Molecular interactions of orthologues of floral homeotic proteins from the gymnosperm Gnetum gnemon provide a clue to the evolutionary origin of ‘floral quartets’


(fax +49 3641 949552; e-mail


Several lines of evidence suggest that the identity of floral organs in angiosperms is specified by multimeric transcription factor complexes composed of MADS-domain proteins. These bind to specific cis-regulatory elements (‘CArG-boxes’) of their target genes involving DNA-loop formation, thus constituting ‘floral quartets’. Gymnosperms, angiosperms’ closest relatives, contain orthologues of floral homeotic genes, but when and how the interactions constituting floral quartets were established during evolution has remained unknown. We have comprehensively studied the dimerization and DNA-binding of several classes of MADS-domain proteins from the gymnosperm Gnetum gnemon. Determination of protein–protein and protein–DNA interactions by yeast two-hybrid, in vitro pull-down and electrophoretic mobility shift assays revealed complex patterns of homo- and heterodimerization among orthologues of floral homeotic class B, class C and class E proteins and Bsister proteins. Using DNase I footprint assays we demonstrate that both orthologues of class B with C proteins, and orthologues of class C proteins alone, but not orthologues of class B proteins alone can loop DNA in floral quartet-like complexes. This is in contrast to class B and class C proteins from angiosperms, which require other factors such as class E floral homeotic proteins to ‘glue’ them together in multimeric complexes. Our findings suggest that the evolutionary origin of floral quartet formation is based on the interaction of different DNA-bound homodimers, does not depend on class E proteins, and predates the origin of angiosperms.


In higher eudicots, the combinatorial interaction of class A, B, C and E floral homeotic genes determines floral organ identity (Coen and Meyerowitz, 1991; Ditta et al., 2004; Pelaz et al., 2000; Theissen, 2001). Briefly, class A and E genes determine the identity of the outer whorl sepals, class A, B and E genes together specify petal identity, class B, C and E genes control stamen development and class C genes together with class E genes function in carpel specification. The vast majority of the floral homeotic genes encode MADS-domain transcription factors. They belong to specific subfamilies, with class B genes being DEFICIENS/GLOBOSA (DEF/GLO)-like genes, class C genes AGAMOUS (AG)-like genes, and class E genes SEPALLATA1 (SEP1)-like genes or AGL6-like genes (Becker and Theissen, 2003; Rijpkema et al., 2009).

MADS-domain proteins bind as dimers to DNA sequence elements termed CArG-boxes (for ‘CC-Arich-GG’, consensus sequence: 5′-CC(A/T)6GG-3′) (Huang et al., 1993; Riechmann et al., 1996b; Schwarz-Sommer et al., 1992). Beyond dimerization, it was proposed by the floral quartet model that floral homeotic proteins act as DNA-bound tetrameric complexes to specify floral organ identity (Theissen, 2001; Theissen and Saedler, 2001). According to this model, the identity of every floral organ is determined by a specific DNA-bound tetrameric complex of MADS-domain proteins. In particular, the model predicts that sepals are specified by a complex of two class A and two class E proteins; petal identity is governed by a complex consisting of two class B, one class A and one class E protein; a tetramer of two class B, one class C and one class E protein functions in stamen specification and, finally, carpels are specified by a complex of two class C and two class E proteins (Krizek and Fletcher, 2005; Theissen, 2001; Theissen and Saedler, 2001). In recent years, experimental evidence confirmed that MADS-domain proteins indeed form multimeric complexes and that these complexes bind to DNA (Ferrario et al., 2003; Honma and Goto, 2001; Immink et al., 2009; Melzer and Theissen, 2009; Melzer et al., 2009; Tonaco et al., 2006).

With the exception of the class A floral homeotic function, the basic principles of floral organ specification probably apply to all angiosperms (Irish, 2009; Kim et al., 2005; Soltis et al., 2007; Theissen et al., 2000; Whipple et al., 2007). However, even though much is known about the mechanisms of floral developmental control in angiosperms, it is still largely unclear how the flower originated during evolution (Bateman et al., 2006; Baum and Hileman, 2006; Frohlich, 2003; Theissen and Melzer, 2007). The origin and evolution of the highly conserved regulatory network controlling floral organ specification is intimately intermingled with the evolutionary origin and diversification of the angiosperm flower in a comprehensive framework of evolutionary developmental biology (‘evo-devo’) (Theissen et al., 2000). To better understand the origin of the flower it is therefore inevitable to study gymnosperms, the closest extant relatives of the angiosperms. Extant gymnosperms comprise conifers, gnetophytes, cycads, and Ginkgo (Bowe et al., 2000; Chaw et al., 2000; Frohlich and Parker, 2000).

Reproductive organs of gymnosperms differ substantially from their angiosperm counterparts. For example, while angiosperm flowers are primarily bisexual and carry a perianth surrounding the reproductive structures, gymnosperm reproductive organs are unisexual and lack a perianth. Also, gymnosperms lack a carpel enclosing the ovules. However, despite these differences, putative orthologues of many genes controlling flower development in angiosperms have been found in gymnosperms. Most importantly, gymnosperms possess DEF/GLO-like genes as well as AG-like genes (Fukui et al., 2001; Futamura et al., 2008; Jager et al., 2003; Mouradov et al., 1999; Rutledge et al., 1998; Sundström et al., 1999; Tandre et al., 1995; Winter et al., 1999). Similar to the situation in angiosperms, AG-like genes from gymnosperms are predominantly expressed in both male and female reproductive organs (Jager et al., 2003; Rutledge et al., 1998; Sundström et al., 1999; Tandre et al., 1995; Winter et al., 1999), while the expression of gymnosperm DEF/GLO-like genes is restricted to male reproductive organs (note that there are no petals in gymnosperms) (Fukui et al., 2001; Mouradov et al., 1999; Sundström et al., 1999; Winter et al., 1999). This indicates that the genetic mechanisms determining reproductive organ identities are similar in all seed plants (i.e. angiosperms and gymnosperms) (Theissen et al., 2000, 2002; Winter et al., 1999). Accordingly, in the lineage that led to the most recent common ancestor of angiosperms and gymnosperms expression of AG-like genes may have been established in reproductive organs to distinguish them from vegetative structures (in which AG-like gene expression is ‘off’). Furthermore, the male-specific expression of DEF/GLO-like genes may have led to the distinction between male and female reproductive structures (Theissen et al., 2000; 2002; Winter et al., 1999).

