Development 1994 Supplement, 107-116 (1994) 107Printed in Great Britain @ The Company of Biologists Limited 1994

Evolution of flowers and inflorescences

Enrico S. Coen* and Jacqueline M. Nugent

John lnnes Centre, Colney, Norwich NR4 7UH, UK*Author for correspondence

SUMMARY

Plant development depends on the activity of meristemswhich continually reiterate a common plan. Permutationsaround this plan can give rise to a wide range of mor-phologies. To understand the mechanisms underlying thisvariation, the effects of parallel mutations in key develop-mental genes are being studied in different species. InAntirrhinum, three of these key genes are: (l)floricaula ffio)a gene required for the production of flowers (2) centror&-dialis (cen), a gene controlling flower position (3) cycloidea(cyc), a gene controlling flower symmetry. Several plantspecies, exhibiting a range of inflorescence types and floralsymmetries are being analysed in detail. Comparativegenetic and molecular analysis shows that inflorescencearchitecture depends on two underlying parameters: abasic inflorescence branching pattern and the positioning

of flowers. Theflo and cen genes play a key role in the posi-tioning of flowers, and variation in the site and timing ofexpression of these genes, may account for many of thedifferent inflorescence types. The evolution of inflorescencestiucture may also have influenced the evolution of floralasymmetry, as illustrated by the cen mutation whichchanges both inflorescence type and the symmetry of someflowers. Conflicting theories about the origins of irregularflowers and how they have coevolved with inflorescencearchitecture can be directly assessed by examining the roleof cyc- and cen-like genes in species displaying variousfloral symmetries and inflorescence types.

Key words: Antirrhinum, Arabidopsis, symmetry

INTRODUCTION

The development of plants depends on the activity of theirmeristeffis, groups of dividing cells located at the growingpoints. These meristems can continue to add new structures tothe plant throughout its life history, giving it the potential forindeterminate growth. They achieve this by generating twotypes of structure on their periphery: primordia which willgrow to form organs such as leaves, petals and stamens; andsecondary meristems, which will form side branches orflowers. In most cases each primordium has a secondarymeristem located in its axil (the angle between the base of theprimordium and the main stem). By changing a few of the keyfeatures of meristeffis, it is possible to account for much of theevolutionary variation in plant form. Here we present a com-parative molecular genetic approach to understanding howsuch changes may account for variation in inflorescence archi-tecture and floral symmetry.

Plants offer several advantages for comparative analysis.Parallel genetic studies have been carried out in severalspecies, allowing comparisons of mutations in key genes.Several species can be transformed, permitting the transfer ofgenes between species to assess how the role of these genesmay have changed during evolution. The indeterminategrowth pattern of plants gives them the potential to producemany different morphologies through limited modificationsin the properties of their meristems. Thus, analysingkey genes controlling meristem behaviour may give

important insights into how diverse plant forms haveevolved.

TYPES OF INFLORESCENCE

Most plants have flowers clustered together in a region termedthe inflorescence. Inflorescences have been classifiedaccording to several criteria (Weberling, 1989). These includewhether the inflorescence is: (1) determinate, where the mainaxis of growth ends in a flower (Fig. la-d), or indeterminate,in which there is no terminal flower (Fig. 1e-i); (2) racemose,having a branching pattern with a single main axis of growth(Fig. le-h), or cymose, having a branching pattern which lacksa main axis (Fig. lc,d); (3) simple, in which no secondarybranches are produced (Fig. 1a,b,e,h,i), or compound, whichproduce secondary and sometimes higher order branches (Fig.lc,d,f,g); (4) bracteose which have leaf-like organs (bracts)subtending flowers, or bractless.

Most of these criteria depend on two underlying parameters,a basic inflorescence branching pattern and the positions offlowers. Since, by definition, an inflorescence must compriseboth flowers and branch points, these two parameters normallycannot be uncoupled and their individual contributions to inflo-rescence architecture assessed separately. However, by usinga molecular genetic approach to analyse inflorescence mutants,we can begin to separate the various parameters and assess

their role in the evolution of inflorescence types.

108 E. S. Coen and J. M. Nugent

Fig. 1. Diagrammatic illustration of various inflorescence types.(a) Solitary terminal flower; (b) simple dichasium; (c) compounddichasium; (d) cyme; (e) raceme; (f) thyrse; (g) compound raceme'(h) corymb; (i) capitulum. Flowers are indicated by solid circles.Bracts are not shown.

