Since all three theories producehierarchical systems, containing nestedsets of subordinated taxa, they permitthe following of information up anddown the phyletic tree. But this is wherethe agreement among the three methodsends. Purely phenetic systems, derivedfrom a single set of arbitrarily chosencharacters, sometimes provide only lowretrieval capacity as soon as other sets ofcharacters are used. The effectiveness ofthe phenetic method could be improvedby careful choice of selected characters.However, the method would then nolonger be "automatic," because any se-lection of characters amounts to weight-ing.

Cladists use only as much informationfor the construction of the classificationas is contained in the cladogram. Theyconvert cladograms, quite unaltered,into classifications, only when the clado-grams are strictly dichotomous. Eventhough cladists lose much information bythis simplistic approach, the informationon lines of descent can be read off theirclassifications directly. However, a ne-glect of all ancestral-descendant infor-mation reduces the heuristic value oftheir classifications. By contrast, sinceevolutionary taxonomists incorporate agreat deal more information in their clas-sifications than do the cladists, they can-not express all of it directly in the namesand ranking of the taxa in their classifica-tions. Therefore, they consider a classifi-cation simply to be an ordered index thatrefers them to the information that isstored elsewhere (in the detailed taxo-nomic treatments).A far more important function of a

classification, even though largely com-patible with the informational one, is thatit establishes groupings about which gen-eralizations can be made. To the extentthat classifications are explicitly basedon the theory of common descent withmodification, they postulate that mem-bers of a taxon share a common heritageand thus will have many characteristicsin common. Such classifications, there-fore, have great heuristic value in allcomparative studies. The validity of spe-30 OCTOBER 1981

cific observations can be generalized bytesting them against other taxa in thesystem or against other kinds of charac-ters (10-12).

Pheneticists, as well as cladists, haveclaimed that their methods of construct-ing classifications are nonarbitrary, auto-matic, and repeatable. The criticisms ofthese methods over the last 15 years (13)have shown, however, that these claimscannot be substantiated. It is becomingincreasingly evident that a one-sidedmethodology cannot achieve all theabove-listed objectives of a good classifi-cation.The silent assumption in the method-

ologies of phenetics and cladistics is thatclassification is essentially a single-stepprocedure: clustering by similarity inphenetics, and establishment of branch-ing patterns in cladistics. Actually a clas-sification follows a sequence of steps,and different methods and concepts arepertinent at each of the consecutivesteps. It seems to me that we mightarrive at a less vulnerable methodologyby developing the best method for eachstep consecutively. Perhaps the stepscould eventually be combined in a singlealgorithm. In the meantime, their sepa-rate discussion contributes to the clarifi-cation of the various aspects of the clas-sifying process.

Establishment of Similarity Classes

The first step is the grouping of speciesand genera by "inspection," that is, by aphenetic procedure. (I use phenetic inthe broadest sense, not in the narrow oneof numerical phenetics.) All of classify-ing consists of, or at least begins with,the establishment of similarity classes,such as a preliminary grouping of plantsinto trees, shrubs, herbs, and grasses.The reason why the method is so oftensuccessful is simply that-other thingsbeing equal-descendants of a commonancestor tend to be more similar to eachother than they are to species that do notshare immediate common descent. Themethod is thus excellent in principle.Numerical phenetics has neverthelessproved to be largely unsuccessful be-cause (i) claims, such as "results objec-tive and strictly repeatable," were notalways justifiable since in practice differ-ent results are obtained when differentcharacters are chosen or different pro-grams of computation are used; (ii) themethod was inconsistent in its claim ofobjectivity since subjective biologicalcriteria were used in the assigning ofvariants (for example, sexes, age class-es, and morphs) to "operational taxo-

nomic units" (OTU's); and, most impor-tantly, the method insisted on the equalweighting of all characters.

It is now evident that no computingmethod exists that can determine "truesimilarity" from a set of arbitrarily cho-sen characters. So-called similarity is acomplex phenomenon that is not neces-sarily closely correlated with commondescent, since similarity is often due toconvergence. Most major improvementsin plant and animal classifications havebeen due to the discovery of such con-vergence (14).

Different types of characters-mor-phological characters, chromosomal dif-ferences, enzyme genes, regulatorygenes, and DNA matching-may leadto rather different grouping. Differentstages in the life cycle may result indifferent groupings.The ideal of phenetics has always been

to discover a measure of total (overall)similarity. Since it is now evident thatthis cannot be achieved on the basis of aset of arbitrarily chosen characters, thequestion has been asked whether there isnot a method to measure degrees ofdifference of the genotype as a whole.Improvements in the method of DNAhybridization offer hope that this methodmight give realistic classifications on aphenetic basis, at least up to the level oforders (15). The larger the fraction of thenonhybridizing DNA, the less reliabiethis method is, because it cannot bedetermined whether the nonmatchingDNA is only slightly or drastically differ-ent.