For class E floral homeotic genes, the situation is slightly more complicated. Until recently, it was only the subfamily of SEP1-like genes for which a class E floral homeotic function had been demonstrated (Ditta et al., 2004; Pelaz et al., 2000). However, studies in petunia, rice and maize indicated that another subfamily of MADS-box genes, termed AGL6-like genes, can also perform a class E floral homeotic function (Li et al., 2010; Ohmori et al., 2009; Rijpkema et al., 2009; Thompson et al., 2009). AGL6-like genes and SEP1-like genes constitute sister clades in many phylogenetic reconstructions (Becker and Theissen, 2003; Nam et al., 2003; Ohmori et al., 2009; Rounsley et al., 1995). Importantly, SEP1-like genes have not been isolated from any gymnosperm so far, suggesting that these genes have been lost in the lineage that led to extant gymnosperms (Theissen and Melzer, 2007; Zahn et al., 2005). In contrast, AGL6-like genes are well known from gymnosperms. Similar to angiosperm class E floral homeotic genes, gymnosperm AGL6-like genes are mainly expressed in reproductive organs (Becker and Theissen, 2003; Mouradov et al., 1998; Shindo et al., 1999; Tandre et al., 1995; Winter et al., 1999). Thus, the common ancestor of AGL6-like and SEP1-like genes might have possessed already a function in floral organ specification (Melzer et al., 2010).

It would be desirable to test this and other hypotheses on gymnosperm reproductive organ specification by genetic methods. Unfortunately, however, the informative phylogenetic position of gymnosperms is diametrically opposed to their amenability to genetic analyses. Long generation times, inefficient genetic transformation procedures and the lack of both known genome sequences and comprehensive mutant collections necessitate the use of indirect methods to understand how reproductive organs are specified in gymnosperms. As the general importance of protein–protein and protein–DNA interactions for MADS-domain protein function is becoming increasingly evident, it appears reasonable to study the interaction properties of MADS-domain proteins from gymnosperms.

Here, we present a comprehensive analysis of protein–protein and protein–DNA interactions of close relatives of floral homeotic proteins from the gymnosperm Gnetum gnemon. All known G. gnemon orthologues of MADS-domain proteins functioning in floral organ development in angiosperms have been included in the analysis. Thus, besides DEF/GLO-like, AG-like and AGL6-like proteins also Bsister proteins were studied. Bsister proteins represent the sister clade of DEF/GLO-like proteins and have been shown to function in ovule and seed development (de Folter et al., 2006; Kaufmann et al., 2005a; Nesi et al., 2002; Prasad et al., 2010).

Our results suggest that the AG-like protein GNETUM GNEMON MADS3 (GGM3) can form floral quartet-like complexes on its own and builds even more stable complexes together with the DEF/GLO-like protein GGM2, thus demonstrating the ability of gymnosperm MADS-domain proteins to form multimeric complexes similar to their angiosperm counterparts. However, in contrast to angiosperms, multimeric complex formation does not depend on orthologues of class E proteins, indicating that the MADS-domain protein interaction network in gymnosperms is less constrained than that in angiosperms.


Predominant heterodimerization rather than homodimerization of G. gnemon MADS-domain proteins

To determine how MADS-domain proteins from G. gnemon interact with each other, GAL4-based yeast two-hybrid assays as well as in vitro pull-down assays were used. The focus of our analyses was on close relatives of floral homeotic proteins, including DEF/GLO-like (GGM2 and GGM15), Bsister (GGM13), AG-like (GGM3) and AGL6-like (GGM9 and GGM11) proteins. For yeast two-hybrid assays interactions were assessed reciprocally by growing colonies on selective medium lacking leucine, tryptophane and histidine. Representative examples of our results are shown in Figure 1(a). Extensive heterodimerization between the different proteins was identified. For example, the DEF/GLO-like protein GGM2 interacted with members from all other subfamilies tested, including the AG-like protein GGM3. Except for GGM2, however, no homodimerization could be detected by yeast two-hybrid assays. About half of the heterodimeric interactions (between GGM2/GGM3; GGM2/GGM9, GGM3/GGM9 and GGM3/GGM13) were only detected in ‘one direction’ (i.e. the interaction was detected when one of the partners was expressed as fusion with the activation domain but not when it was expressed as fusion with the DNA-binding domain of GAL4). Similar observations have been made also in other studies of MADS-domain protein interactions (Immink et al., 2003; Leseberg et al., 2008).

Figure 1.