DEFINING THE PRIMARY INFLORESCENCEBRANCHING PATTERN

The first gene affecting inflorescence development to bestudied in detail was floricaula (flo) in Antirrhinum majus(Carpenter and Coen, 1990; Coen et dl., 1990). Vegetativegrowth in Antirrhinum is characterised by the production oftwo opposite leaves at each node of the main stem. After theplant has produced several vegetative nodes, it switches to thereproductive phase, characterised by the production of singleflowers in the axils of bracts separated by short internodes ona hairy stem (Fig. 2). This type of inflorescence, comprising aseries of axillary flowers affanged along a single main axis, istermed a raceme (Fig. 1e). Inflo mutants, flowers are replacedby shoots (Fi g. 2) that have some characteristics of the repro-ductive phase of growth (bracts, short internodes, hairy stem).These secondary shoots can themselves produce further shoots,termed tertiary shoots, in the axils of their bracts. The overallresult is that the raceme has been replaced by a flo-indepen-dent branching structure that is never terminated by the pro-duction of flowers.

One interpretation of the flo mutant phenotype is that itreflects an alteration in the identity of meristems. In Antir-rhinum there are three types of meristem that are distinguishedby the types and arrangements of organ primordia and thesecondary meristems they produce. The vegetative meristemproduces leaf primordia with secondary vegetative meristemsin their axils. The inflorescence meristem norrnally producesbract primordia with floral meristems in their axils. Floralmeristems produce floral organ primordia but unlike the othermeristem types, do not give rise to secondary meristems. A

further distinguishing feature of floral meristems is that afterthey have produced a defined number of primordia they ceaseto be active. Thus floral meristems are determinate andprovide endpoints to the growth of an axis whereas vegetativeand inflorescence meristems are indeterminate.

In the flo mutant, the main inflorescence meristem producessecondary inflorescence meristeffis, rather than floralmeristeffis, in the axils of its bract primordia. These secondarymeristems can in turn produce further meristems on theirperiphery and this process can continue indefinitely leading toeven higher order meristems. The flo gene is therefore requiredto establish floral meristem identity and in its absence, inde-terminate branching ensues. This early requirement for flo cor-relates with its expression pattern, &S determined by in situhybridisation (Coen et al., 1990). Transcripts of flo are detectedin initiating floral meristems and their subtending bractprimordia but are not observed in either the main infloresencemeristem or in vegetative meristems (Fig. 3). Localisation offlo expression within particular meristems is therefore animportant determinant of which meristems will form flowers.At later stages of development, flo rs expressed transiently inall floral organ primordia except stamens.

One way to alter inflorescence architecture might be tochange the site or timing of flo expression. This can be illus-trated by using wild-type Antirrhinum as a starting point andpredicting the phenotype produced when flo expression isaltered in several ways. A slight delay rn flo activation, suchthat it does not come on immediately in secondary meristemsbut after some tertiary branching is initiated, would give a

cluster of flowers instead of single axillary flowers. Thiscould give rise to a compound inflorescence where eachsecondary branch comprises a small cyme (thyrse, Fig. 1f).Another possiblity would be changing flo expression suchthat it only came on in tertiary rather than secondarymeristems. This would result in each flower being replacedby an indeterminate axillary raceme, giving rise to a

compound raceme (Fig. 1g).

DETERMINATE VERSUS INDETERMINATEINFLORESCENCES

Whereas a delay of flo expression might be expected toproduce a more highly branched inflorescence, ectopicexpression of flo could be involved in producing more compactor determinate inflorescences. One way to convert an indeter-minate inflorescence to a determinate condition would be toreplace the main inflorescence apex by a flower. This is illus-trated by centroradialis (cen) mutants of Antirrhinum (Stubbe,1966) which produce a short raceme, terminated by a singleflower (Fig. 4, note that the shape of the terminal flower isdistinct, see below). Presumably, in wild-type plants this geneprevents expression of flo and other genes needed for flowerdevelopment in the inflorescence apex.