Testing the Naturalness of Taxa

In the first step of the classifying pro-cedure clusters of species were assem-bled that seemed to be more similar toeach other than to species in other clus-ters. These clusters are the taxa werecognize tentatively (16). In order tomake these clusters conform to evolu-tionary theory, two, operationally moreor less inseparable tests, must be made:(i) determine for all species of a cluster(taxon) whether they are descendants ofthe nearest common ancestor and (ii)connect the taxa by a branching tree ofcommon descent, that is, construct acladogram. An indispensable prelimi-nary of this testing is an analysis of thecharacters used to establish the similar-ity clusters.

Character analysis. A careful analysisshows almost invariably that some char-acters are better clues to relationship(have greater weight) than others. Thefewer the number of available charac-

511

on

Nov

embe

r 1,

200

8 w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

ters, the more carefully the weightingmust be done. This weighting is one ofthe most controversial aspects of theclassifying procedure. Investigators whocome to systematics from the outside,say from mathematics, or who are begin-ners tend to demand objective or quanti-tative methods of weighting. There are

such methods, principally ones based on

the covariation of characters, but theyare not nearly as informative as methodsbased on the biological evaluation ofcharacters (17). But such an evaluationrequires an understanding of many as-

pects of the to-be-classified group (thatis, its life history, the inferred selectionpressures to which it is exposed, and itsevolutionary history) that may not beavailable to an outsider. This creates a

genuine dilemma. If strictly taxometricmethods were available that would pro-

duce satisfactory weighting, everyone

would surely prefer them to weightingbased on experience and biologicalknowledge. But so far such methods are

still in their infancy.The greatest difficulty for a purely

phenetic method, indeed for any methodof classification, is the discordance (non-congruence) of different sets of charac-ters. Entirely different classificationsmay result from the use of characters ofdifferent stages of the life cycle as, forinstance, larval versus adult characters.In a study of species of bees, Michener(18) obtained four different classifica-tions when he sorted them into similarityclasses on the basis of the characters of(i) larvae, (ii) pupae, (iii) the externalmorphology of the adults, and (iv) male

genitalic structures. Phenetic delimita-tion of taxa unavoidably necessitates a

great deal of decision-making on the use

and weighting of characters. Often,when new sets of characters becomeavailable, their use may lead to a new

delimitation of taxa or to a change inranking.Determination of the genealogy. Each

group (taxon) tentatively established bythe phenetic method is, so to speak, a

hypothesis as to common descent, thevalidity of which must be tested. Is thedelimited taxon truly monophyletic (19)?Are the species included in this taxonnearest relatives (descendants of thenearest common ancestor)? Have allspecies been excluded that are only su-

perficially or convergently similar?Methods to answer these questions

have been in use since the days of Dar-win, particularly the testing of the ho-mology of critical characteristics of theincluded species. However, Hennig (5)was the first to articulate such methodsexplicitly, and these have been modified

512

by some of his followers. These methodscan be designated as the cladistic analy-sis.Such an analysis involves first the

partitioning of the joint characters of agroup into ancestral ("plesiomorph" inHennig's terminology) characters andderived ("apomorph") characters, thatis, characters restricted to the descend-ants of the putative nearest commonancestor (20). The joint possession ofhomologous derived characters provesthe common ancestry of a given set ofspecies. A character is derived in rela-tion to the ancestral condition of thecharacter. The end product of such acladistic character analysis is a clado-gram, that is, a diagram (dendrogram) ofthe branching points of the phylogeny.Although this procedure sounds sim-

ple, numerous practical difficulties havebeen pointed out (21, 22). Very often thebranching points are inferred by way ofsingle or very few characters and areaffected by all the weaknesses of singlecharacter classifications. More seriousare two other difficulties.

1) Polarity. A derived character is of-ten simpler or less specialized than theancestral condition. For this reason itcan be difficult to determine polarity in atransformation series of characters, thatis, to determine which end of the series isancestral. Tattersall and Eldredge (23)stressed that "in practice it is hard, evenimpossible, to marshall a strong, logicalargument for a given polarity for manycharacters in a given group." Are theyprimitive (ancestral) or derived? Much ofthe controversy concerning the phyloge-ny of the invertebrates, for instance, isdue to differences of opinion concerningpolarity. Hennig tried to elaborate meth-ods for determining polarity but, as oth-ers (24, 25) have shown, with ratherindifferent success. Since characterscome and go in phyletic lines and sincethere is much convergence, the problemof polarity can rarely be solved unequiv-ocally. There are three best types ofevidence for polarity reconstruction.First is the fossil record. Although primi-tiveness and apparent ancientness arenot correlated in every case, neverthe-less as Simpson (26) stressed, "for anygroup with even a fair fossil record thereis seldom any doubt that characters usu-al or shared by older members are almostalways more primitive than those of latermembers." Second is sequential con-straints. Consecutive chromosomal in-versions (as in Drosophila) or sets ofamino acid replacements (and presum-ably certain other molecular events)form definite sequences. Which end ofthe sequence is the beginning can usually

not be read off from the sequence itself,but additional information (polarity ofother character chains, geographical dis-tribution, and the like) often permits anunequivocal determination of the polari-ty. Third is the reconstruction of thepresumed evolutionary pathway. Thiscan sometimes be done by studying evi-dence for adaptive shifts, the invasion ofnew competitors or the extinction of oldones, the behavior of correlated charac-ters, and other biological evidence (11,pp. 886-887; 24). Particular difficultiesare posed when the polarity is reversedin the course of evolution, as docu-mented in the fossil record.

