Gene networks controlling Arabidopsis thaliana flower development




The formation of flowers is one of the main models for studying the regulatory mechanisms that underlie plant development and evolution. Over the past three decades, extensive genetic and molecular analyses have led to the identification of a large number of key floral regulators and to detailed insights into how they control flower morphogenesis. In recent years, genome-wide approaches have been applied to obtaining a global view of the gene regulatory networks underlying flower formation. Furthermore, mathematical models have been developed that can simulate certain aspects of this process and drive further experimentation. Here, we review some of the main findings made in the field of Arabidopsis thaliana flower development, with an emphasis on recent advances. In particular, we discuss the activities of the floral organ identity factors, which are pivotal for the specification of the different types of floral organs, and explore the experimental avenues that may elucidate the molecular mechanisms and gene expression programs through which these master regulators of flower development act.

I. Introduction

The emergence of flowers as reproductive units probably contributed substantially to the evolutionary success of angiosperms, or flowering plants. Darwin described the obscure origin of flowers as an ‘abominable mystery’ because of their sudden appearance in the fossil record (Crepet, 2000). Flowers are thought to have originated from either the male or the female cones of gymnosperms and several theories have been posited as to how this might have occurred (reviewed in Scutt et al., 2006). However, there is little paleobotanical evidence to support a steady and gradual decent with modification of gymnosperm cones to angiosperm flowers (Bateman et al., 2006), and thus flower evolution remains largely enigmatic. Furthermore, the lack of similarity between the reproductive structures of flowering plants and gymnosperms precludes direct comparisons between these two phyla (Fig. 1; Frohlich & Chase, 2007). While in gymnosperms, the male and female reproductive axes are separated and the ovules are exposed (Fig. 1b), angiosperm flowers are typically bisexual and are often composed of four different organ types (Fig. 1a), which are arranged in consecutive circles or whorls (Fig. 3b). In the model plant Arabidopsis thaliana, the outermost whorl (the first whorl) contains four sepals, the second whorl four petals, the third whorl six stamens, which produce pollen, and the fourth whorl two fused carpels, which encompass the ovules (Fig. 1a; Smyth et al., 1990).

Figure 1.

Absence of homology between the organs of angiosperm flowers and gymnosperm cones. (a) An Arabidopsis thaliana flower, as a representative flower of the angiosperm lineage. In angiosperms, the ovules are always enclosed within a gynoecium. In A. thaliana, both male (stamens) and female (gynoecium) reproductive organs are present in the same flower, but in some plant species, these reproductive organs may form on separate flowers. The presence of sepals and petals also varies (Theissen & Melzer, 2007), and in some flowers both can be replaced by tepals, which are undifferentiated organs of the perianth. (b) Male and female pine (Pinaceae) cones represen-tative of the gymnosperm lineage. In male cones, microsporangia, which contain the pollen grains, develop at the base of modified leaf-like organs termed microsporophylls. In female cones, uncovered ovules develop on the surface of megasporophylls instead of being enclosed in a gynoecium. Sporophylls are modified leaf-like organs, thought to be the most closely related gymnosperm structures to carpels (Melzer et al., 2009).

Obtaining insights into the molecular mechanisms underlying gymnosperm and angiosperm reproductive development is at the core of solving Darwin's abominable mystery. Over the past three decades, work in A. thaliana, as well as in several other angiosperm species (including snapdragon (Antirrhinum majus), petunia (Petunia hybrida), rice (Oryza sativa) and others), has resulted in detailed insights into the genetic control of flower development. More recently, researchers have begun to delineate the gene regulatory networks (GRNs) that govern the formation of flowers on a global scale and to generate predictive mathematical models that can be used to simulate regulatory processes during flower development and to drive further experimentation. This work was made possible largely through the introduction of genomic technologies and approaches of systems biology into plant research (Wellmer & Riechmann, 2005; Sablowski, 2010). In this review, we outline some of the main concepts and findings in the field of flower development and then discuss recent progress in elucidating the architecture and composition of the flowering gene networks in A. thaliana.

II. Initiation of flower development

A highly complex GRN strictly regulates the transition from the vegetative to the reproductive phase of plant development, as mistimed flowering can dramatically decrease the number of offspring produced by a plant. Environmental cues such as temperature, photoperiod and nutrient availability as well as endogenous signals (Paul et al., 2008; Wahl et al., 2013) influence the timing of the floral transition (Srikanth & Schmid, 2011; Andres & Coupland, 2012), which leads to the conversion of the shoot apical meristem of a plant into an inflorescence meristem (IM) and subsequently to the formation of flowers. The genetic circuits that integrate these different signals ultimately converge to activate the expression of a small group of so-called floral meristem (FM) identity genes (Liu et al., 2009a; Wellmer & Riechmann, 2010; Siriwardana & Lamb, 2012) in emerging FMs (or floral primordia), which arise from regions of the IM where the phytohormone auxin accumulates as a result of polar transport mechanisms (Fig. 2a; Reinhardt et al., 2000, 2003). Of the FM identity genes, LEAFY (LFY) and APETALA1 (AP1), which both encode transcription factors, are arguably the most important and have been functionally characterized in detail. These regulatory genes were identified from mutants with severe defects in flower development. A loss of LFY function results in the partial conversion of flowers into inflorescence shoots (Weigel et al., 1992) while ap1 flowers (in addition to exhibiting floral organ identity defects; see Section III) produce secondary flowers in the axils of first-whorl organs (Irish & Sussex, 1990). Combining lfy with ap1 mutant alleles leads to a dramatically enhanced phenotype compared with each of the single mutants (Weigel et al., 1992), suggesting that these regulators have some redundant functions in the control of floral development.

Figure 2.

Inflorescence meristem (IM) and floral meristem development in Arabidopsis thaliana. (a) Once the IM (shown in pink) is established, floral meristems are produced on its flanks. Orange floral meristems represent floral stages at which LEAFY (LFY) and APETALA1 (AP1) are both being expressed, whereas the green floral meristem indicates the approximate stage at which the B and C class genes are activated. Numbers on the floral meristems indicate approximate floral stages (Smyth et al., 1990). (b, c) Auxin signaling promotes the expression of LFY. (b) In the absence of auxin, MONOPTEROS (MP) is bound by BODENLOS (BDL), which prevents transcriptional activation of LFY. (c) In the presence of auxin, BDL is polyubiquitylated through interactions with Skp1 (S-phase kinase associated protein)-Cullin-F-box (SCF) complexes containing an auxin receptor such as TRANSPORT INHIBITOR RESPONSE 1 (TIR1). This results in the proteolytic degradation of BDL, allowing transcriptional activation of LFY by MP (Yamaguchi et al., 2013). IAA, indole-3-acetic acid; AuxRE, auxin response element.

