Plant Molecular Biology42: 115–149, 2000.© 2000Kluwer Academic Publishers. Printed in the Netherlands.

115

A short history of MADS-box genes in plants

Günter Theissen∗, Annette Becker, Alexandra Di Rosa, Akira Kanno, Jan T. Kim, ThomasMünster, Kai-Uwe Winter and Heinz SaedlerMax-Planck-Institut für Züchtungsforschung, Abteilung Molekulare Pflanzengenetik, Carl-von-Linn´e-Weg 10,50829 Köln, Germany (∗author for correspondence)

Key words:angiosperm, development, evolution, fern, gymnosperm, MADS-box gene

Abstract

Evolutionary developmental genetics (evodevotics) is a novel scientific endeavor which assumes that changesin developmental control genes are a major aspect of evolutionary changes in morphology. Understanding thephylogeny of developmental control genes may thus help us to understand the evolution of plant and animal form.The principles of evodevotics are exemplified by outlining the role of MADS-box genes in the evolution of plantreproductive structures. In extant eudicotyledonous flowering plants, MADS-box genes act as homeotic selectorgenes determining floral organ identity and as floral meristem identity genes. By reviewing current knowledgeabout MADS-box genes in ferns, gymnosperms and different types of angiosperms, we demonstrate that the phy-logeny of MADS-box genes was strongly correlated with the origin and evolution of plant reproductive structuressuch as ovules and flowers. It seems likely, therefore, that changes in MADS-box gene structure, expression andfunction have been a major cause for innovations in reproductive development during land plant evolution, such asseed, flower and fruit formation.

Introduction: on the origin of novel structuresduring evolution

We explain here what evolutionary developmental ge-netics (evodevotics) is, and how it may help us tounderstand the evolution of diversity and complex-ity in the living world. We present one of the mostimportant corollaries of evodevotics, that changes indevelopmental control genes might be a major causeof evolutionary changes in morphology.

Higher organisms such as plants and animals im-press us with their complexity and their diversity. Takeplants as an example. Every tiny weed you can find ona little walk around the corner is by far more complexthan anything we know from outside the living world,and the diversity of plants is breath-taking ranging, forexample, from huge oak trees to microscopic greenalgae on their bark. Understanding the laws of nature

The MADS homepage: http://www.mpiz-koeln.mpg.de/mads/

that have generated that diversity and complexity is atthe very heart of biology.

Initially, one can try to understand complex organ-isms from an engineer’s point of view – an attitudewhich already has quite some explanatory power. Forexample, interpreting leaves as efficient sun-collectorsexplains why these are generally flat and oriented to-wards the sun. However, functional explanations haveserious limitations in the living world. Why, for exam-ple, do the flowers of some plants have three organs(sepals, petals or tepals) in each whorl of their peri-anth (such as Liliaceae), while others have four (e.g.Brassicaceae) or five (e.g. Rosaceae), if any numberof perianth organs is able to attract pollinators effi-ciently? Why do mammals usually walk on four limbs,insects on six and spiders on eight, if any even numberof limbs allows efficient locomotion on land?

The difficulties with explanations that would sat-isfy engineers in the living world arise from the factthat all features of living organisms are a product bothof necessity and chance during evolution [77]. Some

116

aspects of living creatures merely trace back to chanceevents that became fixed during evolution and cannotbe reduced to anything more meaningful. This is oneof the reasons why living beings can be fully under-stood only from an evolutionary perspective whichtakes their unique ‘history’ into account. Unfortu-nately, it does not mean that evolutionary theory hasalready provided us with a complete understanding ofthe origin of complex and diverse structures in nature.On the contrary, understanding the mechanisms thatgenerated complex organisms, such as oak trees andgreen algae, from bacteria-like ancestors is still one ofthe greatest intellectual challenges. The origin of novelstructures or complete new body plans during evolu-tion has been especially difficult to explain. Some ofthe problems arise from the fact that the ‘classical’, i.e.Darwinian evolutionary theory is a gradualistic one,which assumes that evolution proceeds in a countlessnumber of very small steps, while, on the other hand,partial or intermediate structures might not have anadaptive value.

New ideas are needed to gain a better understand-ing of the origin of complexity and diversity in theliving world or old ones have to be revitalized. Oneof the most promising concepts in that respect is evo-lutionary developmental biology, which has stronghistorical roots reaching back into the 19th century,but is now fashionable again under the term ‘evo-devo’ [37, 41, 124]. Evo-devo assumes that there isa close interrelationship between developmental andevolutionary processes. One of the reasons for this isan astonishing feature of higher organisms: that eventhe most complex organisms are generated by devel-opmental processes that generally start with a singlecell – the fertilized egg-cell (or zygote). Diversity andcomplexity thus do not only have evolutionary originsand causes, but also developmental ones [6]. In thecase of multicellular organisms such as animals andplants, evolution of form is thus the evolution of devel-opmental processes, and any phylogenetic innovationhas to be compatible with the mode of development ina given organismic lineage. This is why developmentmay put serious constraints on evolution, which couldact both as negative forces preventing advantageousalterations as well as positive channels of preferredchange [41].

From the close interdependence of developmentand evolution, one of the most important corollariesof evo-devo can be derived, namely that changes indevelopmental control genes might be a major causeof evolutionary changes in morphology [124]. Un-

derstanding the phylogeny of developmental controlgenes is therefore an important prerequisite for un-derstanding the evolution of plant and animal form(note that we use ‘developmental control gene’ hereas a convenient term for genes which significantlycontribute to developmental processes; for a criti-cal discussion of the term, see [125], and referencestherein). One can assume that the combination ofevolutionary developmental biology with moleculargenetics will provide deep insights into the mecha-nisms behind macroevolution. Since it is the genes thatconnect evolutionary and developmental processes,this novel combination of traditionally separated bi-ological disciplines deserves a new name: evodevotics(for evolutionary developmental genetics). A verystrong molecular genetic aspect clearly distinguishesevodevotics from its historical precursors.

In recent years, it has been discovered that thekey developmental control genes are often membersof a very limited number of multigene families whichencode transcription factors. The paradigm for suchgene families are the homeobox genes [35], whichplay a key role in the specification of the animal bodyplan in both development and evolution [56, 70, 114].Many of the homeobox genes act as homeotic selectorgenes which are involved in differentiating differentbody regions from each other, probably by activat-ing or repressing different sets of downstream genes(‘target or realizator genes’) in different parts of thebody. Unfortunately, studying homeobox genes andanimals alone will not allow us to detect all of the fun-damental laws of macroevolution. All extant animalsprobably are relatively closely related members of amonophyletic group. Their body plans, though verydiverse, were generated in a relatively short periodof time about 540 million years ago (MYA) – hencethat process has been termed the ‘Cambrian explosion’[93]. In many cases, therefore, to distinguish neces-sities of macroevolutionary events from mere chanceevents that have been fixed in evolution, is impos-sible from studying only animals. For example, allanimals specify their body plan in a very similar way,by using a well defined set of homeobox genes (HOXgenes) which are organized in genomic clusters [101,114]. However, the absence ofHOX clusters in plants[73] tells us that the presence of such genes is notan absolute requirement for the evolution of complexmulticellular body plans, a conclusion that could nothave been drawn if only animal evodevotics wouldhave been studied. Therefore, to understand betterthe general rules of the macroevolution of higher or-

117

ganisms, evolutionary lineages should be comparedin which multicellular body plans originated indepen-dently from unicellular ancestors. It seems very likelythat green plants have evolved multicellular develop-ment independently from that of animals and are thusa suitable system to compare with animal development[73].

MADS-box genes and the evodevotics of the flower

We argue that the flower is an ideal model for plantevodevotics, and that the phylogeny of MADS-boxgenes may have played an important role during theorigin and evolution of flower development. We sum-marize what was known about the role of MADS-boxgenes in flower development of some genetic modelsystems – all being higher eudicotyledonous plants –before these genes were studied in a broader series ofphylogenetically informative taxa in order to test theirimportance for flower evolution.

We believe that flowers and their phylogenetic pre-cursors are an ideal model system to study the linkagebetween development, genes and evolution. Floralmorphology is the predominant source of charactersfor angiosperm taxonomy and phylogeny reconstruc-tion [26]. Accordingly, the evolution of floral formhas been studied quite extensively, although importantquestions concerning the origin and diversification offlowers have remained unanswered [21]. For the samereasons, flower development has been studied at highresolution in quite a number of different species (e.g.[28]). The most important advantages of flower evode-votics, however, are provided by genetics. A numberof flowering plant model species, such asArabidopsisthaliana(mouse-ear cress),Petunia hybrida(petunia),Nicotiana tabacum(tobacco) andOryza sativa(rice)can routinely be transformed with genes from otherspecies, so that the conservation of gene function canbe determined by transgenic technology. Moreover,flower development is one of the best understood mor-phogenetic processes of plants on the genetic level.An impressive number of studies in recent years hasculminated in the insight that inflorescence and flowerdevelopment in higher eudicotyledonous floweringplants are determined by a network of regulatory genesthat is organized in a hierarchical fashion (Figure 1)([131]; for reviews, see [88, 123–125]). Close to thetop of that hierarchy are ‘late- and early-floweringgenes’ that are triggered by environmental factors suchas day length, light quality and temperature. These

genes mediate the switch from vegetative to repro-ductive development, perhaps by activating meristemidentity genes. Meristem identity genes ‘control’ thetransition from vegetative to inflorescence and frominflorescence to floral meristems. Within floral meris-tems, cadastral genes set the boundaries of floral organidentity gene functions, thus defining the differentfloral whorls. Some intermediate genes possibly medi-ate between floral meristem and organ identity genes.Floral organ identity genes (homeotic selector genes;‘ABC genes’) specify the organ identity within eachwhorl of the flower by activating ‘realizator genes’.In a classical model, three classes of homeotic geneactivities (‘homeotic functions’) have been proposed,called A, B and C (Figures 1 and 5) [20]. Withinany one of the four flower whorls, expression of Aalone specifies sepal formation. The combination ABspecifies the development of petals, and the combi-nation BC specifies stamen formation. Expression ofthe C function alone determines the development ofcarpels. The model also proposed that the A and Cfunction genes negatively regulate each other (mean-ing that they also exert ‘cadastral’ functions) and thatthe B function is restricted to the second and thirdwhorls independently of A and C functions. In wild-type flowers, the A function is expressed in the firstand second floral whorl, the B function in the sec-ond and third whorl, and the C function in the thirdand fourth whorl. Therefore, sepals, petals, stamensand carpels are specified in whorls one, two, threeand four, respectively (for recent reviews of the ABCmodel, see [103, 124, 132]). The ABC model waslargely based on the analysis ofArabidopsismutants,albeitAntirrhinumwas also considered [20].

Although the ABC model is quite elegant, it failsto explain some complications. Mutations in B and Cfunction genes, for example, have effects in additionto homeotic changes of organ identity. Loss-of-C-function mutants form flowers with an undeterminednumber of floral organs, indicating that C functiongenes not only specify organ identity, but are also nec-essary to confer floral determinacy.Antirrhinum loss-of-B-function mutants lack the fourth floral whorl,suggesting that the B function genes not only specifysecond and third whorl organ identity, but are also nec-essary for fourth whorl formation [128]. Aside fromthat, the B and C function mutants are usually clear-cut. On the contrary, there are notorious problemswith the A function. The flowers of strong loss-of-A-function mutants ofArabidopsis, for example, oftenlack the second whorl, while weaker alleles do not

118

Figure 1. An extremely simplified and preliminary depiction of the genetic hierarchy that ‘controls’ flower development inArabidopsisthaliana. Examples of the different types of genes within each hierarchy level are boxed. MADS-box genes are shown as open squares withthick lines, non-MADS-box genes as circles, and genes whose sequence has not been reported yet as octagons. The position of the geneswas taken from the literature, as cited within this or other reviews [123–125]. Regulatory interactions between the different genes or blocksof genes are symbolized by arrows (activation), double arrows (synergistic interaction) or barred lines (inhibition, antagonistic interaction).Information about these interactions has been compiled from the review articles cited above. For a better overview, by far not all of the genesinvolved in flower development are shown (for review see [88]), and interactions (activation, repression) between the different hierarchy levelsare depicted only globally (for some interactions between individual genes, see e.g. [124]). Absence of lines or arrows between genes meansthat an interaction has not been experimentally demonstrated yet, not that it does not exist. For the downstream genes, just one symbol isshown for every type of floral organ, though whole cascades of many direct target genes and further downstream genes are probably activatedin each organ. The carpel-specific genes shown (AGLs) are only putative examples. Abbreviations used: AG, AGAMOUS; AGL1, 2, 4, 5, 9,11, 13, AGAMOUS-LIKE GENE1, 2, 4, 5, 9, 11, 13; AP1, 2, 3, APETALA1, 2, 3; BEL1, BELL1; CAL, CAULIFLOWER; CO, CONSTANS;ELF1, EARLY FLOWERING1; LD, LUMINIDEPENDENS; LFY, LEAFY; LUG, LEUNIG; NAP, NAC-LIKE, ACTIVATED BY AP3/PI; PI,PISTILLATA; SIN1, SHORT INTEGUMENTS1; SUP, SUPERMAN; UFO, UNUSUAL FLORAL ORGANS; TFL, TERMINAL FLOWER.

119

have a full homeotic conversion of floral organs. Thus,‘ideal’ mutants, in which the first- and second-whorlorgans are homeotically transformed into carpels orstamens, respectively, actually do not exist. Mutantsthat are primarily caused by a loss of the A function areonly known fromArabidopsis. Antirrhinum mutantswith a similar phenotype are due to ectopic expres-sion of a C function gene in whorls 1 and 2 of theflowers [9]. Searches for A function genes in petuniaby a candidate gene approach inspired by results fromArabidopsisalso remained negative [67], suggestingthat the A function is phylogenetically less well con-served than the B and C functions. The confusion withthe A function is a good example of problems thatbecome less enigmatic when considered in an evolu-tionary perspective. It seems that some of the problemswith defining the A function simply reflect the quiterecent and multiple origin of the floral perianth (sepalsand petals). Compared to the perianth organs, stamensand carpels (or their homologues from nonfloweringplants), which are specified by B and C function genes,are evolutionarily more ancient and robust structures(see below).

Based on studies in petunia, the ABC model wasrecently extended by a D function [4]. When ec-topically expressed, the D function genesFBP7 andFBP11from petunia induce the formation of ectopicovules on the perianth organs of transgenic flowers.They have, therefore, been defined as master controlgenes of ovules.

Arabidopsisgenes providing the three homeoticactivities A, B and C are known. The A function iscontributed by bothAPETALA1(AP1) andAPETALA2(AP2), the B function byAPETALA3(AP3) andPIS-TILLATA (PI), and the C function byAGAMOUS(AG). In Antirrhinum, the B function is provided byDEFICIENS (DEF) and GLOBOSA(GLO), the Cfunction by PLENA (PLE). D function genes havebeen mutationally defined only in petunia so far, butsequence similarity suggests that the correspondinggene inArabidopsisis AGL11(Figure 1) [4].

All these genes have been cloned. Except forAP2,all of them share a highly conserved, ca. 180 bp longDNA sequence, called the MADS-box. It encodesthe DNA-binding domain of the respective MADS-domain transcription factors ([20, 111, 115, 123, 132,137]; for recent reviews about MADS-box genes, see[103, 112, 123, 124]). MADS is an acronym for thefour founder proteins MCM1 (from brewer’s yeast,Saccharomyces cerevisiae), AGAMOUS (from Ara-bidopsis), DEFICIENS (fromAntirrhinum), and SRF

(a human protein), on which the definition of this genefamily is based [111].

Within the hierarchical gene network contribut-ing to flower development, MADS-box genes are notonly dominant among the organ identity genes, butare well represented also at other levels, i.e. the lev-els of meristem identity genes, intermediate genes,cadastral genes, and possibly even downstream genes(Figure 1). In contrast to theHOX genes of animals,which are organized in genomic clusters, the MADS-box genes of plants are scattered throughout the entireplant genomes [31, 63].

MADS-domain proteins, like many other eukary-otic transcription factors, have a modular structuralorganization [112]. In the cases of almost all knownseed plant MADS-domain proteins, it is very similar,including a MADS (M), intervening (I), keratin-like(K) and C-terminal (C) domain [66, 97, 123]. Genesencoding this type of protein hence have been termedMIKC-type MADS-box genes [85].

The MADS domain is by far the most highly con-served region of the proteins [97]. In most cases, itis found at the N-terminus of the putative proteins, al-though some plant proteins contain additional residuesN-terminal to the MADS domain (NMIKC-type pro-teins). The MADS domain is the major determinantof DNA binding, but it also performs dimerizationand accessory factor-binding functions [112]. Part ofit folds into a novel structural motif for DNA inter-action, an antiparallel coiled coil ofα-helices thatlies flat on the DNA minor groove [91]. In line withthe conserved nature of their DNA-binding domain,MADS-domain proteins bind to similar DNA sitesbased on the consensus sequence CC(A/T)6GG, whichis called a CArG box (CC-A-rich-GG). CArG boxesare present in the promoter regions of many genesthat are probably regulated by MADS-box genes [112,127].

The I domain, directly downstream of the MADSdomain, comprises ca. 30 amino acids, but is some-what variable in length [66, 85]. It is only relativelyweakly conserved among plant MADS-domain pro-teins [97]. In someArabidopsisMADS-domain pro-teins, it was shown that the I domain constitutes akey molecular determinant for the selective forma-tion of DNA-binding dimers [103]. The K domain,which is not present in any of the animal and fungalMADS-domain proteins known so far [123, 124], ischaracterized by a conserved, regular spacing of hy-drophobic residues, which is proposed to allow forthe formation of an amphipathic helix. It is assumed

120

that such an amphipathic helix interacts with thatof another K domain-containing protein to promotedimerization [103, 112]. The most variable region,both in sequence and length, is the C domain at the Cterminus of the MADS-domain proteins. The functionof this domain is unknown, and it has been shown to bedispensable for DNA binding and protein dimerizationin at least some floral homeotic MADS-domain pro-teins (see, for example, [139]). The C domain could beinvolved in transcriptional activation or the formationof multimeric transcription factor complexes.

According to the reasoning of evodevotics, under-standing the origin and evolution of flower develop-ment depends on an understanding of the origin andevolution of the gene network governing flower de-velopment. Changes in gene number, expression andinteraction thus all could have contributed to the evo-lution of flowers. Since MADS-box genes play suchan important role in the network of flower devel-opment, understanding the phylogeny of MADS-boxgenes might strongly improve our understanding offlower evolution.