2) Kinds of derived characters. Twotaxa may resemble each other in a givencharacter for one of three reasons; be-cause the character existed already inthe ancestry of the two groups before theevolution of the nearest common ances-tor (symplesiomorphy in Hennig's termi-nology), because it originated in thecommon ancestor and is shared by all ofhis descendants (homologous apo-morphy or synapomorphy), or because itoriginated independently by conver-gence in several descendant groups (non-homologous or convergent apomorphy)(27). Since, according to the cladisticmethod, sister groups are recognized bythe possession of synapomorphies, con-vergence poses a major problem. Howare we to distinguish between homolo-gous and convergent apomorphies? Hen-nig was fully aware of the critical impor-tance of this problem, but it has beenquietly ignored by many of his followers.Both grebes and loons, two orders ofdiving birds, have a prominent spur onthe knee and were therefore called sistergroups by one cladist. However, otheranatomical and biochemical differencesbetween the two taxa indicate that theshared derived feature was acquired byconvergence. The reliability of the deter-mination of monophyly of a group de-pends to a large extent on the care that istaken in discriminating between thesetwo classes of shared apomorphy (11,pp. 880-890).There is a third class of derived char-

acters, so-called autapomorphies, whichare characters that were acquired by andare restricted to a phyletic line after itbranched off from its sister group.The pheneticists do not undertake a

character analysis. Cladists and evolu-tionary taxonomists agree with each oth-er in principle on the importance of acareful character analysis. They dis-agree, however, fundamentally in how touse the findings of the character analysisin the construction of classifications,particularly the ranking procedure.

SCIENCE, VOL. 214

on

Nov

embe

r 1,

200

8 w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

The Construction of a Clasification

Cladistic classification. Cladists con-vert the cladogram directly into a cladis-tic classification. In such a classificationtaxa are delimited exclusively by holo-phyly, that is, by the possession of acommon ancestor, rather than by a com-bination of genealogy and degree of di-vergence (19). This results in such incon-gruous combinations as a taxon contain-ing only crocodiles and birds, or onecontaining only lice and one family ofMallophaga.Taxa based exclusively on genealogy

are of limited use in most biologicalcomparisons. Since, as Hull (28) pointedout, cladists really classify charactersrather than organisms, they have tomake the arbitrary assumption that newapomorph characters originate whenevera line branches from its sister line. Thisis unlikely in most cases. Surely thereptilian species that originated the avianlineage lacked any of the ffight special-izations characteristic of modem birds,except perhaps the feathers (29).Two principles govern the conversion

of a cladogram into a cladistic classifica-tion: (i) all branchings are bifurcationsthat give rise to two sister groups, and(ii) branchings are usually connectedwith a change in categorical rank. Cladis-tic classifications are only representa-tions of branching patterns, with com-plete disregard of evolutionary diver-gence, ancestor-descendant relation-ships, and the information content ofautapomorph characters. Because theseaspects of evolutionary change are ne-glected, the cladistic method of classifi-cation "either results in lumping verysimilar forms (parasites and their rela-tives) or in recognizing a multitude oftaxa (perhaps also of other categories)regardless of the extreme similarity ofsome of them. Such simplistic proce-dures do violence to most biological at-tributes other than the pattern of thecladistic branching system, as well as tothe function of a classification for conve-nient information transmittal and stor-age," as Michener remarked (18).These objections show that the meth-

odology of cladistic classification is notsatisfactory. Anyone familiar with thehistory of taxonomy is strangely remind-ed of the principles of Aristotelian logicaldivision when encountering cladisticclassifications with their nrgid dichoto-mies, the mandate that every taxon musthave a sister group, and the principle ofastraight-line hierarchy.There has been much argument over

the relationship between classificationand phylogeny (30). Both cladists and30 OCTOBER 1981

evolutionary taxonomists agree that allmembers of a taxon must have a com-mon ancestor. A phylogenetic analysis,and in particular a clear separation ofhomologous apomorphies from conver-gences, is a necessary component of theclassifying procedure. Classificatoryanalysis often leads to new inferences onphylogeny, and new insights on phyloge-ny may necessitate changes in classifica-tion. These interactions are not in theleast circular (9).

It is quite unnecessary in most cases toknow the exact species that was thecommon ancestor of two diverging phy-letic lines. An inability to specify such anancestral species has rarely impeded pa-leontological research (31, 32). For in-stance, it is of little importance whetherArchaeopteryx was the first real ancestorof modern birds or some other similarspecies or genus. What is important toknow is whether birds evolved from liz-ard-like, crocodile-like, or dinosaur-likeancestors. If a reasonably good fossilrecord is available, it is usually possible,by the backward tracing of evolutionarytrends and by the backward projection ofdivergent phyletic lines, to reconstruct areasonably convincing facsimile of therepresentative of a phyletic line at anearlier time.Simpson (32) has provided us with

cogent arguments about why it is notpermissible to reject information fromthe fossil record under the pretext that itfails to give the phylogenetic connec-tions between fossil and recent taxa withabsolute certainty. Hence, there is nomerit in the suggestion to construct sepa-rate classifications for recent and forfossil organisms. After all, fossil speciesbelong to the same tree of descent asliving species. Indeed, enough evidenceusually becomes available through acareful character analysis to permit rela-tively robust inferences on the mostprobable phylogeny. A number of recentendeavors have been made to develop acladistic methodology that is quantita-tive and automatic. New methods in thisarea are published in rapid successionand it would seem too early to determinewhich is most successful and freest ofpossible flaws (33).