Specific combinations of mutants in the group of FM identity genes can lead to a dramatic delay or an abolition of flowering, again highlighting the inbuilt genetic redundancy during the floral transition. For example, CAULIFLOWER (CAL) and FRUITFUL (FUL), two genes closely related to AP1, are functionally redundant with AP1 in their roles of specifying FMs (Ferrandiz et al., 2000). ap1 cal double-mutant plants undergo a massive over-proliferation of IM-like tissue and flowering is dramatically delayed (Bowman et al., 1993), while introduction of a ful mutation into the ap1 cal background abolishes flowering completely (Ferrandiz et al., 2000). Notably, overexpression of LFY in the ap1 cal ful triple mutant can partially restore flowering, suggesting that a key role for the other FM identity genes is to boost LFY expression levels (Ferrandiz et al., 2000).

The central role of LFY in the establishment of FMs is supported by the recent observation that auxin promotes LFY expression via the auxin response factor MONOPTEROS (MP; Yamaguchi et al., 2013). In the absence of auxin, MP normally interacts with the AUX/IAA protein BODENLOS (BDL), which inhibits its activity (Fig. 2b; Hamann et al., 2002). BDL-mediated repression of MP is lost once BDL is degraded in response to auxin-dependent polyubiquitylation by an Skp1 (S-phase kinase associated protein)-Cullin-F-box (SCF) ubiquitin ligase complex (Cardozo & Pagano, 2004; Mockaitis & Estelle, 2008), which contains an auxin receptor such as TRANSPORT INHIBITOR RESPONSE 1 (TIR1) (Dharmasiri et al., 2005; Kepinski & Leyser, 2005; Fig. 2c). Using chromatin immunoprecipitation (ChIP), both BDL and MP have been shown to occupy a conserved regulatory element in the LFY promoter (Yamaguchi et al., 2013). Furthermore, when inflorescences were treated with a chemical analog of auxin, BDL occupancy decreased dramatically while LFY expression was increased, and upon overexpression of MP or BDL, LFY transcripts were more and less abundant, respectively (Yamaguchi et al., 2013). These data, together with results from a related study (Li et al., 2013), provided molecular insights into how the auxin-dependent outgrowth of primordia from the IM is linked to the specification of FMs and thus the onset of flower development.

Identification of the direct targets of LFY and AP1 has been a central research focus of late, resulting in a much more complete picture of the roles of these regulators during floral initiation. One of the most important functions of LFY is to promote AP1 transcription by direct regulation (Wagner et al., 1999; Winter et al., 2011). The AP1 transcription factor then binds to the regulatory region of LFY to form a positive feedback loop (Wellmer et al., 2006; Kaufmann et al., 2010), leading to a strong and rapid up-regulation of both FM identity genes. Although AP1 expression is controlled by LFY, it is not entirely dependent on it as AP1 transcripts are detectable in lfy mutant inflorescences (Ratcliffe et al., 1999). In fact, several other factors (e.g. LATE MERISTEM IDENTITY2 (LMI2), AGAMOUS-LIKE24 (AGL24) and SHORT VEGETATIVE PHASE (SVP)) are also involved in up-regulating AP1 expression (Pastore et al., 2011; Grandi et al., 2012).

To specify an FM, AP1 and LFY promote the expression of other FM identity genes while concomitantly repressing the expression of negative regulators of flowering. For example, AP1 has been shown to suppress the expression of TERMINAL FLOWER1 (TFL1), TARGET OF EAT1 (TOE1) and TOE3 (Kaufmann et al., 2010). Furthermore, AP1 down-regulates the expression of TEMPRANILLO1 (TEM1) and TEM2 (Kaufmann et al., 2010), which are known to repress the expression of the potent flowering time activator FLOWERING LOCUS T (FT; Castillejo & Pelaz, 2008). LFY also directly suppresses TFL1, TOE3 and TEM1 expression (Parcy et al., 2002; Winter et al., 2011), showing that AP1 and LFY can perform redundant tasks during the initiation of flowering.

AP1 and LFY also control the expression of genes involved in hormone pathways. For example, LFY has been shown to inhibit auxin biosynthesis and to promote auxin signaling in emerging FMs (Li et al., 2013; Yamaguchi et al., 2013), while AP1 regulates, during floral initiation, the expression of several genes involved in gibberellin metabolism and signal transduction (Kaufmann et al., 2010). Together with the large number of transcription factor-coding genes that were identified to act downstream of AP1 and LFY (Kaufmann et al., 2010; Winter et al., 2011), it appears that these FM identity regulators constitute hubs in the flowering GRN and control FM specification by coordinating a multitude of cellular processes and developmental pathways.

III. Specification of floral organs: the ABCE model

Once the FM has been specified, AP1 and LFY activate floral organ identity genes (Lohmann & Weigel, 2002), which specify the four different types of floral organs. The floral organ identity genes were discovered through the study of mutants in A. thaliana and A. majus, which are characterized by homeotic transformations, that is, the replacement of one organ type by another (Fig. 3). Based on the phenotypes of these mutants and their genetic interactions, it was proposed that the affected genes specify either sepals and petals (termed A function), petals and stamens (B function), or stamens and carpels (C function) and thus act in a combinatorial manner to control the development of the four types of floral organs (Bowman et al., 1991; Coen & Meyerowitz, 1991). This so-called ABC model (Fig. 3a) has provided, over the past two decades, the main genetic framework for research on flower development and evolution. It is now known that the underlying cause of the organ transformations in the floral homeotic mutants is the mutation of genes that code for transcription factors (Krizek & Fletcher, 2005). With the exception of the A function gene APETALA2 (AP2), which is a founding member of the AP2/ETHYLENE RESPONSE FACTOR (ERF) transcription factor family (Okamuro et al., 1997), all of the A. thaliana ‘ABC’ genes code for MIKCc-type MADS-domain transcription factors (Yanofsky et al., 1990; Jack et al., 1992; Mandel et al., 1992; Goto & Meyerowitz, 1994; Jofuku et al., 1994; Parenicova et al., 2003). These genes are typically expressed in the organ primordia that they help to specify (Yanofsky et al., 1990; Jack et al., 1992; Mandel et al., 1992; Goto & Meyerowitz, 1994; Wollmann et al., 2010).

Figure 3.