It has been suggested that the indeterminate inflorescence isa derived condition, which required a mechanism for repress-ing terminal flowers to have evolved (Stebbins, 1974). Fur-thermore, because species with indeterminate inflorescencesare present in a wide range of taxa, interspersed with specieshaving a determinate condition, it is believed that a geneticmechanism for repressing terminal flowers must have arisen

lv+$

independently many times. However, it is equally plausiblethat a mechanism for repressing terminal flowers was derivedat an early stage during the evolution of flowering plants andthat in many lineages determinate inflorescences have beensecondarily derived through loss of this mechanism. Is itpossible to distinguish between these two scenarios?According to the multiple gain hypothesis, the mechanism forestablishing indeterminacy might be expected to be differentin distantly related species. However, according to the multipleloss theory, many species could share the same underylingmechanism for establishing indetermin acy . One way to distin-guish between these possibilities is to establish whether thegenes controlling indeterminacy in other species are homolo-gous to cen. For example, the terminal flower (tfl) mutant ofArabidopsis thaliana has a comparable phenotype to cen. It tfl.and cen encode distinct proteins, it would suggest that Ara-bidopsis and Antirrhinum might have acquired indeterminacythrough different routes. If the cen and ffi genes are homolo-gous, it would imply that the indeterminate condition of Antir-rhinum and Arabidopsis may not have been independentlyacquired but was present in their common ancestor.

RACEMOSE VERSUS CYMOSE BRANCHING

The previous examples illustrate how some inflorescencearchitectures might arise through changing the expressionpattern of floral meristem identity genes such as flo.If this werethe only way of altering inflorescence type, flo-Ltke mutants inall species would reveal essentially the same primarybranching pattern. This can be directly assessed in severalcases.

The most intensively studied flo homologue in anotherspecies is the W gene of Arabidopsis (Weigel et aI., 1992;

Schultz and Haughn, 1991). Vegetative growth of wild-typeArabidopsis produces a rosette of leaves separated by shortinternodes. On entering the reproductive phase, the plant boltsto generate an elongated main stem, the lower part of whichhas several small leaves (cauline leaves) and the upper partbears flowers affanged as a raceme. Secondary inflorescencesare initiated within the axils of the cauline leaves (Fig. 2).Inlfy mutants, the plant bolts as norrnal but flowers are replacedby secondary shoots subtended by cauline leaves. In olderplants, the secondary shoots produced at the top of the bolt,display some floral characteristics, showing that some featuresof flower development can be restored, even in the absence oflfi actlity.

The basic branching pattern underlying the Arabidopsisinflorescence, &s revealed by the lfi mutant, is very similar tothat of Antirrhinum (Fig. 2).In both cases there is a main axisof growth with secondary shoots arising in axillary positions.This similarity may reflect the fact that wild-type Arabidopsisand Antirrhinum share the same overall inflorescence archi-tecture, the raceme. A more revealing comparison might be toanalyse the flo-independent branching pattern in species withmarkedly different inflorescence types.

The tomato (Lycopersicon esculentum) inflorescence has a

cymose branching pattern. After vegetative growth, the mainapical meristem becomes a terminal floral meristem associatedwith an adjacent secondary meristem. The secondary meristemterminates with the production of a flower and a tertiary

Evolution of flowers and inflorescences 109

meristem. By repeating this pattern of growth, tertiary andhigher order flowers are successively produced to form a

cymose inflorescence. Although initially in an apical position,the whole inflorescence eventually becomes displaced by an

axillary leafy shoot which continues the main growth of theplant (Fig. 2). After producing about three leaves the axillaryshoot repeats the growth pattern of the primary shoot. Thisgrowth pattern, termed sympodial growth, could be a conse-quence of expressing flo-like genes in the apical meristem andin secondary, tertiary and higher order meristems derived fromit. By committing the apical meristem to producing flowers,the continued growth of the plant would have to occur from an

axillary meristem. According to this view, a flo-like mutant intomato should have an indeterminate leafy shoot with a singleprimary axis of growth, similar to that seen for the flo and lfimutants of Antirrhinum and Arabidopsis. Alternatively,sympodial growth may not be a consequence of flower pro-duction but may reflect the flo-rndependent primary branchingpattern of tomato. In this case, flo-llke mutants in tomatoshould retain sympodial growth.

