On the origin of class B floral homeotic genes: functional substitution and dominant inhibition in Arabidopsis by expression of an orthologue from the gymnosperm Gnetum


  • Kai-Uwe Winter,

    Search for more papers by this author
    • Present address: GE CompuNet Köln, Industriegasse 161e, D-50999 Köln, Germany.

  • Heinz Saedler,

    1. Max-Planck-Institut für Züchtungsforschung, Abteilung Molekulare Pflanzengenetik, Carl-von-Linné-Weg 10, D-50829 Köln, Germany
    Search for more papers by this author
  • Günter Theißen

    Corresponding author
      For correspondence (fax +49 3641 9 491552; e-mail guenter.theissen@uni-jena.de).
    Search for more papers by this author
    • Present address: Friedrich-Schiller-Universität Jena, Lehrstuhl für Genetik, Philosophenweg 12, D-07743 Jena, Germany.

For correspondence (fax +49 3641 9 491552; e-mail guenter.theissen@uni-jena.de).


Class B floral homeotic genes are involved in specifying stamen and petal identity in angiosperms (flowering plants). Here we report that gymnosperms, the closest relatives of the angiosperms, contain at least two different clades representing putative orthologues of class B genes, termed GGM2-like and DAL12-like genes. To obtain information about the functional conservation of the class B genes in seed plants, the representative of one of these clades from Gnetum, termed GGM2, was expressed under the control of the CaMV 35S promoter in Arabidopsis wild-type plants and in different class B mutants. In wild-type plants and in a conditional mutant grown at a permissive temperature, gain-of-function phenotypes were obtained in whorls 1 and 4, where class B genes are usually not expressed. In contrast, loss-of-function phenotypes were observed in whorls 2 and 3, where class B genes are expressed. In different class B gene null mutants of Arabidopsis, and in the conditional B mutant grown at the non-permissive temperature, a partial complementation of the mutant phenotype was obtained. In situ hybridization studies and class B gene promoter test fusion experiments demonstrated that the gain-of-function phenotypes are not due to an upregulation of the endogenous B genes from Arabidopsis, and hence probably involve interactions between GGM2 protein homodimers and class B protein target genes other than the Arabidopsis class B genes itself. To our knowledge, this is the first time that partial complementation of a homeotic mutant by an orthologous gene from a distantly related species has been reported. These data suggest that GGM2 has a function in the gymnosperm Gnetum which is related to that of class B floral organ identity genes of angiosperms. That function may be in the specification of male reproductive organ identity, and in distinguishing male from female reproductive organs.


The origin of the flowering plants has been considered as a serious scientific problem since the time of Charles Darwin. More than a century later, his ‘abominable mystery’ still remains unsolved, and the evolutionary origin of flowers, the well-known reproductive structures of the angiosperms, has also remained enigmatic (Crepet, 1998; Crepet, 2000; Frohlich, 1999; Frohlich and Parker, 2000; Ma and dePamphilis, 2000).

Among the most important reasons given in the literature that have obscured the origin of flowering plants are (i) the large morphological gap between angiosperms and their closest relatives, the gymnosperms, which makes homology assessments very difficult; and (ii) uncertainties in relationships among the different groups of seed plants (angiosperms and gymnosperms), and within the angiosperms (Frohlich and Parker, 2000).

Our understanding of seed plant phylogeny has dramatically improved recently due to phylogeny reconstructions employing diverse molecular markers. For example, a series of studies identified Amborella trichopoda and Nymphaeales (water lilies) as the most basal angiosperms (for a review see Kuzoff and Gasser, 2000). This strongly suggests that the last common ancestor of extant angiosperms already had hermaphroditic flowers with a perianth.

Also, the relationship between the diverse groups of extant seed plants has recently been clarified. The considerable morphological difference between the four groups of extant gymnosperms – gnetophytes, conifers, cycads and Ginkgo – has long been taken as evidence for the gymnosperms being paraphyletic. Most importantly, morphological studies had often identified gnetophytes as the extant sister group of the angiosperms, which was used as a major basis for understanding character evolution leading to flowering plants (for reviews see Donoghue and Doyle, 2000; Doyle, 1994a; Doyle, 1998). More and more molecular data, however, now strongly suggest that extant gymnosperms are a monophyletic group (Bowe et al., 2000; Chaw et al., 1997; Chaw et al., 2000; Frohlich and Parker, 2000; Goremykin et al., 1996; Hasebe et al., 1992; Samigullin et al., 1999) which separated from the lineage that led to angiosperms as early as 300 million years ago (MYA) (Goremykin et al., 1997; Savard et al., 1994; Wolfe et al., 1989). This suggests that, essentially, all well established theories on flower origin have to be reassessed because they explicitly assume a close relationship between gnetophytes and angiosperms (Donoghue and Doyle, 2000; Doyle, 1994a). Thus novel approaches are required to gain a better understanding of flower origin.

It has been suggested that an evolutionary developmental genetics (‘evo-devo’) approach may be especially promising to solve the ‘abominable mystery’ (Frohlich and Meyerowitz, 1997; Theißen and Saedler, 1995; Theißen and Saedler, 1999; Theißen et al., 2000). In the case of multicellular organisms, evolution of form is always the evolution of developmental processes (Gilbert et al., 1996; Gould, 1992). Therefore changes in developmental control genes may be a major starting point for evolutionary changes in morphology (Theißen and Saedler, 1995; Theißen et al., 2000). Understanding the phylogeny of developmental control genes may thus significantly contribute to understanding the evolution of plant form.

Among the most important developmental control genes of flower formation are, arguably, homeotic selector genes termed ‘floral organ identity genes’. The functions of these genes were recognized during the study of homeotic mutants in which the identities of floral organs are changed. In one of the major model systems, Arabidopsis thaliana (thale cress), such mutants come in three classes: A, B and C. Based on these classes of mutants and all combinations of double and triple mutants, the ABC model proposes three classes of combinatorially acting floral organ identity genes called A, B and C, with A specifying sepals in the first floral whorl; A + B, petals in the second whorl; B + C, stamens in the third whorl; and C specifying carpels in the fourth whorl (Coen and Meyerowitz, 1991; Weigel and Meyerowitz, 1994). The model also maintains that the class A and class C genes negatively regulate each other. Based on studies in petunia (Petunia hybrida), the ABC model was later extended by class D genes, specifying ovules (Angenent and Colombo, 1996). Meanwhile, it has been demonstrated by a reverse-genetic approach that yet another class of floral organ identity genes, termed class E (Theißen, 2001) or SEPALLATA genes, is involved in specifying petals, stamens and carpels (Honma and Goto, 2001; Pelaz et al., 2000).

In Arabidopsis, class A genes comprise APETALA1 (AP1) and APETALA2 (AP2). The class B genes are represented by APETALA3 (AP3) and PISTILLATA (PI), and the class C gene is AGAMOUS (AG). Class D genes have been recognized only in petunia so far, where they have been termed FLORAL BINDING PROTEIN7 (FBP7) and FBP11. The class E genes in Arabidopsis comprise SEPALLATA1 (SEP1), SEP2 and SEP3, which have highly redundant functions – any SEP gene can functionally substitute the others.

Recently, the molecular mode of interaction of the floral organ identity genes could be identified (Egea-Cortines et al., 1999; Honma and Goto, 2001). These genes encode proteins that combine to different ternary or quaternary complexes, at least one for each type of floral organ. These protein complexes, representing transcription factors, may exert their function by binding to the promoters of target genes, which they either activate or repress as appropriate for the development of the identities of the different floral organs. According to the ‘quartet model’ (Theißen, 2001; Theißen and Saedler, 2001) two protein dimers of each tetramer recognize two different DNA sites which are brought into close vicinity by DNA bending.

Cloning of all the floral organ identity genes mentioned above revealed that, except for AP2, they all belong to the MADS-box gene family. Therefore the origin of floral organ identity genes and flower origin itself can be understood only in the context of MADS-box gene phylogeny.