Phylogeny reconstructions disclosed that theMADS-box gene family is composed of several de-fined gene clades [26, 85, 97, 123, 124]. Most clademembers share highly related functions and similar ex-pression patterns. For example, the MADS-box genesproviding the floral homeotic functions A, B and Ceach fall into separate clades, namelySQUAMOSA-like (A function),DEFICIENS- or GLOBOSA-like (Bfunction), andAGAMOUS-like genes (C function) [26,97, 123] (for the rules to name MADS-box gene cladesused here, see [123]). The D function genes deter-mining ovule identity [4] also belong to the clade ofAGAMOUS-like genes [123]. Therefore, the establish-ment of the mentioned gene clades was probably animportant event towards the establishment of the flo-ral homeotic functions [123]. Thus the question arisesas to when these gene clades arose during evolutionand how some of their members were transformedinto floral homeotic genes. To answer this, MADS-boxgenes have to be studied in phylogenetically informa-tive taxa. Initially, plant MADS-box genes had beeninvestigated only in a very limited taxonomic range,i.e. the genetic model plants, which are all higher eu-dicots. Meanwhile, however, the situation has changedconsiderably: whileArabidopsisand the like are stillthe favorites of hard-core developmental biologists,quite a number of scientists with evolutionary or agro-nomic interests have started to take plant diversityinto account. In the following sections, we will out-

line what we have learned recently about MADS-boxgenes in non-flowering plants, basal angiosperms, andmonocots. Then we briefly describe some new insightsobtained from the eudicots. We will use these datato reconstruct the evolution of the MADS-box genefamily and its relationship to floral evolution, i.e. wewill tell a short natural history of MADS-box genes inplants. First, however, we will briefly speculate aboutthe origin of plant MADS-box genes.

On the origin and major subdivisions of theMADS-box gene family

We briefly describe what is known about MADS-boxgenes in animals and fungi, and report that homo-logues of MADS-box genes may even exist in bacteria.The MADS-box gene family proper of eukaryotes canbe subdivided into three major clades. Representa-tives of two of these clades (ARG80- and MEF2-likegenes) have only been found in animals and fungi sofar, whereas members of the third group (MIKC-typegenes) seem to be restricted to plants.

The origin of the MADS-box gene family is un-clear. Some bacterial proteins, such as members of theUspA family of stress response proteins known fromEscherichia coliandHaemophilus influenzae, containshort sequence stretches that could be homologousto a part of the MADS domain [87]. However, se-quence similarity between the bacterial proteins andthe MADS domains is so low that special strategiesof sequence database search were needed to detectit. Anyhow, it seems likely that a precursor of theMADS-type DNA-binding domain evolved before theseparation of bacterial and eukaryotic lineages [87]about 2–3.5 billion years ago [69]. Interestingly, acoiled-coil structure is predicted in the downstreamportion of UspA-like proteins [87]. Since the K do-main of plant MADS-domain proteins is also assumedto adopt a coiled-coil structure [66, 112], even the Kdomain may have bacterial roots.

Since MADS-box genes have been found in extantplants, animals and fungi, it is quite safe to assumethat the last common ancestor of these eukaryotic taxa,which existed about one billion years ago, had al-ready at least one gene with a true MADS box [123].The MADS-box gene family can be subdivided intothree major clades,ARG80-like genes (also called the‘SRFgene family’),MEF2-like genes and MIKC-typegenes [45, 85, 123, 124]. WhileARG80- andMEF2-like genes have been found only in animals and fungi

121

so far, MIKC-type genes seem to be restricted to plants(Figure 2). The presence ofARG80- andMEF2-likegenes could represent a synapomorphy of animals andfungi, separating these taxa from plants (Figure 2).However, due to the limited sampling of MADS-boxgenes in any taxon one cannot exclude that, for exam-ple, ARG80- or MEF2-like genes are also present inplants. Moreover, although the hypothesis that animalsand fungi are more closely related to each other thanboth are to plants is supported by quite a number ofmolecular data, there is also evidence for alternativerelationships [130]. The picture drawn in Figure 2 isthus possibly not the last word on this subject.

MADS-box genes in animals and fungi are in-volved in a diverse range of biological activities (re-viewed in [112, 123]). A common denominator ofmost MADS-domain proteins is that they control as-pects of development or cell differentiation. Let ustake theARG80-like genes as an example, which in-cludeARG80andMCM1 from brewer’s yeast and theSRFgenes from animals. WhileARG80is involvedin regulating genes encoding arginine-metabolizingenzymes,MCM1 is involved in a broader range offunctions: in cooperation with different associatedfactors it represses or activates the transcription ofmany genes involved in diverse aspects of the yeastcell cycle and cell growth, metabolism (including thatof arginine) and specialization. The role ofMCM1in the determination of yeast cell type is especiallywell known [112, 123]. Recently, putative orthologuesof MCM1 have also been reported from distant fun-gal relatives of brewer’s yeast, i.e. the fission yeastSchizosaccharomyces pombe(MAP1 gene) and thesmut fungusUstilago maydis(UMC1gene) [61, 136].The SRF (serum response factor) of vertebrates is in-volved in immediate-early gene and muscle-specificgene transcription. Its orthologue fromDrosophila(DSRF) plays a role in tracheal development (reviewedin [112, 123]).

Members of the clade ofMEF2-like genes arekey components in muscle-specific gene regulation inanimals [90], but probably also have functions in non-muscle cells. For more details about animal and fungalMADS-box genes we refer to other reviews on thistopic and the original work cited therein [45, 112,123].

Somewhere in the lineage that led to extant greenplants, MADS-box genes appeared in which theMADS-box was followed by the I-, K- and C-regions,and the MIKC-type genes were born. The molecularmechanism that generated them is unknown. It seems

that extant MIKC-type genes are more closely relatedto MEF2-like genes than toARG80-like genes, imply-ing that the last common ancestor of MIKC-type geneswas moreMEF2- thanARG80-like [123]. Molecularclock analyses and studies on MADS-box genes inferns have helped recently to get better estimates aboutthe time interval in the past when the first MIKC-typegenes appeared, as described below.

MADS-box genes in ferns

We summarize data suggesting that the last commonancestor of ferns and seed plants about 400 MYAcontained at least two different MIKC-type MADS-box genes that were homologues, but not orthologues,of floral homeotic genes. These genes probably hadexpression patterns and functions that were moregeneral than those of the highly specialized floralhomeotic genes from extant flowering plants.

After colonization of land, roughly about 500MYA, land plants (today comprising liverworts, horn-worts, mosses and vascular plants) evolved bodystructures of increasing complexity [57]. Extant vas-cular plants, for example, range from relativelysimple clubmosses (lycopsids), horsetails (equisetop-sids), whisk ferns (Psilotaceae) and ferns (filicopsids)to complex seed plants (spermatopsids), compris-ing gymnosperms and angiosperms (flowering plantssensu stricto) [57]. Although MADS-box gene cDNAshave already been isolated from a moss (MIKC-type;our unpublished data) and a clubmoss (see the citationin [3]), the most basal plants from which MADS-box gene sequences have been published so far areferns [22, 46, 58, 85]. Among the land plants fernsare of considerable scientific interest because theyare very likely the sister group of the seed plants.The two groups diverged about 400 MYA [36, 117].Ferns have several characteristics that are primitivewith respect to vascular plants as a whole [7]. Forexample, they produce naked sporangia at the abax-ial sides of their leaves which lack accessory organssuch as integuments. Ferns thus do not form ovules orseeds, and generally they also do not aggregate theirsporophylls into flower-like structures. Most ferns arehomosporous, i.e. their sporangia produce only onetype of haploid reproductive spores, starting fromdiploid spore mother cells that undergo meiosis. Incontrast to the megaspores of seed plants, the sporesof ferns are shed, and the haploid gametophytes de-veloping from them are entirely independent of the

122

Figure 2. The major clades of MADS-box genes in the evolution of life. A phylogenetic tree of some major taxa of living organisms is shown.The ages (in BYA, billion years ago) given at some nodes of the tree are very rough estimates. Some aspects of the topology of the tree arecontroversial, for example, that fungi are more closely related to animals than to plants (see text). At some internal branches of the tree simplifieddomain structures of representative members of the three major clades of MADS-domain proteins are shown (ARG80- and MEF2-like proteins,MIKC-type proteins); in addition, a representative of a group of putative distant relatives of MADS-domain proteins, i.e. UspA-like proteins,is shown at the eubacterial branch. ‘MADS’, ‘I’, ‘K’ and ‘C’ denote the MADS-, I-, K- and C-domains, respectively. ‘S’ stands for the SAMdomain, present inSRF,ARG80 andMCM1. ‘AD’ symbolizes the presence of a domain with sequence similarity to a part of the MADS domainwithin the UspA-like proteins. The different gene types have been established during the time interval represented by the respective branches ofthe phylogenetic tree, at the latest. This could be concluded from the presence of respective clade members in extant taxa. For example,MEF2-andARG80-like genes have been isolated from animals and fungi so far, but not from plants.

spore-producing plant (the sporophyte). On the game-tophyte, sexual organs (archegonia and antheridia) areformed that produce egg and sperm cells, respectively.Fertilization results in a diploid zygote which developsinto a new sporophytic generation.

The characterization of MADS-box genes in fernshas focused so far onCeratopterisbecause it has somefeatures that qualify it as a plant model system.Cer-atopteris richardii, for example, has a short sexual lifecycle of less than 120 days. Moreover, it behaves likea diploid species and is well suited for genetic anddevelopmental analyses [14].

cDNAS representing more than 15 different ge-nomic loci containing a MADS box have already beenisolated by three different research groups [22, 46, 58,85]. Unfortunately, these groups used three differentsystems of gene nomenclature, which resulted in upto three different names for the same gene. In the fol-lowing section, we always use the gene name that hasbeen published first, but we will also mention synony-

mous names where these exist to facilitate comparisonbetween the different studies.

In one study, cDNAs of 12 different ge-nomic loci, designatedCRM1–CRM12 (for Cer-atopteris MADS1–12) were isolated fromCer-atopteris richardii, Ceratopteris pteroides, or both[22, 85]. CRM8, however, had been published ear-lier under the name ofCERMADS5[58], so weadopt that name here.CERMADS5was later alsocalledCMADS2[46]. Southern blot analysis indicatedthat CRM1–CRM10represent single-copy loci in thegenome ofCeratopteris richardii[22, 85]. cDNAs ofthree additional genes, termedCMADS1, CMADS4andCERMADS3, have been isolated in two other stud-ies [46, 58]. Most cDNAs ofCeratopterisMADS-boxgenes isolated so far show high sequence similarityto typical seed plant MADS-box genes with respectto MADS-domain sequence and overall domain struc-ture, i.e. they can be classified as MIKC-type genes.There is no indication that domain shuffling occurred

123

within the genealogy of these genes. The similar-ity between fern and seed plant MADS-box genesclearly indicates that these genes share a commonancestor from which they were derived by gene du-plications, sequence diversification and fixation [85].The fern genes identified are thus clearly homologuesof the MADS-type floral homeotic genes known fromangiosperms.

To determine the evolutionary relationships be-tween the fern genes and the other known MADS-boxgenes, phylogeny reconstructions were carried out.They disclosed that the genes fromCeratopteriscon-stitute three different gene clades, termedCRM1-,CRM3- andCRM6-like genes, which are interspersedamong seed plant gene clades [46, 85]. TheCRM6-like genes can be further subdivided intoCRM6-likegenessensu strictoandCRM7-like genes (Figure 3).In some phylogenetic trees, monophyly of theCRM6-like genes is not well supported (Figure 3), in someothers theCRM6-like genessensu strictoeven ap-pear separated from theCRM7-like genes [58]. In afew other gene trees, however, even theCRM1- andCRM6-like genes form sister clades [85]. A conserv-ative interpretation of all available data thus leads tothe conclusion that at least two different MIKC-typeMADS-box genes existed already in the last commonancestor of ferns and seed plants [85]. It seems morelikely, however, that at least three or four differentMIKC-type genes were already present in this species.On the other hand, it is obvious from the analysescarried out so far that many of the gene duplicationswhich led to the large number of present-day MIKC-type genes occurred independently in the lineages thatled to extant ferns and seed plants [85]. Although theMADS-box genes fromCeratopteriscan be consid-ered being homologous to the MIKC-type genes fromother plants, including the floral homeotic genes, theyare clearly not orthologues of specific floral homeoticgenes. It seems likely, therefore, that the last commonancestor of ferns and seed plants contained only a rela-tively small number of MIKC-type genes compared tothe large number of genes present in extant seed plantsand ferns [85]. Alternative scenarios are conceivable,but appear far less parsimonious.

Molecular clock estimates suggest that MIKC-typegenes started to diverge about 450–500 MYA, i.e. be-fore the separation of the ferns and the seed plants[96]. The presence of at least two different MIKC-typegenes in the last common ancestor of ferns and seedplants about 400 MYA is in good agreement with thisestimation. Accordingly, it seems likely that the last

common ancestor of MIKC-type genes existed dur-ing the Ordovician, when plants probably started tocolonize the land [96]. Therefore, MIKC-type MADS-box genes probably had already been established inplants more basal than ferns. Cloning of a MIKC-typecDNA from the mossPhyscomitrella patenssupportsthis hypothesis (our unpublished data).

The presence of a short peptide motif at the C-terminal end of the respective proteins suggests a closerelationship between theCRM3-like genes (compris-ing CRM3, also calledCMADS6[46], andCRM9upto now), and theDEF/GLO-like genes [60, and ourunpublished results]. Based on the presence of a N-terminal extension in the derived proteins, a closerelationship betweenCRM6/7-like genes, includingCRM6 (also calledCERMADS2), CERMADS3andCMADS1, and the members of theAG clade has alsobeen postulated [46]. However, we consider the re-spective evidences as weak, since they are based onthe presence of small peptide sequences of limitedsequence similarities. They thus do not necessar-ily define synapomorphies, but also could representhomoplasies (i.e. the recurrences of similarities inevolution). Using phylogeny reconstructions, clear sis-ter group relationships between fern and seed plantMADS-box gene clades have not been established yet.

Since orthologues of floral homeotic genes havenot been isolated so far fromCeratopteris, the ques-tion arises whether such genes actually exist inthis taxon. MADS-box gene cDNA cloning has in-volved three independent research groups and differ-ent cloning techniques. Diverse phases of the fern lifecycle and different plant tissues were used as mRNAsources. Moreover, probes and primers for cloningexperiments were derived, at least in some cases,from Arabidopsisand Antirrhinum floral homeoticgenes. However, the three research groups foundonly CRM1-, CRM3-, CRM6- and CRM7-like genesin quite a redundant fashion [22, 46, 58, 85]. Al-though the possibility remains that orthologues offloral homeotic genes are present inCeratopteris, thisappears less and less likely.

We wanted to verify that the apparent absenceof floral homeotic gene orthologues is not merelya specific feature ofCeratopterisor its close rela-tives. Therefore, we applied cDNA cloning also toOphioglossum, another fern which is only very dis-tantly related toCeratopteris. While Ceratopterisis ahighly derived leptosporangiate fern, the Ophioglos-sales are eusporangiate ferns which branch off nearthe base of the fern tree [95]. cDNAs representing

124

125

126

four different genes could be isolated so far, termedOPM1andOPM3–OPM5(Ophioglossum pedunculo-sumMADS1, 3–5) ([85], and our unpublished data).However, it turned out again that these genes are notmembers of any of the gene clades known from seedplants. WhileOPM3 and OPM4 do not fit into anyclade defined so far, the otherOPM genes seem tobe CRM6- or CRM7-like genes, respectively (Fig-ure 3) [85]. Although bootstrap support for this kindof grouping is often not very high, it suggests thatCRM6- andCRM7-like genes were established at anearly time point in fern evolution (Figure 4). Takentogether, there is no evidence so far that orthologuesof floral homeotic genes (such asSQUA-, DEF-, GLO-or AG-like genes) are present in extant ferns. The factthat no genes from ferns have been isolated that arewithin the gene clades known from seed plants, mightbe correlated to the absence of seed or flowering plantspecific structures, such as ovules, carpels, stamens orfloral perianth organs.

Molecular clock estimates suggested that the lastcommon ancestor of the clade comprisingAGL2-,AGL6- andSQUA-like genes (also termed ‘AP1/AGL9clade’) existed about 370 MYA [96], i.e. after thelineage that led to extant seed plants had already sep-arated from the lineage that led to present-day ferns.This estimation is in agreement with the fact that nodistinctive AGL2-, AGL6- or SQUA-like genes havebeen found in ferns so far, while members of theAGL2and AGL6 clades could be cloned from both gym-nosperms and angiosperms, two seed plant lineageswhich separated about 300 MYA (see below).

Unfortunately, phylogeny reconstructions did notgive specific clues to fern MADS-box gene function,and mutants or transgenic plants in which the ex-pression of these genes is changed are also not yetavailable. Accordingly, the expression of severalCer-atopterisgenes was determined by northern andin situhybridizations to get some idea about their function. Itturned out that most genes are well expressed in thegametophytic as well as the sporophytic phase of thefern life cycle [22, 46, 85]. Exceptions areCRM9andCMADS1, which are much more strongly expressedin the sporophyte than in the gametophyte [22, 46].Exclusive expression in hermaphroditic gametophyteswas reported forCRM3in one study (termedCMADS6there) [46], but this result is controversial, because inother studiesCRM3expression was also observed insporophytes and male gametophytes [22, 85]. Prelim-inary data indicate that in male gametophytes expres-sion of CRM3 is in spermatides that develop within

antheridia. In hermaphroditic gametophytes,CRM3expression was detected in meristematic cells [22].Expression analysis in the sporophyte revealed thatquite a number of genes are expressed in many tissues[22, 46]. An exception isCMADS4, which is predom-inantly expressed in roots [46]. Expression ofCRM3andCRM9, for example, was found in the shoot axis aswell as in fronds of juvenile plants. In cross-sections offertile fronds, expression ofCRM3, CRM6andCRM9was observed, withCRM6expression being relativelystrong in sporangia [22].CMADS1expression was ob-served in the shoot apical meristem, leaf primordia andthe procambium [46]. As the leaves increase in cellnumber,CMADS1signals become stronger in all cellsat the top of the leaf. As tissue systems differentiate,CMADS1expression gradually becomes restricted tothree leaf parts: procambium, sporangium initials, andthe regions that will give rise to the lamina, or pinnae.Signals are also observed in differentiated vascularbundles of the petiole, and in the root apical meristemsand their associated provascular cell files.CMADS1expression can also be observed in developing sporan-gia, but not in the sporangia containing mature spores[46]. The expression patterns ofCRM1 (also cal-ledCerMADS4or CMADS3[46, 58]) andCerMADS5(for synonymous names, see above) are very similar tothose of CMADS1, albeit weaker [46].