 Protein–protein interactions of G. gnemon MADS-domain proteins.
(a) Representative examples of the results obtained with the yeast two-hybrid system. Photographs show colony growth on selective –Leu/–Trp/–His media. For each interaction tested, yeast cells were spotted in 10-fold serial dilution (from left to right). Proteins that were expressed as fusions with the GAL4 activation domain (AD) are shown horizontally, proteins expressed as GAL4 binding domain (BD) fusions are shown vertically.
(b) Protein–protein interactions tested by in vitro pull-down assays. His6-tagged GGM proteins (‘His6-GGMn’) were used as bait, bound to Ni–NTA beads and incubated with untagged, radioactively labelled GGM prey proteins as noted above the gel. On the left side of each gel, 2 μl of in vitro translated prey protein solution were loaded (input) for size comparison. ‘Δ’ denotes a negative control in which the ‘bait’in vitro translation extract was programmed with an expression vector that did not contain a cDNA insert. Bands are numbered according to the number of the respective GGM protein. In one case GGM9 bait protein is marked with an asterisk to indicate that it has also been radioactively labelled and is thus visible on the gel.
(c) Comparison of the interaction data obtained from in vitro pull-down and yeast two-hybrid assays. Grey triangles indicate an interaction between two proteins. Interactions that have been found in one direction only in the yeast two-hybrid assay are not differentiated from the ones that have been found reciprocally. The weak growth of GGM11-AD/GGM3-BD transformants observed in some yeast two-hybrid experiments (Figure 1a) was not detected reproducibly; it was therefore not counted as indicating an interaction between GGM3 and GGM11 here. n.d., not do-able (the interaction could not be determined in pull-down assays because of unspecific binding of the proteins to the Ni-NTA agarose beads). n.t., not tested (the interaction was not tested in pull-down assays).

To establish whether the interactions found with the yeast two-hybrid system can be confirmed by another method, in vitro pull-down assays were employed. For this purpose, one protein was in vitro translated as a His6-tagged fusion, bound to nickel charged agarose beads and subsequently incubated with a non-tagged, radioactively labelled partner. Representative results are shown in Figure 1(b).

In the 17 cases in which both yeast two-hybrid and pull-down assays were successfully applied, equivalent results (interaction or no interaction) were found in 14 cases (82%; Figure 1c). One heterodimer (GGM2/GGM9) was detected with the yeast two-hybrid assay only, another one (GGM3/GGM11) only with pull-down assays (Figure 1c). Remarkably, homodimerization was not observed with both methods except for GGM2/GGM2 dimers, which were only detected with the yeast two-hybrid assay, however (Figure 1c). It is also important to note that none of the methods revealed any interaction partner for the DEF/GLO-like protein GGM15.

Many G. gnemon MADS-domain proteins can bind to DNA both as homodimers and heterodimers

MADS-domain proteins are known to bind as dimers to DNA sequences termed CArG-boxes. We thus used electrophoretic mobility shift assays (EMSAs) to analyse DNA-binding activity of different combinations of G. gnemon MADS-domain proteins. To estimate DNA-binding, probe A, a 51-bp long double-stranded oligonucleotide containing a CArG-box derived from the regulatory intron of AG was used. Except GGM15 all of the tested proteins were capable of binding to this DNA probe (Figure 2). GGM3, GGM11 and GGM13 formed protein–DNA complexes with a gel electrophoretic mobility similar to that of the GGM2–DNA complex, for which there is good evidence that it represents a homodimer bound to a CArG-box (Figure 2) (Winter et al., 2002b). This finding strongly indicates that also GGM3, GGM11 and GGM13 bind to CArG-boxes as homodimers. This result was confirmed by using mixtures of full-length and C-terminally deleted versions of the same protein obtained by co-translation, as generally described (Huang et al., 1996; Melzer et al., 2009). In these cases, three different retarded bands were observed, representing DNA probes that have bound (from low to high mobility) homodimers of full-length proteins, heterodimers of one full length and one truncated protein, and homodimers of truncated proteins, respectively (Figure 2). In contrast, GGM9 constituted a protein–DNA complex of unusually low electrophoretic mobility (Figure 2). This observation cannot be easily attributed to the molecular weight or charge of GGM9, which is very similar to that of the other proteins analysed. It is unclear whether the low electrophoretic mobility is due to unanticipated changes in DNA conformation upon GGM9 binding, due to an unusual structure of the GGM9 protein, or due to binding of GGM9 to DNA as a multimer rather than a dimer, even though the latter scenario appears the most likely one.

Figure 2.

 DNA binding of GGM proteins.
Proteins applied are noted above the gel. ‘ΔC’ is used to indicate C-terminal deleted proteins. Heterodimeric complexes are indicated by white triangles. Homodimers are indicated by black triangles. GGM9–DNA complexes, for which the stoichiometry of binding is unclear, are indicated by double arrows. The GGM9–GGM2–DNA complex is interpreted to represent a heterodimer as its electrophoretic mobility is similar to that of the GGM11–GGM2–DNA complex. Also, the co-translation of GGM3 and GGM9 yielded a protein–DNA complex with an electrophoretic mobility slightly lower than that of GGM3–GGM3–DNA complexes and thus is likely to represent a GGM3–GGM9 heterodimer. Co-translation of GGM15 with other GGM-proteins yielded only complexes with an electrophoretic mobility almost identical to the previously identified homodimers. We thus concluded that GGM15 did not form protein–DNA complexes with any of the partners tested. Homodimeric complexes are not always visible when heterodimer formation was tested, possibly because the vast majority of protein was assembled into heterodimeric complexes. For unknown reasons, both GGM2ΔC and GGM11 alone occasionally formed two protein–DNA complexes. ‘Δ’ indicates negative controls in which in vitro translation lysate programmed with a vector that did not contain a cDNA insert was added. ‘free DNA’ indicates the un-bound DNA probes. ‘M’ denotes lanes in which a radioactively labelled DNA marker (NEB 100 bp DNA ladder) was applied.

To assess the formation of DNA-binding heterodimers, proteins were co-translated and afterwards co-incubated with probe A. Heterodimerization was inferred when a band of intermediate mobility compared with the two DNA-bound homodimers was observed. When the homodimers formed by the two co-translated full length proteins had almost the same electrophoretic mobility, a C-terminal deleted construct of one of the proteins was used. Representative examples of EMSA results are shown in Figure 2. The DEF/GLO-like protein GGM2 was capable of forming DNA-binding heterodimers with several partners (GGM9, GGM11 and GGM13), whereas the AG-like protein GGM3 interacted with GGM9 and GGM11 to constitute a DNA-binding heterodimer. In addition, an interaction between GGM11 and the Bsister protein GGM13 was detected. Notably, GGM2 and GGM3, which interacted in yeast two-hybrid as well as pull-down assays, did not form a DNA-binding heterodimer in our hands (Figures 2 and 3a).