All well characterized plant MADS-box genes belong to a single clade of genes with a conserved structural organization, the so-called MIKC-type domain structure (reviewed by Becker et al., 2000; Theißen et al., 2000), including a MADS (M-), intervening (I-), keratin-like (K-) and C-terminal (C-) domain. The MADS-domain is by far the most highly conserved region of the proteins (Purugganan et al., 1995). In most cases it is found at the N-terminus of the putative proteins. The MADS-domain is the major determinant of DNA-binding, but it also performs dimerization and accessory factor-binding functions (Pellegrini et al., 1995; Zachgo et al., 1995). MADS-domain proteins bind to DNA sites based on the consensus sequence 5′-CC(A/T)6GG-3′, which is called a CArG-box (for CC-A rich-GG). The I-domain, located directly downstream of the MADS-domain, is only relatively weakly conserved (Becker et al., 2000; Purugganan et al., 1995). It may constitute a key molecular determinant for the selective formation of DNA-binding dimers (Riechmann and Meyerowitz, 1997). The K-domain is characterized by a conserved, regular spacing of hydrophobic residues, which is proposed to allow for the formation of an amphipathic helix. It is assumed that such an amphipathic helix interacts with that of another K-domain containing protein to promote dimerization (Riechmann and Meyerowitz, 1997; Zachgo et al., 1995). The most variable region is the C-domain at the C-terminal end of the MADS-domain proteins. In some MADS-domain proteins it is involved in transcriptional activation (Cho et al., 1999), or in the formation of ternary or quaternary complexes (Egea-Cortines et al., 1999).

Phylogeny reconstructions demonstrated that the MADS-box gene family is composed of several defined gene clades (Doyle, 1994b; Münster et al., 1997; Purugganan et al., 1995; Theißen and Saedler, 1995; Theißen et al., 1996; Theißen et al., 2000). Most clade members share similar functions and expression patterns. Specifically, the class A, B, C + D and E (SEPALLATA) floral homeotic genes each fall into separate clades: SQUAMOSA- (SQUA-, class A); DEFICIENS- or GLOBOSA- (DEF- or GLO-, class B); AGAMOUS- (AG-, class C + D); and AGL2-like genes (class E) (Doyle, 1994b; Purugganan et al., 1995; Theißen, 2001; Theißen et al., 1996; for rules of naming MADS-box gene clades see Theißen et al., 1996).

The recent results in phylogeny reconstructions strongly suggest that the establishment of these gene clades was an important prerequisite for the origin of the floral organ identity genes and their functions (Theißen et al., 1996; Theißen et al., 2000). Gene clades were identified containing putative orthologues from both angiosperms and gymnosperms. Among these are the AG-like genes (sensuTheißen et al., 1996), comprising the floral homeotic class C and class D genes or their gymnosperm orthologues (Hasebe, 1999; Rutledge et al., 1998; Tandre et al., 1995; Winter et al., 1999); and the DEF/GLO/GGM2/DAL12/CJMADS1-like genes, being class B genes or orthologues thereof (Becker et al., 2000; Fukui et al., 2001; Mouradov et al., 1999; Sundström et al., 1999; Winter et al., 1999). For simplicity, members of the AG clade will henceforth be termed ‘C genes’, and DEF-, GLO-, GGM2-, DAL12- and CJMADS1-like genes will be called ‘B genes’, referring to the genealogy of these genes and not necessarily to their function (in contrast, for example, to the term ‘class C gene’, which means a certain type of floral organ identity gene). GGM2- and DAL12-like genes are two clades of B genes that have so far been found only in gymnosperms, such as the gnetophyte Gnetum gnemon and the conifer Picea abies, while members of the clades of DEF- and GLO-like genes appear to be restricted to angiosperms.

Recently, it was found that the B genes have a sister clade (termed GGM13-like or Bsister genes) with hitherto overlooked members in both gymnosperms and angiosperms (Becker et al., 2000; Becker et al., 2002).

Despite serious attempts, obvious B and C gene orthologues from ferns or more basal plants have not yet been isolated. It appears that the clades that gave rise to floral organ identity genes were established 300–400 MYA (Becker et al., 2000). There is evidence that the expression patterns of the members of these clades have been conserved since that time. The class C and D genes of angiosperms are exclusively expressed in reproductive organs (stamens and carpels, or ovules, respectively), and their putative orthologues from gymnosperms are also expressed in both male as well as female reproductive units, but not in vegetative organs (Rutledge et al., 1998; Tandre et al., 1995; Tandre et al., 1998; Winter et al., 1999). The class B genes of angiosperms are expressed in stamens (male organs) and in petals which may have been derived from stamens during evolution, but not in carpels (female organs). Their orthologues from Gnetum and conifers are exclusively expressed in male reproductive units (Fukui et al., 2001; Mouradov et al., 1999; Sundström et al., 1999; Winter et al., 1999).

Given the extreme morphological diversity of seed plant reproductive structures, which has seriously obscured flower origin so far, the finding that gymnosperms have putative orthologues of floral organ identity genes deserves attention. Taken together with the conserved expression patterns of these genes, it suggests that the B and C genes of gymnosperms have functions that are somehow similar to those of their angiosperm orthologues, despite the fact that the respective reproductive structures themselves look so different.

As outlined above, the class C and D genes of angiosperms specify the identity of reproductive organs. Therefore it may well be the function of the expression of C genes in gymnosperms to distinguish between reproductive organs (where expression is on) and non-reproductive organs (where expression is off), and to specify ovules. The class B genes of angiosperms specify stamens (male organs), but not carpels (female organs). Thus it could be the function of the expression of class B gene orthologues in gymnosperms to distinguish between male reproductive organs (where expression is on) and female reproductive organs (where expression is off) (Theißen et al., 2000). Differential expression of B genes may thus represent the ancestral sex-determination mechanism of all seed plants (Winter et al., 1999), and all B genes may thus have equivalent functions in this respect. In our terminology a gene is a functional equivalent of another gene in another organism if these genes do approximately the same in a very similar functional and molecular context within the different organisms. In the case of transcription factors, this means that they regulate orthologous target genes, make protein–protein interactions with orthologous proteins, and so on, although novel interactions may also have been established during evolution, or some interactions may have been abolished. Operationally, definition of functional equivalence largely depends on the demonstration of similarity in conventional or transgenic mutant phenotypes, in expression patterns, or in abilities to interact with DNA or proteins.

Unfortunately, functional equivalence of gymnosperm genes is difficult to test experimentally using mutant phenotypes, because gymnosperms are difficult to work with molecularly and genetically. They have large genomes, they are all woody species with very long life cycles and thus need a lot of space and time to develop, they are difficult to transform with DNA, and only very few mutants are known.

As an alternative, one may try to demonstrate that a gene of interest is able to substitute its putative orthologue in another organism in transgenic experiments. Arabidopsis plants ectopically expressing C genes from conifers resemble plants ectopically expressing AG itself, suggesting that the conifer genes can substitute some functional aspects of the class C floral homeotic genes in a flowering plant background, at least in the given experimental set up (Rutledge et al., 1998; Tandre et al., 1998). This is clearly compatible with the view that the conifer C genes are functional equivalents of their angiosperm orthologues.

Here we show that the gnetophyte gymnosperm G. gnemon has two types of B genes, termed GGM2 and GGM15. We demonstrate that expression of GGM2, driven by the CaMV 35S promoter, not only can partially substitute for the Arabidopsis B genes AP3 and PI in overexpression experiments, but also can partially substitute these genes in different class B floral organ identity gene mutant backgrounds. In addition, we report that expression of GGM2 in Arabidopsis leads to a partial, but specific, loss of class B floral homeotic gene activity. Our data strongly suggest that GGM2, if expressed in Arabidopsis, specifically interacts with at least some of the partners of the Arabidopsis B genes and thus may have highly related interaction partners in Gnetum. GGM2 and the Arabidopsis B genes thus probably have equivalent functions. The implications of these findings for our understanding of flower origin are discussed.