Schematic representation of the ABCE model of flower development in Arabidopsis thaliana and of the homeotic transformations in the corresponding mutants. (a) A graphical representation of the ABCE model. (b–f) Schematic representations of the floral organ types present in a wild-type A. thaliana flower and in flowers of the homeotic mutants. (b) A wild-type flower contains sepals, petals, stamens and carpels, from the outer to the inner whorl. (c) Strong apetala1 (ap1) mutants form bracts in the first whorl instead of sepals and do not contain second-whorl organs. (d) Strong ap2 mutants have carpelloid organs in the first whorl and do not contain organs in the second whorl. (e) Strong ap3 and pistillata (pi) mutants have sepals in the second whorl instead of petals, and stamens are replaced by carpels (or filaments; not shown) in the third whorl. (f) Strong agamous (ag) mutants do not form reproductive organs. Their flowers show reiterations of consecutive whorls containing sepals, petals, and then again petals. se, sepal; pe, petal; st, stamen; ca, carpel; br; bract.

von Goethe (1790) proposed that floral organs are modified leaves and, indeed, leaf-like organs arise instead of floral organs when all three classes of floral organ identity genes are mutated (Bowman et al., 1991). However, ectopic expression of the floral organ identity factors does not result in the transformation of leaves into floral organs (Honma & Goto, 2001), suggesting that flower-specific co-factors must exist that are required for their activity. Through reverse genetic analyses, it was found that four closely related and partially redundant SEPALLATA genes (SEP1 to SEP4), which also encode MADS-domain transcription factors and exhibit flower-specific expression, function together with the floral organ identity genes. In fact, removing the activities of all SEP genes simultaneously leads to a conversion of floral organs into leaf-like structures (Pelaz et al., 2000; Ditta et al., 2004). Also, when SEP genes are ectopically expressed in conjunction with certain combinations of floral organ identity genes, leaves can be converted into floral organs (Honma & Goto, 2001; Pelaz et al., 2001). These genetic interactions formed the basis of the expanded ABCE model, which states that the SEP genes (conferring E function) encode flower-specific co-factors that are required for organ specification in all floral whorls (Theissen & Saedler, 2001; Lohmann & Weigel, 2002; Krizek & Fletcher, 2005).

IV. Floral patterning: activating the floral organ identity genes in specific domains

As mentioned in Section III, the expression of the floral organ identity genes is directly promoted by LFY, which is expressed throughout the emerging FM (Parcy et al., 1998; Siriwardana & Lamb, 2012). At the same time, LFY suppresses the expression of negative regulators of the floral organ identity genes, further contributing to their activation. For example, the protein EMBRYONIC FLOWER1 (EMF1) is involved in the polycomb group-mediated suppression of B and C class genes outside of FMs (Calonje et al., 2008), and LFY down-regulates EMF1 expression (Winter et al., 2011). For the direct activation of the floral organ identity genes, LFY appears to require co-factors. Recently, SEP3 and LFY have been shown to interact in vitro and it appears that a SEP3/LFY-containing complex activates B and C class gene expression (Liu et al., 2009b). Although SEP3 behaves as an LFY co-factor, it is expressed in the inner three floral whorls during early flower development (Mandel & Yanofsky, 1998) and therefore cannot significantly contribute to the domain-specific activation of the floral organ identity genes.

As described in Section II, AP1 expression is directly promoted by LFY, and AP1 transcripts initially accumulate throughout the FM (Wagner et al., 1999). During early flower development, AP1 expression becomes restricted to the outer two whorls where it functions as an A class gene in the specification of sepals and petals (Mandel et al., 1992). The suppression of AP1 in the center of the flower is (directly or indirectly) mediated by the C class gene AGAMOUS (AG), as in strong ag mutants, AP1 transcripts are present in the center of the flower (Gustafson-Brown et al., 1994).

Regulation of B function, which activity is mediated in A. thaliana by the MADS-domain transcription factors APETALA3 (AP3) and PISTILLATA (PI) that forms as obligate heterodimers (Riechmann et al., 1996a; Yang et al., 2003), is more complex. Although AP3/PI expression is reduced in lfy mutants, it is not abolished (Weigel & Meyerowitz, 2002), indicating that other factors regulate their expression. Moreover, overexpression of LFY alone is not sufficient to activate the expression of AP3 in seedlings (Parcy et al., 1998). The F-box protein UNUSUAL FLORAL ORGANS (UFO) proved to be an excellent candidate for a co-regulator of B class gene expression, as plants without functional copies of UFO have homeotic transformations similar to those of B class mutants (Wilkinson & Haughn, 1995; Lee et al., 1997). The first indication that UFO behaves as an LFY co-factor came from experiments in which a fusion between a dominant transcriptional repressor domain and UFO was overexpressed in plants (Chae et al., 2008). This resulted in transgenic lines that strongly resembled lfy mutants, suggesting that UFO is present at the same regulatory sites as LFY (Chae et al., 2008). As mentioned in Section II, F-box proteins like UFO are part of SCF-type ubiquitin ligase complexes that polyubiquitylate their targets, marking them for degradation via the 26S proteasome. UFO has been shown to physically interact with ASK1 (Arabidopsis Skp1-like) (Samach et al., 1999; Wang et al., 2003), another component of SCF complexes; it has also been demonstrated that ASK1 genetically interacts with LFY, PI and AP3 (Zhao et al., 2001; Ni et al., 2004). Therefore, it was proposed that UFO mediates the degradation of a factor that inhibits LFY-driven gene expression (Zhao et al., 2001). Subsequently, it was found that LFY and UFO proteins interact in vitro and in vivo. Furthermore, it was shown that LFY is required to direct UFO to the AP3 promoter (Chae et al., 2008). These data led to an alternative proposal where UFO mediates the polyubiquitylation of LFY after a number of rounds of transcription, thereby promoting the turnover of LFY. This proteolysis of LFY is thought to increase transcriptional activity by allowing ‘unspent’ LFY transcription factors to bind to the promoters of target genes. While this mode of activity has not yet been demonstrated for LFY, several examples of transcription factor degradation being necessary for transcriptional activation have indeed been found in yeast, mammals and plants (Muratani & Tansey, 2003; Spoel et al., 2009).

The expression of AP3/PI is also promoted by AP1, as indicated by the absence of their transcripts in lfy ap1 double-mutant flowers (Weigel & Meyerowitz, 1993). By fusing AP1 to a strong transcriptional activation domain, it was shown that B class gene expression can be induced by AP1; however, as with LFY, this also appears to require the activity of UFO (Ng & Yanofsky, 2001; Lamb et al., 2002), as well as that of SEP3 (Sridhar et al., 2006; Kaufmann et al., 2010; Wu et al., 2012).