The expression of most fern genes in both majorphases of the life cycle is in remarkable contrast tothe situation in seed plants, where expression of aMIKC-type gene in the gametophytic phase has beendemonstrated to date only in a single case, theAGL17-like geneDEFH125fromAntirrhinum[140], althoughmany MADS-box genes are expressed in stamens,carpels or ovules. Expression in both sporophytes andgametophytes suggests a more ubiquitous function ofthe fern genes in the control of development or celldifferentiation than the temporally and spatially quiterestricted functions of the homeotic genes determiningfloral organ identity of angiosperms.

MADS-box genes with a relatively ubiquitous ex-pression in the sporophytic phase do also exist inseed plants. Examples are most members of the cladeof TM3-like genes andAGL3, an AGL2-like gene(reviewed in [123]). As indicated above, a closerelationship between the fern gene clades and theTM3- or AGL2-like genes from seed plants cannotbe demonstrated. However, the organs in which theorgan identity genes of seed plants are specificallyexpressed were very likely not present in the lastcommon ancestor of ferns and seed plants. It seems

127

plausible, therefore, that the rather ubiquitous ex-pression of most fern MADS-box genes and of someMADS genes from seed plants represents the ances-tral state of MIKC-type gene expression. The highlyorgan-specific expression of the floral homeotic genesof angiosperms is thus very likely a derived conditionthat was achieved during the processes in which someMIKC-type genes were recruited as floral homeoticgenes. If so, spatiotemporal restriction of gene expres-sion was an important aspect during the co-option ofMADS-box genes as homeotic selector genes of spe-cialized plant organs. This gene recruitment must haveoccurred in the lineage that led to seed plants after thelineage that led to extant ferns had already branchedoff. It has been speculated that the restriction ofMADS-box gene expression may have been caused bythe evolution of other genes that regulate the MADS-box genes, such as relatives ofLEAFY or CURLYLEAF [46]. However, these changes in expression pat-terns could also have been caused by mutations incis-regulatory elements controlling MADS-box geneexpression [124]. In some precedent cases, concerninganthocyanin biosynthesis and growth form in maize,the molecular basis of evolutionary changes in geneexpression in plants has been clarified recently. Inthese cases it turned out thatcis-regulatory elements,not trans-acting factors, were responsible for changesin gene expression (examples cited in [6]). It has evenbeen argued that modifications in thecis-regulatory re-gions of transcriptional regulators represent a predom-inant mode for the evolution of novel plant forms [23].Besidestrans-acting factors, evolutionary changes inMADS-box gene promoters should therefore be seri-ously considered as a possible cause for the changes inMADS-box gene expression during evolution.

Besides the rather ubiquitous spatiotemporal ex-pression of most genes, severalCeratopterisMADS-box genes also display some other features that areatypical of seed plant MADS-box genes. For exam-ple, there is evidence that the primary transcripts ofa relatively large fraction of genes, includingCRM1,CRM4 (also calledCerMADS1), CRM6 (also calledCerMADS2) andCRM9are alternatively spliced [22,58]. For comparison, although more than 150 differ-ent MIKC-type genes have been reported so far fromseed plants (Figure 3), alternative splicing has beenreported only in a single case [62]. However, alterna-tive splicing is typical ofMEF2-like MADS-box genesfrom animals (for reviews, see [90, 123]) and has alsobeen documented in cases of some transposon-likeelements containing a MADS box which have been

isolated from the flowering plant, maize (see below)[31, 78, 79]. Alternative splicing, therefore, may rep-resent an ancient mechanism to increase the diversityof protein products from individual MADS-box genesthat has been reduced in seed plants. One should notbe too surprised, however, if alternative splicing playsa more important role in seed plants than currentlythought.

Another unusual feature of some fern MADS-boxgenes concerns their structure. While the majorityof fern cDNAs have the potential to encode perfect(N)MIKC-type proteins, cDNAs of several other loci,includingCRM11, CRM12and CMADS5, also showhigh sequence similarity to MADS-box gene cDNAs,but do not contain continuous open reading frames,due to the presence of in-frame stop codons or nu-cleotide insertions or deletions [22, 46]. Whether therespective genomic loci have a function is unclear. Inprinciple, they could encode truncated proteins thatwork as transcriptional modulators. They even mayencode full-length proteins generated by programmedframeshifting (ribosome hopping). Also a functionapart from the protein level is conceivable. Alterna-tively, these loci may simply represent nonfunctionalpseudogenes that got into the vicinity of promotersand are therefore transcribed. In mammalian genomes,nonfunctional pseudogenes are often created throughreverse transcription of mRNA and integration of thecopy DNA into the genome. A similar mechanismmight work in ferns. This hypothesis is supported bythe fact that, in contrast toCRM1–10, Southern hy-bridizations revealed severalCRM12-like loci in theCeratopterisgenome even under high-stringency hy-bridization conditions [22]. Analysis of genomic locisuch asCRM11andCRM12might give further cluesto their origin. It is interesting to note that our ob-servations are not unprecedented. It has been reportedthat the majority of genomic clones of homologues tothe chlorophylla/b-binding (CAB) protein that havebeen isolated from the homosporous fernPolystichummunitumare defective. A major cause is, again, thepresence of in-frame stop codons and nucleotide inser-tions or deletions [94]. Whether the probably defectiveCABgenes are transcribed has not been reported. Oneof the explanations for theCAB gene defects is genesilencing upon polyploidization [94]. However, sincewe found multiple copies for only a minority ofCer-atopteris MADS-box genes, this does not seem tobe a likely explanation for the structurally aberrantMADS-box gene cDNAs reported here.

128

129

Figure 4 (top). MADS-box genes in the evolution of vascular plants. A phylogenetic tree of major taxa of vascular plants is shown. The ages(in MYA, million years ago) given at some nodes of the tree are rough (and in part controversial) estimates based on different studies. Thetopology of the tree is also controversial: that Gnetales are more closely related to conifers than to angiosperms could be concluded frommolecular data [15, 39, 133, 134], but is in contrast to widely accepted interpretations of morphological data. At the left side of the root andsome internal branches of the tree three important stages in the evolution of the megasporangium are schematically depicted. From bottom totop: a sporangium that is not covered by an integument, a condition still found in extant ferns; a sporangium that is covered by an integument(ovule); and a sporangium that, in addition, is surrounded by a carpel. The gene names besides the branches denote gene subfamilies, not singlegenes. These have been established during the time interval represented by the respective branches of the phylogenetic tree, at the latest. Thiscould be concluded from the presence of respective subfamily members in extant taxa. For example,AG-, AGL2-, AGL6-, DEF/GLO-, GGM13-,STMADS11- andTM3-like genes have been isolated from angiosperms and gymnosperms, but not from ferns. At the root of the phylogenetictree, the domain structure of a typical MIKC-type MADS-box gene is shown. Our analyses have demonstrated that the last common ancestorof ferns and seed plants already had at least two genes of that type [85]. Abbreviations of genes or gene subfamilies:AG, AGAMOUS; AGL2,6, 12, 15, 17, AGAMOUS-like gene 2, 6, 12, 15, 17;CRM1, 3, 6, 7, CeratopterisMADS-box gene 1, 3, 6, 7;DEF, DEFICIENS; DEF/GLO,a precursor of bothDEF- andGLO-like genes;GGM4-7, 10, 13, Gnetum gnemonMADS-box gene 4–7, 10, 13;GLO, GLOBOSA; OPM3,4, Ophioglossum pedunculosumMADS-box gene 3, 4;SQUA, SQUAMOSA; STMADS11, Solanum tuberosumMADS-box gene 11;TM3, 8,tomato MADS-box gene 3, 8.Figure 5(bottom). How the land plants learned the floral ABC. Different states of the ABCD model (some of them hypothetical) of floral organspecification are plotted onto a phylogenetic tree of major taxa of vascular plants. The organs specified by the different homeotic functions areindicated above the models. At the branch leading to the angiosperms (eudicots, monocots and basal angiosperms), different ancestral versionsof the ABCD model that might have been present at the base of the angiosperms are shown. These versions have been suggested (from left toright) in this work, or in [6] or [3], respectively. The ages (in MYA, i.e. million years ago) given at some nodes of the tree are rough estimates (asin Figure 4). At the right side of some internal branches of the tree, gene subfamilies, not individual genes, are indicated (e.g., ‘SQUA’ meansthe clade ofSQUA-like genes). The relationships between representatives of these gene subfamilies and homeotic functions are symbolized byarrows. For example, aSQUA-like gene (i.e.AP1) provides the A function inArabidopsis. (Note thatSQUAitself is not an A function gene!).A DEF- and aGLO-like gene possibly provide the B function in all angiosperms, whileAG-like genes provide both the C and the D function.The different relationships have been established during the time interval represented by the respective branches of the phylogenetic tree, at thelatest. Abbreviations used: A, B, C, D, the floral homeotic functions;AG, AG-like genes; C/D, a precursor of floral homeotic functions C and D;DEF, DEF-like genes;DEF/GLO, a precursor of bothDEF- andGLO-like genes; FM, function in the specification of floral meristems;GLO,GLO-like genes; IM, function in the specification of inflorescence meristems;SQUA, SQUA-like genes.

A third unusual observation was made with theintracellular localization ofCRM9mRNA. In all tis-sues whereCRM9 expression was detected,in situhybridization studies gave a strong signal in the nu-cleus, while in the cytoplasm, hybridization signalswere much lower, if present at all [22]. Thus it seemsthat the majority ofCRM9 mRNA is retained in thenucleus and cannot be translated. It could be, there-fore, that formation of CRM9 protein is not (only)regulated transcriptionally, but (also) by nuclear ex-port of CRM9mRNA. It could also be, however, thatCRM9represents a nonfunctional gene, or thatCRM9does not function at the protein level. Whether nuclearexport is linked to alternative splicing is unknown sofar. It is conceivable, for example, that not all of thedifferent splice variants can be exported, implying thatalternative splicing would control nuclear export.

Nuclear retention of mRNA is also not unprece-dented in ferns. It has been reported that phytochrome(PHY1) mRNA in the fernAdiantum capillus-venerisis predominantly nuclear in location in light-grownyoung leaves (croziers), while the mRNA in dark-grown tissue appears uniformly in both nucleus andcytoplasm [89]. These findings support the view thatferns have included nuclear export of mRNA into theirrepertoire of gene regulation.

Finally, let us gather the facts and try to recon-struct the last common ancestor of extant ferns andseed plants. Very likely, it had no ovules or floral or-gans but, like extant ferns, had naked sporangia and anindependent gametophytic generation. It already hadmore than one MIKC-type MADS-box gene, but prob-ably fewer than extant ferns or seed plants. None ofthe MIKC-type genes was an orthologue of a specificfloral homeotic gene. These genes probably had quitea ubiquitous expression during the life cycle of theplant, possibly involving the gametophytic as well asthe sporophytic phase. It seems likely that these geneswere not organ identity genes, but had more generalroles in the transcriptional control of development orcell differentiation, i.e. more comparable to the role ofMCM1 in the life of yeast.

MADS-box genes in gymnosperms

We summarize data suggesting that the last commonancestor of extant gymnosperms and angiosperms,about 300 MYA, already had at least 7 differ-ent MADS-box genes, i.e.AG-, AGL2-, AGL6-,DEF/GLO-, GGM13-, STMADS11- and TM3-likegenes. Probably, most of these genes were already

130

involved in specifying reproductive organs, such asovules, in the sporophyte. Expression of an ancestralversion of the homeotic C and D functions, provided byanAG-like gene, was probably used to distinguish re-productive from non-reproductive (vegetative) organs.In addition, expression of an ancestral B function, pro-vided by a basalDEF/GLO-like gene, was possiblyused to distinguish between male and female repro-ductive organs. Thus, orthologues of floral homeoticgenes and a precursor of the ABCD system of flo-ral organ specification (BC/D system) had probablyalready been established at the evolutionary base ofextant seed plants.

The term ‘gymnosperm’ (meaning ’naked seed’)indicates that we are now dealing with plants thatdevelop seeds, but in which these seeds are not en-closed within a carpel as in angiosperms (see below).Gymnosperms and angiosperms together constitutethe taxonomic group of seed plants (spermatopsids,spermatophytes). Seed plants have become the mostsuccessful land plants, probably because of the se-lective advantage the formation of seeds gives theseplants over all others [117]. Most likely the reason isthat seeds are unrivaled in their capacity to dispersethe next generation. Seeds are just ripened ovules,and ovules can be defined as integumented indehiscentmegasporangia [117]. They consist of an envelope,the integument(s), with a micropyle, and a megaspo-rangium (the nucellus) inside of which a megagame-tophyte develops. There is evidence that seed plantsevolved from gymnosperm-like plants with a fern-like mode of reproduction called progymnosperms [7].Therefore, the pollen sacs and nucelli of seed plantsare probably homologous to fern sporangia. The tran-sition from the naked dehiscent sporangia of fern-likeancesters to ovules characterizes one of the most im-portant steps in land plant evolution (Figure 4). Itinvolved several key innovations, such as the evolutionof heterospory [117]. According to molecular data,the last common ancestor of extant seed plants existedabout 300 MYA – recent estimations range from 285to 348 MYA [40, 108] –, and earliest fossil evidenceof gymnosperms dates back about 350–365 MYA [7,121]. Gymnosperms are, therefore, phylogeneticallymuch older than angiosperms (see below).

Extant gymnosperms comprise four groups: coni-fers, gnetophytes, cycads andGinkgo. Only a fewMADS-box gene cDNAs have been isolated from cy-cads andGinkgoso far (Figure 3; and our unpublisheddata), since the focus of MADS-box gene research ingymnosperms has been on conifers (due to their eco-

logical and commercial importance) and gnetophytes(because they are often considered a sister group of theangiosperms).

Conifer MADS-box gene cDNAs have been re-ported from spruce (Picea abies, Picea mariana) andpine species (Pinus radiata, Pinus resinosa) [64, 82–84, 105, 116, 119]. Phylogeny reconstructions re-vealed that the genes for which full-length cDNAshave been obtained so far all fall into gene cladeswell known from angiosperms, namelyAG-, AGL2-,AGL6-, DEF/GLO- andTM3-like genes [81, 84, 116,119, 123, 124, 134] (see also Figure 3). However, PCRcloning of a 61 bp segment using degenerate primerstargeted to the MADS box suggested the presence ofover 27 MADS-box genes within black spruce (Piceamariana), including several for which no orthologousangiosperm MADS-box gene has been identified yet[105].

In contrast to many angiosperm flowers, which arehermaphroditic, the investigated conifers are monoe-cious species that have truly unisexual reproductiveaxes. The female strobili (or seed cones) are com-pound axes consisting of two-scaled units with asterile bract and a seed-bearing (ovuliferous) scale.In contrast, the male strobili are simple structurescomposed only of microsporophylls [119]. Expressionstudies indicated that the MADS-box genes identi-fied so far are transcribed in male and female strobili.Some are also expressed in vegetative organs, such asthe AGL6-like genePRMADS3from Monterey pine(Pinus radiata), which is also transcribed in needleprimordia [81]. The transcripts of theTM3-like geneDAL3 from Norway spruce (Picea abies) were alsofound in vegetative shoots, but not in embryos, seedsor seedlings [119]. TheAGL6-like geneDAL1 is alsoexpressed in vegetative shoots in their first year of de-velopment, but not in the epicotyl, including the apicalmeristem, of the seedling [118]. Byin situ hybridiza-tion, PRMADS1–3from Monterey pine were found tobe expressed in groups of cells that form ovuliferousscale and microsporophyll primordia [81]. Similarly,expression of theAG-like gene,DAL2, from Norwayspruce was detected in ovuliferous scales, but not inbracts, the cone axis or the apical meristem [120]. Ex-pression of its orthologue from black spruce,SAG1,was found to be very similar in female cones [105].In male cones,SAG1expression was detected at alow level in the tissue that makes up the tapetal layer[105]. These data suggest thatAG-, AGL2- andAGL6-like genes of conifers are all involved in reproductivestructure formation.

131

To test whether the structural similarity and phy-logenetic relatedness betweenAG from the floweringplant ArabidopsisandDAL2 from Norway spruce iscoupled to a similarity in function, an analysis of theeffect of DAL2 expression under the control of theconstitutive 35S promoter in transgenicArabidopsisplants was made [120]. Most transformants showedphenotypic alterations that seem not very informativewith respect toDAL2 function, such as curled rosetteleaves and early flowering. Some transformants, how-ever, had homeotic changes of flower organ identity. Inthese plants, the sepals had gained female charactersat the margins, such as ovule-like structures and papil-lary cells characteristic of stigma. Petals had obtainedmale characteristics: they appeared to be transformedinto filamentous organs or stamen-like organs witha filament-like proximal part capped with an anther-like structure. The third- and fourth-whorl organswere mainly unaffected by expression of the transgene[120]. The transformants thus resembleArabidopsisplants ectopically expressingAGorthologues from an-giosperms, such asAG itself or BAG1 from Brassicanapus[68, 75]. Since the ABC model predicts that theC function antagonizes the A function, the observedphenotype can be expected in case of a loss of the Afunction or an ectopic expression of the C function inthe first and second whorl of theArabidopsisflower.The results obtained withAG andBAG1have demon-strated thatAG or its close relatives are sufficient toprovide ectopically the homeotic C function. The sim-plest explanation for the results withDAL2, therefore,would be thatDAL2 activity in the perianth organscan functionally substitute forAG activity in ectopicexpression experiments. This functional substitutionwould imply several partial functions, i.e. suppressingA gene activity, directing carpel identity to the outer-most whorl, and interacting with B class genes (AP3,PI) in directing stamen identity to the second whorl oforgans in transgenic flowers. However, it could alsobe that expression ofDAL2 results in an extension ofAG expression into the perianth whorls, for example,becauseDAL2 protein is able to activate theAG pro-moter, or becauseDAL2 turns off theArabidopsisAfunction. If so, the homeotic transformation of whorl1 and whorl 2 organs would be the result of ectopicAG expression, or of the formation of functional AG-DAL2 heterodimers rather thanDAL2 alone. In eithercase, the data indicate that DAL2 is able to interactwith components of the regulatory context ofAG, andthat thus these kinds of interactions have been con-served over at least 300 million years (the logic of such

conclusions has also been illustrated elsewhere [124]).It might be that MADS, I and K domains of DAL2 areneeded for these interactions, explaining why these areso similar to those of AG. However, complementationof anAG loss-of-function mutant with theDAL2 genecould provide a more stringent test for the extent towhichDAL2 is able to substitute theAGfunction in theArabidopsiscontext. Results very similar to the onesreported here forDAL2 have also been obtained withits black spruce orthologue,SAG1[105].