Evolutionary classification. The taxo-nomic task of the cladist is completedwith the cladistic character analysis. Thegenealogy gives him the classificationdirectly, since for him classification isnothing but genealogy. The evolutionarytaxonomist carries the analysis one stepfurther. He is interested not only inbranching, but, like Darwin, also in thesubsequent fate of each branch. In par-ticular, he undertakes a comparative

study of the phyletic divergence of allevolutionary lineages, since the evolu-tionary history of sister groups is oftenstrikingly different. Among two relatedgroups, derived from the same nearestcommon ancestor, one may hardly differfrom the ancestral group, while the othermay have entered a new adaptive zoneand evolved into a novel type. Eventhough they are sister groups in the ter-minology of cladistics, they may deservedifferent categorical rank, because theirbiological characteristics differ to suchan extent as to affect any comparativestudy. The importance of this consider-ation was stated by Darwin (1, p. 420):"I believe that the arrangement of thegroups within each class, in due subordi-nation and relation to the other groups,must be strictly genealogical in order tobe naturl, but that the amount of differ-ence in the several branches or groups,though allied in the same degree in bloodto their common progenitor, may differgreatly, being due to the different de-grees of modification which they haveundergone, and this is expressed by theforms being ranked under different ge-nera, families, sections or orders." Dar-win refers then to a diagram of threeSilurian genera that have modern de-scendants; one has not even changed-generically, but the other two have be-come distinct orders, one with three andthe other with two families.The question as to what extent an

analysis of degrees of divergence is pos-sible, is still debated. The cladist makesonly "horizontal" comparisons, catalog-ing the synapomorphies of sister groups.The evolutionary taxonomist, however,also makes use of derived characters thatare restricted to a single line of descent,so-called autapomorph characters (Fig.1), which are apomorph characters re-stricted to a single sister group. Theimportance of autapomorphy is well il-lustrated by a comparison of birds withtheir sister group (34). Birds originatedfrom that branch of the reptiles, theArchosauria, which also gave rise to thepterodactyls, dinosaurs, and crocodil-ians. The crocodilians are the sistergroup of the birds among living organ-isms; a stem group of archosaurians rep-resents the common ancestry of birdsand crocodilians. Although birds andcrocodilians share a number of synapo-morphies that originated after the archo-saurian line had branched off from theother reptilian lines, nevertheless croco-dilians are on the whole very similiar toother reptiles, that is, they have devel-oped relatively few autapomorph charac-ters. They represent the reptilian"grade," as many morphologists call it.

513

on

Nov

embe

r 1,

200

8 w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

Birds, by contrast, have acquired a vastarray of new autapomorph characters inconnection with their shift to aerial liv-ing. Whenever a clade (phyletic lineage)enters a new adaptive zone that leads toa drastic reorganization of the clade,greater taxonomic weight may have to beassigned to the resulting transformationthan to the proximity of joint ancestry.The cladist virtually ignores this ecologi-cal component of evolution.The main difference between cladists

and evolutionary taxonomists, thus, is inthe treatment of autapomorph charac-ters. Instead of automatically giving sis-ter groups the same rank, the evolution-ary taxonomist ranks them by consider-ing the relative weight of their autapo-morphies as compared to their syna-pomorphies (Fig. 1). For instance, one ofthe striking autapomorphies of man (incomparison to his sister group, the chim-panzee) is the possession of Broca's cen-ter in the brain, a character that is close-ly correlated with man's speaking abili-ty. This single character is for mosttaxonomists of greater weight than vari-ous synapomorphous similarities or evenidentities in man and the apes in certainmacromolecules such as hemoglobinsand cytochrome c. The particular impor-tance of autapomorphies is that theyreflect the occupation of new niches andnew adaptive zones that may have great-er biological significance than synapo-morphies in some of the standard macro-molecules.

I agree with Szalay (35) when he says:"The loss of biological knowledge whennot using a scheme of ancestor-descen-dant relationship, I believe, is great. Infact, whereas a sister group relationshipmay . . . tell us little, a postulated andinvestigated ancestor-descendant rela-tionship may help explain a previouslyinexplicable character in terms of itsorigin and transformation, and subse-quently its functional (mechanical) sig-nificance." In other words, the analysisof the ancestor-descendant relationshipsadds a great deal of information thatcannot be supplied by the analysis ofsister group relationships.