Numerous regulators of the C class gene AG have been identified. These include repressors, such as the polycomb group protein CURLY LEAF, that prevent AG expression outside of the flower through epigenetic mechanisms (Goodrich et al., 1997; Calonje et al., 2008), as well as activators that promote the transcription of AG in the center of the flower (Liu & Mara, 2010). For example, overexpression of WUSCHEL (WUS), encoding a homeodomain transcription factor, in flowers resulted in the ectopic formation of stamens and carpels (Lenhard et al., 2001; Lohmann et al., 2001), an effect that also required LFY activity (Lenhard et al., 2001). When the regulatory regions of AG were fused to the uidA reporter gene, β-glucuronidase activity was stronger in the presence of both WUS and LFY than in the presence of either WUS or LFY alone (Lohmann et al., 2001). Gel-shift assays further showed that WUS can bind to sequences in the second intron of AG that are in close proximity to known LFY binding sites, although WUS and LFY do not appear to interact directly with each other (Lohmann et al., 2001). WUS is required for stem cell maintenance and is expressed in a small number of cells in the center of the FM (Laux et al., 1996; Mayer et al., 1998). Given that WUS is only expressed in a subset of the cells expressing AG (Yanofsky et al., 1990; Drews et al., 1991; Mayer et al., 1998), it is not clear how AG expression is promoted by WUS throughout its full domain. However, it has been shown recently that WUS protein can move within meristems from the region where it is being synthesized into neighboring cells (Yadav et al., 2011). Thus, the activation of AG may depend, at least in part, on a non-cell-autonomous function of WUS.

Several other transcriptional regulators boost the expression of AG in the center of FMs, including the basic leucine zipper (bZIP) transcription factor PERIANTHIA (PAN). PAN binds directly to the second intron of AG (Das et al., 2009; Maier et al., 2009) and appears to activate AG expression, as flowers overexpressing a fusion between PAN and a transcriptional repressor domain are indeterminate (Das et al., 2009), a hallmark of ag mutant flowers (Bowman et al., 1991). PAN seems to act redundantly with other factors to maintain high levels of AG expression, which are required to maintain floral determinacy (Prunet et al., 2009).

The original ABC model predicted that A and C function genes act in a mutually antagonistic manner (Bowman et al., 1991). In fact, AP2 suppresses AG in the outer two floral whorls; however, AG does not appear to control AP2 expression levels, as AP2 transcripts do not accumulate in the center of ag mutant flowers (Wollmann et al., 2010). Rather, it has been shown that this function is mediated by the microRNA miR172, which prevents the accumulation of AP2 mRNA and protein in the third and fourth whorls (Chen, 2004; Wollmann et al., 2010). Perturbation of AP2 activity in the third whorl resulted in the conversion of stamens into petals (Wollmann et al., 2010), implying that AP2 is sufficient to confer petal identity to floral organs, although it is not strictly necessary for the specification of petals (Krogan et al., 2012). Recent genetic evidence implies that the transcriptional co-repressor family TOPLESS/TOPLESS RELATED (TPL/TPR) and HISTONE DEACTYLASE 19 (HDA19) are also required to suppress AG expression, and in situ hybridizations confirmed that the AG expression domain expands in these mutant backgrounds (Krogan et al., 2012). Remarkably, expression of a fusion of the AP2 DNA-binding domain to the transcriptional co-repressor TPL under the control of the TPL regulatory sequence was sufficient to restore AP2 function in a strong ap2 mutant background. Yeast two-hybrid and pull-down assays demonstrated that AP2 and TPL physically interact, while bimolecular fluorescence complementation (BiFC) and yeast two-hybrid assays showed that TPL interacts with the histone deacetylase HDA19 (Krogan et al., 2012), whose activity leads to gene repression. Furthermore, ChIP experiments confirmed that AP2, TPL and HDA19 bind to the second intron of AG, which contains important regulatory elements. All of these data together indicate that an AP2-TPL-HDA19-containing complex is required to prevent AG expression in the outer two whorls. Using ChIP, the AP2-TPL-HDA19 complex members were also shown to bind to the regulatory regions of AP3 and SEP3. Furthermore, expression of AP3, PI and SEP3 was detected in the outer whorls of plants lacking functional copies of AP2, TPL or HDA19 (Krogan et al., 2012). Together, these data show that AP2-containing complexes act as spatial regulators of gene expression during floral patterning (Fig. 4).

Figure 4.

Floral organ boundary maintenance by the ABCE proteins in A. thaliana flowers. First-whorl and second-whorl floral organ fate is promoted by the A-function proteins APETALA1 (AP1) and AP2. AP1 also interacts with SEUSS (SEU) and LEUNIG (LUG), while AP2 forms a complex with TOPLESS (TPL) and HISTONE DEACTYLASE 19 (HDA19). AP1 and AP2 repress C function in the first two whorls. These same protein complexes have been shown to suppress the expression of the B class homeotic regulators AP3 and PISTILLATA (PI), as well as the E class gene SEPALLATA 3 ( SEP3) in sepals. AP3 and PI have been shown to transcriptionally suppress the expression of carpel-specification genes, such as CRABS CLAW (CRC), SHATTERPROOF1 (SHP1) and SHP2 in the third whorl. AGAMOUS (AG) suppresses, directly or indirect, AP1 expression in the third and fourth whorls while miR172 prevents AP2 mRNA accumulation in the inner two whorls. Whorls are represented by colored rectangles, from light orange (first whorl) to red (fourth whorl). Blue shading represents organ boundaries. Boxes with smooth outlines represent promoter activities, while boxes with jagged outlines represent RNA accumulation. Circles represent proteins and blunt-end lines indicate repression.

V. Mechanistic insights into floral organ identity factor function

Through the use of genomic technologies, such as microarray analysis and ChIP coupled to next-generation DNA sequencing (ChIP-Seq), it has been shown that the floral organ identity factors influence the expression of thousands of downstream genes, through either direct or indirect regulation (Ito et al., 2004; Wellmer et al., 2004, 2006; Gomez-Mena et al., 2005; Mara & Irish, 2008; Kaufmann et al., 2009, 2010; Jiao & Meyerowitz, 2010; Wuest et al., 2012). To directly control transcription, the MADS-domain transcription factors bind to DNA motifs with a similarity to the sequence 5′-CC(A/T)6GG-3′, known as a CArG-box (Riechmann et al., 1996a). Given that CArG-box motifs are extremely common in the A. thaliana genome (over 15 000 unique coding sequences are associated with a CArG-box in the A. thaliana genome) and that imperfect CArG-boxes (i.e. with mismatches to the consensus sequence) have also been shown to be functional (de Folter & Angenent, 2006), it is unlikely that the presence of this sequence in the promoter of a given gene is the solitary factor that dictates a transcriptional response (de Folter & Angenent, 2006). Indeed, genome-wide studies revealed that only a subset of genes that are associated with MADS-domain protein binding exhibit altered expression when their activities are perturbed (Table 1; Kaufmann et al., 2010; Wuest et al., 2012). Similar observations have been made for several non-MADS-domain transcription factors from plants (Busch et al., 2010; Yant et al., 2010; Winter et al., 2011) as well as from other eukaryotic model organisms (Farnham, 2009). Whether or not these ‘transcriptionally silent’ binding events are functional is currently unclear. However, in mammals, transcription factors are thought to be able to scan the entire genome for binding availability in a matter of minutes (Hager et al., 2009). Given that ChIP merely identifies a snapshot of transcription factor binding without any kinetic information, the association of a transcription factor with sites it displays affinity for is perhaps unsurprising. Also, transcriptional control by the floral organ identity factors may require the presence of additional regulators at target gene promoters to elicit a transcriptional response, and it is from this perspective that the so-called floral quartet model provides a functional framework.