The falsiflora (fa) mutant of tomato lacks a flowering inflo-rescence and produces a highly branched system of shootsinstead (Stubbe, 1963, Fig. 2). The mutant still exhibitssympodial growth, showing that this growth pattern is inde-pendent of flower production. Preliminary genetic andmolecular results indicatethatfa may be the tomato homologueof flo. However, the fa mutation reveals a primary branchingpattern that is quite distinct from that reveale d by flo or lfy.Unlike flo and lfy, the fa branching system that replaces theinflorescence, lacks a main axis of growth, even though it pro-liferates extensively. This suggests that the wild-type tomatoinflorescence can be distinguished from that of Antirrhinum orArabidopsis by two features: a cymose primary branchingpattern which is flo-independent and the production of terminalflowers which is flo-dependent. Both of these features can beseen as reflecting early meristem behaviour.

BRACTEOSE VERSUS BRACTLESSINFLORESCENCE

In addition to overall architecture, another feature that distin-guishes inflorescences is the presence or absence of bracts sub-tending the flowers. Antirrhinum has flowers subtended bybracts (bracteose infloresence) whereas Arabidopsis flowersare bractless (Fig. 2). The absence of bracts is a general featureof the family Cruciferae, to which Arabidopsis belongs and isthought to be a derived condition. In W mutants, shoots orabnormal flowers often with subtending bracts, are producedwhere bractless flowers would normally arise (Schultz andHaugho, 1991; Weigel et a1., 1992). This demonstrates that thebractless condition is to a large extent W-dependent. Theanalysis of flo and lfy expression indicates how the bractlesscondition may have evolved.

In Antirrhinum, flo is expressed in floral meristems and theirsubtending bract primordia. In Arabidopsis, W is onlyexpressed in floral meristems. However, when bract primordiaare allowed to form tn Arabidopsis, as in W mutants, theyaccumulate /, RNA, giving an expression pattern similar tothat of wild-type Antirrhinum (Fig. 3).

One interpretation of these results is that originally W was

1 10 E. S. Coen and J. M. Nugent

Antinhinum

hlelllqz

Fig.2

ffi flo/lfy expression

witdUW

Antinhinum

expressed in flower meristems andbract primordia, as in Antirrhinum.Subsequently, in a commonancestor of the Cruciferae, Wacquired the role of incorporatingcells that would normally go on toform the bract, into the floralmeristem. The bracts are thereforesuppressed, not by inhibiting theactivity of a bract-forming region,but by incorporating the cells fromsuch a region into the flower. Theadvantage of recruiting all cells toform a flower without the attendingbract might be to accelerate flowerdevelopment. This is illustrated inlfi mutants, where the developmentof floral meristems subtended bybracts is retarded relative to wild-type, presumably because they startoff with fewer cells (Fig. 3).

In some members of the Cru-ciferae, the role of lfy in cell recruit-ment may be partly separated fromits role in flower production. Forexample, in Alyssum, the secondaryinflorescences that are producedjust below the main raceffie, are notsubtended by cauline leaves (Fig.5). One possible explanation is that

Fig. 2. Diagrammatic illustrations andphotographs comparing wild-type and

flo mutant of Antirrhinum (top panel),wild-type and /71 mutant ofArabidopsis (middle panel) and wild-type and fa mutant of tomato (bottompanel). In all cases, wild type is shownon the left. In the diagrammaticillustrations, leaves are shown inbrown, bracts or cauline leaves ingreen, flowers as blue circles andabnormal "flowers" as green circles. Inthe photographs, only the upper part ofthe plant is shown for Antirrhinum andArabidopsis and only singleinflorescences are shown for tomato.

wild tyW

Arabidopsis

leafy

Arabi:dopsis

Evolution of flowers and inflorescences 1 1 1

Fig. 3. Expression pattern of flo inwild-type Antirrhinum and /71 in wild-type and mutant Arabidopsis.Abbreviations: I, inflorescencemeristem; F, floral meristem; B, bractprimordium; s, sepal primordium; FI,meristem with both floral andi nfl orescence character.

Fig. 4. Photograph of the inflorescence of an Antirrhinum cen mutant shown from the side. Notethe symmetrical terminal flower.

112 E. S. Coen and J. M. Nugent

in these cases, W is acting to incorporate cells, that wouldnormally go on to form the cauline leaf, into the meristem ofthe secondary inflorescence. Within the main raceme ofAlyssum, the lfy gene acts as it does in Arabidopsis, recruitingcells into the secondary meristem and switching its identity tothat of a floral meristem.