By definition, C class genes are involved in spec-ifying stamen and carpel identity. Since there areno stamens and carpels in gymnosperms, the ques-tion arises as to which functionDAL2/SAG1 ful-fills in the conifer context. Note that even success-ful heterologous transformation studies, as describedabove, may not always answer such questions! SpruceDAL2/SAG1mutants that could give an answer arealso not available. Expression studies suggest thatDAL2/SAG1is involved in the determination of ovulif-erous scale or ovule identity, and of male reproductiveorgan identity. The ability to convert petals into sta-mens inArabidopsisis consistent with the notion thatDAL2/SAG1might be able to interact with B classgenes in specifying male reproductive organs. Thepresence ofDEF/GLO-like genes in conifers could bepredicted from phylogeny reconstructions [120], butis now also supported by gene cloning (see below).ThusDAL2/SAG1might interact with one or severalDEF/GLO-like genes from spruce in order to specifymale reproductive organ identity. Expression studiesand transgenic experiments both suggest, therefore,that DAL2/SAG1function is more similar to that ofangiosperm C function than D function genes (whoseexpression is restricted to ovules, implying that theyare not expressed in male reproductive organs, andwhose function is in specifying ovule identity). At firstglance, this may seem a paradox, since ovules, in con-trast to carpels, are present in all gymnosperms andare thus very likely phylogenetically older. Specifyingovules (i.e. D function), therefore, should be a moreancient function ofAG-like genes than specifying sta-mens and carpels (i.e. C function). One has to takeinto consideration, however, thatDAL2/SAG1functionmight be ancestral to both C and D functions. So howcan the early evolution ofAG-like gene function inseed plants be conceived? Only one type ofAG-likegene has been isolated so far from any gymnospermspecies (Figure 3). Phylogeny reconstructions sug-gest that these genes are basal to both the C- andD-function genes from angiosperms (Figure 3). We

132

suggest, therefore, that it could have been the ancestralfunction of these genes to distinguish reproductive or-gans such as male sporophylls and ovuliferous scales,including ovules (where expression is on) from vege-tative organs, including cone bracts (where expressionof these genes is off). Genes such asDAL2/SAG1maystill provide such a function today. Later, at the levelof angiosperms, a gene duplication and diversifica-tion event might have resulted in the fixation of twodifferent genes. While one gene type (C class genes;AG-like genessensu stricto) specialized in specify-ing stamens and carpels, the other (D class genes;FBP7/FBP11/AGL11-like genes) became restricted tospecify ovule identity (Figure 5).

Gnetophytes (Gnetales) are an enigmatic group ofseed plants with only three genera,Gnetum, EphedraandWelwitschia. Most phylogenetic analyses of mor-phological data agree that among the groups of extantseed plants, the gnetophytes are the sister group ofthe angiosperms [21, 24, 25]. According to this view,angiosperms and gnetophytes are members of a cladecalled ‘anthophytes’, to emphasize their shared pos-session of flower-like reproductive structures [21].Since answers to the still unresolved question of an-giosperm origin are intimately connected to the identi-fication of their sister group among extinct and extanttaxa [21, 39], gnetophytes have found much scientificinterest. However, some recent phylogeny reconstruc-tions based on molecular data do not support an antho-phyte clade; instead, they favor monophyly of extantgymnosperms, albeit with low bootstrap support, im-plying that gnetophytes are more closely related toconifers than to angiosperms [15, 39].

cDNA sequences of 13 different single-copyMADS-box genes of the gnetophyteGnetum gnemonhave been published so far ([133, 134]). Phylogenyreconstructions indicated that seven of them are mem-bers of novel gene subfamilies, for which membersfrom dicots have not been published so far (see Fig-ure 3). In one case (GGM13), however, a highlyrelated sequence has been isolated recently from amonocotyledonous flowering plant (our unpublishedresults). Due to the limited knowledge about the num-ber and type of MADS-box genes in any plant species(includingArabidopsis) it remains to be seen if ortho-logues of the other genes are also present in floweringplants. Alternatively, the respective gene clades orig-inated within the gymnosperms after the lineage thatled to the angiosperms had already branched off. Mostof the genes are expressed in male and/or femalestrobili, but not in leaves, suggesting that they have

functions similar to the floral meristem or organ iden-tity genes of angiosperms ([133], and our unpublishedresults).

Phylogeny reconstructions revealed that the othersix genes (GGM1, 2, 3, 9, 11, 12) fall into well de-fined gene clades known already from angiosperms,i.e. STMADS11[13], TM3-, DEF/GLO-, AG-, orAGL6-like genes, respectively ([134], and our unpub-lished data). They are thus putative orthologues of therespective genes from angiosperms. Among them isGGM2, the first DEF/GLO-like gene (B class geneorthologue) reported from a gymnosperm [133, 134](Figure 3). The presence of aDEF/GLO-like gene,however, is not a synapomorphy uniting floweringplants and gnetophytes, since genes belonging to thatclade have meanwhile also been found in two coniferspecies, Norway spruce and Monterey pine [84, 116](Figure 3). Whether these genes are more closely re-lated toDEF- or GLO-like genes, or are basal to both(as suggested by Figure 3), could not be clarified un-equivocally by the construction of phylogenetic genetrees so far (our unpublished results). Analysis of theexon-intron structure ofGGM2, however, supportsthe latter hypothesis (our unpublished results). At thistime, therefore, we favor the hypothesis that there wasonly oneDEF/GLO-like gene in the last common an-cestor of extant gymnosperms and angiosperms. Thegene duplication that generated distinctDEF andGLOclades may have happened in the angiosperm lineageafter the lineage that led to extant gymnosperms hadalready branched off (Figures 4 and 5). A close rela-tionship betweenGGM2and the other members of theDEF/GLO clade is not only supported by phylogenyreconstruction, but also by the presence of a ‘paleoAP3 motif’ at the C-terminal end of the GGM2 proteinand a ‘derived PI motif’ in a subterminal position (ourunpublished results; for the definition of the motifs,see [60]). Moreover, in sequence alignments GGM2shares a highly specific character state at an indel(insertion-deletion) position with all other DEF- andGLO-like proteins (our unpublished data). However,both features have also been found forGGM13 (ourunpublished results), which according to phylogenyreconstructions is a slightly more distant relative ofDEF- andGLO-like genes (Figure 3).

Analogous to the observations withDAL2/SAG1,expression of theAG-like geneGGM3 was found inmale as well as female strobili ofGnetum, but notin leaves. TheTM3-like geneGGM1 showed, as ex-pected, a more ubiquitous expression in male andfemale strobili and in leaves. However, expression

133

of the DEF/GLO-like geneGGM2 was found to berestricted to male strobili [133, 134]. It could havebeen an ancient function of members of that geneclade, therefore, to distinguish between male (wheregene expression is on) and female reproductive struc-tures (where expression is off) (see Figure 5). It iseasy to imagine how the floral homeotic B functionof angiosperms evolved from such a gene, since italso distinguishes male reproductive organs (i.e. sta-mens, expressing B plus C function) from femaleones (i.e. carpels, expressing only C function). It alsoseems plausible that this gene function was recruitedto specify petals (expressing A and B function) whenthese were ’derived’ from stamens in some lineages ofangiosperms (see below) [3, 60].

Phylogeny reconstructions indicated that in allcases where gene subfamily members are availablefrom angiosperms, gnetophytes and conifers, i.e.within the AG-, AGL6-, DEF/GLO- and TM3-likegenes, the genes fromGnetumalways form subcladestogether with conifer genes, to the exclusion of theangiosperm genes (Figure 3). This finding providesmolecular evidence for the hypothesis that gneto-phytes are more closely related to conifers than to an-giosperms (Figure 4). The conclusion is in contradic-tion to the anthophyte theory and to widely acceptedinterpretations of morphological data for almost a cen-tury [5, 21, 24]. The sister group relationship betweengnetophytes and conifers makes it likely that manyof the angiosperm-like features of Gnetales, such asthe flower-like appearance of reproductive structures,reduced female gametophytes, double-integumentedovules, dicotyledonous seeds, vessels in the secondarywood, net-veined leaves and the presence of double-fertilization, are homoplasies rather than homologouscharacter states. With respect to angiosperm origins,gnetophytes are thus possibly less informative than of-ten thought [21, 24, 25]. It could be, however, thatthe parallel appearance of the mentioned charactersin angiosperms and gnetophytes was facilitated bya common developmental potential that was alreadypresent in the last common ancestor of (gnetophytes+ conifers) and angiosperms (or even of all extantseed plants, if extant gymnosperms represent a mono-phyletic group [15]). It seems an exciting hypothesisthat a set of MADS-box genes might have been partof the developmental potential facilitating convergentevolution in different seed plant lineages. Therefore,this last common ancestor is of considerable evolu-tionary interest, so let us try to reconstruct it withrespect to some morphological features and MADS-

box genes, taking together the data reviewed above. Tosimplify things, we assume that extant gymnospermsare really a monophyletic group, and that gene typesthat have been found in angiosperms as well as ingnetophytes or conifers were thus present in the lastcommon ancestor of all extant seed plants.

Like ferns, the most recent common ancestor of ex-tant seed plants probably had an elaborate two-phaselife cycle with a dominating sporophytic generation.In contrast to most ferns, however, the sporophyteproduced two types of spores, micro- and megas-pores, and the megagametophytes developing fromthe megaspores were not independent, but remainedwithin the ovules of the sporophyte. After fertiliza-tion, the ovules developed into seeds. The sporophyteperhaps had unisexual reproductive axes. Figure 3 anddata published elsewhere [134] show that there arefive different well-defined clades containing MADS-box gene members from both gymnosperms and an-giosperms, indicating that at least five different MIKC-type genes existed in the last common ancestor ofcontemporary seed plants, namely at least one repre-sentative of each of the clades ofAG-, AGL2-, AGL6-,DEF/GLO- andTM3-like genes. In addition, there wasmost likely a sixth gene closely related toGGM13,and a seventh gene closely related toGGM12, be-cause putative orthologues for these genes also wereisolated from angiosperm species ([13], and our un-published data). Probably most of these genes werealready involved in specifying reproductive organs ofthe sporophyte. The last common ancestor of extantseed plants probably used an ancestral version of thehomeotic C and D functions (C/D function), pro-vided by anAG-like gene, to distinguish reproductivefrom non-reproductive organs. In addition, it possiblyused an ancestral B function provided by at least oneDEF/GLO-like gene to distinguish between male andfemale reproductive organs. Thus a precursor of theABCD system of floral organ specification had proba-bly been established already as a BC/D system at thebase of extant seed plants, while it was completelyabsent in the last common ancestor of ferns and seedplants (Figure 5). The data on MADS-box genes inferns suggest that there was a relatively small pool ofMIKC-type genes in the last common ancestor of fernsand seed plants. Therefore it is likely that descendantsof that pool of genes were generated by gene duplica-tion, diversification and fixation, and were recruited inthe lineage leading to seed plants to give rise to floralhomeotic genes. It is conceivable, therefore, that inthe time interval prior to the radiation of extant seed

134

plants, but subsequent to their divergence from fern-like ancestors, i.e. between 300 and 400 MYA, someif not most clades of MADS-box genes known fromangiosperms had been established (Figure 4). There isa striking temporal coincidence between the appear-ance of these genes and the occurrence of seed plantsand the seed habit. For example, the oldest knownseed plant (Elkinsia) has been preserved in the fos-sil record of that time interval (Late Devonian, about365 MYA), and different intermediate stages in theevolution of the ovule have been found in the fossilrecord of the Lower Carboniferous, about 350 MYA[121]. We are not aware that such a clear coincidencebetween the appearance of new types (clades) of de-velopmental control genes (such asAG-like genes)and the appearance of novel morphological structures(such as ovules and seeds) has ever been reported forthe macroevolution of a non-plant system. Since theextant descendants of these genes are expressed inovules, ovuliferous scales, or seeds, and thus prob-ably are involved in controlling the development ofthese structures, it seems quite possible that the es-tablishment of the new clades of MADS-box genes atthe time of ovule and seed ‘invention’ was not just acoincidence, but an important functional step in theevolutionary establishment of these structures.

Progymnosperms, i.e. plants that already hadgymnospermous wood but still a pteridophytic, free-sporing mode of reproduction, also existed in thatcritical time interval 300–400 MYA, since their fossilshave been found from Middle Devonian to Early Car-boniferous (Tournaisian) [7]. It is intriguing to think,therefore, that during this time the establishment ofAG-like genes in progymnosperms might have beenan important aspect to confer ovules to plants that stillhad a pteridophytic mode of reproduction, but other-wise were already gymnosperm-like [85]. The progen-itor of extant seed plants, established at this time, wasthe starting point for the evolution of the enormousmorphological diversity we see in present-day seedplants.

Due to the large morphological gaps between thedifferent seed plant groups (extant and fossil), ho-mologies between their reproductive structures areoften difficult to assess [24]. This is especially truefor the floral organs of angiosperms compared to theorgans of the reproductive units of the gymnosperms.It is one of the reasons why definite answers to thequestion of what a flower actually is and from whichorgans of which gymnosperms its organs were derivedhave been lacking (for a review, see [21]). However,

since homologous organs should generally express or-thologous developmental control genes, we have goodreasons to assume that MADS-box genes are suitabletools to test assumptions about structural and develop-mental homologies among the reproductive structureswithin the diverse seed plant groups [24]. For example,some evolutionary models suggest that angiospermpetals are homologous to the outer integument ofGne-tum reproductive units [24]. If so, orthologues of Bfunction genes such asGGM2 should be expressedin the outer integument ofGnetum, which exists inmale as well as female strobili. However,GGM2is notexpressed in female strobili at all [133, 134].GGM2expression in male strobili is also not in the integu-ments surrounding the antherophores, but only in theantherophore itself [134]. Expression of theAG-likegeneGGM3 in the Gnetumouter integuments [133,134] makes it also appear unlikely that they are homol-ogous to perianth organs of angiosperms, but would becompatible with an alternative model due to which theouter integument ofGnetumis homologous to the in-tegument of angiosperm ovules [24] or even to carpels.In line with this, SAG1, one of the conifer ortho-logues ofGGM3, is especially strongly expressed inthe integuments of the ovules [105].

For several reasons, however, these conclusions arestill preliminary. For example, orthology between re-spective genes from gymnosperms and angiospermsshould be tested more rigorously, and independentco-option (recruitment) of genes into nonhomologousdevelopmental processes cannot be excluded, so thatmore genes should be analyzed (for discussion of thatproblem, see [1]). However, we believe that the strongcorrelation between MADS-box gene phylogeny andthe evolution of certain morphological structures (e.g.ovules) promises that studies such as the ones in-dicated here will help to clarify the origin of theflower.

It has often been argued that there are insuperablemorphological gaps between angiosperms and gym-nosperms which are even more difficult to overcomethan the gap between ferns and seed plants. Withrespect to MADS-box genes and the system of re-productive organ specification, we obviously see theopposite: while there are probably no orthologues offloral homeotic genes in ferns, there are clearly somein gymnosperms (Figures 3, 4 and 5). At the levelof molecular developmental control, the reproductiveunits of gymnosperms are thus more similar to theflowers of angiosperms than morphological studiesmay have suggested.

135

MADS-box genes in basal angiosperms

Basal angiosperms are crucial for our understand-ing of flower origin. Although we do not know yethow the ‘first flower’ looked, we are quite sure thatthe last common ancestor of extant angiosperms al-ready had at least 9 different MADS-box genes. Thesewere distinct representatives of the clades ofDEF- andGLO-like genes, anAGL15-like gene, and the set ofgenes that was already present in the last common an-cestor of extant seed plants (besides aDEF/GLO-likegene,AG-, AGL2-, AGL6-, GGM13-, STMADS11-andTM3-like genes).

Our considerations have now reached the floweringplants. Since flowers are often defined as short, spe-cialized axes bearing closely aggregated sporophylls,gymnosperms and even some pteridophytes (such asclubmosses) may also produce ‘flowers’. It is neces-sary to clarify, therefore, that when we use the term‘flowering plants’ within this review, we mean the an-giosperms (i.e. flowering plantssensu stricto). Theterm angiosperm means ‘vessel seed’. Besides sta-mens with two pairs of pollen sacs, the most usefuldiagnostic morphological feature of angiosperms is acarpel enclosing the ovule/seed [21]. The carpel is themorphological basis for fruit development. From thenaked sporangia of ferns via the integumented mega-sporangia (ovules) of gymnosperms (resulting inseeds) to the angiosperm ovules enclosed in carpels(resulting in seeds within fruits) we see a clearmacroevolutionary tendency to cover the megaspo-rangium and its derivatives (see Figure 4).

The angiosperm mode of reproduction has provenvery succesful, because flowering plants now domi-nate the vegetation of most ecosystems on land, andthey consist of more species than all other groups ofland plants combined (about 250 000–300 000) [21].One probable reason for the angiosperms’ success isthat fruits provide additional possibilities for an effec-tive distribution of seeds, for example by the help ofanimals. In many cases animals are also important foroutcrossing during sexual reproduction. The capacityto outcross effectively is the second major advantageof the angiosperms. It is facilitated by flower types thatefficiently attract diverse pollinators (bees, beetles,birds, etc.), depending on the angiosperm species.

The sudden appearance and considerable diversi-fication of the angiosperms within the fossil recordof the Early Cretaceous, about 130–90 MYA, seemsstill almost the same ‘abominable mystery’ as it wasto Charles Darwin more than a century ago. As al-

ready mentioned in the section on gymnosperms, theorigin of the flower has also remained a mystery.Homologies between organs within gymnosperm andangiosperm reproductive units are unclear, and thelong-standing question of whether angiosperm flowersderive from a simple branch or from multiple branches(euanthial vs. pseudanthial scenario) is still unresolved[21]. We have noticed, however, that according toconsiderations outlined by Doyle [24], the prelimi-nary expression data of theGnetumgenesGGM2andGGM3 (see the gymnosperm section) suggest organhomologies that fit to a pseudanthial rather than aneuanthial model of flower origin [134].