Gymnosperms contain two different clades of B genes

Recently, two close relatives of class B floral homeotic genes, GGM2 and GGM15, have been reported from the gymnosperm model system G. gnemon (Becker et al., 2000; Winter et al., 1999). In order to clarify the relationships between these genes and all other known B genes, phylogeny reconstructions have been carried out. The phylogenetic trees obtained (Figure 1; Figure S1) reveal that both genes belong to different clades, which contain members from both gnetophytes and conifers. This suggests that the last common ancestor of these taxa already contained both a GGM2- and a GGM15-like gene. GGM15-like genes will henceforth be termed DAL12-like genes (Sundström et al., 1999) according to the rules of MADS-box gene clade nomenclature outlined elsewhere (Theißen et al., 1996). Membership of GGM2- and DAL12-like genes in a B gene superclade is supported by diverse phylogeny reconstructions (Figure 1; Mouradov et al., 1999; Sundström et al., 1999; Winter et al., 1999). Two other features support that notion. Sequence alignments with other MADS-domain proteins reveal that the conceptual GGM2- and DAL12-like proteins, like all other B proteins, have a characteristic deletion within the I-domain (Figure S2). Moreover, the GGM2- and DAL12-like proteins contain a subterminal ‘PI motif-derived’ sequence, and some of them also a terminal ‘PaleoAP3 motif’ at their C-terminal ends (Supplementary Figure 3: http://www. blackwell-science.com/tpj/). PI motifs and derived sequences have so far been found only in B proteins; and PaleoAP3 motifs only in a subset of B proteins from gymnosperms and angiosperms (Figure S3; Kramer and Irish, 2000; Kramer et al., 1998; Mouradov et al., 1999; Sundström et al., 1999; Winter et al., 1999).

Figure 1.

Phylogeny of B and Bsister genes of angiosperms and gymnosperms.

A highly simplified phylogenetic tree of the DEF-, GLO-, GGM2-, DAL12-, CJMADS1- (B) and GGM13-like (Bs) genes is shown, with DEF, GLO and GGM13 symbolizing gene clades, not individual genes. The complete tree of these plant MADS-box genes is available as Supplementary Figure 1 (http://www.blackwell-science.com/tpj/). The numbers beside some nodes give bootstrap percentages. Genus names of species from which the gymnosperm B genes were isolated are given in parentheses behind the gene names. Symbols at certain nodes define putative apomorphies, as described in the legend. See text for further explanations.

Gymnosperm B genes are ancestral to both types of class B floral homeotic genes

The tree in Figure 1 suggests, with relatively high bootstrap support, that the angiosperm B genes (the DEF- and GLO-like genes) form a clade so that the gymnosperm B genes are basal to all angiosperm B genes. However, controversial results have been obtained in previous studies (Mouradov et al., 1999; Sundström et al., 1999; Winter et al., 1999). To clarify the relationships of the different B genes, we therefore analysed the exon/intron structures of GGM2 and GGM15.

The exon/intron patterns of plant MADS-box genes show a high extent of conservation, but also characteristic differences exist in the length of exons and positions of introns (Tandre et al., 1998). In most plant MADS-box genes from which the genomic structure is known, exons 5 and 6 coding for the downstream end of the K-domain and the upstream part of the C-domain are 42 base pairs long. This structure thus appears to represent the basal (plesiomorphic) condition among MIKC-type MADS-box genes. In contrast, in all DEF- and all GLO-like genes from higher eudicots, exon 6 is 45 bp long. Additionally, in all GLO-like genes examined so far exon 5 is only 30 bp long.

The complete sequences and exon/intron structures of GGM2 and GGM15 were determined by genomic PCR, sequencing, and comparison to respective cDNA sequences [genomic sequences have been deposited under Accession Nos AJ421709(GGM2) and AJ421710(GGM15)]. Exons 5 and 6 of GGM2 were both found to be 42 bp in length. The same condition could also be shown for the putative orthologous genes DAL11 and DAL13 isolated from Norway spruce (Sundström et al., 1999), corroborating the view that the clade of GGM2-like genes is ancestral to the superclade of DEF- and GLO-like genes. For GGM15 and its putative orthologue from spruce, DAL12 (Sundström et al., 1999), only one exon could be identified at the homologous position. This exon is 75 bp in length in the case of GGM15, and 113 bp in the case of DAL12 (Sundström et al., 1999). Therefore uncertain changes, possibly involving the loss of the intron between exons 5 and 6, have to be postulated to explain the exon/intron organization of DAL12-like genes.

Our current hypothesis about the relationships and the structural evolution of the B genes, based on phylogeny reconstructions using protein sequences (Supplementary Figure 1: http://www.blackwell-science.com/tpj/) and parsimonious assumptions about character evolution, can be summarized as follows (Figure 1). Within the MIKC-type MADS-box genes a gene duplication and sequence diversification led to the establishment of the last common ancestor of the B and Bsister gene superclade. This B + Bsister gene ancestor is characterized by three synapomorphic characters: a shortened I region; a PI motif; and a PaleoAP3 Motif. A further gene duplication then separated the Bsister genes and the B genes (Becker et al., 2000; Becker et al., 2002; Figure 1). Within the B genes, four basal clades can be distinguished: a clade uniting the DEF- and GLO-like genes from angiosperms; the GGM2-like genes; the DAL12-like genes; and the CJMADS1-like genes (the latter three clades contain only gymnosperm genes). The DEF- and GLO-like genes have an exon 6 which is 45 rather than 42 nucleotides long, a feature that is conserved in both clades. Additionally, the GLO-like genes are distinguished from the DEF-like genes by an exon 5 which is 30 instead of 42 bp long. Moreover, the PaleoAP3 motif has been lost within this clade. Whereas the DAL12-like genes may have lost an intron, the intron distribution within the GGM2-like genes seems not to have changed. Within the GGM2-like genes the conifer genes DAL11, DAL13 and PRDGL and, possibly independently thereof, also the CJMADS genes, lost the PaleoAP3 motif (Figure 1; Supplementary Figure 3: http://www.blackwell-science.com/tpj/). In addition to the characteristic deletion within the I-domain, PRDGL and GGM15 show insertions within the I-domain (Supple mentary Figure 2: http://www.blackwell-science.com/tpj/), which probably evolved independently (Figure 1).

Which of the gymnosperm gene clades, if any, is more closely related to the clade of DEF- and GLO-like genes from angiosperms could not be resolved, because bootstrap support for any topology concerning the relationships of DEF/GLO-, GGM2-, DAL12- and CJMADS1-like genes is weak, and other informative characters are not available. However, our phylogenetic trees clearly indicate that none of the gymnosperm clades is a direct sister group of either the DEF- or the GLO-clade (Figure 1). As no member of the DEF- or GLO-like genes could be isolated from a gymnosperm plant, and no angiosperm GGM2-, DAL12- or CJMADS1-like gene is known so far, the gene duplication that separated the DEF- and the GLO-like genes probably occurred within the lineage that led to the angiosperms, after the gymnosperms had split off. Thus the GGM2-, or the DAL12-, or the CJMADS1-like genes, or a clade of two or three of these clades together, may be orthologous to the floral homeotic class B genes. On the following, we test for a representative of one of these gymnosperm clades, GGM2, functional equivalence between a gymnosperm B gene and class B floral homeotic genes.

Ectopic expression of GGM2 in Arabidopsis containing wild-type class B genes

In the dicotylodonous model plant Arabidopsis thaliana, both class B genes, AP3 and PI, have to be expressed ectopically to achieve complete homeotic conversions in the first and the fourth whorl to petaloid and stamenoid organs, respectively (Krizek and Meyerowitz, 1996). However, flowers of plants ectopically expressing AP3 already exhibit a partial conversion of fourth whorl carpels to stamens (Jack et al., 1994). The resulting mosaic organs consist of both stamen and carpel tissues. An (at least partially) active fourth whorl class B homeotic gene function is achieved in the respective experimental set up due to endogenous fourth whorl expression of PI also in wild-type Arabidopsis (see Figure 6a). In contrast, flowers of plants ectopically expressing PI show a partial first whorl conversion of sepals into petals due to a background expression of AP3 in the first floral whorl in wild-type Arabidopsis (as illustrated in Figure 6a). To examine whether GGM2 is able to achieve alterations similar to any of these homeotic changes, and therefore is able to substitute any class B gene in the given experimental set up, we constructed transgenic Arabidopsis plants expressing GGM2 under the control of the CaMV 35S promoter employing Agrobacterium-mediated transformation. The phenotypic effects of GGM2 expression during development of transgenic plants were analysed in the T2 and T3 generations by scanning electronic microscopy. The expression patterns of the endogenous class B genes AP3 and PI and of the transgene GGM2 were determined by in situ hybridization.