The emergence of the bractless condition may reflect analteration in the W itself and/or alterations in target genesacting downstream of lfy. One way to test this would be totransform W mutants with flo and observe its expressionpattern and phenotypic consequences. Preliminary resultsindicate that flo does not complement the lfi mutant, suggest-ing that some functionally important differences exist betweenthese genes (R. Elliott, J. Nugent, R. Carpenter and E. Coen,unpublished results). Further experiments, such as swappingdomains between lfy and flo,, may shed light on where the keydifferences reside.

EVOLUTION OF FLORAL SYMMETRY

Flowers are classified as being either irregular, having onlyone plane of mirror symmetry or regular, having more thanone plane of symmetry. The most intensive genetic analysis offloral symmetry has been carried outin Antirrhinum, which has irregularflowers that are markedly asymmet-rical along their dorsoventral axis.Wild-type Antirrhinum flowersconsist of five petals that are unitedfor part of their length to form a tubeending with five lobes. The petallobes are of three types: two largedorsal (upper) petals, two side petalsand a ventral (lowest) petal. Theflower is also irregular with respectto the stamens. Five stamens areinitiated, alternate with the petalsand are also of three types: the dorsalstamen is aborted and the two sidestamens are shorter than the twoventral stamens. Mutations incycloidea (cyc) give regular flowerswith five-fold symmetry in certaingenetic backgrounds (Fig. 6). Allfive petals and stamens resemble theventral petal and stamens of thewild-type. It has been proposed thatthe irregular phenotype of wild-typeflowers is dependent on cyc activityestablishing an axis of dorsoventralasymmetry. The activity of cyc ispredicted to be greatest in the dorsalregions of the flower meristem andto decline towards the more ventralregions. This would account for theventralised phenotype of cycmutants (Carpenter and Coen,1990).

Early flowering plants are thoughtto have had regular flowers and

irregularity is considered to be a derived condition. Theanalysis of genes like cyc allows us to address severalimportant questions about the genetic basis of this evolution-ary change: what role might cyc have played in the ancestralspecies with regular flowers and how was cyc subsequentlyrecruited to establish dorsoventral asymmetry? How manytimes has irregularity evolved and has it always involve d cyc?Before considering these questions it is important first toconsider the current view of how irregularity has evolved,largely based on comparative morphology.

Irregularity is thought to have evolved independently manytimes, perhaps arising on as many as 25 separate occasions(Stebbins, 197 4). The alternative to this multiple gain hypoth-esis, is that irregularity arose only a limited number of timesand was subsequently lost several times in independentlineages. According to this view, irregularity may be muchmore ancient than is commonly believed. Three types ofargument can be used to evaluate these two hypotheses:

( 1) The multiple gain hypothesis is claimed to be the mostparsimonious way to explain the broad phylogenetic distribu-tion of regularity as compared to the more sporadic occuffenceof irregularity. However, this is partly a circular argumentbecause the phylogenies it is based upon have been constructedusing morphological data that includes floral symmetry and

Fig. 5. Photograph of the Alyssum inflorescence. Arrow shows a secondary inflorescence notsubtended by a cauline leaf.

Evolution of flowers and inflorescences 1 13

Fig. 6. Photographs of wild-type and cyc mutant of Antirrhnum (A) and comparable peloric mutants from Linaria vulgaris (B), Saintpaulia (C)

and Sin ningia speciosa (D). For all species, the wild type is on the left and mutant on the right.

correlated characters. An objective evaluation can only be

made if the phylogenies are independent of the morphologicalcharacter being assessed.

Even with a more objective phylogeny, arguments based onparsimony are problematical because they depend on knowingthe relative probabilities of gaining or losing irregularity. Thisis illustrated in Fig. 7, which shows the relationships betweenfamilies in a portion of the subclass Asteridae and is based onmolecular data (Olmstead et &1., 1993). A traditional viewwould be that the common ancestor of these species wasregular, implying that irregularity evolved a minimum of 3