Current hypotheses of angiosperm evolution haveidentified two large clades (monocots and eudicots,see below) embedded within a poorly defined basalassemblage of magnoliid dicots (Magnoliidae) [21],which we call ‘basal angiosperms’ here. There is agreat diversity of floral structure and biology amongbasal angiosperms. Both large, multiparted bisexualflowers and small, simple, frequently unisexual flow-ers are widespread, and variation in the number andarrangement of floral parts is extreme [21]. This,and the substantial morphological gap between gym-nosperms and angiosperms (see above), has preventedidentification of the basic condition of the angiospermflower. Did the ‘first flower’ more look like aMagnoliaflower with its numerous elaborate tepals, or like oneof Sarcandra glabrawith a single bract, stamen andcarpel [21]?

Our inability to reconstruct the ‘first flower’ im-plies that we do not know the succession of steps inthe evolution of the molecular ‘control’ of flower for-mation. How did the BC/D system of reproductiveorgan specification possibly present in gymnospermschange into the ABCD model of floral organ identity?However, educated guesses about plausible interme-diate steps and the implications for MADS-box genephylogeny can be made (Figure 5). A most primitiveflower might just have been composed of one or morestamens and carpels (including ovules) without a peri-anth, such as the flowers ofSarcandra. We only needB, C and D function genes expressed in a suitable com-binatorial way along a single reproductive shoot axisto specify the respective organs. Perianthless flow-ers are not prominent among the different suggestionsof what the earliest flower might have looked like.However, since the identity of the organs of peri-anthless flowers could be completely specified withhomeotic functions that were possibly present alreadyin gymnosperms (Figure 5), we argue that such simple

136

flowers should be seriously considered as a plausi-ble model for the ‘first flower’. In fact, such flowersmight help to bridge the enormous gap between gym-nosperm and angiosperm reproductive structures (seethe gymnosperm section).

The more convential models assume that the mostancestral flower already had a perianth. One hypothe-sis suggests that the ancestral condition was a single,petaloid whorl expressing both A- and B-functiongenes (Figure 5) [3]. According to this hypothesis, thecalyx whorl, expressing only the A function, was lateradded externally to protect flower buds from preda-tion. Other ancestral ABCD models assume that thebasal flower had one or more sepaloid perianth whorlsspecified by A-function genes. Petals, and thus the dis-tinction between corolla and calyx, could have evolvedlater by the outward extension of B function into theinner of two perianth whorls (Figure 5) [6].

These models provoke several other questions:how were sepals or petals generated? Where did theA function come from? It is widely accepted that,in contrast to the reproductive organs (stamens andcarpels), which evolved only once, sterile perianthorgans originated several times independently withinthe angiosperms, although details are unresolved ([29,60], and references therein). Similar organs, such aspetals, are thus not necessarily homologous (in themeaning of ‘derived from a common ancestor’, i.e.historically orthologous, as defined elsewhere [3]). Ithas been concluded from morphological evidence thatpetals have been derived many times independentlyfrom stamens, for example, several times within thelower eudicots and at least once at the base of thehigher eudicots. Such petals are called andropetals.Among the basal angiosperms, the Nymphaeales prob-ably have andropetals. A second type of petals,bracteopetals, may have been derived from sepals orsterile organs subtending the flowers. Most basal an-giosperms are assumed to have bracteopetals, such asthe Magnoliales, Piperales and Aristolochiales ([60],and references therein). As outlined elsewhere, how-ever, there are severe conceptional problems withthese simple views [3]. For example, historical, po-sitional and process homology (the latter meaning thattwo structures are specified by the same type of genes)should be distinguished. Moreover, with respect to ho-mology, it would be more appropriate to distinguishbetween orthology and paralogy (for a detailed dis-cussion of this topic in floral development, see [3]).If not stated otherwise, we use the terms homology,orthology and parology here only in their historical

dimension, meaning, for example, that a feature is ho-mologous in different lineages if it was already presentin the last common ancestor of these lineages.

The origin of the A function is another notoriousproblem. It seems that there are different kinds ofgenes behind it even within the higher eudicots (seebelow), so a simple and general answer might notbe possible. We have noted, however, that in caseof Arabidopsis, the two A-function genes also func-tion as floral meristem identity genes. Determiningfloral meristem identity might be a function that isneeded earlier than the specification of floral organidentity. This is surely true for ontogeny, but mightalso be true in the case of evolution, because theearliest flowers did not necessarily have a perianthspecified by A-function genes (see above), but verylikely already needed some floral meristem identityfunction to distinguish floral from vegetative tissue.Therefore, we suggest that at least in some cases, theA function could be a derivative of the function deter-mining floral meristem identity (Figure 5). In line withthis, A-function genes or their orthologues from otherspecies (such asSQUAMOSAfrom Antirrhinum) areoften expressed in several whorls of the flower and innon-floral organs, not just in sepals and petals [50, 52].

It is obvious that studies onSQUA-, DEF-, GLO-andAG-like MADS-box genes in basal angiospermswith different floral structures (simple and complexones) may help to understand the evolution of theABCD model. It will be interesting to examine, forexample, how the independent derivations of petalsduring angiosperm evolution are reflected in the useof organ identity genes. Have independently derivedpetals always recruitedDEF- andGLO-like genes tospecify petal identity? Or have other types of genestaken over that function in lower angiosperms? Ifthe latter is true,DEF- and GLO-like genes are notnecessarily expressed in the petals of some lower an-giosperms! Petal specification genes which are notDEF- or GLO-like genes may seem more likely in thecase of bracteopetals than in the case of andropetals,because bracts or sepals do not expressDEF- andGLO-like genes, while stamens do.

The origin of the firstSQUA-like gene is an-other open question. Did it appear already withinearly angiosperms before the lineages that led to ex-tant monocots and eudicots separated, or even at thegymnosperm level? How was it derived – by gene du-plication from anAGL2/AGL6/SQUAancestral gene,as phylogeny reconstructions may suggest (Figure 3)[85, 97, 123]? DoSQUA-like genes also specify sepals

137

or petals in species other thanArabidopsis(e.g., inbasal angiosperms), or was the A function a role ac-quired later bySQUA-like genes in the lineage that ledto Arabidopsis?

During the course of a study devoted to the role ofDEF- and GLO-like genes in petal and stamen evo-lution, Kramer et al. [60] have isolated respectivecDNA clones not only from higher and lower eudi-cots, but also from the MagnoliaceaeMichelia figoandLiriodendron tulipifera, and the PiperaceaePeper-omia hirtaandPiper magnificum, which are all basalangiosperms. It was during the course of this workthat the already mentioned GLO and DEF specificmotifs were detected (see the fern and gymnospermsections). The study by Krameret al. documents ahigh frequency of gene duplications within both theDEF andGLO lineages. The authors assume that atthe base of theDEF/GLO lineage was aCRM3-likefern gene containing a paleoAP3 motif [60], but sup-port for this hypothesis by phylogeny reconstructionsis weak at best (see the fern section). There is hardlyany doubt, however, that aDEF/GLO-like gene en-coding a terminal paleoAP3 motif and a subterminalPI motif was already established when gymnospermsstarted to diverge (see the gymnosperm section). A keyancestral gene duplication occurred near the base ofthe angiosperms, resulting in the distinct lineages ofDEF-like genes (which retained a highly conservedpaleoAP3 motif, while the PI motif diverged morestrongly) andGLO-like genes (which have lost the pa-leoAP3 motif, but maintained a highly conserved PImotif). A second major duplication event occurred inthe DEF lineage near the base of the higher eudicotradiation. It resulted in a euAP3 lineage (includingAP3 and DEF) in which the euAP3 motif replacesthe paleoAP3 motif, and aTM6 lineage, in which thepaleoAP3 motif is maintained. This duplication eventmay reflect the origin of a petal-specificDEF functionin the higher eudicot lineage at the time when theseplants recruited petals from stamens [60]. Many otherindependent gene duplication events in the differentangiosperm lineages followed the creation of the sep-arateDEF, GLOandTM6 clades, resulting in pairs ofhighly related paralogues within several species suchas lily and rice (see Figure 3).

The functions of the different DEF and GLO spe-cific motifs have not yet been reported, nor have theexpression patterns of theDEF- andGLO-like genesfrom basal angiosperms been published. With thesegenes in hand, however, the tools are available nowto test some of the hypotheses outlined above.

AGL15 is an interesting type of MADS-box genewhich is expressed in developing embryos and thusmight be involved in ‘controlling’ embryogenesis [47,104]. Embryos are formed by all land plants – hencethey are also called embryophytes –, making it con-ceivable thatAGL15-like genes may even exist innonseed plants. However, evidence for that is miss-ing: AGL15-like genes have been published so far onlyfrom Brassicaceae species (Figure 3). The isolation ofanAGL15-like gene from the basal angiospermMag-nolia (Hilde Fischer, personal communication) at leastsuggests that this type of gene was already present nearthe base of the flowering plants (Figure 4).

Summing up, we do not know what the first flowerlooked like, but it probably employed a genetic sys-tem for floral reproductive organ specification thathad already been established at the gymnosperm level(BC/D system). It is relatively clear that the first flow-ering plant already had at least nine different MADS-box genes (seven gene types known already from thelast common ancestor of extant seed plants, plus anAGL15-like gene and separate lineages ofDEF- andGLO-like genes; Figure 4). It is very likely that someof these genes provided part of the molecular basisfor the enormous diversification of the flower structureduring angiosperm evolution.

MADS-box genes in monocots

Monocots comprise taxa with quite different inflores-cences and flowers, such as grasses and lilies. Somedata suggest that the B and C functions in these flowerswork quite similarly to those of the eudicots. However,the B function in grasses specifies lodicules ratherthan petals in the second whorl of the flower. In liliesand their close relatives the B function is very likelynot only expressed in the second and third, but alsoin the first floral whorl. Therefore, lilies and their rel-atives have a simple perianth (perigon) composed oftwo whorls containing petaloid organs called tepals.In addition to the MADS-box genes that were alreadypresent in the last common ancestor of extant an-giosperms, the last common ancestor of monocots andeudicots about 200 MYA also already had anAGL17-and aSQUA-like gene. Thus it contained at least 11different MADS-box genes.

Among the angiosperms, the monocotyledons (Lil-iopsida) are defined as a monophyletic group by theirsingle cotyledon and some other features [21]. Ac-cording to molecular estimates, the monocot lineage

138

separated from the other angiosperms about 160–200MYA [40, 135]. However, the oldest known fossils ofmonocots were deposited just about 90 MYA [33].

Monocots are of great interest here for at least tworeasons: the structural diversity of their flowers and in-florescences, and the commercial importance of theirflowers, seeds and fruits. The importance of mono-cots for human culture could be one of the reasonswhy MADS-box genes have been studied in quite anumber of diverse species, including cereal grassessuch as maize (Zea maysssp. mays), rice (Oryzasativa), wheat (Triticum aestivum), and sorghum(Sorghum bicolor), a lily species (Lilium regale), atulip species (Tulipa gesneriana), asparagus (Aspara-gus officinalis), and an orchid species (Aranda× deb-orah). Compared to our knowledge about MADS-boxgenes in nonflowering plants, a great deal is knownabout these genes in monocots, so that a comprehen-sive review is not possible here. Therefore, we justgive an overview of present knowledge, with a focuson recent progress.

Although the first MADS-box gene cDNA re-ported from a monocot was from an orchid [65], themajority of MADS research focuses now on the ce-reals maize and rice [10–12, 17, 18, 31, 32, 42,53–55, 71, 72, 110, 126]. These two species not onlyfeed the world to a large extent, but are also suitablemodel systems for plant genetics and molecular biol-ogy. Comparatively little is known about MADS-boxgenes in sorghum [42] and wheat [86].

Like other typical grasses, cereals produce tiny,wind-pollinated flowers that are distinct from the flow-ers of other taxa. Although the flowers themselves aresimplified and small, they are generally assembled intocomplex higher-order structures (spikelets, inflores-cences). Let us take maize as an example. In contrastto rice, which forms hermaphroditic flowers, maize isa monoecious species, i.e. it generates male and fe-male inflorescences separately on the same plant. Themale inflorescence (tassel) develops in a terminal po-sition, whereas the female inflorescences (ears) growin the axils of vegetative leaves. The unisexual flowertypes of the tassel and ear are both derived from aninitially bisexual state through the abortion of pistilprimordia in the tassel and stamen primordia in theear [16]. The three stamens or carpels (pistil) of eachmaize flower are surrounded by a pair of bract-likeorgans called palea (inner) and lemma (outer), thusconstituting structures called florets. In the flowers ofthe tassel lodicules are also formed, two knob-like pe-rianth organs which are needed to open the florets at

anthesis. Two florets – an upper and a lower one – aretogether enclosed by another pair of bract-like organscalled glumes, thus forming spikelets. In the spikeletsof female inflorescences, only the upper floret devel-ops due to abortion of the lower floret tissues at earlydevelopmental stages. In the spikelets of male inflo-rescences, both florets develop to maturity. Spikeletsare formed in pairs along the ear and tassel inflores-cence, with one spikelet being pedicellate, the othersessile. The inflorescences and flowers of maize andother grasses are thus in some respect similar, in othersdifferent from the flowers of other taxa (for reviewsabout floret, spikelet and inflorescence structures incereals, see e.g. [16, 19, 109]). It is an interestingquestion, therefore, how these similarities and dissim-ilarities are reflected in the structure, expression andfunction of MADS-box genes.

Since the stamens, carpels and ovules of mono-cots and eudicots are probably homologous organs,it seems likely that they are specified by orthologousB, C and D class homeotic genes, i.e.DEF-, GLO-and AG-like MADS-box genes. Concerning C classgenes, phylogeny reconstructions suggest an ortholo-gous relationship betweenZAG1/ZMM2 from maizeon the one hand andAG fromArabidopsison the other[110, 126]. Employing a reverse genetics approach, aputative null allele ofZAG1was identified [71]. Sur-prisingly, the floral phenotype did not show a homeotictransformation of reproductive organs into nonrepro-ductive ones, which would have been expected from aC function gene. Rather, supernumerary carpels wereobserved, indicating a loss of floral meristem deter-minacy. Besides specifying stamen and carpel identity(C function),AGalso plays a role in establishing floralmeristem determinacy. PossiblyZAG1has only the lat-ter aspect of theAG function. SinceZAG1is expressedin stamen and carpel primordia, however, it seemsmore likely that there is a redundancy in C functionbetweenZAG1 and ZMM2 [71, 109] or otherAG-like genes (see below). Expression patterns suggestthatZAG1 is more important for carpel development,while ZMM2should be more important for stamen de-velopment.ZMM2 mutants andZAG1/ZMM2 doublemutants will show whether these hypotheses are cor-rect. However, even this will probably not be the endof the story. It is well known that, due to the segmentalallotetraploid origin of the maize genome, genes thatare single copy in diploid plant species are representedby a pair of genes in maize [34, 126]. While most genepairs in maize trace back to diversification events thatstarted either about 11 or 21 MYA, theZAG1/ZMM2

139

gene pair is much more ancient (60 MYA) [34]. Itcould well be, therefore, thatZAG1andZMM2 haveduplicate loci in the maize genome, and that evenmore than two genes are involved in providing theC function and floral meristem determinacy. A pu-tative duplicate locus ofZMM2 was cloned recently(our unpublished results). In phylogeny reconstruc-tions this gene forms a clade together withZMM2andOSMADS3from rice, to the exclusion ofZAG1,suggesting an ancestrally orthologous relationship be-tween the gene pair constituted byZMM2 and itsduplicate locus, andOSMADS3(our unpublished re-sults). Interestingly, ectopic expression ofOSMADS3in tobacco (Nicotiana tabacum) driven by the CaMV35S promoter resulted in phenotypic alterations mim-icking the results of ectopic expression ofAG [55],indicating thatOSMADS3can substitute a C functiongene in transgenic experiments (for a more thoroughinterpretation of these kinds of experiments, see thegymnosperm section). It seems reasonable to hypoth-esize that not onlyOSMADS3, but alsoZMM2 and itsduplicate locus represent C function genes in grasses.

Thanks to the cloning of theSILKY1gene, con-siderable progress was also made in understanding theB function in maize [109]. Thesilky1 mutant has aphenotype strikingly similar to the B function mu-tants of eudicots. Loss of B function inArabidopsisor Antirrhinum leads to a conversion of second-whorlpetals into sepals and third-whorl stamens into carpels.In the silky1 mutant of maize, the stamens in thetassel develop as carpel-like structures, and the lod-icules are replaced by palea-like organs. In the ear of asilky1mutant, the stamens are converted to carpel-likestructures, and the normal program of stamen abor-tion is bypassed [109].SILKY1is expressed in organprimordia which give rise to lodicules and stamens.According to sequence analysis,SILKY1 is probablyan orthologue of the B class geneDEF [109]. Thedata suggest that with respect to the B function, theABCD model can be applied to grass species (Fig-ure 5). Moreover, they also suggest that lodiculesare homologous to eudicot petals and that the paleais homologous to eudicot sepals (and is not a pro-phyll, as other interpretations have it). It could also be,however, that orthologous genes have been recruitedindependently for the specification of organs that arehistorically not orthologous (in the terminology usedby Albertet al. [3]).

The mutant phenotype ofOSMADS4, a GLO-likegene from rice, is very similar to that ofsilky1 [54],suggesting that, as in higher eudicots, both aDEF-

and aGLO-like gene are necessary to provide the floralhomeotic B function in monocots. Whether DEF- andGLO-like proteins interact in monocots in the sameway as they do in eudicots is unknown so far.

It is not yet known whether MADS-box genesare involved in providing the A function in monocots(as theSQUA-like geneAP1 does inArabidopsis).Clearly, however,SQUA-like genes are present inmaize, sorghum and lily (Figure 3) ([10, 31, 42, 72],and our unpublished results). Consequently, this geneclade had already been established in the last com-mon ancestor of monocots and eudicots (Figure 4).AncestralSQUA-like genes could have been involvedin specifying inflorescence or floral meristem identity(see the basal angiosperm section). The relatively largenumber of these genes in some extant species suggeststhat they later may have been recruited for several dif-ferent functions (for results supporting that hypothesissee also the eudicot section). For example, cDNAsrepresenting 5 differentSQUA-like genes have alreadybeen isolated from maize ([10, 31, 72], and our unpub-lished results). For some of them, expression patternssuggest that they are not involved in establishing floralmeristem identity, but function at later stages of floretdevelopment [10].