Figure 6.

Models of floral homeotic genes and proteins actions.

(a) Modified ABC models of plants that express PI, AP3 or GGM2 throughout the flower, as indicated. AG, AP1-AP3 and PI denote organ-identity genes from Arabidopsis; GGM2, a B gene from Gnetum gnemon. 35S::X denotes fusions of the CaMV 35S promoter and the coding region of gene X. Weak expression of AP3 in the first and of PI in the fourth floral whorl is indicated. Numbers at the top indicate floral whorls; letters at the bottom, floral homeotic functions, with A specifying sepals; A + B, petals; B + C, stamens; and C, carpels.

(b) Protein quartets that may form under the conditions indicated. Arrows symbolize target genes. For details, see text.

The eight independently transformed GGM2-transgenic T1 lines all showed BASTA resistance and an identically altered flower phenotype, described in more detail later. Within 24 T2 lines, 17 showed a 3 : 1 segregating mutant flower phenotype. The copy number of the transgenes within these lines was determined by Southern blot hybridization (data not shown). Four lines with only one or two copies of the transgene were used for further analyses in following generations.

35S::GGM2 transgenic plants showed phenotypic features of both 35S::PI (Krizek and Meyerowitz, 1996) and 35S::AP3 plants (Jack et al., 1994). No vegetative phenotype was observed; structural and developmental changes were restricted to inflorescences and flowers. In the first whorl of flowers of GGM2 transgenic plants, the sepals are stretched out (Figure 2d) compared to wild-type flowers (Figure 2a) and show petaloid cell structures along the margins (compare Figure 2(b), wild type, with Figure 2(e). These organ alterations are very similar to the phenotypes described for plants ectopically expressing the Arabidopsis class B gene PI (compare Figure 2d, this work, with Figure 1b of Krizek and Meyerowitz, 1996). The 35S::GGM2 plants described here also show aberrant development of the fourth whorl carpels resulting in organs with reduced stigmatic tissues, poorly developed valves (Figure 2j), and mosaic tissues consisting of typical carpel cells and thin elongated filament-like cells (compare Figure 2f with Figure 2c).

Figure 2.

Transgenic Arabidopsis ectopically expressing GGM2.

Wild-type flower organ phenotypes (a–c) are compared with phenotypes of 35S::GGM2 transgenic plants (d–j). Scanning electron micrographs are shown in (b,c,e,f,j).

(a) Wild-type Arabidopsis flower.

(b) SEM of abaxial surface of a wild-type sepal. Inset, abaxial surface of a wild-type petal.

(c) Surface of a wild-type gynoecium style. Inset, the surface of the filament of a wild-type stamen.

(d) Early flower of 35S::GGM2 plants with sepals stretched out.

(e) Sepal with petaloid margin (arrows).

(f) Surface of a transgenic flower gynoecium. Arrows point to a mixture of carpeloid (lower left) and filament-like cells (upper right).

(g) Inflorescence of a 35S::GGM2 plant showing different phenotypes of earlier and later flowers. Earlier flowers (upper edges) show stretched out sepals and disturbed carpel development, but well developed petals and stamens. Late flowers show a loss of class B gene function phenotype with small, greenish petals and non-fertile stamens.

(h) A close-up of later flowers showing a loss of class B gene function phenotype.

(j) A late flower with small sepaloid petals, disturbed stamen development, and a gynoecium with abnormally developed valves and reduced stigmatic tissue.

Further phenotypic alterations were observed in the second and third whorl of 35S::GGM2 transgenic flowers developing later within an inflorescence (Figure 2g,h). In addition to the features described above, the petals of such flowers were restricted in development (Figure 2h,j). These modified petals were as small as sepals and had a greenish colour. Additionally, the stamens in the third whorl were restricted in growth and did not mature properly, so that self-pollination was impossible and sterile flowers were obtained. These developmental alterations represent a specific class B gene mutant phenotype, but a loss-of-function rather than a gain-of-function one.

To find out whether these changes were due to co-suppression, the expression of the endogenous class B genes AP3 and PI was observed by in situ hybridization in wild-type and GGM2 transgenic plants (Figure 3a,b,d,e). AP3 transcripts could be detected in the same way in petal and stamen primordia in the wild type (Figure 3d) as well as in GGM2 transgenic flowers (Figure 3a). No transcripts were detected in the fourth whorl (Figure 3a,d); and although flowers at very early stages of development show AP3 expression in the first whorl (Jack et al., 1992), the developmentally more advanced flowers shown here do not (Figure 3a,d). The expression pattern of AP3 appeared to be unchanged by ectopic expression of GGM2 (compare Figure 3a with Figure 3d).

Figure 3.

Expression of AP3, PI and GGM2 in 35S::GGM2 transgenic and wild-type inflorescences as detected by in situ hybridization.

All pictures show longitudinal sections of young Arabidopsis inflorescences. (a–c) Inflorescences of 35S::GGM2 transgenic plants; (d–f) inflorescences of wild-type plants. Sections hybridized to (a,d) an AP3 antisense probe; (b,e) a PI antisense probe; (c,f) a GGM2 antisense probe.

No changes were also observed for the expression of PI in 35S::GGM2 transgenic plants. PI expression was detected in the organs of the second and third whorl, and in younger flower stages also in fourth whorl organs of both wild-type (see Goto and Meyerowitz, 1994) and transgenic flowers (Figure 3b,e). In transgenic plants, strong GGM2 transcription was detected in all four whorls of the flower as well as in vegetative tissues (Figure 3c), as expected due to the strength and ubiquitous action of the 35S promoter. Only a faint background was visible in wild-type plant tissues (Figure 3f), indicating that our GGM2 hybridization probe does not cross-hybridize with the endogenous class B genes, or any other Arabidopsis gene. Taken together, we did not find any indication of upregulation of AP3 or PI in the first or fourth floral whorl, or any decrease in accumulation of mRNA of these genes in the second or third floral whorl of GGM2-expressing plants.

To further test the hypothesis that the observed loss-of-function phenotypes in second- and third-whorl organs of 35S::GGM2 plants are not due to a gene-silencing mechanism, Arabidopsis plants were transformed with several 35S::GGM2 derivatives. These derivatives have additional stop codons upstream of the ordinary translation termination point and hence can produce only truncated GGM2 proteins. One construct contains the complete GGM2 open reading frame (encoding amino acids 1–232), but has a stop codon within the MADS-box. Respective transgenic plants produce an almost perfect GGM2 transcript, but a drastically truncated translation product (only aa 1–30), and thus no functional GGM2 protein. Other constructs have open reading frames missing the 3′ part of the C-terminal region (aa 1–187 present) or the complete C-terminal region (aa 1–163 present). All transgenic plants obtained from these experiments had perfect wild-type flowers (data not shown).