times independently (see (+) in Fig. 7). The alternative, is thatthe ancestor was irregular and implies that regularity wasderived a minimum of 4 times (see (-), Fig. 7). The two expla-nations are about equally parsimonious, assuming that theprobabilities of gaining or losing irregularity are equal.However, this is unlikely to be true because the probabilitiesdepend on two biological variables: the genetic facility withwhich irregularity can be gained or lost and the adaptive con-sequences of the change. The cyc mutation of Antirrhinumillustrates an alteration from the irregular to the regularcondition. To our knowledge, no mutations have been

described in species with regular flowers that render theminegular. This would argue that it is easier to lose irregularitythan to gain it. However, this must be tempered by adaptiveconsiderations. The cyc mutation would clearly confer a

selective disadvantage because bees, the primary pollinators ofAntirrhinum, would be unable to enter and cross-pollinate themutant flowers efficiently. It is therefore very difficult to

establish overall whether gain or loss is the more likely expla-nation for the phylogeny in Fig. 7.

(2) Another ilrgument used in favour of the multiple gainhypothesis is that irregularity is a more specialised adaptationfor pollination and is therefore more likely to be derived.However, although irregularity may have originally evolvedfrom the regular state, it need not always be the derivedcondition. Evolution is not a unidirectional process and spe-

cialised characters can be lost as well as gained.(3) Irregular flowers occur in many different guises. For

example, the irregular flowers of mint, pea and Antirrhinum,are all structurally very different. This has been taken as

evidence in support of the multiple gain hypothesis as it seems

to suggest that many independent mechanisms for generatingirregularity have evolved. However, it is possible that some ofthe different types of irregular flowers share the same under-lying mechanism for generating asymmetry. The differencesmay simply reflect the imposition of irregularity on differentframeworks of floral development.

It is impossible to resolve these issues purely on the basis oftaxonomic information. Isolating genes like cyc that are

involved in controlling floral symmetry allows a more directapproach to addressing these problems. The molecular basis ofboth the irregular and the regular condition could be addressed

by comparing the activity of cyc-like genes in a range ofspecies using a combination of genetic, molecular and trans-genic techniques.

Mutants that give regular instead of irregular flowers, termedpeloric mutants, have been described in several species in

114 E. S. Coen and J. M. Nugent

Nintiana

Lycopersicon

Schizanthus

Conwlvulus

Borago

Echium

Antirrhinum

Linaria

Aanthus

Galeobdolon

Saintpaulia

Sinningia

Gentiana

Asclepias

Spigelia

Chioaw

Solanaceae

Convulvulaceae

Boraginaceae

Scrophularlaceae

Acanthaceae

Labiatae

Gesnerlaceae

Gentianaceae

Apocynaceae

Loganlaceae

Rubiaceae

Lamlales s.l.

o-l

$joo-lvrVrvJvvvtvroooo

oooMv-MMMv'oooo

v

t

o"v

addition to Antirrhinum. The first peloric mutant was describedby Linnaeus in Linaria vulgaris (toadflax), a close relative ofAntirrhinum (Fig. 68, Linnaeus, 1749). As with c!c, thismutant confers a ventralised phenotype, all petals resemble thelowest petal of the wild type, which is easily distinguished inLinaria by the presence of a spur. Peloric mutants with ven-tralised phenotypes have also been described for Saintpaulia(African violets, Fig. 6C) and Sinningia (gloxinias, Fig. 6D),both members of the family Gesneriaceae (Fig. 7). Althoughgenetic analysis of peloric mutants has not yet been carried outin species other than Antirrhinum, it is tempting to speculatethat they reflect alterations Ln cyc-like genes. This can be testedby analysing the structure and expression of cyc homologuesin these mutants.

All of these examples of peloric mutants are in species thatare quite closely related. According to the phylogeny presentedin Fig. I these species fall within a monophyletic group (theLamiales s.l.; Olmstead et al., 1993) that includes only specieswith irregular flowers. It is likely that the common ancestor ofthis group had irregular flowers that depended on cyc-like geneactivity. However, it is unclear whether irregul arrty in speciesthat lie outside the Lamiales is also cyc-dependent. Accordingto the multiple gain hypothesis, the mechanism for establish-ing irregularity might be expected to be different outside thisgroup. The alternative hypothesis, that irregul arrty is moreancient, predicts that irregul arrty in some species outside theLamiales should also be cyc-dependent. One way to distinguishbetween these possibilities is to determine the role of cychomologues in species with irregular flowers, such as Schizan-thus (butterfly flower), Echium (Fig. 7) or even more distantlyrelated species. If cyc-ltke genes are involved in controllingirregularity in these species, it then raises the question ofwhether cyc was recruited once in a common ancestor of allthese species or whether it was recruited independently severaltimes. This might be resolved by studying the role, if any, ofcyc-like genes in species with regular flowers and how thiscompares with its role in irregular species.