The large number ofSQUA-like genes in maizemay seem surprising. However, recent increases ingene number in some clades (by gene duplication, inmaize also by ancient allotetraploidy) are a commontheme in grasses. For example, 8 differentAGL2-like genes have been isolated from maize so far (ourunpublished results), and for 4 of them putative or-thologues from rice or sorghum have already beenfound (Figure 3). These genes have a broad rangeof expression patterns, suggesting also a functionaldiversification [10–12, 123]. For example,ZMM6 ex-pression is initially restricted to just one primordiumout of each pair of developing spikelet primordia, sug-gesting that this gene is involved in determining thealternative identity of spikelet primordia (pedicellatevs. sessile spikelet) [11, 122]. Expression ofZMM8and ZMM14 is detectable only in the upper, not inthe lower, floret of each developing spikelet, suggest-ing that these genes determine the alternative identityof the upper vs. the lower floret within each spikeletprimordium. Alternatively, these genes may be in-volved in conferring determinacy to the spikelet orupper floret meristem [12]. The timing ofZMM6 andZMM8/ZMM14expression may determine the numberof spikelets at a certain position on the inflorescenceaxis, or the number of florets per spikelet, respec-

140

tively. These genes could thus have played a role inregulating spikelet or floret number in grasses duringevolution [10, 12, 122]. It seems reasonable, there-fore, that in the phylogenetic lineage which led tograsses, an ancestral member of theAGL2 clade hasbeen amplified and its descendants have been recruitedfor the establishment of novel positional information(concerning spikelet and floret number and arrange-ment). These positional values are characteristic ofgrass inflorescences and are not found within the sim-ple inflorescences of some dicotyledonous plants suchasArabidopsis.

cDNAs of more than 30 different MADS-boxgenes, belonging to more than 10 different subfami-lies, have been cloned from maize so far ([10, 31, 32,72, 110, 126], and our unpublished results). Amongthem is anAGL17-like gene, but also several genes forwhich orthologues from nonmonocots have not beenreported yet. In these more than 30 different genes anespecially interesting class of sequences is not evenincluded, the Transposed MADS-box elements ofZeaNo. 1 (TMZ1elements; also calledZEM genes;ZAG4is also a representative of this class of sequence el-ements) [31, 72, 78, 79]. TheTMZ1 elements havemany features that are typical of transposons, such asvarying copy numbers and genomic locations in differ-ent maize lines, 13 bp perfect terminal-inverted repeats(TIRs) at their flanks and 3 bp target sequence dupli-cations. The last two features are both characteristicof theEn/Spmtransposon family [31, 79]. TheTMZ1elements are the only plant sequences containing aMADS box published so far without a MIKC-typedomain structure. Remarkably, they contain MADSboxes nearly identical to those of the members oftheAGAMOUSclade, encoding MADS domains thatfall well into the AG clade in phylogeny reconstruc-tions based on MADS-domain sequences [123]. TheseMADS boxes are flanked, however, by sequencesthat are absolutely unrelated to sequences ofAG-likegenes. The most plausible scenario for the origin andevolution of theTMZ1elements thus seems to be thatan En-like transposable element captured a MADSbox of anAG-like gene somewhere in the lineage thatled to maize, and was then distributed in the genomesof maize and its relatives [31, 79].TMZ1 elementshave not been reported so far from outside the genusZea.

The Liliaceae have flowers that superficially lookvery different from the flowers of grasses. For exam-ple, often they are very large and have a simple, yetshowy perianth (perigon) composed of two whorls of

organs, each containing three petaloid tepals. Flow-ers with a similar perianth structure are also knownfrom some basal angiosperms (such asCabomba, be-longing to the Nymphaeales). Floral structures likethese could be easily explained by a modified ABCDmodel in which the expression of the B function hasexpanded to whorl 1 (Figure 5) [129]. Tulip mutantsare known that strongly support this hypothesis. Aputative loss-of-B-function mutant is called ‘Viridi-flora’. It has flowers where the tepals in whorls 1and 2 are homeotically transformed into sepaloid orleaf-like structures. The 6 stamens of the mutant tulipflower are transformed into carpel-like structures. Aputative loss-of-C-function mutant is also known. Ithas tepal-like structures in whorls 1, 2 and 3, andfrom the center of the flower a new flower structurearises which again has tepal-like structures in all 3outer whorls [129]. It seems not unlikely that anAG-like gene is affected in the loss-of-C-function mutant.Similarly, aDEF- or aGLO-like gene may be mutatedin the loss-of-B-function mutant. However, one shouldtake into consideration that the perianths of monocotsand eudicots possibly evolved independently (see [60],and references therein). But even then a recruitmentof DEF- andGLO-like genes for the specification ofperianth organ identity – independently of a very sim-ilar event in the eudicots – would seem the most likelyscenario to explain the petaloid character of perianthorgans in Liliaceae. However, alternative scenarioscannot be excluded yet (see also the basal angiospermsection).

A number of cDNAs representing MADS-boxgenes from the lily speciesLilium regale have beencloned recently (Figure 3), and cloning of the re-spective orthologues from tulip is well underway (ourunpublished results). Therefore, the hypotheses out-lined above can be rigorously tested soon. Expressionof a DEF-like gene in both perianth whorls and in thestamen whorls of lily support the hypothesis that thepetaloid character of all tepals is due to the expressionof B-function genes in perianth whorls 1 and 2 (ourunpublished results; see Figure 5).

MADS-box genes in eudicots

The flowers of eudicots are typically composed of abipartite perianth with sepals in the first whorl andpetals in the second whorl, followed by stamen andcarpel whorls. The ABCD system of floral organ spec-ification seems highly conserved within the higher

141

eudicots, except that an A functionsensu strictomaybe provided by different genes in different species ormay even not exist in some species. In lower eudicots,the temporal and spatial pattern of floral homeoticgene expression is more diverse. ForArabidopsiswenow have definite proof, provided by mutant analy-sis, that MADS-box gene function is not restrictedto flower development, but reaches from root to fruitdevelopment.

Eudicots are defined by the production of triapertu-rate or triaperturate-derived pollen [21]. They can befurther subdivided into the lower eudicots, compris-ing the Ranunculidae, basal Hamamelidae and basalRosidae, and the higher eudicots, made up of thebulk (about 75%) of the angiosperm species, includingthe major genetic model species such asArabidop-sis, Antirrhinum, and Petunia ([60], and referencestherein).

The structure, function and phylogeny of eudi-cotyledonous MADS-box genes has already been ex-tensively reviewed [74, 103, 123, 124]. Therefore,we rather want to concentrate on addressing somegeneral evolutionary issues and on describing recentbreakthroughs in understanding.

The gene phylogeny in Figure 3 reveals a complexbut intriguing pattern of lineage-specific increases inthe number of MADS-box genes during flowering-plant evolution. Most angiosperm subfamilies ofMIKC-type genes characterized to date contain at leasttwo members (putative recent paralogues) found in asingle eudicotyledonous plant species, documentingcontinued diversification and fixation of MIKC-typegenes during eudicot evolution [103, 123]. The evolu-tion of MADS-box genes obviously did not come toa standstill, even after the establishment of the ABCDmodel within the eudicots. Since the flower ground-plan is quite fixed in eudicots (e.g. most flowers havea bipartite perianth with sepals and petals followed bystamen and carpel whorls) the ABCD functions andthe genes encoding them might also be conserved. Anumber of ongoing studies on MADS-box genes indiverse eudicot species indicates that this is indeed thecase for the BCD functions and the respective genesof higher eudicots (e.g. [138], and many unpublisheddata). In contrast, the A function is less well definedand seems more flexible (e.g. [138], and many un-published data; see also the basal angiosperm andmonocot sections). This holds true even in comparisonbetweenArabidopsisand the other model plants.

In Antirrhinum, for example, no loss-of-functionphenotype has so far been identified which can sep-

arate the determination of the first floral whorl (as-sumed to depend on the A function) from the deter-mination of the flower itself [80]. This suggests that inAntirrhinum, the A function is not simply a derivativeof the function providing floral meristem identity, assuggested forArabidopsis(see above). It rather seemsthat these two functions are still fully linked inAn-tirrhinum and thus cannot be separated (which maybe the evolutionarily more ancient condition). As sug-gested by a precursor of the ABC model [111], theproduction of sepals in the first whorl may be inherentin the establishment of floral meristem identity. Ac-cordingly, in the rare cases where flowers are formedin plants that have theSQUAgene mutated –SQUAis theAntirrhinumorthologue of the A function geneAP1 from Arabidopsis– organ specification defectsare not apparent.

The A function is also different in petunia fromthat inArabidopsis. There is evidence that the petuniaorthologue of the non-MADS geneAP2 from Ara-bidopsisdoes not perform an A function, althoughanalysis of the blind mutant indicates that such a func-tion exists [67]. Expression ofAP2and its homologuesfrom petunia suggest an ancestral function in ovuleor seed development [52, 67]. Such an origin of theA function is not contradictory to our suggestion thatit might have been derived from a function in thespecification of floral meristem identity. One possi-bility is that things could be different forSQUA- andAP2-like genes. However,AP2also works as a floralmeristem identity gene, and the function in the speci-fication of floral meristem identity could just representan intermediate evolutionary step: since ovules areevolutionarily older than flowers, and these are proba-bly older than the eudicot perianth, orthologues ofAP2may have evolved step by step from genes involvedin seed formation, via genes that are also involved inspecifying floral meristem identity to genes that alsoplay a role in specifying organ identity in the perianth.

Although the BCD part of the ABCD model canbe considered to be relatively strongly conserved,some variations are known, especially with respect tothe genes that provide these functions, as has beenreviewed already [103, 123]. Recently it has been re-ported thatDEF andGLOorthologues (i.e. putative Bfunction genes) of some species of the lower eudicotsubclass Ranunculidae show expression patterns thatsignificantly deviate from the ones of higher eudicots[59]. While expression of these genes in stamens ofthe ranunculid species examined is as in higher eudi-cots, in some species (Dicentra eximiaand Papaver

142

nudicaule) transcripts or proteins ofDEF- andGLO-like genes accumulate in young petal primordia, butat later stages of develoment expression diminishes,becomes restricted to petal tips and margins, or evencompletely disappears [59]. It has been suggested thatthese data indicate that theDEF- andGLO-like genesof lower eudicots do not function in the same manneras their orthologues from higher eudicots to specifypetal identity [59]. It has also been pointed out thatthese findings may reflect several independent evo-lutionary derivation events of petals from stamens,during the courses of whichDEF- andGLO-like geneswere recruited to new tissue- or cell-type-specific rolesin the petals [59]. However, as outlined in the monocotsection,DEF- or GLO-like genes may specify petal orlodicule identity in monocots, a clade which separatedfrom the lineage that led to eudicots earlier than thespeciation events that gave rise to lower and highereudicots. Moreover, during very early developmentalstages, the petal primordia of all ranunculid speciesexamined show expression ofDEF- and GLO-likegenes [59]. It thus also could be that the develop-mental program that specifies the identity of petaloidorgans is a synapomorphy of all extant monocots andeudicots (or even all extant angiosperms), and that theloss of requirement ofDEF- andGLO-like genes forthe maintenance of petal identity during late develop-mental stages reflects a secondary evolutionary eventwithin some lineages that led to lower eudicots.

Despite the general conservation of the eudicotyle-donous flower groundplan, some deviations are alsowell known here. For example, unisexual flowersevolved several times independently. Thus dioeciousplants (where male and female flowers are borne onseparate individuals) and monoecious plants (wheremale and female flowers are borne on the same indi-vidual) were established independently in many lin-eages. Unisexual flowers are usually produced duringontogeny from potentially hermaphroditic flowers bysuppression of the development of either male orfemale organs in a particular whorl. However, the gen-eral impression now is that during flower development,organ abortion or suppression events and MADS-boxgene expression (and thus floral organ identity) areunder independent control mechanisms. For example,in the case of the dioecious plantSilene latifolia ithas been found that the putative B- and C-type floralhomeotic genes are expressed in male flowers (wherethe gynoecium does not differentiate) and in femaleflowers (where stamen primordia degenerate duringdevelopment) in the same whorls with similar timing

[44]. Independence of reproductive organ arrest andMADS-box gene expression was also reported for themonoecious plant cucumber (Cucumis sativus) [92],and holds also true for the putative C class geneZMM2from the monoecious monocot maize.ZMM2 is ex-pressed in stamen and carpel primordia throughouttheir development, though stamen primordia abort inthe female inflorescences and carpel primordia in themale ones [10, 12]. The situation is different, how-ever, in the dioecious dicot sorrel (Rumex acetosa),because the expression of the putative C function genebecomes undetectable here as soon as the inappropri-ate set of organs (stamens in female flowers, carpels inmale flowers) cease to develop [2]. However, absenceof C function gene expression may well be a conse-quence of the arrest in organ development rather thanits cause.

The perianth ofRumexis another interesting case:rather than having typical sepals and petals in whorls1 and 2, respectively, first- and second-whorl organsare both sepaloid, and the second whorl does notexpress the putativeDEF orthologues and functionalequivalentsRAD1andRAD2, so there is probably nohomeotic B function in the second whorl [2]. It isconceivable, therefore, that elimination of B functionexpression in the second whorl caused the sepaloidphenotype of the petals ofRumex acetosaand per-haps also of other species, including many that arewind-pollinated. Note that the situation inRumexthusis somehow opposite to that in the Liliaceae, wherewe have petaloid organs (tepals) in the first and sec-ond whorl, probably due to ectopic expression of Bfunction genes in the first whorl of the flower (see themonocot section).

Moreover, there is evidence that an incompleteloss of C function can result in an increase in repro-ductive organs in an otherwise normalArabidopsisflower [76]. Changes in the strength or spatiotemporalpattern of floral homeotic gene expression may thushave played an important role during the phylogeneticdiversification of the eudicotyledonous flower.

To understand better the role of the ABCD genesduring flower development, their upstream regulatorsand their target genes must be identified. Recently,there have been breakthroughs on both topics. Onewas cloning of theCURLY LEAF(CLF) gene fromArabidopsis[38]. Functional defects inCLF lead toleaf curling, which is also caused by ectopic expres-sion of the C function geneAGAMOUS. In clf mutants,AG is indeed ectopically expressed in leaves (and someother parts of the plant), and the available evidence

143

indicates that it may be the normal function ofCLFto repressAG in inappropriate parts of the plant dur-ing relatively late stages of development. However,CLF is obviously not involved in initially establish-ing the spatial pattern ofAG transcription. It surelycame as a surprise that the CLF protein has exten-sive sequence similarity to the protein product of theDrosophilaPolycomb-group geneEnhancer of zeste(E(z)) [38]. In Drosophila, Polycomb-group genes arepart of the ‘memory system’ that maintains the initialspatial patterns of inactive or active homeotic selec-tor genes through many rounds of cell division. Thesimilarity between CLF and E(Z) suggests that bothproteins share a common ancestor and have been con-served since the animal and plant lineages split, andthat plants possess Polycomb-group gene functions.This is very remarkable, because the homeotic targetgenes in plants and animals encode different kindsof transcription factors, i.e. either MADS-domain orhomeodomain proteins [124]. It seems likely, there-fore, that the use of Polycomb-group proteins as re-pressors of homeotic genes has evolved independentlyin plants and animals.

Although candidate genes had been obtained pre-viously, the first gene that was experimentally shownto be a direct target of a floral homeotic gene wasidentified just recently [106]. A steroid-regulated ver-sion of the class B protein AP3 had been expressedin an ag ap3mutant plant. The differential displaytechnique was then used to identify an mRNA that wasup-regulated by steroid treatment (i.e. after providingthe B function). To detect only direct target genes,indirect effects were blocked by a protein synthesisinhibitor. The identified mRNA corresponds to a genethat contains a NAC domain, named after the foundingmembers of this gene family (NO APICAL MERIS-TEM (NAM) from petunia,ATAF1-2, CUP-SHAPEDCOTYLEDONS2(CUC2) from Arabidopsis), henceit was calledNAP (NAC-LIKE, activated byAP3/PI)[106]. The expression pattern ofNAPand the pheno-types caused by its misexpression indicate that it playsa role in the transition between growth by cell divisionand cell expansion in stamens and petals [106].

The first MADS-box genes in plants were detectedby scientists who were mainly interested in flower de-velopment [115, 137]. It is only natural, therefore,that during the early days of plant MADS science,investigations were focused on this topic. However,several studies that demonstrated the transcription ofa number of MADS-box genes outside floral organssuggested relatively early that members of this gene

family play regulatory roles beyond flower develop-ment, i.e. during root, leaf, fruit, seed and embryo de-velopment (e.g. [49, 66, 104, 124]). The existence ofMADS-box genes in gymnosperms and ferns furtherdemonstrated that the role of these genes in plants isnot restricted to flower development. Mutant analysishas provided proof that someArabidopsisMADS-boxgenes have nonfloral functions. For example,AGL1andAGL5are a pair of recently duplicated paraloguesof AG-like genes. These two genes encode function-ally redundant proteins that are required for the properdevelopment of the fruit dehiscence zone, becausein agl1 agl5double mutants, the mature siliques failto dehisce [63]. Since indehiscent fruits evolved sev-eral times independently even within the Brassicaceae(Klaus Mummenhoff, pers. comm.), it will be inter-esting to find out whether mutations in orthologuesof AGL1/AGL5 are involved in some of these evo-lutionary changes.AGL8 – now calledFRUITFULL(FUL) – is anotherArabidopsisgene that is involved infruit development. The gene is required for the normalpattern of cell division, expansion and differentiationduring morphogenesis of the silique [43]. The majorpart of the silique is provided by the carpel valves, andit could well be thatFUL is a valve identity gene [63].

The genes discussed above are interesting exam-ples of gene duplications and functional diversifica-tions within gene subfamilies. The class C geneAG,the putative class D geneAGL11, and theAGL1/5pair of recent paralogues involved in fruit develop-ment are all members of the clade ofAG-like genes,which was established probably 300–400 MYA (seeabove). This implies that the gene duplications thatled to these genes occurred within the past 400 mil-lion years. The place of action (fruits) of some of thedescendants of the gene duplication events (AGL1/5)were not yet established when the firstAG-like geneappeared, thus documenting involvement of new genesin the appearance of new structures and functions.

An even more striking example for functionalchange is provided by theFUL gene. Together withAP1 and CAL, FUL belongs to the clade ofSQUA-like genes.SQUA-like genes are typically expressedin inflorescence or floral meristems, and, accordingly,work as meristem identity genes [123]. Althoughfulsingle mutants do not have an inflorescence-specificphenotype, it is also one of the first genes expressedafter the transition to flowering at the inflorescenceapex [63]. And indeed,ap1 cal ful triple mutantsshow an extreme enhancement of theap1 calpheno-type, indicating partial functional redundancy between

144

AP1 CALandFUL during flower initiation. It seemslikely, therefore, that the role in meristems reflects theancestral function ofSQUA-like genes, and thatFULwas recruited later for an additional function duringthe course of fruit evolution.