It is sometimes claimed that the analy-sis of ancestor-descendant relationshipslacks the precision of cladistic sistergroup comparisons. However, as wasshown above and as is also emphasizedby Hull (36), the cladistic analysis isactually full of uncertainties. The slightpossible loss of precision, caused by theuse of autapomorphies, is a minor disad-vantage in comparison with the advan-tage of the large amount of additionalinformation thus made available.The information on autapomorphies

514

9

A b d

Fig. 1. Cladogram of taxa A, B, and C.Cladists combine B and C into a single taxonbecause B and C share the synapomorphcharacter b. Evolutionary taxonomists sepa-rate C from A and B, which they combine,because C differs by many (c through k)autapomorph characters from A and B andshares only one (b) synapomorph characterwith B.

permits the conversion of the cladograminto a phylogram. The phylogram differsfrom the cladogram by the placement ofsister groups at different distances fromthe joint common ancestry (branchingpoint) and by the expression of degree ofdivergence by different angles. Both ofthese topological devices can be translat-ed into the respective categorical rankingof sister groups. These methods (37)generally attempt to discover the short-est possible "tree" that is compatiblewith the data. Yet, anyone familiar withthe frequency of evolutionary reversalsand of evolutionary opportunism, real-izes the improbability of the assumptionthat the tree constructed by this so-called "parsimony method" corre-sponds to the actual phylogenetic tree."To regard [the shortest tree method] asparsimonious completely misconceivesthe intent and use of parsimony in sci-ence" (38).

It is not always immediately evidentwhether a tree construction algorithm isbased on cladists principles or on themethods of evolutionary classification. Ifthe "special similarity" on which thetrees are based are strictly synapomor-phies, then the method is cladistic. Ifautapomorphies are also given strongweight, then the method falls under evo-lutionary classification.The particular aspect of the method of

evolutionary taxonomy found most un-acceptable to cladists is the recognitionof "paraphyletic" taxa. A paraphyletictaxon is a holophyletic group from whichcertain strikingly divergent membershave been removed. For instance, theclass Reptilia of the standard zoologicalliterature is paraphyletic, because birds

C and mammals, two strikingly divergent<' descendants of the same common ances-k tor of all the Reptilia, are not included.

Nevertheless, the traditional class Repti-lia is monophyletic, because it consistsexclusively of descendants from thecommon ancestor, even though it ex-cludes birds and mammals owing to thehigh number of autapomorphies of theseclasses. The recognition of paraphyletictaxa is particularly useful whenever therecognition of definite grades of evolu-tionary change is important.

The Ranking of Taxa

Once species have been grouped intotaxa the next step in the process ofbiological classification is the construc-tion of a hierarchy of these taxa, the so-called Linnaean hierarchy. The hierar-chy is constructed by assigning a definiterank such as family or order to eachtaxon, subordinating the lower catego-ries to the higher ones. It is a basicweakness of cladistics that it lacks asensitive method of ranking and simplygives a new rank after each branchingpoint. The evolutionary taxonomist, fol-lowing Darwin, ranks taxa by the degreeof divergence from the common ances-tor, often assigning a different rank tosister groups. Rank determination is oneof the most difficult and subjective deci-sion processes in classification. One as-pect of evolution that causes difficultiesis mosaic evolution (39). Rates of diver-gence of different characters are oftendrastically different. Conventionallytaxa, such as those of vertebrates, aredescribed and delimited on the basis ofexternal morphology and of the skeleton,particularly the locomotory system.When other sets of morphological char-acters are used (for example, sense or-gans, reproductive system, central ner-vous system, or chromosomes), the evi-dence they provide is sometimes con-flicting. The situation can becomeworse, if molecular characters are alsoused. The anthropoid genus Pan (chim-panzee), for instance, is very similar toHomo in molecular characters, but mandiffers so much from the anthropoid apesin traditional characters (central nervoussystem and its capacities) and occupa-tion of a highly distinct adaptive zonethat Julian Huxley even proposed toraise him to the rank of a separate king-dom-Psychozoa.

It has been suggested that differentclassifications should be constructed foreach kind of character, or at least formorphological and molecular characters.Yet there is already much evidence that

SCIENCE, VOL. 214

on

Nov

embe

r 1,

200

8 w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

the acceptance of several classificationsbased on different characters would leadto insurmountable complications. Bytaking all available data into consider-ation simultaneously, a classification canusually be, constructed that can serveconveniently as an all-purpose classifica-tion or, as Hennig (5) called it, "a gener-al reference system."

It is usually possible to derive morethan one classification from a phylo-gram, because higher taxa are usuallycomposed of several end points of. thephylogram, and different investigatorsdiffer by the degree to which they lumpsuch terminal branches into a singlehigher taxon (40). An example is thephylogram of the higher ferns on which,as Wagner (41) has shown, six differentclassifications have been founded (Fig.2) and many more are possible. Theextent to which investigators "split" or"lump" higher taxa, thus, is of consider-able influence on the classifications theyproduce.