Expression of GGM2 in Arabidopsis class B mutants

To find out whether the GGM2 cDNA is able to complement the class B mutants of Arabidopsis thaliana, the transgene was ectopically expressed in the conditional temperature-sensitive mutant ap3-1 and in mutants containing the strong alleles ap3-3 or pi-1. ap3-1 homozygous mutant plants showed the class B mutant phenotype at non-permissive temperatures of 26°C and higher (Figure 4a), whereas at permissive temperatures of 16°C and lower they were fertile and resembled wild-type plants (Figure 4c; Bowman et al., 1989). The mutant phenotype of ap3-1 plants shows sepaloid organs instead of petals in the second whorl of the flowers, and organs ranging from apparently normal stamens via carpeloid stamens to non-fused carpels in the third whorl (Bowman et al., 1989). According to preliminary data, plants homozygous for ap3-1 and transgenic for 35S::GGM2 (ap3-1; 35S::GGM2 plants), when cultured at the non-permissive temperature (26°C), developed more stamens or stamenoid organs and fewer carpels or carpeloid organs than ap3-1 plants. However, analysis of a higher number of plants will be required to quantify this effect more accurately. In any case, the stamens of the ap3-1; 35S::GGM2 plants were shorter than wild-type stamens and were not fertile (Figure 4b). In addition, such plants showed similar mutant features to 35S::GGM2 plants with a wild-type background. The sepals were stretched out and the carpels had poorly differentiated valves and stigmatic tissue. The second-whorl organs appeared sepaloid and thus unchanged compared to non-transgenic ap3-1 plants (Figure 4b). ap3-1; 35S::GGM2 plants grown at the permissive temperature of 16°C (Figure 4d) appeared very similar to plants grown at the non-permissive temperature. In contrast to non-transgenic ap3-1 plants grown at 16°C (Figure 4c), the second-whorl organs were thus not rescued by growth at a permissive temperature and stamens were not fertile. Therefore the transgene appears to interfere with the class B gene function, as observed similarly in 35S::GGM2 transgenic plants with a wild-type background.

Figure 4.

Arabidopsis class B gene mutants ectopically expressing GGM2.

Scanning electron micrographs are shown in (g,h) and (l,m).

(a) Flower of an ap3-1 mutant plant grown at non-permissive temperature of 26°C.

(b) Flower of an ap3-1 mutant ectopically expressing GGM2 at 26°C. Sepals are stretched out. Petals appear green and are small as sepals. Third-whorl organs are stamen-like or missing, and the gynoecium shows poorly developed valves and reduced stigmatic tissue.

(c) Flower of an ap3-1 mutant plant grown at permissive temperature of 16°C. Flowers show a wild-type phenotype.

(d) Flower of an ap3-1 mutant ectopically expressing GGM2 at 16°C. Flowers show the same phenotype as depicted in (b). Although grown at permissive temperature, sepaloid petals and disturbed stamen development indicate a B loss-of-function phenotype.

(e) Flower of an ap3-3 mutant plant. Petals are sepaloid and carpels instead of stamens develop in the third whorl.

(f) Flower of an ap3-3 mutant ectopically expressing GGM2. Filament-like organs develop in the third whorl.

(g) Filament-like organs in the third whorl of GGM2 expressing ap3-3 mutant plants as depicted in (f). Some organs are topped with stigmatic tissue (lower right).

(h) Close-up of a filament-like organ as depicted in (g).

(j) Flower of a pi-1 mutant plant. Petals are sepaloid. While the third whorl is missing, the gynoecium is increased in diameter and consists of more than two carpels.

(k) Flower of a pi-1 mutant plant ectopically expressing GGM2. Filament-like organs develop in the third whorl, similar to those shown in (f–h).

(l) Filament-like third-whorl organ of a pi-1 flower expressing GGM2 as shown in (k).

(m) Filament-like third-whorl organ topped with stigmatic tissue.

ap3-3 contains a stop codon within the MADS-box (Jack et al., 1992) and thus represents a putative null allele. Plants mutant for this allele develop sepals instead of petals in the second whorl of their flowers and carpels, instead of stamens in the third whorl – they display full homeotic conversions of second- and third-whorl organs (Figure 4e). Mixed organs like those known from the ap3-1 mutants are not observed in ap3-3 flowers. To express GGM2 ectopically in the ap3-3 background, we crossed a homozygous ap3-3 plant with a 35S::GGM2 plant with wild-type background. Progeny plants were selected for the transgene by BASTA resistance and checked by PCR and sequencing of the AP3 locus for homozygozity of the ap3-3 mutation.

In contrast to normal ap3-3 mutants, plants homozygous for ap3-3 and transgenic for 35S::GGM2 developed carpeloid structures in the third whorls of their flowers only in exceptional cases (Figure 4f). However, functional stamens could not be observed either. Rather, filament-like structures, sometimes topped by stigmatic tissue (Figure 4g), but usually not topped at all (Figure 4h), were found very frequently. The cell structure of these organs was very similar to that of filament cells of wild-type stamens (compare Figure 4h with the inset of Figure 2c). However, the filament-like organs of ap3-3; 35S::GGM2 flowers were usually thinner than the filaments of wild-type stamens.

Very similar organs could be observed in plants homozygous for the strong pi-1 allele (Bowman et al., 1989; Bowman et al., 1991; Goto and Meyerowitz, 1994) and transgenic for 35S::GGM2 (Figure 4k,l). In addition to filament-like organs (Figure 4l), mixed organs made out of carpeloid and filament-like structures (Figure 4m) also developed in the third flower whorl of pi-1; 35S::GGM2 plants. In a minority of plants these filament-like structures even appeared to be topped by antheroid tissues (data not shown). Homozygous pi-1 plants do not develop third-whorl organs but show an increased diameter of the fourth whorl gynoecium, which is usually composed of more than two carpels (Bowman et al., 1989; Figure 4j). The weaker alleles pi-2 and pi-3 develop the usual four whorls and show carpeloid or filament-like organs within the third whorl (Bowman et al., 1991). Thus the flower phenotype of pi-1; 35S::GGM2 plants resembles the phenotypes of plants carrying the weaker pi alleles pi-2 or pi-3, revealing partial complementation of pi-1 by 35S::GGM2. In situ hybridization studies did not reveal any changes of the expression of AP3 and PI in pi-1; 35S::GGM2 plants compared to pi-1 plants, but also demonstrated strong expression of the GGM2 gene in these transgenic lines (data not shown).

Expression of AP3::GGM2 in transgenic Arabidopsis class B mutants

To analyse whether the GGM2 homodimer is able to bind to and to activate the AP3 promoter, we transformed wild-type Arabidopsis plants with a construct comprising the GGM2 cDNA driven by the AP3 promoter. The offspring of these plants was then crossed with homozygous ap3-3 and pi-1 plants. The resulting plants were selfed and the segregating progeny generation was characterized by BASTA selection, PCR and sequence analysis to look for the presence of the transgene and of class B gene loss-of-function alleles, as described above. None of the plants homozygous for the ap3-3 or the pi-1 mutation, and transgenic for the AP3::GGM2 construct, differed phenotypically from non-transgenic ap3-3 and pi-1 plants, respectively. Thus AP3::GGM2 constructs were not able to complement the strong loss-of-function mutants. In situ hybridization experiments were carried out to analyse the expression of the transgene and the endogenous class B genes. In AP3::GGM2 plants homozygous for the ap3-3 mutation, only weak expression of AP3 (Figure 5a) and PI (Figure 5b) could be detected, and no GGM2 transcript was found (Figure 5c). In contrast, transgenic AP3::GGM2 plants heterozygous for ap3-3 showed strong expression of the endogenous class B genes AP3 (Figure 5d) and PI (Figure 5e) and also of the transgene GGM2 (Figure 5f). Similar observations have been made with pi-1; AP3::GGM2 plants. Plants homozygous for the pi-1 mutation showed no or only weak expression of AP3 (Figure 5g), PI (Figure 5h) and the transgene GGM2 (Figure 5j), whereas transgenic heterozygous pi-1 plants expressed both endogenous class B genes, AP3 (Figure 5k) and PI (Figure 5l), as well as the transgene GGM2 (Figure 5m) in the second and third whorls of the flowers. As indicated by the heterozygous plants, the AP3–PI heterodimer is required to keep the expression of the endogenous class B genes and of the transgene upregulated.

Figure 5.

Expression of AP3, PI and GGM2 in AP3::GGM2 transgenic inflorescences of B-class mutants as detected by in situ hybridization.

All pictures show longitudinal sections of young Arabidopsis inflorescences transgenic for AP3::GGM2. (a–c) Inflorescences of ap3-3 homozygous plants; (d–f) inflorescences of ap3-3 heterozygous plants; (g–j) inflorescences of pi-1 homozygous plants; (k–m) inflorescences of pi-1 heterozygous plants. Sections hybridized to: (a,d,g,k) an AP3 antisense probe; (b,e,h,l) a PI antisense probe; (c,f,j,m) a GGM2 antisense probe.