Figure 3.

 A scenario for the evolution of floral quartets from gymnosperms to angiosperms.
(a) Comparison of the data obtained from the yeast two-hybrid and in vitro pull down assays with those from the EMSA experiments. Grey triangles indicate an interaction between the proteins. For yeast two-hybrid and pull down assays, an interaction was counted only when observed with both methods. As in Figure 1, interaction tests not do-able or interactions not tested in pull-down assays are designated by n.d. and n.t., respectively.
(b) Rationale of conclusions concerning dimeric and higher order interactions. The G. gnemon DEF/GLO-like protein GGM2 and the AG-like protein GGM3 interacted in yeast two-hybrid and pull down assays, but did not form DNA-binding heterodimers with each other. In contrast, DNA-binding of GGM2 and GGM3 homodimers was detected. Thus, a heterotetrameric complex of GGM2 and GGM3 is proposed. Such a complex might function in specification of male organs. A homotetramer composed of GGM3 might specify female organs.
(c) Complexes proposed to specify carpels and stamens in A. thaliana. In contrast to G. gnemon, both of the hypothesized complexes contain SEP1-like (SEP) proteins in addition to DEF/GLO-like (AP3 and PI) and AG-like (AG) proteins.
(d, e) Comparison of the heterodimeric MADS-domain protein interaction network from A. thaliana (d) and G. gnemon (e). Proteins are indicated by ovals and protein–protein interactions by lines. Proteins are grouped according to their affiliation to different subfamilies. For G. gnemon, an interaction was shown when detected in yeast two-hybrid and/or pull down assays. Data from A. thaliana have been taken from de Folter et al. (2005) and Yang et al. (2003b).

In all cases (i.e. homodimers and heterodimers) binding of the proteins to the labelled DNA was strongly reduced when a 100-fold excess of unlabelled probe A was added, but not when a 100-fold excess of an oligonucleotide not containing a CArG-box was added (data not shown). This demonstrated sequence-specificity of DNA-binding of the MADS-domain proteins.

GGM2 and GGM3 loop DNA in floral quartet-like complexes

Our findings that both the DEF/GLO-like protein GGM2 and the AG-like protein GGM3 can form DNA-binding homodimers, but constitute heterodimers that are not capable of binding to DNA suggest that the interaction between GGM2 and GGM3 usually takes place in a multimeric, floral quartet-like complex. The GGM2–GGM3 interaction may mediate the interaction between DNA-bound homodimers of GGM2 and GGM3 by looping the DNA between the two homodimers (Figure 3b).

To explore whether GGM2 and GGM3 indeed loop the intervening DNA upon binding to two CArG-boxes, DNase I footprint, a sensitive assay for the formation of floral quartet-like complexes (Melzer and Theissen, 2009; Melzer et al., 2009), was applied. We first assayed GGM2 and GGM3 separately to determine their ability to form looped complexes without a partner protein (Figure 4a). Incubation of GGM3 with probe B, that carries two CArG-boxes that are spaced by six helical turns yielded a pattern typical for looped DNA, with sites of enhanced and diminished sensitivity between the two CArG-boxes spaced in approximately 5 bp intervals (Figure 4a, compare lanes 12 and 13; Figure 4b). For GGM2-bound DNA, sites of enhanced and diminished sensitivity between the two CArG-boxes were barely detectable (Figure 4a, compare lanes 6 and 7; Figure 4b). This situation indicates that GGM2 alone induces loop formation only very weakly, at best. In contrast, when GGM2 and GGM3 were co-incubated (Figure 4a, compare lanes 8 and 7; Figure 4b) the pattern of sites with enhanced and diminished sensitivity was more pronounced compared with GGM3 or GGM2 alone (Figure 4b). As protein–DNA complexes were separated from free DNA prior to analysis of the digestion patterns, this indicates that a DNA-bound complex containing GGM2 and GGM3 induces DNA loops that are more stable than those induced by GGM3 alone (Figure 4b).

Figure 4.

 DNase I footprint assays.
(a) DNA probes carrying two CArG-boxes were incubated with in vitro translated proteins as noted above the lanes. ‘B’ and ‘C’ indicates the DNA probe being used (i.e. CArG-boxes were spaced by 6 and by 6.5 helical turns, respectively). A SEP3 protein that lacks the K3 and C-domain (SEP3ΔK3C) and does not induce DNA-looping (Melzer et al., 2009) was used as a negative control for DNA-loop formation. Lane numbers are marked at the bottom of the figure. ‘Δ’ denotes lanes in which reticulocyte extract programmed with vector that did not contain a cDNA insert was added to the DNA probe. ‘Uncut DNA’ denotes lanes in which DNA not treated with DNase I was added. G+A and C+T sequencing reactions of the probes are shown for comparison. The CArG-box regions are indicated.
(b) Quantitative analysis of DNase I footprint signals shown in (a). The plots show the change of sensitivity to DNase I digestion after protein binding in single base pair-steps, using free DNA as a reference. Values were corrected for differences in DNA loading as described (Melzer et al., 2009).