Comparison of the Three Majior Schools

Each school believes that its classifi-cation is the "best." Pheneticists as wellas cladists claim that their respectivemethods have also the great merit ofgiving automatically nonarbitrary re-sults. These claims cannot be substanti-ated. To be sure grouping by pheneticcharacters and determination of holo-phyly by cladistic analysis are valuablecomponents of the procedure of biologi-cal classification. The great deficiency ofboth phenetics and cladistics is the fail-ure to reflect adequately the past evolu-tionary history of taxa.What needs to be emphasized once

more is the fact that groups of organismsare the product of evolution and that noclassification can hope to be satisfactorythat does not take this fact fully intoconsideration. Both pheneticists and cla-dists are ambiguous in their attitude to-ward the evolutionary theory. The phe-neticists claim that their approach iscompletely theory-free, but they never-theless assume that their method willproduce a hierarchy of taxa that corre-sponds to descent with modification. Onthe basis of this assumption, they alsoclaim to be "evolutionary taxonomists"(42), but the fact that different pheneticprocedures may produce very differentclassifications and that their procedure is

tionary rates, and rates of evolutionarydivergence) from their consideration (43)and tend increasingly not to classify spe-cies and taxa, but only taxonomic char-acters (28) and their origin. The connec-tion with evolutionary principles is ex-ceedingly tenuous in many recent cladis-tic writings.By contrast, the evolutionary taxono-

mists, as -indicated by the name of theirschool and by well-articulated state-ments of some of its major representa-tives (7), expressly base their classifica-tions on evolutionary theory. They aimto construct classifications that reflectboth of the two major evolutionary pro-cesses, branching and divergence (clado-genesis and anagenesis). They make fulluse of information on shifts into newadaptive zones and rates of evolutionarychange and believe that the resultingclassifications are a key to a far richerinformation content.Although the three schools still seem

rather fundamentally in disagreement, asfar as the basic principles of classifica-tioh are concerned, the more moderaterepresentatives have quietly incorporat-ed some of the criteria of the opposingschools, so that the differences amongthem have been partially obliterated. Forinstance, Farris' (44) clustering of spe-cial similarities is a phenetic method

Fig. 2. Six differentpossible classifica-tions of ferns, basedon the same dendro-gram. Each filled cir-cle is a genus, andeach open circle is afamily. The differ-ences are due towhich and how manygenera are combinedto make up the fam-ilies. [From W. H.Wagner (41, figure 7)].

not influenced by evolutionary consider-ations refutes this assertion. The cladistsexclude most of evolutionary theory (forexample, inferences on selection pres-sures, shifts of adaptive zones, evolu-30 OCTOBER 1981

based on the weighting of characters.The evolutionary school uses pheneticcriteria to establish similarity classes andto construct a classification, and cladis-tic criteria to test the naturalness of taxa.Comparing what McNeill (45) says infavor of phenetics (appropriately modi-fied) and Farriss (44) against it, we findthat the gap has narrowed. I have nodoubt that moderates will be able todevelop an eclectic methodology, onethat contains a proper balance of phenet-ics and cladistics that will produce farmore "natural classifications" (16) thanany one-sided approach that relies exclu-sively on a single criterion, whether it beoverall similarity, parsimony of branch-ing pattern, or what not. Evolutionarytaxonomy, from Darwin on, has beencharacterized by the adoption- of aneclectic approach that makes use of simi-larity, branching pattern, and degree ofevolutionary divergence.

Classification and Information Retrieval

Biological classifications have two ma-jor objectives: to serve as the basis ofbiological generalizations in all sorts ofcomparative studies and to serve as thekey to an information storage system.Up to this point, I have concentrated on

515

on

Nov

embe

r 1,

200

8 w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

those aspects of classifying that help tosecure a sound basis for generalizations.This leaves unanswered the question ofwhether achievement of this first objec-tive is, or is not, reconcilable withachievement of the second objective. Isthe classification that is soundest as abasis of generalizations also most conve-nient for information retrieval? This, in-deed, seems to have been true in mostcases I have encountered. However, wecan also look at this problem from anoth-er side.

It is possible at nearly each of thethree major steps in the making of aclassification to make a choice betweenseveral alternatives. These choices maybe scientifically equivalent, but somemay be more convenient in aiding infor-mation retrieval than others. If wechoose one of them, it is not necessarilybecause the alternatives were "falsi-fied," but rather because the chosenmethod is "more practical." In this re-spect, biological classifications are notunique. Scientific theories are nearly al-ways judged by criteria additional totruth or falsity, for instance, by theirsimplicity or, in mathematics, by their"6elegance." Therefore, it can be assert-ed that convenience in the use of aclassification, including its function askey to information retrieval, is not nec-essarily in conflict with its more purelyscientific objectives (46-48).

References and Notes

1. C. Darwin, On the Origin of Species (Murray,London, 1859).

2. For an illuminating survey of the thinking of thatperiod see F. A. Bather [Proc. Geol. Soc. Lon-don 83, LXII (1927)].

3. R. R. Sokal and P. H. A. Sneath, Principles ofNumerical Taxonomy (Freeman, San Francisco,1963). A drastically revised second edition waspublished in 1973.

4. The method was first published under the mis-leading name phylogenetic systematics, butsince it is based on only a single one (branching)of the various processes of phylogeny, the termscladism or cladistics have been substituted andare now widely accepted.