Table 1. Percentage of high-confidence target genes among genome-wide transcription factor-DNA localization studies
Protein of interestAntibodyGenotypeTissue% DEG and boundNo. genes boundReferences
  1. cal, cauliflower; DEG, differentially expressed gene; GR, glucocorticoid; WUS, WUSCHEL.

  2. The experimental design (i.e. protein of interest, antibody used for immunoprecipitation, tissue and genotype) is indicated. The ‘percentage of DEG and bound’ genes was calculated by determining the overlap between genes that responded to a change in the transcription factor's activity and the genes associated with transcription factor binding, and dividing that number by the number of genes associated with transcription factor binding. The ‘No. genes bound’ refers to the number of genes that were within the vicinity of a binding peak, as defined in each individual study. In Kaufmann et al. (2010), c. 44% of genes that were bound by APETALA1 (AP1) showed some transcriptional response; however, only 10.8% of genes were defined as being ‘robustly’ differentially expressed. In Wuest et al. (2012), given that AP3/PISTILLATA (PI) are known to act as obligate heterodimers, the chromatin immunoprecipitation coupled to next-generation DNA sequencing (ChIP-Seq) and transcriptome studies performed on AP3 and PI were combined to define a set of high-confidence targets (22.3% of all bound genes). In Winter et al. (2011), c. 41% of genes bound by LEAFY (LFY) were rapidly differentially expressed upon LFY-GR activation. Some putative direct targets in Busch et al. (2010) and Winter et al. (2011) were omitted from the calculations, as they were not represented on the microarray platforms used.

AP1αAP135Spro:AP1-GR ap1-1 cal-1Inflorescence meristem10.82298Kaufmann et al. (2010)
WUSαWUSWild typeSeedlings6.8159Busch et al. (2010)
AP2αAP2Wild typeLeaf7.72135Yant et al. (2010)
LFYαLFY35Spro:LFY-GRSeedlings22980Winter et al. (2011)
LFYαLFYWild typeWhole inflorescences24.6662Winter et al. (2011)
AP3/PIαGFP35Spro:AP1-GR AP3pro:AP3-GFP ap1-1 cal-1 ap3-3/pi-1Stage c. 5 flowers22.31558Wuest et al. (2012)

1. The floral quartet model

Aside from the highly conserved MADS DNA-binding domain, MIKCc-type MADS-domain proteins contain at least three other functional domains, termed the I, K and C domains (Parenicova et al., 2003). The K domain has been shown to be important for protein–protein interactions and is linked to the C domain, which has been suggested to act as a trans-activation domain, by the weakly conserved I domain (Riechmann & Meyerowitz, 1997; Gramzow & Theissen, 2010). Riechmann et al. (1996a,b) showed that AP1 and AG homodimers as well as AP3/PI heterodimers can form, in addition to higher order complexes composed of AP1/AP3/PI or AG/AP3/PI (Riechmann et al., 1996a). Thus, it was proposed that the floral organ identity factors might target diverse sets of genes by exhibiting partner specificity. As mentioned above, the SEP genes were identified as being required for floral organ identity gene function (Pelaz et al., 2000; Ditta et al., 2004), but the mechanism underlying their activities was not initially clear. Through yeast two-hybrid and co-immunoprecipitation experiments, the SEP proteins were shown to interact with the MADS-domain floral organ identity factors (Honma & Goto, 2001). These findings led to the proposal that tetrameric complexes, consisting of SEP proteins with specific combinations of the floral organ identity factors, could selectively influence the transcription of bound target genes (Fig. 5; Theissen & Saedler, 2001). Several lines of evidence have been produced that are consistent with this scenario, including results obtained from in vitro DNA-binding assays, fluorescence resonance energy transfer (FRET) and fluorescence lifetime imaging (FLIM) experiments, affinity purification and mass spectrometry, BiFC of in vivo samples and genome-wide DNA–protein interaction studies (Riechmann et al., 1996a; Immink et al., 2009; Kaufmann et al., 2009, 2010; Smaczniak et al., 2012; Wuest et al., 2012).

Figure 5.

Models for higher order MADS-domain protein complexes and their binding to DNA. Please note that the different scenarios shown are not mutually exclusive. (a) Dimers of MADS-domain proteins (represented by two ovals) may require the DNA binding of an additional transcription factor(s) (shown as a hexagon) to activate gene expression. In this model, a physical interaction between these different transcriptional regulators is not a prerequisite for activation of the target genes. (b) Dimers of MADS-domain proteins can directly interact with other transcription factors that bind to neighboring cis regulatory elements. (c) The transcriptional activity of MADS-domain transcription factor dimers may require interaction with an additional regulator (green circle) that does not bind DNA. (d) Floral quartets may mediate DNA looping to influence transcription in vivo (left). In some cases, DNA looping is thought to require interaction with additional regulators that bind to distinct regulatory elements in the promoter of target genes (right). TFBS, transcription factor binding site.

The floral quartet model predicts that quaternary complexes composed of MADS-domain proteins recognize two different CArG-boxes at the same time to regulate the expression of downstream targets (Melzer et al., 2010). However, it has been demonstrated that multimeric complexes can bind single CArG-boxes or two CArG-boxes at the same time (Fig. 5a–d; Smaczniak et al., 2012). Moreover, it has been shown that SEP3 homotetramers can mediate DNA looping, which suggests that regulation of gene expression may require distal enhancers (Fig. 5d; Melzer et al., 2009).

Although the floral quartet model posits an explanation for how the floral organ identity factors bind DNA, it does not explain how target genes are selectively regulated. In this context, it is important to note that when the DNA-binding domains between AP1/AP3/PI/AG were swapped, their DNA-binding properties were not altered (Riechmann et al., 1996a,b). Furthermore, when the amino-terminal halves of the MADS-domains of AP1, AP3, PI and AG were replaced with the corresponding regions of mammalian MADS-domain proteins, which have DNA-binding specificities distinct from those of the floral organ identity factors, their overexpression could (partially) rescue the defects of the floral homeotic mutants (Riechmann & Meyerowitz, 1997). These data strongly suggest that the DNA-binding specificity conferred by the MADS-domain is only one input for transcriptional regulation by the floral organ identity factors.

Data from genome-wide localization studies of floral MADS-domain transcription factors provided clues as to how the floral organ identity factors might specifically regulate downstream targets (Kaufmann et al., 2009; Wuest et al., 2012). Aside from CArG-box motifs, putative binding sites for several other transcription factor families were identified in close proximity to the genomic regions bound by the MADS-domain transcription factors. These include binding motifs for basic helix-loop-helix; bZIP; TEOSINTE BRANCHED1, CYCLOIDEA, and PCF (TCP); and zinc-finger transcription factors (Kaufmann et al., 2009; Wuest et al., 2012). Therefore, the combined activities of the floral organ identity factors and additional transcriptional regulators are probably necessary to determine when and where the expression of downstream genes is promoted or suppressed (Fig. 5a–d).