Fig. 7. Phylogenetic tree showingrelationships between species andfamilies of a portion of the subclassAsteridae. Examples of species andfamilies having regular flowers areindicated by symmetrical symbolswhereas those with irregular flowers areindicated by asymmetrical symbols. Infamilies having a mixture of specieswith irregular and regular flowers, bothsymbols are shown, the larger symbolindicating the predominant conditionwithin the family. If the ancestor of thegroup was regular, irregul arrty wasderived independently at least 3 times inthe group (indicated by +). If theancestor of the group was irregular, thiscondition was lost at least 4 times in thegroup (indicated by -).

COEVOLUTION OF INFLORESCENCE STRUCTUREAND FLORAL SYMMETRY

There is a strong correlation between floral asymmetry andinflorescence architecture: irregular flowers are commonlyassociated with indeterminate racemose inflorescenceswhereas regular flowers can occur in both racemose andcymose inflorescences (Stebbins, I9l4). This correlation maybe a consequence of either selective or developmental con-straints. For example, if the adaptive value of the irregularcondition depends on the flowers being presented to pollina-tors on a racemose inflorescence, this would mean thatselection was involved. Alternatively, if the genetic mecha-nisms for generating asymmetry are dependent on theracemose condition, a developmental constraint would beinvolved.

The cen mutant of Antirrhinum, which has a determinateinflorescence, provides strong evidence for developmentalconstraints. The cen mutant produces two types of flower:axillary flowers that are irregular and a terminal flower that isregular (Fig. 4). This appears to be a general property of cen-like mutants in a wide range of species with irregular flowers,such as Digitalis (Scrophulariaceae), Salvia grandiflora andGaleobdolon luteum (Labiatae), Delphinium elatum andAconitum varie gatum (Ranunculaceae); in all of these cases theterminal flower produced in the mutants is regular (Peyritsch,1870, I8l2). The terminal flower of cen resembles the axillaryflowers of cyc mutants (Fig. 6), suggesting that cyc is not activein the apical meristem. The production of cyc-dependentasymmetry seems to require that floral meristems are inaxillary positions, and have a particular cellular environment.Axillary meristems are in an asymmetric environment, with theinflorescence apex above them and a bract below. Thispolarised environment could provide necessary cues for estab-lishing the cyc system. By contrast, a terminal flower meristemis in a symmetrical environment and may lack the cuesrequired to activate cyc in an asymmetrical manner. Accord-

ingly, species producing only terminal flowers might be unableto make them irregular. However, there are apparent excep-tions to this view because some species, such as Sc'ftii.ctrtthu,s(Fig. 7) have irregular flowers occupying terminal positions.Molecular and phenotypic analysis of these species may f urtherour understanding of the relationship between flower positionand symmetry.

Other aspects of the coevolution of inflorescence type andfloral symmetry appear to reflect selective constraints. Mostmembers of the Cruciferae have regular flowers borne onextended racemes, as shown for Arabidopsis and Alystr,,r,(Figs 2 and 5). However, in some members of this family, suchas lberis, the inflorescence axis is shorter and produces a

broad dome of flowers (corymb, Figs lh and 8). The overalleffect of the corymb is to simulate a single large regularflower. Within the corymb flowers are irregular, the twoventral petals of each flower beingmuch larger than the dorsal ones.This is presumably an adaptationthat prevents the petals from over-lapping and protruding into thecentre of the inflorescence. Thusselection may have played a role inthe coevolution of the compressedinflorescence axis and the irregular-ity of the flowers.