If the patterns of sites with enhanced and diminished sensitivity are really caused by protein-induced DNA looping, they are expected to depend on the stereo-specific orientation of the CArG-boxes. If the CArG-boxes are spaced by 6.5 instead of 6 helical turns, protein-induced looping requires energetically costly twisting of the DNA and thus is expected to happen less likely. Indeed, for GGM2 co-incubated with GGM3, especially the pattern of sites with diminished sensitivity was weaker when probe C was used, on which the two CArG boxes were spaced by 6.5 helical turns (Figure 4a, compare lanes 8 and 7 with lanes 9 and 10; Figure 4b). This supports our conclusion that GGM2 and GGM3 together induce DNA looping upon binding. In contrast, when GGM2 or GGM3 were separately incubated with probe C, no considerable change in sensitivity was detected compared with experiments where probe B was used (Figure 4a, compare lanes 5 and 4 to lanes 6 and 7, as well as lanes 11 and 10 to lanes 12 and 13; Figure 4b). This provides further evidence that GGM2 alone does only weakly loop DNA. For GGM3, this lack of stereo-specificity in binding was unexpected. However, the well pronounced sites of enhanced and diminished sensitivity towards DNase I digestion indicate that GGM3 loops the DNA upon binding. As similar results were obtained for probes B and C we suggest that GGM3 loops DNA irrespective of the stereo-specific orientation of the CArG-boxes. This lack of stereo-specificity in binding might be explained by an intrinsic flexibility of the GGM3 protein that enabled complex formation even when the two CArG-boxes were separated by 6.5 helical turns.

When probe B was applied together with GGM2 or with GGM3 in EMSAs, a protein–DNA complex of low electrophoretic mobility was detected (marked by asterisks in Figure 5a, lanes 13 and 15). Using a mixture of full length and C-terminal truncated GGM3 indicated that this complex consists of four proteins bound to probe B (Figure 5b). Surprisingly, however, a complex of four proteins bound to DNA was already detected when GGM3 was incubated with probe D, which is of the same length as probe B but carries only one CArG-box (Figure 5a, lanes 6–8; Figure 5b). In contrast, GGM2 bound only as a dimer to probe D (Figure 5a, lanes 2–4). We speculate that GGM3-homotetramers bind on probe D to the CArG-box and to a second unknown site to which affinity is very weak. Nevertheless, this second site appears to be required, because GGM3 does not bind to a shorter DNA probe with one CArG-box as a tetramer, but only as a dimer (Figure 2). Binding to this second site might facilitate a homotetramer formation of GGM3 that is sufficiently stable to compensate for the absence of a second CArG-box for DNA-binding. This stability in homotetramer formation might also provide an additional explanation as to why a lack of stereo-specificity in binding was observed in DNase I footprint experiments.

Figure 5.

 DNA binding of GGM2 and GGM3 to long DNA fragments carrying one or two CArG boxes.
Band assignment is as in Figure 2. DNA probes are noted below the gel. Probes applied are symbolized as black lines with grey rectangles representing the CArG-boxes. Probe D carries one CArG-box; probe B carries two CArG-boxes. In (a) complex formation of GGM2 and GGM3 on probe B and D is shown. Triangles mark single dimers bound to the DNA-probes, bands marked by asterisks presumably represent complexes with four proteins bound to a DNA fragment. Circles denote complexes of unknown nature that were not observed reproducibly. In (b), different ratios of GGM3 and GGM3ΔC were used to demonstrate that four GGM3 proteins bind to probe B and D. GGM3ΔC:GGM3 ratios were obtained by mixing plasmid templates for in vitro translation as noted above the gel. Protein–DNA complexes are indicated with lowercase letters. Possible complex compositions are illustrated by the icons on the right. Full and partial circles represent GGM3 and GGM3ΔC, respectively.

Also, as is evident from Figure 5(a) (lane 8), the C-terminal part of the protein is dispensable for GGM3 multimerization. Similar observations have been made previously for other MADS-domain proteins as well (Immink et al., 2009; Melzer et al., 2009; Yang and Jack, 2004).

Interestingly, when GGM2 and GGM3 were applied together, binding behaviour to probe B as well as to probe D was similar as for GGM3 alone (Figure 5a, compare lanes 5 and 6, as well as lanes 14 and 15). However, the electrophoretic mobility of the respective protein–DNA complexes was slightly higher than that of GGM3 homotetramers, indicating that a heterotetramer of GGM2 and GGM3 was reconstituted.


In this work we explore the interaction properties of gymnosperm MADS-domain transcription factors that are closely related to floral homeotic proteins and are hence supposed to have similar functions in specifying organ identity. To determine the pattern of protein–protein interactions, yeast two-hybrid and in vitro pull-down assays were used. The experimental premises for two proteins to interact are rather different for the two methods. Thus, a positive result obtained with both methods likely indicates a genuine interaction between MADS-domain proteins.

All of the genes encoding the proteins analysed here are expressed in reproductive structures, while expression in leaves was not detected (Becker et al., 2000; 2002; 2003; Winter et al., 1999). In general, genes of GGM proteins that interacted are co-expressed, indicating that the interactions determined here can potentially take place in G. gnemon.

The expression of the AGL6-like gene GGM9 is approximately equally strong in male and female cones (Winter et al., 1999), whereas GGM11, the second AGL6-like gene from G. gnemon, is expressed markedly stronger in female cones than in male cones (Winter et al., 1999). In addition, GGM11 but not GGM9 mRNA is detectable in the sterile ovules of male cones (Becker et al., 2003). In line with this, we found that GGM11, but not GGM9 interacts with the female-specific Bsister protein GGM13. Together, this suggests that GGM11 plays a more prominent role in female organ development than GGM9.

The DEF/GLO-like genes GGM2 and GGM15 are specifically expressed in male cones. Even though the expression patterns of both genes are similar, especially in early stages of development GGM2 seems to be expressed throughout the male reproductive unit whereas GGM15 expression is more spatially restricted to the antherophore (Winter et al., 1999; Becker et al., 2003). Thus, it was hypothesized that GGM2 is more generally involved in specification of male reproductive units whereas the function of GGM15 is restricted to the antherophore. Our data are consistent with this hypothesis in suggesting that GGM2 has a diverse set of interaction partners whereas interaction partners for GGM15 remain to be found.