2. Co-factors and co-factor candidates

While interactions between MADS-domain proteins are well established in A. thaliana and other angiosperms (Immink et al., 2010), candidates for proteins that can associate with them, and may be required for their function, have been identified only sporadically until recently. For example, yeast-two hybrid screens and genetic analyses indicated that LEUNIG (LUG) and SEUSS (SEU), which are components of a transcriptional co-repressor complex, can form higher order regulatory complexes with AP1 or SEP3 to suppress AG expression in the outer two floral whorls (Fig. 4; Conner & Liu, 2000; Franks et al., 2002; Sridhar et al., 2006). A recent proteomic analysis of floral MADS-domain protein complexes confirmed the interactions between AP1 and SEU while also revealing several other candidate co-factors (Smaczniak et al., 2012). For instance, using affinity purification and mass spectrometry, BELLRINGER (BLR), a homeodomain transcription factor, was shown to interact with AP1. Yeast three-hybrid assays further indicated that AP1/SEP3 or AP1/SEP4 heterodimers interact with BLR (Smaczniak et al., 2012). Furthermore, using ChIP assays, it was shown that BLR binds to regulatory regions of LFY, AP2 and TOE1 (Smaczniak et al., 2012) that are also targeted by AP1 during floral transition (Kaufmann et al., 2010). Therefore, BLR/AP1 complexes probably cooperate to mediate the initiation of flowering.

A number of proteins involved in chromatin modification also appear to interact with the floral organ identity factors. These include CHROMATIN REMODELING 4 (CHR4), CHR11 and CHR17, which associate with AP1, AP3, PI, AG and SEP3 (Smaczniak et al., 2012). In vitro studies indicated that interactions between PI and CHR4, CHR11 and CHR17 are stabilized in the presence of double-stranded DNA, while chr11 chr17 double mutants displayed several floral defects consistent with a role in organ specification (Smaczniak et al., 2012). The ATPase chromatin remodelers BRAHMA (BRM) and SPLAYED (SYD) associate with AP1 and SEP3 (Smaczniak et al., 2012). BRM and SYD are partially functionally redundant and appear to play a role in activating B and C class gene expression (Wagner & Meyerowitz, 2002; Hurtado et al., 2006). In fact, LFY and SEP3 have been shown to interact with BRM and SYD to promote the expression of the floral organ identity genes (Wu et al., 2012). The B and C function regulators also associate with SYD and BRM, although the functional significance of this interaction is currently unclear (Smaczniak et al., 2012). The histone H3 lysine 27 (H3K27) demethylase RELATIVE OF EARLY FLOWERING 6 (REF6) appears to form a complex with AP1 (Smaczniak et al., 2012). Overexpression of REF6 produces pleiotropic defects including strong floral phenotypes such as carpelloid sepals (Lu et al., 2011). Although AP1 has not yet been shown to influence H3K27 methylation levels, AG activity reduces the degree of H3K27 tri-methylation at the promoter of KNUCKLES (KNU; Sun et al., 2009), a key regulator of FM development, suggesting that the floral organ identity factors are indeed involved in the control of epigenetic modifications.

The results outlined above imply that the floral organ identity factors interact with many other proteins to control the transcription of target genes. However, the protein interactions described thus far do not indicate how transcriptional regulation of different sets of genes is achieved. A recent study of cardiac cell fate in Drosophila implied that DNA sequence specificity of a given transcription factor is only required to a certain degree to control gene expression during development (Junion et al., 2012). Instead, it was proposed that a subset of transcription factors associate with cis-regulatory elements with extensive motif flexibility and that these transcription factors then recruit the remainder of the transcriptional complex. Although the general applicability of this model has yet to be tested, it provides an attractive scenario for the activity of the floral organ identity factors and for other regulators involved in the control of flower development.

VI. Gene expression programs controlled by the floral organ identity factors

As described above, there is now ample evidence that the activity of a small number of regulatory proteins is sufficient to superimpose floral organ identity upon an underlying developmental program that specifies leaves. However, the mechanistic basis of this organ transformation remains largely unknown. A combination of transcriptomics and genome-wide localization studies can be used to identify direct and indirect targets of transcription factors (Malone & Oliver, 2011; Furey, 2012), and these types of studies have been performed for a number of floral organ identity factors. They showed that genes that are controlled by these master regulators have diverse functions (Gomez-Mena et al., 2005; Mara & Irish, 2008; Kaufmann et al., 2010; Wuest et al., 2012), suggesting that the regulation of a multitude of developmental pathways is involved in floral organ specification. In this section, we will review recent progress in the understanding of the gene expression programs controlled by the floral organ identity factors. We will start by discussing some of the technical approaches and innovations that have made these advances possible.

1. Experimental approaches for dissecting the flowering GRN

Cell specification is driven by differential gene expression, which is largely determined by the interaction between cis-regulatory elements and transcription factors (Wellmer & Riechmann, 2005). Complex combinations of transcription factor binding sites can exist within the regulatory regions of genes, and spatio-temporal expression changes of transcription factors further complicate the dissection of the GRNs that control cellular differentiation. Advanced techniques such as RNA-Seq or ChIP-Seq are now available that allow, on a genome-wide scale, the assessment of gene expression changes and transcription factor occupancy during development or in response to a perturbation of a regulatory gene. These novel approaches have led to detailed insights into the structure and composition of GRNs that would probably have remained elusive using traditional genetic and molecular techniques.

In the field of flower development, the application of these methods has been hampered by the fact that flowers of the main model A. thaliana are small, and young floral primordia are minute and difficult to dissect (Smyth et al., 1990). Furthermore, A. thaliana flowers are initiated sequentially by the IM so that all flowers of an inflorescence are at different developmental stages (Fig. 2a; Smyth et al., 1990), making the collection of sufficient amounts of tissue for further analysis prohibitively difficult. The so-called floral induction system offers easy access to relatively homogenous populations of young flower buds, allowing the assessment of the GRNs underlying A. thaliana flower development through the use of genomic and proteomic technologies (Wellmer et al., 2006). The system is based on the overexpression of a fusion between AP1 and the hormone-binding domain of the rat glucocorticoid receptor in an ap1 cal double-mutant background (Wellmer et al., 2006). As mentioned in Section II, ap1 cal plants overproduce IM-like tissue in place of flowers. Upon treatment with a steroid hormone, the AP1-GR (glucocorticoid) fusion translocates from the cytoplasm to the nucleus where it can influence transcription (Wellmer et al., 2006; Kaufmann et al., 2010). Once the AP1-GR protein is activated, synchronous flowering is initiated throughout the inflorescence-like meristems. This system has been extensively used to investigate the molecular mechanisms underlying early flower development through a wide range of methods including genomic as well as proteomic approaches (Wellmer et al., 2006; Das et al., 2009; Sun et al., 2009; Jiao & Meyerowitz, 2010; Kaufmann et al., 2010; Wuest et al., 2012).