5. W. Hennig's original statement is Grundzdgeeiner Theorie der Phylogenetischen Systematik(Deutscher Zentralverlag, Berlin, 1950). Agreatly revised second edition (reprinted in1979) is Phylogenetic Systematics, D. D. Davisand R. Zangerl, Eds. (Univ. of Illinois Press,Urbana, 1966); see also W. Hennig (47). Anindependent phylogenetic analysis of characterswas made by T. P. Maslin [Syst. Zool. 1, 49(1952)]. For an overview of the more significantrecent literatture see D. Hull (36) and J. S. Farris[Syst. Zool. 28, 483 (1979)].

6. Some cladists in recent years have relaxed therequirements of strict dichotomy and have per-mitted tri- and polyfurcations or have quietlyabandoned dichotomy by admitting empty inter-nodes in the cladogram. Polyfurcations can betranslated into several alternate bifurcations[see J. Felsenstein, Syst. Zool. 27, 27 (1978)],and this makes the automatic conversion of thecladogram into a classification of sister groupsimpossible.

7. The classical statement of this theory is to befound in C. Darwin (1, pp. 411-434). G. G.Simpson [Principles of Animal Taxonomy (Co-lumbia Univ. Press, New York, 1961)] and E.Mayr (48) provide comprehensive modern pre-sentations of this theory. Several critical recentanalyses are: W. Bock (11); C. D. Michener(18); Syst. Zool. 27, 112 (1978); P. D. Ashlock(12).

8. F. Darwin, Life and Letters of Charles Darwin(Murray, London, 1887), vol. 2, p. 247.

9. D. Hull, Evolution 21, 174 (1967); see also W.Bock (11).

10. F. E. Warburton, Syst. Zool. 16, 241 (1967); W.Bock, ibid. 22, 375 (1973).

11. W. Bock, in Major Patterns in Vertebrate Evo-lution, M. K. Hecht, P. C. Goody, B. M. Hecht,Eds. (NATO Advanced Study Institute Series,Plenum, New York, 1977), vol. 14, pp. 851-895.

12. P. D. Ashlock, Syst. Zool. 28, 441 (1979).13. I shall not, at this time, recount the almost

interminable controversies among the threeschools. For critiques of phenetics see E. Mayr(48, pp. 203-211), L. A. S. Johnson [Syst. Zool.19, 203 (1970)], and D. Hull [Annu. Rev. Ecol.Syst. 1, 19 (1970)]. Some of the weaknessespointed out by these early critics have beencorrected in the 1973 edition of Sokal andSneath (3) and by J. S. Farris (44). For critiquesof cladistics see E. Mayr (21), R. R. Sokal (22),G. G. Simpson (32), D. Hull (36), P. D. Ashlock[Annu. Rev. Ecol. Syst. 5, 81 (1974)1; and L. vanValen (49).

14. A particularly illuminating example is the break-ing up of the plant group Amentiferae, which hasbeen shown to consist of taxa secondarily adapt-ed for wind pollination [R. F. Thorne, Brittonia25, 395 (1973)]. Examples among animals ofradical reclassifications are the Rodentia, para-sitic bees, certain beetle families, and the turbel-larians.

15. C. G. Silaley, in preparation.16. There have been arguments since before the

days of Linnaeus about how to determinewhether or not a system, a classification, is"natural.", William Whewell, at a time beforeDarwin, had proclaimed his theory of commondescent, expressed the then prevailing pragmat-ic consensus, "The maxim by which all systemsprofessing to be natural must be tested is this:that the arrangement obtained from one set ofcharacters coincides with. the arrangement ob-tained from another set" [W. Whewell, Philos.Inductive Sci. 1, 521 (1840)]. Interestingly, thecovariance of characters is still perhaps the bestpractical test of the goodness of a classification.Since Darwin, of course, that classification isconsidered most natural that best reflects theinferred evolutionary history of the organismsinvolved.

17. For a tabulation and analysis of such qualitativemethods of weighting, see E. Mayr (48, pp. 220-228).

18. C. D. Michener, Syst. Zool. 26, 32 (1977).19. I use the word monophyletic in its traditional

sense, as a qualifying adjective of a taxon.Various definitions of monophyletic have beenproposed but all of them for the same concept, aqualifying statement concerning a taxon. A tax-on is monophyletic if all of its members arederived from the nearest common ancestor [E,Haeckel, Naturliche Schopfungsgeschichte(Reimer, Berlin, 1868)]. Cladists have attemptedto turn the situation upside down by placing alldescendants of an ancestor into a taxon. Mono-phyletic thus becomes a qualifying adjective fordescent, and a taxon is not recognized by itscharacteristics but only by its descent. Thetransfer of such a well-established term asmonophyletic to an entirely different concept isas unscientific and unacceptable as if someonewere to "redefine" mass, energy, or gravity byattaching these terms to entirely new concepts.P. D. Ashlock [Syst. Zool. 20, 63 (1971)] hasproposed the term holophyletic for the assem-blage of descendants ofa common ancestor. Seealso P. D. Ashlock (12, p. 443).