The only case of two genes encoding proteins that interact in our assays but that are not co-expressed is GGM2 and GGM13. It cannot be ruled out that these genes are co-expressed in a spatially or temporally very restricted way that escaped attention so far, especially as previous gene expression analyses have not been very detailed. However, as DEF/GLO-like and Bsister proteins originated by a duplication of a presumably homodimerizing ancestor, the GGM2–GGM13 interaction could also be a remnant of this ancestral homodimerizing property.

Some MADS-domain proteins of G. gnemon form floral quartet-like complexes

Even though certain interactions were reliably detected by yeast two-hybrid and pull-down assays, comparison of these data sets with the EMSA results yielded some notable differences (Figure 3a). Most importantly, heterodimers of GGM2 and GGM3 as well as of GGM3 and GGM13 were observed in yeast two-hybrid and pull-down assays, but the respective proteins did not form DNA-binding heterodimers in EMSAs. However, DNA-binding homodimers of GGM2, GGM3 and GGM13 were detected in EMSAs. We thus speculate that the interaction between GGM2 and GGM3 as well as the interaction between GGM3 and GGM13 occurs in multimeric complexes to mediate the interaction between two DNA-bound homodimers. In line with this, DNase I footprint data suggest that GGM2 and GGM3 can bind to DNA fragments carrying two closely spaced CArG-boxes by looping the intervening DNA. We thus conclude that homodimers of GGM2 and GGM3 bind to separate sites on the DNA and that a tetramer is formed by the interaction between GGM2 and GGM3 (Figure 3b). DNase I footprint and EMSA data further suggest that GGM3 alone is capable of forming homotetramers on appropriate DNA fragments. This property was not or only to a much weaker extent observed for GGM2.

As we observed DNA-bound GGM3-homotetramers, one may also expect a GGM3–GGM3 interaction to be detected in yeast two-hybrid or pull-down assays. Such an interaction could mediate the association of two GGM3 homodimers. However, we did not observe a GGM3–GGM3 interaction in the absence of DNA-binding, possibly because it is too weak to be detected with the methods used here. In contrast, when two GGM3 homodimers are bound in vicinity to each other on the same DNA-fragment, their local concentration is increased and hence an interaction is supposed to happen more likely, leading to tetramer formation.

Taken together, it appears likely that at least some gymnosperm MADS-domain proteins possess the ability to form tetrameric, DNA-bound complexes that are similar to angiosperm floral quartets. In angiosperms, tetrameric complexes composed of AG-like and SEP1/AGL6-like proteins or of AG-like, SEP1/AGL6-like and DEF/GLO-like proteins are postulated to function in carpel and stamen specification, respectively (Figure 3c) (Theissen, 2001; Theissen and Saedler, 2001; Melzer et al., 2010). In addition, complexes containing Bsister-, AG-like and SEP1/AGL6-like proteins are hypothesized to function in ovule development (Kaufmann et al., 2005a). Our data are compatible with the hypothesis that in G. gnemon homotetramers of GGM3 function in female organ specification whereas GGM2–GGM2/GGM3–GGM3–DNA complexes confer male organ identity (Figure 3b). Furthermore, we hypothesize that a GGM3–GGM3/GGM13–GGM13 complex is involved in ovule development.

Conservation and diversity among MADS-domain protein interactions

In A. thaliana (and possibly also in most other angiosperms), few dimeric interactions occur between MADS-domain proteins belonging to the different clades studied here. Rather SEP1-like and possibly also AGL6-like proteins act as mediators that ‘glue’ MADS-domain proteins from different clades together in multimeric complexes (Figure 3d) (Immink et al., 2009). In contrast, our data indicate that in G. gnemon extensive dimeric interactions between MADS-domain proteins belonging to different subfamilies occur (Figure 3e). We propose that this reflects the property of gymnosperm MADS-domain proteins to form multimeric complexes also without SEP1-like or AGL6-like proteins. This is supported by evidences indicating that the G. gnemon AG-like protein GGM3 alone or in combination with the DEF/GLO-like protein GGM2 is capable of forming multimeric complexes in the absence of an additional mediator.

The question arises as to why and how differences in multimer formation between angiosperms and gymnosperms were established during evolution. It might well be that nothing but neutral changes led to these differences. Interaction patterns specifying certain developmental programs may change during evolution without affecting the developmental program itself (Tanay et al., 2005; Tsong et al., 2006). The regions determining interaction strength and specificity of plant MADS-domain proteins are often rather small (Immink et al., 2009; Melzer and Theissen, 2009; Riechmann et al., 1996a; Yang and Jack, 2004; Yang et al., 2003a,b). Therefore, destroying or establishing a heterodimer interaction surface or switching from multimerization to dimerization and back might require only a few of mutational changes, thus creating a flexible network of protein interactions.

However, if changes in interaction behaviour reflect neutral changes, the question arises as to why the interaction network of MADS-domain proteins appears to be more or less conserved at least in higher eudicots (de Folter et al., 2005; Immink et al., 2003; Leseberg et al., 2008). Due to the intimate link between the floral homeotic proteins and the evolution of the flower we are tempted to hypothesize that the dependence on SEP1/AGL6-like proteins for multimer formation in eudicots provided some selective advantage over the less hierarchical organised network in gymnosperms.

For example, the need for SEP1/AGL6-like proteins in eudicots to constitute multimeric complexes and hence to specify reproductive organ identity might introduce an additional level of developmental robustness. This robustness would be similar to the obligate heterodimerization of DEF/GLO-like proteins observed in higher eudicots. In this case, the floral homeotic B function can only be performed if both partners, a DEF-like and a GLO-like protein, are present, heterodimerize, and establish an autoregulatory feedback loop (Schwarz-Sommer et al., 1992). Computational analyses suggest that the evolution of obligate heterodimerization has increased the robustness of the system against stochastic noise in gene expression (Lenser et al., 2009). Along these lines, by making the formation of reproductive organs depending from yet another partner, the system might be more robust against random fluctuations of any of the selector proteins. Also, the necessity to express several proteins at once to determine organ identity decreases the risk of heterochrony and heterotopy (i.e. organ formation at the wrong time or place), and thus may contribute to the canalization of flower development (see also Winter et al., 2002b).