From genome-wide analyses of gene expression in young and more mature flowers, it has been concluded that the topology of the GRN underlying flower development changes considerably over time (Wellmer et al., 2004, 2006; Gomez-Mena et al., 2005; Wuest et al., 2012). However, the stage-specific activities of the floral organ identity factors and of other floral regulators are currently not well defined. Also, given that the floral organ identity genes are expressed in more than one type of floral organ and that their expression patterns vary over time (Yanofsky et al., 1990; Drews et al., 1991; Jack et al., 1992; Goto & Meyerowitz, 1994; Gustafson-Brown et al., 1994), the current lack of organ- or tissue-specific information on gene expression during flower development greatly hampers the understanding of the gene networks mediating organ morphogenesis. Although fluorescence-activated cell sorting (FACS) and laser capture microdissection (LCM) have been successfully used to isolate specific cell types in plants (Birnbaum et al., 2003; Wuest et al., 2010; Liu et al., 2011), these methods are extremely time-consuming and offer suboptimal yields and purity of specific cell types (Wang et al., 2012). Translating ribosome affinity purification (TRAP; Heiman et al., 2008) has also been used to produce organ- or cell-specific expression data in plants. TRAP entails epitope-tagging of ribosomal subunits so that polysomes can be immunoprecipitated and the associated mRNAs can be analyzed. Jiao & Meyerowitz (2010) used this technique in an elegant study to isolate transcripts present in the AP1, AP3 and AG expression domains at two stages of early flower development using the floral induction system. By overlapping the over-represented transcripts from each expression domain, organ- and domain-specific transcripts were inferred.

An alternative method to TRAP utilizes affinity-based isolation of tagged nuclei from which either chromatin or RNA can be purified (Deal & Henikoff, 2010). Isolation of nuclei tagged in specific cell types (INTACT) has been used to identify differences in chromatin features and expression profiles of hair and non-hair cells in the A. thaliana root (Deal & Henikoff, 2010). In contrast to TRAP, which only allows analysis of polysome-bound RNA, INTACT also facilitates the analysis of chromatin through the isolation of nuclei. The yields of the tagged cell type(s) are extremely high and the technical set-up required to perform INTACT is relatively simple and cost-effective. By using the INTACT method in combination with the floral induction system described above, it should be possible to obtain data on gene expression and transcription factor occupancy with high spatial and temporal resolution.

2. Global insights into floral organ identity factor function

Genome-wide studies have revealed that many of the genes that are regulated by the floral organ identity factors encode transcriptional regulators (Table 2; Gomez-Mena et al., 2005; Mara & Irish, 2008; Kaufmann et al., 2010; Wuest et al., 2012), which are involved in controlling a multitude of diverse processes, from hormone signaling to boundary formation. There is now mounting evidence that, in some cases, these regulatory proteins perform subsets of the functions initially attributed to the floral organ identity factors. Arguably the best example of such a mechanism is the transcription factor-coding gene NOZZLE/SPOROCYTELESS (NZZ/SPL), which is a regulator of sporogenesis (Schiefthaler et al., 1999; Yang et al., 1999). Both AG and AP3/PI have been shown to promote NZZ/SPL transcription through direct interactions with sites in the 3′ (Ito et al., 2004) and 5′ (Wuest et al., 2012) regions of the gene, respectively, and ectopic expression of NZZ/SPL can induce microsporogenesis in petals of ag-1 mutant flowers (Ito et al., 2004). Thus, it appears that the B and C function regulators control the most central aspect of reproductive floral organ formation through the activation of a single master regulatory gene.

Table 2. Percentage of transcription factor-coding genes identified in microarray experiments on floral tissues
GenotypesComparisonTissue% DEGs encoding TFsReferences
  1. ag, agamous; ap, apetala; cal, cauliflower; DEG, differentially expressed gene; GR, glucocorticoid; pi, pistillata; TF, transcription factor; DEX, dexamethasone; WT, wild type.

  2. The experimental design (i.e. genotype, samples being compared and tissue type) is indicated. Transcription factors represent c. 6% of the coding genes in the Arabidopsis thaliana genome (Riechmann et al., 2000).

WT/ap1-1/ap2-2/ap3-3/pi-1/ag-3RNA from wild-type samples was compared with RNA from mutant samplesWhole inflorescences5.5Wellmer et al. (2004)
35Spro:AG-GR ap1-1 cal-1RNA from nontreated samples was compared with RNA from DEX-treated samplesap1-1 cal-1 inflorescences26Gomez-Mena et al. (2005)
35Spro:AP1-GR ap1-1 cal-1RNA from mock-treated samples was compared with RNA from DEX-treated samplesap1-1 cal-1 inflorescences13.4Wellmer et al. (2006)
35Spro:AP3-GR 35Spro:PI ap3-3RNA from nontreated samples was compared with RNA from DEX-treated samplesWhole inflorescences10.9Mara & Irish (2008)
35Spro:AP1-GR ap1-1 cal-1/35Spro:AP1-GR ap1-1 cal-1 ap3-3/35Spro:AP1-GR ap1-1 cal-1 pi-1RNA from B-function mutant samples was compared with RNA from non-B-function mutant samples after DEX treatmentap1-1 cal-1 inflorescences13Wuest et al. (2012)

The floral organ identity factors also directly modulate the expression of genes with general roles in leaf and shoot development. One example of this is JAGGED (JAG), which encodes a zinc-finger transcription factor that regulates the proliferative phase of lateral organ growth. Plants without a functional copy of JAG have serrated leaves and floral organ margins (Dinneny et al., 2004, 2006; Ohno et al., 2004) as a consequence of impaired anisotropic growth (Schiessl et al., 2012). ChIP-Seq performed on whole inflorescences and young floral buds identified JAG regulatory sequences as being bound by SEP3 and AP3/PI, respectively (Kaufmann et al., 2009; Wuest et al., 2012). Moreover, JAG mRNA levels responded to an induction of AG activity (Gomez-Mena et al., 2005) and the removal of AP3/PI activity (Wuest et al., 2012), respectively, suggesting that B and C functions control JAG expression. Therefore, it appears that B and C function regulators have assimilated JAG, which performs a more general role during development, into the floral GRN.