20. Terms like apomorph, synapomorph, derived,ancestral, and so forth always refer to charac-ters of taxa at all levels. A genus may havesynapomorphies with another genus, and somay an order with another order. It is thisapplicability of the same criteria for taxa of allranks that permits the construction of the Lin-naean hierarchy.

21. E. Mayr, Z. Zool. Syst. Evolutionsforsch. 12, 94(1974); reprinted in, E. Mayr, Evolution and theDiversity of Life (Harvard Univ. Press, Cam-bridge, Mass., 1976), pp. 433-478.

22. R. R. Sokal, Syst. Zool. 24, 257 (1975).23. I. Tattersall and N. Eldredge, Am. Sci. 65, 204

(1977).24. D. S. Peters and W. Gutmann, Z. Zool. Syst.

Evolutionsforsch. 9, 237 (1971).25. 0. Schindewolf, Acta Biotheor. 18, 273 (1968);

H. K. Erben, Verh. Dtsch. Zool. Ges. 79, 116(1979). -

26. G. G. Simnpson (32); see also L. van Valen (49).27. For a diagram of these three categories of mor-

phological resemblance see figure 1 in W. Hen-nig (47).

28. "Cladistic classifications do not represent theorder of branching of sister groups, but the orderof emergence of unique derived characters" [seeD. Hull (36)].

29. G. G. Simpson [The Major Features of Evolu-tion (Columbia Univ. Press, New York, 1953),p. 348] discusses the fallacy of the cladist as-sumption.

30. Phylogeny is equated by cladists with cladogen-esis (branching), while the evolutionary taxono-mist subsumes both branching and evolutionarydivergence (anagenesis) under phylogeny.

31. C. W. Harper, J. Paleontol. 50, 180 (1976).32. G. G. Simpson, in Phylogeny of the Primates,

W. Pluckett and F. S. Szalay, Eds. (Plenum,New York, 1975), pp. 3-19.

33. J. H. Camin and R. R. Sokal, Evolution 19, 311(1965); W. MS Fitch and E. Margoliash, Science155, 279 (1967); W. M. Fitch, in Major Patternsof Vertebrate Evolution, M. K. Hecht, P. C.Goody, B; M. Hecht, Eds. (NATO AdvancedStudy Institute Series, Plenum, New York,1977), vol. 14, pp. 169-204.

34. There are literally hundreds of cases to illustratethis situation. I use again the classical case ofbirds and crocodilians because even a nonbiolo-gist will understand the situation if such familiaranimals are used. The holophyletic classificationof the lice (Anoplura) derived from one of thesuborders of the Mallophaga is ariother particu-larly. instructive example [K. C. Kini and H. W.Ludwig, Ann. Entomol. Soc. Am. 71, 910(1978)].

35. F. S. Szalay, Syst. Zool. 26, 12 (1977).36. D. Hull, ibid. 28, 416 (1979).37. J. W. Hardin, Brittonia 9, 145 (1957); W. H.

Wagner, in Plant Taxonomy: Methods and Prin-ciples, L. Benson, Ed. (Ronald, New York,1962), pp. 415-417; A. G. Kluge an.d J. S. Farris,Syst. Zool. 18, 1 (1969); J. S. Farris, Am. Nat.106, 645 (1972).

38. L. H. Throckmorton, in'Biosystematics in Agri-culture, J. A. Romberger, Ed. (Wiley, NewYork, 1978), p. 237. Others who have ques-tioned the validity of the so-called parsimonyprinciple are M. Ghiselin [Syst. Zool. 15, 214(1966)] and W. Bock (11).

39. Unequal rates of evolution for different struc-tures or for any other components of pheno-types or gehotypes are designated mosaic evolu-tion.

40. See E. Mayr (48, pp. 238-241) on the differencesbetween splitters and lumpers.

41. W. H. Wagner, "The construction of a classifi-cation," in Systematic Biology (Publication1692, National Academy of Sciences, Washing-ton, D.C., 1969), pp. 67-90.

42. R. R. Sokal in (22), "I have yet to meet anonevolutionary taxonomist."

43. Several leading cladists have recently publishedantiselectionist statements.

44. J. S Farris, in Major Patterns in VertebrateEvolutiort, M. K. Hecht, P. C. Goody, B. M.Hecht, Eds. (NATO Advanced Study InstituteSeries, Plenum, New York, 1977), vol. 14, pp.823-850.

45. J. McNeill, Syst. Zool. 28, 468 (1979).46. For criteria by which to judge the practical

usefulness of biological classifications, see E.Mayr (48, pp. 229-242).

47. W. Hennig, Annu. Rev. Entomol. 10, 97 (1965).48. E. Mayr, Principles of Systematic Zoology

(McGraw-Hill, New York, 1969).49. L. van Valen, Evol. Theory 3, 285 (1978).50. Drafts were read by P. Ashlock, J. Beatty, W.

Bock, W. Fink, C. G. Hempel, and D. Hul, toall of whom I am indebted for valuable sugges-tions and critical comments, not all ofwhich wasI able to accept.

SCIENCE, VOL. 214516

on

Nov

embe

r 1,

200

8 w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from