SeveralArabidopsisgenes are expressed preferen-tially or exclusively in roots [104, 141]. One of them,the AGL17-like geneANR1, could be identified as akey determinant of developmental plasticity in roots[141]. Transgenic plants in whichANR1was repressedno longer responded to NO−3 -rich zones in the soil bylateral root proliferation, which is in contrast to thebehavior of wild-type roots. Could there be a better ex-ample than that to demonstrate that MADS-box genefunction, even within eudicots, reaches far beyondflower development?

Another exciting case is the expression of aDEF-like (NMH7) and anAGL17-like gene (NMHC5) dur-ing nodule development of alfalfa (Medicago sativa;reviewed in [27]). The definition of the exact role ofthese genes in nodule development, however, awaitsthe isolation of respective loss-of-function mutants.

Outlook for future studies

We briefly mention major gaps in our understanding ofplant MADS-box gene evolution, and discuss some in-novative directions of future research. We suggest howcooperation between researchers interested in MADSevodevotics might be facilitated by a MADS homepageprovided on the worldwide web.

There is still a long way to go until we will under-stand the role of MADS-box genes in plant evolutionin satisfactory detail. We hope that the previous sec-tions, however, have indicated that we are moving inthe right direction. Still there are severe gaps in ourknowledge. Concerning the major steps of land plantevolution, we know very little about MADS-box genesin mosses, and nothing about these genes in liverwortsand algae, for example. Since liverworts have recentlybeen identified as the earliest land plants [100], andthe green charophycean algae are the sister group ofall land plants [57], these taxa should be studied withhighest priority. We also know nothing about the func-tion of MADS-box genes in any nonseed plant, butgenerating gene knockouts by homologous recombi-nation in the mossPhyscomitrella patensmight be oneway to change that soon.

Gene cloning and sequencing, phylogeny recon-structions, expression studies and mutant analysis

have made it possible to correlate MADS-box genephylogeny with the evolution of plant morphology.For example, the contribution of gene duplicationsand gene recruitments to the evolution of the genenetworks controlling plant morphology thus becameobvious. Comparison of mutant phenotypes betweendifferent taxa (e.g. floral homeotic mutants ofAra-bidopsisand Antirrhinum) which are caused by or-thologous genes indicated the equivalence of genefunctions in often distantly related species. The ex-pression of orthologous genes in heterologous back-grounds (e.g. coniferAG-like genes inArabidopsis)indicated to what extent genes from other taxa cansubstitute for the homologous function within the hostplant. However, the conditions of the experimentaldesign have to be carefully taken into consideration.For example, functional substitution in mis- or over-expression studies might be less demanding than thecomplementation of loss-of-function mutants (see thegymnosperm section). However, experiments usingthe DEF gene to complementAP3 mutants have al-ready been succesful, although the donor and acceptorspecies (Antirrhinum and Arabidopsis, respectively)are relatively distantly related eudicots [51, 107].

The replaceability of gene functions by heterolo-gous transgenes is a stringent test for the conservationof gene function. Therefore, more and more of thesekinds of experiments will probably be carried out, in-volving more and more species. For an optimal use ofsuch experiments it is essential that the phylogeneticrelationships between the genes of interest are clari-fied. For example, interspecific comparisons betweenorthologous genes are generally more useful thancomparisons between non-orthologous genes, and thecomplementation of mutants should also be triedpreferably with orthologous genes. To facilitate the de-termination of the phylogenetic relationship betweenany MADS-box gene of interest and the published setof genes, a ‘MADS homepage’ has been establishedand made accessible via the worldwide web (URL:http://www.mpiz-koeln.mpg.de/mads/). Among otherthings, it contains a list of published MADS-box genesfrom which information about these genes can be ac-cessed. Phylogenetic trees which clarify the evolution-ary relationship between a new gene of interest and allpublished genes will be made on demand. The onlycondition for having such an analysis carried out isthat a permission is given to present the phylogenetictree including the position of the new gene (and indi-cating the species it was isolated from) on the MADShomepage. The gene sequence itself will be kept confi-

145

dential if this is requested. (For details, see the MADShomepage.) The MADS homepage may thus becomean information node at which scientists who have iso-lated a new MADS-box gene might be able to identifycolleagues who have cloned orthologues from otherspecies.

Another way to minimize the problem of identify-ing orthologous genes is to study monophyletic groupsof closely related species. Especially the phylogeneticsurroundings ofArabidopsisandAntirrhinumseem in-teresting in that respect, not only because so much isknown about the model plants that can be comparedto other species, but also because substantial mor-phological variation exists among the respective taxa[28]. Studies in which the role of MADS-box genesin the variability of floral morphology is investigatedhave been initiated for relatives of bothArabidopsisand Antirrhinum [8, 102]. These studies address animportant issue: that the function of flowering-plantMADS-box genes is largely known from studyingmutants. It is not yet known whether variability atthe same loci that, upon mutation, change the flowerstructure or traits such as time to flowering, also un-derlies natural variability in these traits. It is essential,therefore, to determine the amount of natural variabil-ity at MADS-box gene loci at the molecular level innatural populations of plants. Such studies have justbegun, using theArabidopsisgenesCAL, AP3andPIas model systems [98, 99].

During the course of this review, changes inMADS-box gene function during evolution have be-come obvious. Changes in gene expression pattern,often caused by mutations incis-regulatory elementswithin promoter regions [23], may be a major reasonfor many of these changes. Therefore, comparativestudies of promoters of orthologous genes with differ-ent expression patterns would be very interesting. Un-fortunately, we know only very little about promoterfunctions of MADS-box genes. However, detailedstudies on the promoters of theAG and AP3 genesfrom Arabidopsispublished recently [48, 113, 127]indicate that the situation is significantly improving.

Other changes in the connections of the MADS-box genes to other genes within the gene networkshave probably also taken place during evolution. Forexample, target genes may have changed (e.g. whenthe B function genes started to specify not only re-productive organs, but also petals), and also protein-protein interactions may be phylogenetically dynamic(e.g. while the DEF- and GLO-like proteins of eu-dicots only form heterodimers, the single ancestor

of both proteins may have formed homodimers, orheterodimers with other proteins). Thus the biochem-ical properties of the proteins encoded by MADS-boxgenes have probably changed during evolution. In or-der to work properly, a MADS-domain protein, likemost other transcription factors, generally has to fulfilseveral subfunctions: it must form homo- or het-erodimers, or even higher-order protein complexes;it must bind to DNA; and it must activate or repressthe basal transcriptional machinery. To study protein-protein and protein-DNA interactions, a number ofin vitro and in vivo assays are available and have al-ready been applied to MADS-domain proteins fromthe genetic model plants. These techniques includeelectrophoretic mobility shift assays (EMSAs), theyeast two-hybrid system, DNA footprinting assays andrandom DNA binding site selection experiments (e.g.[30, 49]). In order to find out how the interaction withDNA and proteins changed during MADS-domainprotein evolution, all these techniques should be ap-plied to series of orthologous proteins from distantlyrelated species.

There is obviously still a long way to go until weunderstand the role of MADS-box genes in plant evo-lution in satisfactory detail. But every step on the waywill be a great intellectual pleasure.

Acknowledgements

We thank Peter Engström, Hilde Fischer, Dirk Gassen,Mitsuyasu Hasebe, Katrin Henschel, Vivian Irish,Elena Kramer, Bill Martin, Aidyn Mouradov, JensPahnke and Marty Yanofsky for communicating un-published data or manuscripts in press. We are in-debted to Aidyn Mouradov, Peter Engström and Mit-suyasu Hasebe, who provided us with constructivecriticism on preliminary versions of the gymnospermor the fern section, respectively. Many thanks alsoto Kirsten Bomblies and two anonymous reviewersfor their very helpful comments on the manuscript.We also would like to thank Wolfram Faigl and Su-sanne Werth for skillful technical assistance. Part ofthis work was supported by a grant from the DeutscheForschungsgemeinschaft (DFG) to G.T. (grant Th417/3-1). One of us (A.B.) was supported by a fellow-ship from the DFG (Graduiertenkolleg ‘MolekulareAnalyse von Entwicklungsprozessen bei Pflanzen’).A.K. was supported by a Japanese Society for the Pro-motion of Science (JSPS) Postdoctoral Fellowship forResearch Abroad.

146

Note added in proof

According to Dr Mushigian, one of the authors ofRef. 87, UspA now appears to be an ATP-binding,not a DNA-binding protein, so that the homology be-tween a part of the MADS-domain and a stretch of theUspA protein discussed above and elsewhere [87] isdoubtful.

We apologize for not citing all of the relevant papers of our col-leagues because of space constraints. Please note, however, that amore comprehensive list of publications about MADS-box genesis provided at the ’MADS homepage’ (URL: http://www.mpiz-koeln.mpg.de/mads/).

References

1. Abouheif E, Akam M, Dickinson WJ, Holland PWH, MeyerA, Patel NH, Raff RA, Roth VL, Wray GA: Homology anddevelopmental genes. Trends Genet 13: 432–433 (1997).

2. Ainsworth C, Thangavelu M, Crossley S, Buchanan-Wollaston V, Parker J: Male and female flowers fromthe dioecious plantRumex acetosashow different patternsof MADS-box gene expression. Plant Cell 7: 1583–1598(1995).

3. Albert VA, Gustafsson MHG, Di Laurenzio L: Ontogeneticsystematics, molecular developmental genetics, and the an-giosperm petal. In: Soltis DE, Soltis PS, Doyle JJ (eds),Molecular Systematics of Plants II, pp. 349–374. KluwerAcademic Publishers, Boston, MA (1998).

4. Angenent GC, Colombo L: Molecular control of ovule devel-opment. Trends Plant Sci 1: 228–232 (1996).

5. Arber EAN, Parkin J: Studies on the evolution of the an-giosperms: the relationship of the angiosperms to the Gne-tales. Ann Bot 22: 489–515 (1908).

6. Baum DA: The evolution of plant development. Curr OpinPlant Biol 1: 79–86 (1998).

7. Beck CB: Origin and Evolution of Gymnosperms. ColumbiaUniversity Press, New York (1988).

8. Bowman JL, Smyth DR: Patterns of petal and stamen reduc-tion in Australian species ofLepidiumL. (Brassicaceae). IntJ Plant Sci 159: 65–74 (1998).

9. Bradley D, Carpenter R, Sommer H, Hartley N, Coen E:Complementary floral homeotic phenotypes result from op-posite orientations of a transposon at theplena locus ofAntirrhinum. Cell 72: 85–95 (1993).

10. Cacharrón J: MADS-Box-Gene inZea mays: Verglei-chende Expressionsuntersuchungen an Modellen paralogerund orthologer Genpaare. Ph.D. thesis, Mathematisch-Naturwissenschaftliche Fakultät der Universität zu Köln,Germany (1998).

11. Cacharrón J, Fischer A, Saedler H, Theissen G: Expressionpatterns of MADS-box genes in maize as studied byin situhybridization. Maize Genet Coop Newsl 69: 37–38 (1995).

12. Cacharrón J, Saedler H, Theissen G: Expression of theMADS-box genesZMM8 andZMM14 during inflorescencedevelopment ofZea maysdiscriminates between the upperand the lower floret of each spikelet. Dev Genes Evol 209:411–420 (1999).

13. Carmona MJ, Ortega N, Garcia-Maroto F: Isolation andmolecular characterization of a new vegetative MADS-boxgene from Solanum tuberosumL. Planta 207: 181–188(1998).

14. Chasan R: Ceratopteris: a model plant for the 90s. Plant Cell4: 113–115 (1992).

15. Chaw S-M, Zharkikh A, Sung H-M, Lau T-C, Li W-H:Molecular phylogeny of extant gymnosperms and seed plantevolution: analysis of nuclear 18S rRNA sequences. Mol BiolEvol 14: 56–68 (1997).

16. Cheng PC, Greyson RI, Walden DB: Organ initiation and thedevelopment of unisexual flowers in the tassel and ear ofZeamays. Am J Bot 70: 450–462 (1983).

17. Chung Y-Y, Kim S-R, Finkel D, Yanofsky MF, An G: Earlyflowering and reduced apical dominance result from ectopicexpression of a rice MADS box gene. Plant Mol Biol 26:657–665 (1994).

18. Chung Y-Y, Kim S-R, Kang H-G, Noh Y-S, Park MC, FinkelD, An G: Characterization of two rice MADS box geneshomologous toGLOBOSA. Plant Sci 109: 45–56 (1995).

19. Clifford HT: Spikelet and floral morphology. In: Soder-strom TR, Hilu KW, Campbell CS, Barkworth ME (eds),Grass Systematics and Evolution, pp. 21–30. SmithsonianInstitution Press, Smithsonian Institution, Washington, DC(1987).

20. Coen ES, Meyerowitz EM: The war of the whorls: geneticinteractions controlling flower development. Nature 353: 31–37 (1991).

21. Crane PR, Friis EM, Pedersen KR: The origin and earlydiversification of angiosperms. Nature 374: 27–33 (1995).

22. Di Rosa A: Molekularbiologische Untersuchungen zumUrsprung homöotischer Gene in Pflanzen am Beispielder MADS-Box-Genfamilie aus dem FarnCeratopterisrichardii. PhD thesis, Mathematisch-NaturwissenschaftlicheFakultät der Universität zu Köln, Germany (1998).

23. Doebley J, Lukens L: Transcriptional regulators and theevolution of plant form. Plant Cell 10: 1075–1082 (1998).

24. Doyle JA: Origin of the angiosperm flower: a phylogeneticperspective. Plant Syst Evol (Suppl) 8: 7–29 (1994).

25. Doyle JA: Seed plant phylogeny and the relationships ofGnetales. Int J Plant Sci 157 (Suppl): S3–S39 (1996).

26. Doyle JJ: Evolution of a plant homeotic multigene fam-ily: towards connecting molecular systematics and moleculardevelopmental genetics. Syst Biol 43: 307–328 (1994).

27. Doyle JJ: Phylogenetic perspectives on nodulation: evolvingviews of plants and symbiotic bacteria. Trends Plant Sci 3:473–478 (1998).

28. Endress PK: Evolution and floral diversity: the phylogeneticsurroundings ofArabidopsisandAntirrhinum. Int J Plant Sci153: S106–S122 (1992).

29. Endress PK: Floral structure and evolution of primitive an-giosperms: recent advances. Plant Syst Evol 192: 79–97(1994).

30. Fan H-Y, Hu Y, Tudor M, Ma H: Specific interactions be-tween the K domains of AG and AGLs, members of theMADS domain family of DNA binding proteins. Plant J 12:999–1010 (1997).

31. Fischer A, Baum N, Saedler H, Theissen G: Chromoso-mal mapping of the MADS-box multigene family inZeamaysreveals dispersed distribution of allelic genes as wellas transposed copies. Nucl Acids Res 23: 1901–1911 (1995).

32. Fischer A, Saedler H, Theissen G: Restriction fragmentlength polymorphism-coupled domain-directed differentialdisplay: a highly efficient technique for expression analy-

147

sis of multigene families. Proc Natl Acad Sci USA 92:5331–5335 (1995).

33. Gandolfo MA, Nixon KC, Crepet WL, Stevenson DW, FriisEM: Oldest known fossils of monocotyledons. Nature 394:532–533 (1998).

34. Gaut BS, Doebley JF: DNA sequence evidence for the seg-mental allotetraploid origin of maize. Proc Natl Acad SciUSA 94: 6809–6814 (1997).

35. Gehring WJ: The homeobox in perspective. Trends BiochemSci 17: 277–280 (1992).

36. Gifford EM, Foster AS: Morphology and Evolution of Vas-cular Plants, 3rd ed. Freeman, New York (1988).

37. Gilbert SF, Opitz JM, Raff RA: Resynthesizing evolutionaryand developmental biology. Dev Biol 173: 357–372 (1996).

38. Goodrich J, Puangsomlee P, Martin M, Long D, MeyerowitzEM, Coupland G: A Polycomb-group gene regulateshomeotic gene expression inArabidopsis. Nature 386: 44–51(1997).

39. Goremykin V, Bobrova V, Pahnke J, Troitsky A, Antonov A,Martin W: Noncoding sequences from the slowly evolvingchloroplast inverted repeat in addition torbcL data do notsupport Gnetalean affinities of angiosperms. Mol Biol Evol13: 383–396 (1996).

40. Goremykin VV, Hansmann S, Martin WF: Evolutionaryanalysis of 58 proteins encoded in six completely sequencedchloroplast genomes: revised molecular estimates of twoseed plant divergence times. Plant Syst Evol 206: 337–351(1997).

41. Gould SJ: Ontogeny and phylogeny – revisited and reunited.BioEssays 14: 275–279 (1992).

42. Greco R, Stagi L, Colombo L, Angenent GC, Sari-Gorla M,Pè ME: MADS box genes expressed in developing inflores-cences of rice and sorghum. Mol Gen Genet 253: 615–623(1997).

43. Gu Q, Ferrándiz C, Yanofsky MF, Martienssen R: TheFRUITFULL MADS-box gene mediates cell differentiationduring Arabidopsis fruit development. Development 125:1509–1517 (1998).

44. Hardenack S, Ye D, Saedler H, Grant S: Comparison ofMADS box gene expression in developing male and femaleflowers of the dioecious plant white campion. Plant Cell6:1775–1787 (1994).

45. Hasebe M, Banks JA: Evolution of MADS gene family inplants. In: Iwatsuki K, Raven PH (eds), Evolution and Di-versification of Land Plants, pp. 179–197. Springer-Verlag,Tokyo (1997).

46. Hasebe M, Wen C-K, Kato M, Banks JA: Characterizationof MADS homeotic genes in the fernCeratopteris richardii.Proc Natl Acad Sci USA 95: 6222–6227 (1998).

47. Heck GR, Perry, SE, Nichols, KW, Fernandez DE: AGL15,a MADS domain protein expressed in developing embryos.Plant Cell 7: 1271–1282 (1995).

48. Hill TA, Day CD, Zondlo SC, Thackeray AG, Irish VF:Discrete spatial and temporalcis-acting elements regu-late transcription of theArabidopsisfloral homeotic geneAPETALA3. Development 125: 1711–1721 (1998).