It is well possible that also AGL6-like proteins from gymnosperms are capable of mediating multimeric complex formation, although this remains to be determined experimentally. For instance, one of the AGL6-like proteins, GGM9, shows an unusual low electrophoretic mobility in EMSA compared with other protein–DNA complexes (Figure 2), which might indicate that GGM9 alone is already capable of forming multimeric complexes. Also, both AGL6-like proteins, GGM9 and GGM11, formed DNA-binding dimers with GGM2 and GGM3 (Figures 2 and 3a). Thus, the interaction between GGM2 and GGM3 might also function to constitute a DNA-bound GGM2–GGM9/GGM3–GGM9 or GGM2–GGM11/GGM3–GGM11 tetramer. This tetramer would be similar to a complex proposed to specify stamen identity in angiosperms.

However, from the network topology in G. gnemon it is not evident that gymnosperm AGL6-like proteins act in a similar way as hubs as SEP1/AGL6-like proteins do in A. thaliana (Figure 3d,e). Therefore, even if G. gnemon AGL6-like proteins are capable of forming multimeric complexes with DEF/GLO-like and AG-like proteins, there may be, in contrast to what is postulated for eudicots, no strict dependence on these additional mediators of complex formation. Thus, in contrast to the obligate dependence of multimer formation on SEP1/AGL6-like proteins in higher eudicots, the potential facultative role of gymnosperm AGL6-like proteins in mediating multimer formation may keep the system unconstrained and hence may not lead to the same selective advantages based on developmental robustness and canalization of development.

It is clear that more data need to be gathered to test these simple hypotheses, and that more complex scenarios currently cannot be excluded. It is possible, for example, that complexes which contain only gymnosperm AG-like or AG- and DEF/GLO-like proteins are only able to regulate a part of the target gene spectrum that confers reproductive organ identity, and that complexes also including AGL6-like proteins are necessary for controlling the complete set of target genes. It remains an important task for future research to study protein–DNA interactions of different complexes with different CArG-box combinations, including promoter regions of potential target genes from G. gnemon. Also, it has not been assessed yet whether the complexes proposed here possess transcription activation potential. The available evidence suggests that in A. thaliana, SEP1-like proteins but not AG- or DEF/GLO-like proteins can activate transcription (Honma and Goto, 2001). Hence it will be interesting to see whether AG-like, DEF/GLO-like or AGL6-like proteins from G. gnemon have transcription activation potential.

Homodimerization versus heterodimerization of MADS-domain proteins

Almost all floral homeotic proteins tested so far form homodimers that bind to CArG-boxes (Kaufmann et al., 2005b). This also holds true for most of the MADS-domain proteins from G. gnemon tested here, supporting the notion that homodimerization upon DNA binding is a common and ancient feature of MADS-domain proteins (Kaufmann et al., 2005b). In contrast, except for GGM2, we were not able to demonstrate homodimerization with the yeast two-hybrid system or in vitro pull-down assays. Difficulties in detecting homodimerization of MADS-domain proteins using the yeast two-hybrid system have already frequently been reported (see de Folter et al., 2005; for example). The problem might be inherent to GAL4-based two-hybrid systems, as the GAL4 DNA-binding domain (GAL4-BD) alone already dimerizes and thus a potential dimer between GAL4-BD chimeras might be favoured over a dimer consisting of a GAL4-BD and a GAL4-activation domain chimera (Smirnova et al., 1999). However, the failure to detect homodimers with the pull-down assay but not with EMSAs cannot be easily attributed to problems inherent to these techniques. Rather, these findings indicate that homodimerization in free solution is relatively weak but is stabilized by DNA-binding of the proteins. It should also be noted, however, that in living plant cells plant-specific cofactors might stabilize homodimerization of certain MADS-domain proteins also in solution (Immink et al., 2002).

Variations in B protein interactions

The two DEF/GLO-like proteins from G. gnemon tested here showed extremely divergent interaction patterns. Whereas GGM2 interacted with all proteins examined, we were neither able to detect homodimerization nor heterodimerization for GGM15, despite its close relationship to GGM2.

One of the putative orthologues of GGM2 in Picea abies (Norway spruce) is DAL11, whereas the putative orthologue of GGM15 is DAL12 (Winter et al., 2002a). Similarly to our results, DAL11 interacted with several other proteins in yeast two-hybrid assays, whereas DAL12 did not (Sundström and Engström, 2002). It could well be, therefore, that lack of interaction with the ‘usual suspects’ might be a general feature of the proteins belonging to the DAL12-subclade. The fact that GGM15 and DAL12 are from distantly related gymnosperms (gnetophytes and conifers, respectively) indicates that these proteins have been strongly conserved and hence should have an important function. When ectopically expressed in A. thaliana, both, DAL11 and DAL12 conferred similar phenotypes (Sundström and Engström, 2002), corroborating the view that DAL12 and, by inference, also GGM15 are functional proteins. How their function is exerted remains elusive, however. Possibly plant-specific cofactors are required for GGM15 and DAL12 to interact with other proteins. However, SDS-PAGE analysis of in vitro translated proteins consistently yielded a weaker signal for GGM15 compared with the other proteins (data not shown). Therefore, we cannot completely rule out that the production of GGM15 in heterologous systems is impaired, and that this impeded the detection of interaction partners.