The coevolution of inflorescencesand flowers is further illustrated bythe mixed inflorescences of some ofthe Compositae. The inflorescenceaxis within this group is highly com-pressed to form a dense cluster offlowers (florets), termed a capitulum(Fig. li). In some species, such as

daisy or chrysanthemum it consistsof two types of florets: ray floretsthat are highly irregular and occupythe periphery and disc florets that areregular and occupy the central domeof the capitulum (Fig. 8). This isreminiscent of the cen phenotype ofAntirrhinum, where the centralterminal flower is regular and theaxillary flowers are irregular. In bothcases, the activation of genes con-trolling irregularity seems to be

restricted to a peripheral zonearound the inflorescence apex. In thecase of a daisy, this zone gives riseonly to the outermost florets whereasin Antirrhinunz it produces all of theaxillary flowers. Nevertheless, theproduction of inflorescences with a

mixture of regular and irregularflowers is not found as a wild-typecondition in plants with extendedracemes like Antirrhinum but iscommon in species with a

capitulum. This may reflect a

selective advantage of mixed inflo-rescences (such as mimicking a large

Evolution of flowers and inflorescences 1 15

flower) when the flowers are tightly clustered within a

capitulum.The similarity between cett and daisy-like inflorescences

raises the question of whether c'r'c'-like genes are involved inthe control of irregularity within the Compositae, a family thatis quite distantly related to Arttirrhinum (e.g. it lies outside thephylogeny shown in Fig. 7).lt is possible to test this by lookingfor, and analysing the expression of, c"\'('-like genes in speciesfrom this group. This analysis may be helped by exploitingmutants described in some species of the Compositae thataf'fect floret development. Mutants in species such as Chrysan-themum have been described that result in a capitulum com-prising only irregular flowers (Fig.8).These might be

explained in terms of an extension of r'\'c'-like gene activity intothe central dome. Sinrilarly, mutants that give only regularflowers rnight result from a loss of c'\'c'-like gene activity.

Fig. 8. Photographs of the coryrnb inflorescence f}om Iberi.s (top panel) and the capitulurn fromwild-type Chn,,sunthentum (bottom lefi) with outer irregular florets (white) and inner regularflorets (yellow). A mutant Chn'sctnthernunr, with only irregular florets is shown for comparison(bottom right).

1 16 E. S. Coen and J. M. Nugent

These examples illustrate the importance of studying theevolution of symmetry and inflorescence architecture together.The use of comparative molecular genetic analysis of genes

such as flo, cen and cyc should start to reveal how this coevo-lution may have occurred.

We would like to thank Rosemary Carpenter, Desmond Bradley,Pilar Cubas and Richard Olmstead for critical comments and discus-sions of the manuscript and researchers at Kew Gardens for helpfuldiscussions.

REFERENCES

Carpenter, R. and CoeD, E. S. (1990). Floral homeotic mutations produced bytransposon-mutagenesis in Antirrhinum majus. Genes Dev. 4, 1483-1493.

Coen, E. S., Romero, J. M., Doyle, S., Elliott, R., Murphy, G. andCarpenter, R. (1990). floricaula: a homeotic gene required for flowerdevelopment in Antirrhinum majus. Cell 63, 13 1 I - 1322.

Linnaeus, C. (1749). De Peloria. Diss. Uppsala:Amoenitates Acad.Olmstead, R. G, Bremer, 8., Scott, K. M. and Palmer, J. D. (1993). A

parsimony analysis of the Asteridae sensu lato based on rbcL sequences.

Ann. Missouri Bot. Gard.80, 700-722.Peyritsch, J. (1870). Uber pelorienbildungen. Sitzunger Akad. Wiss 62, 1-

27.Peyritsch, J. 0872). Uber Pelorienbildungen. Sitzunger Akad. Wiss 66, 1-

35.Schultz, E. A. and Haughn, G. 'W. (1991). LEAFY a homeotic gene that

regulates inflorescence development in Arabidopsis. Plant Cell 3, 77 1-78 1.

Stebbins, G. L. (1974). Flowering Plants, Evolution above the Species Level.

MA: Harvard University Press.

Stubbe, H. (1963). Mutanten der Kulturtomate Lycopersicon esculentumMiller IY . Die Kulturpflanze, 11,603-644.

Stubbe, H. ( I 966). Genetik und Zytologie von Antirrhinum L. sect. AntirrhinumVeb. Jena: Gustav Fischer.

Weberlitrg, F. (1989). Morphology of Flowers and Inflorescences. Cambridge:Cambridge University Press.

Weigel, D., Alvarez, J., Smyth, D. R., Yanofsky, M. F. and MeyerowitzrB.M. ( 1992). LEAFY controls floral meristem identity in Arabidopsis. Cell 69,843-859.