Basal branching of gymnosperm B genes

B genes constitute a clade which is moderately to strongly supported by phylogeny reconstruction (Winter et al., 1999); a clade of B plus Bsister genes is quite well supported by both bootstrap values and a number of putative synapomorphies (Becker et al., 2000; Becker et al., 2002). B genes comprise angiosperm as well as gymnosperm genes, but the relationships within the B gene clade proved difficult to resolve. In previous studies, different types of data (DNA or protein sequences), different principles of phylogeny reconstruction (distance and maximum parsimony methods), and diverse data sets have been used, with variable outcomes. In most analyses gymnosperm B genes seemed to be monophyletic, and either part of the clade of GLO-like genes (Sundström et al., 1999), or basal to both DEF- and GLO-like genes (Becker et al., 2000; Winter et al., 1999), but these hypotheses were only weakly supported. An alternative hypothesis (Sundström, 2001) suggests that the gymnosperm B genes are not monophyletic, but that distinct members of both DEF- (DAL12) and GLO-like genes (DAL11, DAL13) exist in gymnosperms. This hypothesis fits with the presence/absence of PaleoAP3 motifs in B genes from Picea abies (Norway spruce) (Supplementary Figure 3: http://www.blackwell-science.com/tpj/), but we are not aware of any gene tree that would support such a scenario.

The difficulties with obtaining an unambiguous B gene tree suggest a rather low signal-to-noise ratio in the data used for phylogeny reconstructions. All previous analyses were based on relatively small data sets with respect to B genes, but also included MADS-box genes from other subfamilies. In contrast, we increased sampling of B genes here, and improved the sequence alignment and thus the quality of the trees by excluding all sequences which do not belong to the B or Bsister genes. Thus we achieved a tree which, in our view, represents the most solid hypothesis on B gene evolution available so far.

The tree on B gene evolution presented here (Figure 1) is not completely resolved, but most branches have moderate to high bootstrap support. Moreover, applying the parsimony principle to the evolution of informative characters such as the presence or absence of certain sequence motifs, or length polymorphisms in some exons and introns, supports the model of B gene evolution presented here. Its most important aspect within the context of the present study is the support of the hypothesis that gymnosperm B genes are basal to both types of B genes from angiosperms (DEF- and GLO-like genes). This led us to ask the intriguing question as to whether the gymnosperm B gene GGM2 can substitute any of the two types of class B genes from angiosperms, or even both.

GGM2 can partially substitute AP3 as well as PI in ectopic expression experiments

Arabidopsis plants in which GGM2 was ectopically expressed throughout the flowers (35S::GGM2) showed petaloid sepals as well as stamenoid carpels (Figure 2d–f). Whereas the latter is characteristic for plants that express AP3 throughout the flower (35S::AP3), the former is typical for plants that ectopically express PI (35S::PI). Thus GGM2 can substitute for both single endogenous class B floral homeotic genes from Arabidopsis in experiments where expression is driven by the CaMV 35S promoter (Figure 6a).

Two simple explanations are conceivable here: (a) that GGM2-GGM2 homodimers substitute partially for the AP3–PI heterodimer everywhere in the flower; or (b) that AP3-GGM2 and GGM2-PI heterodimers substitute for the AP3–PI heterodimer in the first and fourth whorl of the flower, respectively. Hypothesis (b) appears quite plausible due to the wild-type expression of AP3 in the first and of PI in the fourth floral whorl (Figure 6a). There is strong evidence, however, based on the yeast two-hybrid system and electrophoretic mobility shift assays, that the GGM2 protein is able to homodimerize, while AP3-GGM2 and GGM2-PI complexes could not be observed in such experiments (Winter et al., 2002). Therefore we strongly favour scenario (a) over (b).

Although GGM2-GGM2 thus may be able to partially substitute for AP3–PI heterodimer function in the organs of the first and fourth floral whorls during ectopic expression experiments, it cannot substitute completely for the class B protein heterodimer. This is indicated by the incompleteness of the homeotic transformations of first- and fourth-whorl organs towards second- and third-whorl organ identity (Figure 2d–f). In 35S::AP3; 35S::PI double transgenic lines, complete homeotic transformations of sepals to petals and of carpels to stamens have been obtained (Krizek and Meyerowitz, 1996).

Incompleteness of the homeotic transformations in 35S::GGM2 plants may be easily explained by the fact that Gnetum and Arabidopsis, and thus GGM2 and AP3–PI, are separated by at least 300 million years of independent evolution. During that time a considerable number of non-silent mutations have accumulated in the coding regions of the genes, some of which may now impair interactions of the GGM2 protein with other partners (proteins, target genes), with which AP3–PI complexes usually interact and which are vital for the class B gene function.

In contrast, the fact that GGM2 can partially substitute for both AP3 and PI in ectopic expression experiments seems more remarkable to us, as it suggests that the GGM2 homodimer can bind to and correctly control (activate or repress) at least some of the target genes of AP3–PI required for the specification of petal or stamen identity. This view is corroborated by the finding that in 35S::GGM2 plants the expression patterns of AP3 and PI are not changed (Figure 3), indicating that the phenotypic effects are not brought about in an indirect way via activation of AP3 and PI by the GGM2 protein. Thus the GGM2 protein may confer the effect directly by binding to at least some of the target genes of AP3–PI.

This hypothesis is further supported by the analysis of transgenic plants in which the expression of the GGM2 gene is under the control of the AP3 promoter (Figure 5). In ap3-3/ap3-3 plants containing the AP3::GGM2 construct, only weak expression of AP3 and PI could be detected and no GGM2 transcript was found. In contrast, ap3-3/+ plants containing the AP3::GGM2 construct showed strong expression of the endogenous class B genes AP3 and PI and also of the transgene GGM2. Similar observations were made with pi-1; AP3::GGM2 plants. Potential functionality of the AP3 promoter fragment used is demonstrated by the strong transcription of GGM2 in plants that have at least one functional allele of each class B gene. However, although early activity of the class B gene promoters is independent of class B proteins (Schwarz-Sommer et al., 1992), in class B gene null mutants the AP3::GGM2 construct does not lead to a detectable accumulation of GGM2 transcript. This suggests that GGM2 protein (which may be produced initially) does not activate the AP3 promoter, further supporting the notion that GGM2 does not activate all class B gene targets. Moreover, it makes it even more unlikely that GGM2 activity in transgenic Arabidopsis is conferred indirectly via activation of the native class B genes, including AP3.

The fact that GGM2 can partially substitute both kinds of class B genes present in Arabidopsis, rather than only AP3 or PI, fits well to the outcome of phylogeny reconstructions suggesting that the GGM2-like genes separated from the angiosperm genes before these split into DEF- and GLO-like genes via gene duplication (Figure 1).

Partial complementation of class B null mutants by GGM2

35S::GGM2 plants that were homozygous for putative null alleles of class B genes showed very similar phenotypes in that filament-like structures developed in the third floral whorl, suggesting that class B floral homeotic gene function has been partially rescued. This interpretation is supported by the fact that 35S::GGM2 plants carrying the strong allele pi-1 resemble plants homozygous mutant for the weaker alleles pi-2 or pi-3, plants that have residual PI function and thus class B gene activity.

The similarity of the phenotypes obtained may be surprising, given that the class B gene mutant itself has quite different phenotypes with respect to third-whorl organs (carpeloid in ap3-3, missing in pi-1). The fact that GGM2 can partially complement both classes of B gene, however, is in line with data suggesting that GGM2 can, at least in part, substitute for both AP3 and PI in ectopic expression experiments (see above).

Also, the hypothesis that the class B gene function is directly contributed by the GGM2 gene product, rather than by an indirect mechanism involving ectopic upregulation of AP3 and PI, is strongly corroborated by the complementation of strong alleles. As the ap3-3 allele carries a nonsense mutation within the MADS-box, it probably cannot produce functional gene product, so any mechanism that requires active AP3 gene product – such as activation of AP3 by GGM2 – can be almost excluded in the case of plants that are homozygous for ap3-3, such as the ones investigated here. Analogous arguments may apply to pi-1/pi-1 plants, as the pi-1 allele contains a nonsense mutation within the I-region and thus may also represent a null allele.