Sharp gene expression boundaries exist during floral organ specification in eudicots; however, the same appears not to be not true for at least some basal angiosperms (Theissen & Melzer, 2007), where, for example, the expression of the B class genes is found outside of the petal and stamen domains (Kim et al., 2005). This spatially expanded B class gene expression is reflected in the altered morphology of flowers of basal angiosperms which show gradual transitions from outer tepals to inner tepals, inner tepals to stamens, and stamens to carpels (Theissen & Melzer, 2007). These observations are summarized by the ‘fading borders’ model, which proposes that the floral organ identity genes are expressed strongly at the center, and weakly at the edges, of their expression domains (Soltis et al., 2007). This then leads to a scenario where weak and partially overlapping expression domains of different classes of floral organ identity genes result in the production of chimeric or intermediate organs (Soltis et al., 2007). Although there is currently only limited experimental support for this model (Kim et al., 2005; Chanderbali et al., 2010), it provides an attractive hypothesis for the molecular mechanism underlying the differences in floral morphology between basal angiosperms and other flowering plants.

In eudicots, such as A. thaliana, several genetic pathways promote the formation of sharp organ boundaries and of spatially well-defined expression domains of key floral regulators. Interestingly, the floral organ identity factors seem to play a decisive role in this process. For example, ChIP-Seq and transcriptome analysis showed that AP3 and PI directly suppress the expression of genes with known functions in carpel and ovule development (Wuest et al., 2012). These data were further supported by the finding that an important regulator of carpel development, CRABS CLAW (CRC), is precociously expressed in the third whorl of B function mutants (Bowman & Smyth, 1999). Furthermore, artificial microRNAs (amiRNAs) that specifically target the transcripts of AP3 or PI were used to perturb the function of these genes at various stages of flower development, and this resulted in abrupt transformations of stamens into carpels, without any chimeric intermediates (Wuest et al., 2012). Together these data indicate that B function behaves as a molecular switch that determines whether female or male reproductive organs are specified, and this mechanism may underlie the formation of bisexual axes in angiosperm flowers (Theissen et al., 2002).

Another factor contributing to the formation of sharp organ boundaries in A. thaliana might be the formation of feedback loops upon floral organ identity gene activation (Gomez-Mena et al., 2005; Theissen & Melzer, 2007). This type of regulation would lead to amplification of small differences, resulting in a switch-like ‘on–off’ behavior of gene expression. In support of this idea, the B and C function regulators in A. thaliana have been shown to boost their own expression (Goto & Meyerowitz, 1994; Gomez-Mena et al., 2005; Wuest et al., 2012). Also, they directly regulate several other genes that are required to maintain floral organ boundaries, such as RABBIT EARS (RBE) and SUPERMAN (SUP), which themselves control floral organ identity gene expression (Sakai et al., 1995, 2000; Takeda et al., 2004; Krizek et al., 2006; Wuest et al., 2012). These mechanisms, together or separately, could have been central to the transition between the putative fading borders situation found in flowers of some basal angiosperms and the sharp borders observed in eudicot flowers (Theissen & Melzer, 2007).

VII. Network modeling

Mathematical models can accurately describe how genetic interactions translate into phenotypic traits (Hammer et al., 2006). Once these models have been generated, discrepancies between experimental data and the proposed model can be identified and investigated. Furthermore, modeling can be used to analyze causality, as parameters can be explicitly defined and explored by simulation (Prusinkiewicz & Runions, 2012). Initially, mathematical models for flower development were based on experimental data derived from a small number of genes (Espinosa-Soto et al., 2004). More recently, production of comprehensive floral GRNs has become possible as a result of the availability of functional genomics tools. However, the ability to translate these large-scale data sets into GRNs and to mathematically model their dynamic behavior is currently limited (Moreno-Risueno et al., 2010).

A discrete model of the ABCE model of floral organ identity specification was generated based on logical rules that were derived from existing experimental data (Espinosa-Soto et al., 2004). The model showed that all possible starting conditions of the network converge to one of five steady states (i.e. IM, sepals, petals, stamens or carpels). Recovery of steady states that correspond to described expression patterns was also possible in mutant backgrounds, supporting the robustness of the model (Espinosa-Soto et al., 2004). It was noted that the steady states of reproductive organs appeared to be more stable than those of sepals or petals. This observation can be correlated with the fact that the presence and phenotypes of perianth organs are more variable than those of reproductive organs in angiosperms (Theissen & Melzer, 2007). The model also correctly predicted that AG is involved in a positive feedback loop to maintain its own expression (Gomez-Mena et al., 2005).

Although gene network models are useful to predict genetic interactions, they typically do not include information on cell division patterns during morphogenesis. Time-lapse imaging of biological samples, commonly referred to as live imaging, allows spatio-temporal assessments of development to be made (Roeder et al., 2011). The expression patterns of selected regulators can be studied and, once the acquired images have been processed and analyzed, mathematical models can be generated. Live imaging was used, for example, to investigate the appearance of different cell sizes in the A. thaliana sepal, which depend on whether or not a cell undergoes endoreduplication (Roeder et al., 2010). A computational model predicted that the decision to divide and to arrest division was made largely at random. By incorporating the probabilities of each decision, which were calculated from the in vivo imaging data, the authors’ model could recapitulate the dividing pattern of a wild-type sepal. Furthermore, when these probabilities were altered in accordance with in vivo imaging data from mutant lines, the model accurately reflected cell division patterns produced in plants mutant for cell cycle regulators (Roeder et al., 2010).

These examples show that mathematical modeling can be useful in predicting developmental outcomes. In fact, several developmental systems have already been extensively modeled using network analysis and live imaging approaches to facilitate hypothesis-driven research (Roeder et al., 2011; Ryu et al., 2013). Although impressive models have been generated that help to explain how the shape and growth of lateral floral organs are achieved (Cui et al., 2010; Green et al., 2010), predictive models that can accurately describe the dynamic behaviors of large-scale networks required for tissue differentiation and organ specification remain elusive.

VIII. Conclusions

Over the past 25 yr, considerable progress has been made in deciphering the GRNs and molecular mechanisms underlying the formation of flowers. Despite these advancements, several important aspects of flower development remain poorly understood. These include processes such as boundary formation as well as the differentiation and maturation of floral organs. Several lines of investigation must be followed to address these knowledge gaps and to further unravel the structure and composition of the flowering gene network. First, the activities of the known floral regulators must be analyzed with stage- and tissue-specific resolution. The availability of new experimental approaches, such as the TRAP or INTACT method, should dramatically facilitate this work. Secondly, the regulatory complexes that control gene expression during flower development must be characterized. To this end, advanced proteomics techniques are now available. A combination of these different approaches and the development of more sophisticated computational models with predictive functions should lead in the coming years to major advances in our understanding of the GRNs underlying the formation of flowers in A. thaliana as well as in other angiosperm species.


Work in our laboratory is supported by grants from Science Foundation Ireland. We thank three anonymous reviewers for helpful suggestions.