49. Huang H, Tudor M, Weiss CA, Hu Y, Ma H: TheArabidopsisMADS-box geneAGL3 is widely expressed and encodes asequence-specific DNA-binding protein. Plant Mol Biol 28:549–567 (1995).

50. Huijser P, Klein J, Lönnig W-E, Meijer H, Saedler H, Som-mer H: Bracteomania, an inflorescence anomaly, is causedby the loss of function of the MADS-box genesquamosainAntirrhinum majus. EMBO J 11: 1239–1249 (1992).

51. Irish VF, Yamamoto YT: Conservation of floral homeoticgene function betweenArabidopsisand Antirrhinum. PlantCell 7: 1635–1644 (1995).

52. Jofuku KD, den Boer BGW, Van Montagu M, Okamuro JK:Control of Arabidopsisflower and seed development by thehomeotic gene APETALA2. Plant Cell 6: 1211–1225 (1994).

53. Kang H-G, An G: Isolation and characterization of a riceMADS box gene belonging to theAGL2 gene family. MolCell 7: 45–51 (1997).

54. Kang H-G, Jeon J-S, Lee S, An G: Identification of class Band class C floral organ identity genes from rice. Plant MolBiol 38: 1021–1029 (1998).

55. Kang H-G, Noh Y-S, Chung Y-Y, Costa MA, An K, An G:Phenotypic alterations of petal and sepal by ectopic expres-sion of a rice MADS box gene in tobacco. Plant Mol Biol 29:1–10 (1995).

56. Kappen C, Ruddle FH: Evolution of a regulatory gene family:HOM/HOX genes. Curr Opin Gen Dev 3: 931–938 (1993).

57. Kenrick P, Crane PR: The origin and early evolution of plantson land. Nature 389: 33–39 (1997).

58. Kofuji R, Yamaguchi K: Isolation and phylogenetic analy-sis of MADS genes from the fernCeratopteris richardii. JPhytogeogr Taxon 45: 83–91 (1997).

59. Kramer EM, Irish VF: Evolution of genetic mechanismscontrolling petal development. Nature 399: 144–148 (1999).

60. Kramer EM, Dorit RL, Irish VF: Molecular evolution ofgenes controlling petal and stamen development: duplica-tion and divergence within theAPETALA3andPISTILLATAMADS-box gene lineages. Genetics 149: 765–783 (1998).

61. Krüger J, Aichinger C, Kahmann R, Bölker M: A MADS-box homologue inUstilago maydisregulates the expressionof pheromone-inducible genes but is nonessential. Genetics147: 1643–1652 (1997).

62. Kyozuka J, Harcourt R, Peacock WJ, Dennis ES: Eucalyptushas functional equivalents of theArabidopsis AP1gene. PlantMol Biol 35: 573–584 (1997).

63. Liljegren SJ, Ferrándiz C, Alvarez-Buylla ER, Pelaz S,Yanofsky MF: ArabidopsisMADS-box genes involved infruit dehiscence. Flowering Newsl 25: 9–19 (1998).

64. Liu J-J, Podila GK: Characterization of a MADS box gene(Accession No. Y09611) from immature female cone of redpine (PGR 97-032). Plant Physiol 113: 665 (1997).

65. Lu Z-X, Wu M, Loh C-S, Yeong C-Y, Goh C-J: Nucleotidesequence of a flower-specific MADS box cDNA clone fromorchid. Plant Mol Biol 23: 901–904 (1993).

66. Ma H, Yanofsky MF, Meyerowitz EM:AGL1-AGL6, anAra-bidopsisgene family with similarity to floral homeotic andtranscription factor genes. Genes Dev 5: 484–495 (1991).

67. Maes T, Van de Steene N, Van Montagu M, Gerats T: TheAP2-like genes ofPetunia hybrida. Flowering Newsl 25: 35–40 (1998).

68. Mandel MA, Bowman JL, Kempin SA, Ma H, MeyerowitzEM, Yanofsky MF: Manipulation of flower structure intransgenic tobacco. Cell 71: 133–143 (1992).

69. Martin WF: Is something wrong with the tree of life? BioEs-says 18: 523–527 (1996).

70. McGinnis W, Kuziora M: The molecular architects of bodydesign. Scient Am 270 (2): 36–42 (1994).

71. Mena M, Ambrose BA, Meeley RB, Briggs SP, YanofskyMF, Schmidt RJ: Diversification of C-function activity inmaize flower development. Science 274: 1537–1540 (1996).

72. Mena M, Mandel MA, Lerner DR, Yanofsky MF, SchmidtRJ: A characterization of the MADS-box gene family inmaize. Plant J 8: 845–854 (1995).

148

73. Meyerowitz EM: Plants and the logic of development. Ge-netics 145: 5–9 (1997).

74. Meyerowitz EM: Genetic and molecular mechanisms of pat-tern formation inArabidopsisflower development. J PlantRes 111: 233–242 (1998).

75. Mizukami Y, Ma H: Ectopic expression of the floral homeoticgene AGAMOUS in transgenicArabidopsis plants altersfloral organ identity. Cell 71: 119–131 (1992).

76. Mizukami Y, Ma H: Separation ofAG function in floralmeristem determinacy from that in reproductive organ iden-tity by expressing antisenseAG RNA. Plant Mol Biol 28:767–784 (1995).

77. Monod J: Le hasard et la nécessité. Éditions du Seuil, Paris(1970).

78. Montag K, Salamini F, Thompson RD:ZEMa, a member ofa novel group of MADS box genes, is alternatively spliced inmaize endosperm. Nucl Acids Res 23: 2168–2177 (1995).

79. Montag K, Salamini F, Thompson RD: The ZEM2 familyof maize MADS box genes possess features of transposableelements. Maydica 41: 241–254 (1996).

80. Motte P, Wilkinson M, Schwarz-Sommer Z: Floral meristemidentity and the A function inAntirrhinum. Flowering Newsl25: 41–43 (1998).

81. Mouradov A, Glassick TV, Hamdorf BA, Murphy LC, MarlaSS, Yang Y, Teasdale R: Family of MADS-box genes ex-pressed early in male and female reproductive structures ofMonterey pine. Plant Physiol 117: 55–61 (1998).

82. Mouradov A, Glassick T, Teasdale R: Isolation and charac-terization of a new MADS-box cDNA fromPinus radiata(Accession No. U76726) (PGR 97-027). Plant Physiol 113:664 (1997).

83. Mouradov A, Glassick T, Vivian-Smith A, Teasdale R: Iso-lation of a MADS box gene family fromPinus radiata(accession No. U42399 and U42400) (PGR 96-002). PlantPhysiol 110: 1047 (1996).

84. Mouradov A, Hamdorf B, Teasdale RD, Kim J, Winter K-U, Theissen G: ADEF/GLO-like MADS-box gene from agymnosperm:Pinus radiatacontains an orthologue of an-giosperm B class floral homeotic genes. Dev Genet (in press)(1999).

85. Münster T, Pahnke J, Di Rosa A, Kim JT, Martin W, SaedlerH, Theissen G: Floral homeotic genes were recruited fromhomologous MADS-box genes preexisting in the commonancestor of ferns and seed plants. Proc Natl Acad Sci USA94: 2415–2420 (1997).

86. Murai K, Murai R, Ogihara Y: Wheat MADS box genes, amultigene family dispersed throughout the genome. GenesGenet Syst 72: 317–321 (1997).

87. Mushegian AR, Koonin EV: Sequence analysis of eukaryoticdevelopmental proteins: ancient and novel domains. Genetics144: 817–828 (1996).

88. Okada K, Shimura Y: Genetic analyses of signalling in flowerdevelopment usingArabidopsis. Plant Mol Biol 26: 1357–1377 (1994).

89. Okamoto H, Silverthorne J, Wada M: Spatial patterns of phy-tochrome expression in young leaves of the fernAdiantumcapillus-veneris. Plant Cell Physiol 38: 1397–1402 (1997).

90. Olson EN, Perry M, Schulz RA: Regulation of muscle dif-ferentiation by the MEF2 family of MADS box transcriptionfactors. Dev Biol 172: 2–14 (1995).

91. Pellegrini L, Tan S, Richmond TJ: Structure of serum re-sponse factor core bound to DNA. Nature 376: 490–498(1995).

92. Perl-Treves R, Kahana A, Rosenman N, Xiang Y, SilbersteinL: Expression of multipleAGAMOUS-like genes in male andfemale flowers of cucumber (Cucumis sativusL.). Plant CellPhysiol 39: 701–710 (1998).

93. Philippe H, Chenuil A, Adoutte A: Can the Cambrian explo-sion be inferred through molecular phylogeny? Development(Suppl): 15–25 (1994).

94. Pichersky E, Soltis D, Soltis P: Defective chlorophylla/b-binding protein genes in the genome of a homosporous fern.Proc Natl Acad Sci USA 87: 195–199 (1990).

95. Pryer KM, Smith AR, Skog JE: Phylogenetic relationshipsof extant ferns based on evidence from morphology andrbcLsequences. Am Fern J 85: 205–282 (1995).

96. Purugganan MD: The MADS-box floral homeotic gene lin-eages predate the origin of seed plants: phylogenetic andmolecular clock estimates. J Mol Evol 45: 392–396 (1997).

97. Purugganan MD, Rounsley SD, Schmidt RJ, Yanofsky M:Molecular evolution of flower development: diversificationof the plant MADS-box regulatory gene family. Genetics140: 345–356 (1995).

98. Purugganan MD, Suddith JI: Molecular population-geneticsof the Arabidopsiscauliflower regulatory gene: nonneutralevolution and naturally occuring variation in floral homeoticfunction. Proc Natl Acad Sci USA 95: 8130–8134 (1998).

99. Purugganan MD, Suddith JI: Molecular population geneticsof floral homeotic loci: departures from the equilibrium-neutral model at theAPETALA3andPISTILLATAgenes ofArabidopsis thaliana. Genetics 151: 839–848 (1999).

100. Qiu Y-L, Cho Y, Cox JC, Palmer JD: The gain of three mi-tochondrial introns identifies liverworts as the earliest landplants. Nature 394: 671–674 (1998).

101. Raff RA: The Shape of Life. Genes, Development, and theEvolution of Animal Form. University of Chicago Press,Chicago, IL (1996).

102. Reeves PA, Olmstead RG: Evolution of novel morphologi-cal and reproductive traits in a clade containingAntirrhinummajus(Scrophulariaceae). Am J Bot 85: 1047–1056 (1998).

103. Riechmann JL, Meyerowitz EM: MADS domain proteins inplant development. Biol Chem 378: 1079–1101 (1997).

104. Rounsley SD, Ditta GS, Yanofsky MF: Diverse roles forMADS box genes inArabidopsisdevelopment. Plant Cell 7:1259–1269 (1995).

105. Rutledge R, Regan S, Nicolas O, Fobert P, Coté C, BosnichW, Kauffeldt C, Sunohara G, Séguin A, Stewart D: Char-acterization of anAGAMOUShomologue from the coniferblack spruce (Picea mariana) that produces floral homeoticconversions when expressed inArabidopsis. Plant J 15:625–634 (1998).

106. Sablowski RWM, Meyerowitz EM: A homolog ofNOAPICAL MERISTEMis an immediate target of the floralhomeotic genesAPETALA3/PISTILLATA. Cell 92: 93–103(1998).

107. Samach A, Kohalmi SE, Motte P, Datla R, Haughn GW: Di-vergence of function and regulation of class B floral organidentity genes. Plant Cell 9: 559–570 (1997).

108. Savard L, Li P, Strauss SH, Chase MW, Michaud M, Bous-quet J: Chloroplast and nuclear gene sequences indicate LatePennsylvanian time for the last common ancestor of extantseed plants. Proc Natl Acad Sci USA 91: 5163–5167 (1994).

109. Schmidt RJ, Ambrose BA: The blooming of grass flowerdevelopment. Curr Opin Plant Biol 1: 60–67 (1998).

110. Schmidt RJ, Veit B, Mandel MA, Mena M, Hake S, YanofskyMF: Identification and molecular characterization ofZAG1,

149

the maize homolog of theArabidopsisfloral homeotic geneAGAMOUS. Plant Cell 5: 729–737 (1993).

111. Schwarz-Sommer Z, Huijser P, Nacken W, Saedler H, Som-mer H: Genetic control of flower development by homeoticgenes inAntirrhinum majus. Science 250: 931–936 (1990).

112. Shore P, Sharrocks AD: The MADS-box family of transcrip-tion factors. Eur J Biochem 229: 1–13 (1995).

113. Sieburth LE, Meyerowitz EM: Molecular dissection of theAGAMOUScontrol region shows thatciselements for spatialregulation are located intragenically. Plant Cell 9: 355–365(1997).

114. Slack JMW, Holland PWH, Graham CF: The zootype and thephylotypic stage. Nature 361: 490–492 (1993).

115. Sommer H, Beltrán J-P, Huijser P, Pape H, Lönnig W-E,Saedler H, Schwarz-Sommer Z:Deficiens, a homeotic geneinvolved in the control of flower morphogenesis inAntir-rhinum majus: the protein shows homology to transcriptionfactors. EMBO J 9: 605–613 (1990).

116. Sundström J, Carlsbecker A, Svensson M, Svenson M, Ur-ban J, Theissen G, Engström P: MADS-box genes activein developing pollen cones of Norway spruce (Picea abies)are homologous to the B-class floral homeotic genes inangiosperms. Dev Genet (in press).

117. Stewart WN, Rothwell GW: Paleobotany and the Evolutionof Plants, 2nd ed. Cambridge University Press, Cambridge,UK (1993).

118. Tandre K: Molecular approaches to the developmental bi-ology of Norway spruce,Picea abies. Acta UniversitatisUpsaliensis, Uppsala, Sweden (1997).

119. Tandre K, Albert VA, Sundas A, Engström P: Conifer ho-mologues to genes that control floral development in an-giosperms. Plant Mol Biol 27: 69–78 (1995).

120. Tandre K, Svenson M, Svensson ME, Engström P: Conser-vation of gene structure and activity in the regulation of re-productive organ development of conifers and angiosperms.Plant J 15: 615–623 (1998).

121. Taylor TN, Taylor EL: The Biology and Evolution of FossilPlants. Prentice Hall, Englewood Cliffs, NJ (1993).

122. Theissen G, Cacharrón J, Fischer A, Saedler H: MADS-boxgenes in the evolution of maize (Zea maysssp.mays). In:Hong JC, Kim MJ (eds), Proceedings of the 2nd Korea-Germany Joint Symposium in Plant Biotechnology, pp. 1–12.Gyeongsang National University of Korea, Chinju, Korea(1994).

123. Theissen G, Kim J, Saedler H: Classification and phylogenyof the MADS-box multigene family suggest defined roles ofMADS-box gene subfamilies in the morphological evolutionof eukaryotes. J Mol Evol 43: 484–516 (1996).

124. Theissen G, Saedler H: MADS-box genes in plant ontogenyand phylogeny: Haeckel’s ‘biogenetic law’ revisited. CurrOpin Genet Dev 5: 628–639 (1995).

125. Theissen G, Saedler H: Molecular architects of plant bodyplans. Prog Bot 59: 227–256 (1998).

126. Theissen G, Strater T, Fischer A, Saedler H: Structuralcharacterization, chromosomal localization and phylogeneticevaluation of two pairs ofAGAMOUS-like MADS-box genesfrom maize. Gene 156: 155–166 (1995).

127. Tilly JJ, Allen DW, Jack T: The CArG boxes in the promoterof the Arabidopsisfloral organ identity geneAPETALA3mediate diverse regulatory effects. Development 125: 1647–1657 (1998).

128. Tröbner W, Ramirez L, Motte P, Hue I, Huijser P, LönnigWE, Saedler H, Sommer H, Schwarz-Sommer Z:GLOBOSA:a homeotic gene which interacts withDEFICIENS in thecontrol of Antirrhinum floral organogenesis. EMBO J 11:4693–4704 (1992).

129. van Tunen AJ, Eikelboom W, Angenent GC: Floral organo-genesis inTulipa. Flowering Newsl 16: 33–37 (1993).

130. Veuthey A-L, Bittar G: Phylogenetic relationships of fungi,plantae, and animalia inferred from homologous comparisonof ribosomal proteins. J Mol Evol 47: 81–92 (1998).

131. Weigel D, Meyerowitz EM: Genetic hierarchy controllingflower development. In: Bernfeld M (ed), Molecular Basis ofMorphogenesis, pp. 91–105. John Wiley, New York (1993).

132. Weigel D, Meyerowitz EM: The ABCs of floral homeoticgenes. Cell 78: 203–209 (1994).

133. Winter K-U: Charakterisierung von MADS-Box-Genen derGymnospermeGnetum gnemonL. Diploma thesis, Univer-sität Bonn, Germany (1997).

134. Winter K-U, Becker A, Münster T, Kim JT, Saedler H, Theis-sen G: MADS-box genes reveal that gnetophytes are moreclosely related to conifers than to flowering plants. Proc NatlAcad Sci USA 96: 7342–7347 (1999).

135. Wolfe KH, Gouy M, Yang Y-W, Sharp PM, Li W-H: Dateof the monocot-dicot divergence estimated from chloroplastDNA sequence data. Proc Natl Acad Sci USA 86: 6201–6205(1989).

136. Yabana N, Yamamoto M:Schizosaccharomyces pombemap1+ encodes a MADS-box-family protein required forcell-type-specific gene expression. Mol Cell Biol 16: 3420–3428 (1996).

137. Yanofsky MF, Ma H, Bowman JL, Drews GN, FeldmanKA, Meyerowitz EM: The protein encoded by theArabidop-sis homeotic geneagamousresembles transcription factors.Nature 346: 35–39 (1990).

138. Yu D, Kotilainen M, Pöllänen E, Mehto M, Elomaa P, Helar-iutta Y, Albert VA, Teeri TH: Organ identity genes andmodified patterns of flower development inGerbera hybrida(Asteraceae). Plant J 17: 51–62 (1999).

139. Zachgo S, de Andra Silva E, Motte P, Tröbner W, SaedlerH, Schwarz-Sommer Z: Functional analysis of theAntir-rhinumfloral homeoticDEFICIENSgene in vivo and in vitroby using a temperature-sensitive mutant. Development 121:2861–2875 (1995).

140. Zachgo S, Saedler H, Schwarz-Sommer Z: Pollen-specificexpression ofDEFH125, a MADS-box transcription factorin Antirrhinumwith unusual features. Plant J 11: 1043–1050(1997).

141. Zhang H, Forde BG: AnArabidopsis MADS box genethat controls nutrient-induced changes in root architecture.Science 279: 407–409 (1998).