Functional substitutions of homeotic genes from distantly related species have found considerable attention in the scientific literature. These experiments, however, typically employed only ectopic expression of a gene. It has been reported, for example, that Arabidopsis plants ectopically expressing the C genes DAL2 or SAG1 from the conifers Picea abies or Picea mariana, respectively, resemble plants ectopically expressing the putative Arabidopsis orthologue AGAMOUS itself, suggesting that the conifer genes can substitute some functional aspects of the class C floral homeotic genes in a flowering plant background, at least in the given experimental set up (Rutledge et al., 1998; Tandre et al., 1998). Another spectacular example along these lines of reasoning involves the PAX6-like genes of animals. Transgenic fruitflies (Drosophila melanogaster) that ectopically express the eyeless gene, driven by a suitable promoter, develop ectopic eyes on almost any of the appendages of the fly; this, together with the eyeless loss-of-function phenotype, led to the suggestion that eyless represents a ‘master control gene’ of eye development. Ectopic eye formation was also obtained when the ectopically expressed gene eyeless was substituted in these experiments by orthologous genes from animals as distantly related as mouse (Mus musculus), sea squirt (Phallusia mammillata), and squid (Loligo opalescens) (reviewed by Gehring and Ikeo, 1999).

In all these cases, however, it was not reported whether the functional substitution is a direct or indirect effect, that is, whether the heterologous homeotic proteins bind directly to the target genes, or whether the conifer C genes primarily activate the Arabidopsis AG gene and the heterologous PAX6 genes primarily the Drosophila EYELESS gene. The problem is that many transcription factors, certainly including the class B genes (Goto and Meyerowitz, 1994; Schwarz-Sommer et al., 1992), but perhaps also comprising the PAX6-like genes, are subject to autoregulatory control. Thus it is quite conceivable that DAL2/SAG1 activates the native AG gene in Arabidopsis, which then regulates its downstream target genes. Similarly, the mouse, sea squirt and squid PAX6-like proteins may activate the EYELESS gene of Drosophila, which then activates the appropriate target genes. The obvious experimental test would be to try and rescue stamen and carpel development in ag nulls by expression of DAL2 or SAG1; and to try and rescue eye development in eyeless null mutants by expression of PAX6-like genes from mouse, sea squirt or squid. As far as we know, however, such experiments have not been reported so far (see Harris, 1997).

To the best of our knowledge, the partial complementations of putative class B gene null mutants by expression of GGM2 reported here are the first experiments that exclude indirect effects via activation of native class B genes, thus demonstrating that complementation of homeotic genes over large phylogenetic distances – the last common ancestor of angiosperms and gymnosperms existed about 300 MYA – are possible.

Dominant inhibition of class B homeotic gene function by GGM2

Relatively late-developing flowers of transgenic Arabidopsis plants with a wild-type genetic background, but ubiquitously expressing GGM2, had small and greenish petals and reduced, non-fertile stamens, suggesting that the class B gene floral homeotic function is specifically disturbed in these plants. A very similar phenotype was observed with ap3-1; 35S::GGM2 plants, carrying a temperature-sensitive AP3 allele; in this case plants grown at a permissive temperature looked almost like those grown at a non-permissive temperature. The developmental alterations observed represent a specific class B gene mutant phenotype, but a loss-of-function rather than a gain-of-function one.

Several lines of evidence strongly suggest that the observed loss of class B gene function is not based on a gene-silencing mechanism. First, sequence similarity between the GGM2 and the AP3 or PI cDNAs is less than 50% and thus appears to be too low for gene-silencing mechanisms as we know them (Matzke and Matzke, 1995). In addition, the transcriptional as well as post-transcriptional gene-silencing mechanisms known usually lead to a strong reduction in mRNA accumulation of the affected gene, which we did not observe for AP3 or PI (Figure 3). Finally, minor sequence variants of the GGM2 cDNA which, however, truncated the open reading frame, completely abolished the loss of class B function phenomenon, strongly suggesting that the effect requires GGM2 protein. As truncation of a part of the C-terminal region already suffices to abolish the effect, almost full-length GGM2 protein may be required to cause the loss of class B gene function.

It is well known that modification of transcription factors, such as mutations in the activation domain, can change them from activators to dominant inhibitors of transcription (Paz-Ares et al., 1990). It thus seems an attractive hypothesis that, under some developmental circumstances, GGM2 protein specifically binds to class B gene targets, but then inhibits rather than activates the transcription of these genes. MADS- and I-domain, plus (sometimes) part of the K-domain, are usually sufficient for dimerization and DNA-binding of MIKC-type MADS-domain proteins (Riechmann and Meyerowitz, 1997; Zachgo et al., 1995), while the C-terminal domain is required for the formation of higher-order protein complexes such as tetramers or ‘quartets’ (Egea-Cortines et al., 1999; Theißen, 2001). Thus the formation of multimeric protein complexes involving GGM2 may be required for the loss of class B gene function. Only these complexes may be able specifically to recognize some class B gene targets, and just these complexes may be stable enough to inhibit the transcription of these genes (Figure 6b).

However, deletion of part of the C-terminal domain reduced DNA-binding of GGM2, but not of some other MADS-domain proteins, in gel-retardation assays (C. Weiser and G.T., unpublished results); in addition, the C-terminal domain could also be important for protein stability in vivo. Thus the requirement of the complete GGM2 coding region for transgenic effects does not conclusively reveal the necessity of multimer formation for suppression of the class B gene function. The data are also compatible with GGM2 exerting its effects as a protein dimer (or, less likely, a monomer). Under an alternative hypothesis, even DNA binding may not be required: full-length (rather than truncated) versions of GGM2 may bind and thus titrate some factors that are specifically required for the function of AP3–PI complexes within the cell; hence these cells show a loss of class B gene function.

Is GGM2 a gymnosperm functional equivalent of class B floral homeotic genes?

We have shown here that GGM2 can partly substitute for AP3 and PI in ectopic expression experiments, and that this is probably not an indirect effect involving the ectopic upregulation of AP3 and PI, but rather directly caused by the GGM2 protein. This view is strongly supported by the finding that some aspects of class B gene substitution were observed even in putative class B gene null mutants.

Functional equivalence between genes from different organisms (see above) does not necessarily imply that the genes can mutually replace each other, as trivial incompatibilities can occur in systems whose components did not co-evolve for hundreds of millions of years. However, if functional substitution is observed, it may be used as a strong argument in favour of functional equivalence, because its explanation by mere chance or coincidence will, in most cases, be the less likely alternative. The capability of GGM2 to substitute for the function of class B floral homeotic genes in different experimental set ups thus strongly supports hypotheses about functional equivalence between the gymnosperm and angiosperm genes.

However, the question is: in what precise sense and to what extent do the class B genes of angiosperms and GGM2 have equivalent functions? A prerequisite for concluding functional equivalence is expression in homologous organs, as homologous genes may have different functions when expressed at different sites. The class B genes of angiosperms specify petal and stamen identity during flower development, and their expression is generally focused on these organs, but gymnosperms have neither petals nor stamens. However, their male reproductive units (microsporophylls) are the only structures in which the putative orthologues of class B genes from conifers and Gnetum are expressed (Mouradov et al., 1999; Sundström et al., 1999; Winter et al., 1999), so these may be homologous to angiosperm stamens. Thus it could well be the function of GGM2 to specify microsporophyll identity in Gnetum, so that the functions of GGM2 and the angiosperm class B genes may be equivalent in that respect.