In summary, crucial features of the molecular network controlling floral organ identity in angiosperms are evident in gymnosperms as well. Though gymnosperms lack SEP1-like proteins, the ancestral functions of proteins from this subfamily and AGL6-like proteins might have been largely redundant. Thus, the gymnosperm network may be close to an ancestral state of protein–protein interactions among seed plant MADS-domain proteins. Therefore, it might have been the selective loss of some interactions on the one hand and the establishment of hubs on the other hand that constituted the present day network in flowering plants and that paved the way for the evolution of the angiosperm flower.

Experimental procedures

Vector construction and protein in vitro translation

Six MADS-box genes from G. gnemon were selected for this study: GGM2 (GI:5019428), GGM3 (GI:5019430), GGM9 (GI:5019455), GGM11 (GI:6468291), GGM13 (GI:5019463) and GGM15 (GI:10880310). For yeast two-hybrid assays, the full length coding sequences were cloned into the pGBKT7 and the pGADT7 vectors. Except GGM3, all coding sequences were cloned using EcoRI/BamHI recognition sites. GGM3 was cloned using SmaI/SacI (pGADT7) and NcoI/BamHI (pGBKT7) recognition sites. For EMSAs, full length coding sequences as well as the C-terminal deletion derivatives GGM2ΔC (conceptual translation yields a protein of 163 aa in length), GGM3ΔC (208 aa), GGM9ΔC (204 aa), GGM11ΔC (200 aa), and GGM13ΔC (187 aa) were cloned into pSPUTK using NcoI and BamHI recognition sites. The pSPUTK GGM2 construct has been published previously (Winter et al., 2002b). For in vitro pull-down assays, bait proteins were expressed with an N-terminal hexa-histidine tag (His6-tag) by cloning the respective coding sequences into pIVEX1.4WG using NcoI and SmaI recognition sites. pSPUTK constructs containing full-length coding sequences were used to translate prey proteins.

For in vitro translation the TNT SP6 and TNT T7 quick coupled transcription/translation system (Promega) was used according to the manufacturer’s instructions. For EMSA and DNase I footprint assays, proteins where co-translated when tested for an interaction. For pull-down assays, bait and prey proteins were translated separately. The correct size of the proteins produced was checked for every cDNA template at least once by SDS-PAGE (Figure S1).

Yeast two-hybrid assay

Yeast two-hybrid analyses were performed with the Matchmaker Gal4 Two-Hybrid System 3 (Clontech ( The pGBKT7 constructs were transformed into yeast strain AH109, and pGADT7 constructs into strain Y187. After transformation, yeasts were mated essentially as described (Causier and Davies, 2002). Subsequently, similar amounts of yeast cells were dissolved in 90 μl TE buffer and serially diluted to up to 1:1000 in TE buffer. Two μl of each dilution were spotted on –Leu/–Trp/–His plates (supplemented with 3 mm 3-AT) to test for an interaction and on –Leu/–Trp plates for control of yeast growth. Interactions were usually scored after 8–9 days of incubation at 22°C. A β–galactosidase assay was performed essentially as described (Duttweiler, 1996) on colonies grown on –Leu/–Trp/–His plates (data not shown). Proteins were assessed to interact if yeast growth was observed in the three dilutions and if the lacZ assay was positive. All interactions were tested at least in duplicate.

DNA binding site probes

DNA probes used have been described previously (Melzer and Theissen, 2009; Melzer et al., 2009). Sequences of the probes are listed in Table S1. For EMSAs, probes were labelled with [α-32P] dATP using the Klenow fragment (exonuclease minus) and purified with the QIAquick nucleotide removal kit (QIAGEN,

The DNA fragments used for DNase I footprint assays were prepared by PCR amplification using primers binding to the T3 and T7 promoter region of the plasmids encoding the probes. The PCR product was digested using XhoI and radioactively labelled with T4 polynucleotide kinase and [γ-32P] ATP. One of the labelled ends was removed using XbaI digestion and gel purification.


EMSA experiments were done as described (Melzer et al., 2009), except that 2 μl in vitro translated protein and approximately 1–2 ng labelled DNA probes were added to the binding buffer (For the experiments shown in Figure 5b, 6–7 ng of labelled DNA were used.). Distilled water was used to scale the reaction volume to 12 μl. Incubation time was approximately 90 min.

In vitro pull-down assay

25 μl of in vitro translated bait protein solution was incubated with 5 μl of Ni–NTA magnetic agarose beads (QIAGEN) in 100 μl binding buffer (50 mm NaH2PO4, 1 m NaCl, 40 mm imidazole) at 15°C for 1 h with gentle shaking. In vitro translation lysate programmed with DNA templates that did not contain a cDNA insert was used as control to determine non-specific binding of prey proteins to the Ni–NTA agarose beads. Beads were washed two times with interaction buffer (binding buffer supplemented with 0.01% Tween20) and re-suspended in 100 μl interaction buffer. Subsequently, 25 μl prey protein solution was added and incubated at 15°C for 1 h with gentle shaking. After washing two times with 100 μl interaction buffer, bait and prey proteins were eluted with 20 μl elution buffer (50 mm NaH2PO4, 1 m NaCl, 250 mm imidazole, and 0.01% Tween20) and separated on 12% SDS–polyacrylamide gels. Signals were detected by autoradiography.

DNase I footprint assays

DNase I footprint assays and subsequent quantitative analysis was carried out as generally described (Melzer et al., 2009), including separating of protein–DNA complexes from free DNA after DNase I cleavage.


We are grateful to Kerstin Kaufmann for very valuable support during initial stages of this project. We thank two anonymous reviewers for their helpful comments on an earlier version of this manuscript. YQW received a scholarship from the International Leibniz Research School for Microbial and Biomolecular Interactions (ILRS Jena). RM received a fellowship from the Studienstiftung des deutschen Volkes. Part of this work was supported by grants TH 417/5-1 and -2 from the Deutsche Forschungsgemeinschaft (DFG) to GT.