This is compatible with the notion that it could well be the ancestral function of B genes in both gymnosperms and angiosperms to distinguish between male reproductive organs (where expression is on) and female reproductive organs (where expression is off) (Theißen et al., 2000). Differential expression of B genes may thus represent the primary sex-determination mechanism of all seed plants (Winter et al., 1999). The presence of orthologues of class B and class C genes in diverse gymnosperms suggests that the system for specification of reproductive organ identity in angiosperms was recruited from a similar system – perhaps involved in sex determination and the specification of reproductive organs – which was already present in the last common ancestor of all extant seed plants about 300 MYA (Theißen et al., 2000). The capability of GGM2 to substitute for the function of class B floral homeotic genes in different experimental set ups strongly suggests that GGM2, if expressed in Arabidopsis, specifically interacts with at least some of the partners of the Arabidopsis B genes, and thus may have highly related interaction partners in Gnetum. Among the homologous interaction partners are certainly target genes, but probably also other transcription factors. For example, in G. gnemon GGM2 could bind to the AGAMOUS-like protein GGM3 (Winter et al., 1999) and other MADS-domain proteins, thus constituting protein quartets that specify male reproductive organ identity (Figure 6b; Theißen, 2001; Theißen and Saedler, 2001). Similar complexes may also be formed in 35S::GGM2 Arabidopsis plants, but containing AG rather than GGM3 (Figure 6b).

Experimental procedures

Sequence alignments and construction of phylogenetic trees

Phylogenetic trees were constructed based on a set of MADS-domain protein sequences comprising most published DEF- and GLO-like proteins (as defined elsewhere: Theißen et al., 1996; Theißen et al., 2000), all closely related gymnosperm proteins published so far, and a set of GGM13-like (or Bsister) proteins (Becker et al., 2000; Becker et al., 2002; Mouradov et al., 1999; Sundström et al., 1999; Winter et al., 1999). Multiple sequence alignments were generated by using the pileup program of the gcg package (version 10.0) with a gap creation penalty of 8 and a gap extension penalty of 2 (default parameters). Based on alignments of the ‘170 domain’ (Theißen et al., 1996), distance matrices were generated using the protein distance algorithm, version 3.55c, which is based on the PAM model of amino acid transition. Phylogenetic trees were constructed as described (Münster et al., 1997), employing the neighbour-joining method, version 3.5, as implemented by the phylip program package. To statistically evaluate the tree topology, 100 bootstrap samples were generated as described (Münster et al., 1997). The tree presented as Supplementary Figure 1: http://www.blackwell-science.com/tpj/ was rooted by the phylip program retree, using the clade of GGM13-like sequences as an outgroup (Becker et al., 2000; Becker et al., 2002).

Genomic sequencing

Genomic DNA (1 µg) isolated from Gnetum gnemon leaf material was used as a template to amplify overlapping sequences spanning the complete loci of GGM2 and GGM15. A standard PCR program and the ‘Expand Long Template Kit’ (Roche, Mannheim, Germany) were used. Sequences of primers used are available upon request. The PCR products obtained were purified employing the PCR-Purification Kit (Qiagen, Hilden, Germany), and sequenced on both strands as described elsewhere (Münster et al., 2001).

In situ hybridization

For in situ hybridization experiments, PCR fragments of the I and K regions (≈200 bp) of the GGM2, AP3 and PI cDNAs were generated by use of primers containing a T7 promoter sequence. These fragments were used as templates for synthesizing digoxigenin-labelled RNA probes employing the DIG-RNA-labeling Kit (Roche) and T7 RNA polymerase. Arabidopsis plant material was fixed in 4% (w/v) paraformaldehyde in 1 × PBS (Sambrook et al., 1989) for 1 h at room temperature, following an overnight incubation at 4°C. Embedding of plant material, pretreatment and hybridization of the slides was carried out as described elsewhere (Winter et al., 1999). Final washing of the hybridized slides was done twice in 0.3 × SSPE at 55°C for 30 min. Immunological detection was carried out following DeBlock and Debrouwer (1993) using the Anti-DIG Fab fragment antibody (Roche).

Generation and analysis of Arabidopsis transformants

An NcoI–BamHI fragment containing the GGM2 coding region was generated using PCR amplification and cloned into the pRT100 vector (Töpfer et al., 1993) containing the CaMV 35S promoter. To generate the AP3::GGM2 construct, a HincII–NcoI PCR fragment spanning 1.7 kb of the AP3 promoter (pAP3 in Hill et al., 1998) was amplified by PCR and used to replace the 35S promoter sequence of the pRT100 vector. EcoRI PCR fragments, containing both the promoter sequence and the coding region of GGM2, were then cloned into the T-DNA vector pBAR-A (G. Cardon, MPIZ Köln, personal communication) containing the bar gene, which confers resistance to the herbicide BASTA. Agrobacterium tumefaciens GV3101 (Van Larabeke et al., 1974) was used to transform Arabidopsis thaliana (ecotype Columbia) plants by an infiltration method published by Bechtold et al. (1993). Plants were grown at 20°C, 16 h light/8 h darkness, and putative T1 transformants were selected by spraying with a 0.1% BASTA solution.

Seeds of the A. thaliana (Landsberg erecta) mutants ap3-1 (CS3085), ap3-3 (CS3086) and pi-1 (CS77) were obtained from the Arabidopsis Stock Center (USA).

The temperature-sensitive ap3-1 mutant plants were cultivated at 16°C (permissive temperature) and transformed as described above. The ap3-3 and pi-1 plants were crossed with T3 transgenic plants containing the constructs described above. To check progeny plants for homozygosity of ap3 or pi mutant alleles, genomic DNA of every single plant was isolated following a protocol provided by Edwards et al. (1991), and the respective AP3 or PI loci were amplified by PCR and sequenced.

Phenotypes were followed for at least three plant generations, and presence of the transgene was confirmed by DNA gel-blot hybridization. Transcription of the transgene in T3 plants was analysed by in situ hybridization. Flower phenotypes were examined using a Zeiss DSM940 (Zeiss, Oberkochen) scanning electron microscope essentially as described (Schwarz-Sommer et al., 1992).

Supplementary Material

The following material is available from http://www. blackwell-science.com/products/journals/suppmat/TPJ/TPJ1375/TPJ1375sm.htm

Figure S1. Phylogenetic tree showing the relationships between B and Bsister proteins. Proteins from gymnosperms are indicated by boxes, all other proteins are from angiosperms. The numbers next to some nodes give bootstrap percentages, shown only for relevant nodes. Some clade names are given at the right.

Figure S2. Comparison of domain structures of B proteins with those of other MADS-domain proteins from gymnosperms and angiosperms. Conceptual amino acid sequences of MADS-box genes were aligned by the GCG computer program. Asterisks at the sequence ends mark stop codons. Shaded boxes indicate MADS- and K-domains. In between these two domains is the I-region, in which a sequence indel which is specific for B and Bsister proteins is highlighted by a shaded bold line. A double head arrow highlights the sequence insertions specific for PRDGL from Pinus and GGM15 from Gnetum.

Figure S3. B and Bsister proteins share conserved motifs in their C-terminal regions. An alignment of C-terminal regions of some B and Bsister proteins from angiosperms and gymnosperms is shown. (I) DEF-like proteins from higher eudicotyledonous flowering plants (from top to bottom: AP3 from Arabidopsis thaliana, DEF from Antirrhinum majus, PMADS1 from Petunia hybrida). (II) DEF-like proteins from lower eudicots (PtAP3 from Pachysandra terminalis; RbAP3-1 from Ranunculus bulbosus) or a basal angiosperm/magnoliid dicot (PhAP3 from Peperomia hirta). (III) Bsister proteins, from a gymnosperm (GGM13 from Gnetum gnemon) and a basal angiosperm (AeAP3-2 from Asarum europaeum). (IV) B proteins from gymnosperms (GGMn from Gnetum gnemon; DALn from Picea abies; PrDGL from Pinus radiata). (V) GLO-like proteins from higher eudicots (GLOBOSA from Antirrhinum majus; FBP1 from Petunia hybrida, PI from Arabidopsis thaliana) or lower eudicots, respectively (CpPI from Caltha palustris, RbPI-1 from Ranunculus bulbosus). Gene names are according to Kramer and Irish (2000). Regions defining motifs identified by Kramer et al. (1998) have been boxed. Within boxes of the alignment the most frequent amino acids at any position are highlighted in boldface.

Accession numbers of genomic sequences reported in this paper: AJ421709(GGM2); AJ421710(GGM15).