Reflections on the ABC model of flower development

History

Humans have long been fascinated by flowers, perhaps initially due to their presence as a sign or harbinger of food. Thus, it is not surprising that variations in flower structure are described in the oldest surviving botanical literature. For example, Theophrastus wrote (circa 300 BC), “Among roses there are many differences, in the number of petals, … in beauty of colour, and in sweetness of scent. Most have five petals, but some have twelve or twenty, and some a great many more than these; for there are some, they say, which are even called “hundred-petalled.” [vol II, book VI 3 to 5, p39 (Theophrastus 1916)]. A few hundred years later, Pliny also described (possibly the same) rose variants: “The essential points of difference in the rose are the number of the petals, the comparative number of thorns on the stem, the colour, and the smell. The number of the petals, which is never less than five, goes on increasing in amount, till we find one variety with as many as a hundred, and thence known as the “centifolia” [Book XXI Chap 10 (Plinius the Secondus 1634)]. As China is the center of diversification of the genus Rosa, some variants likely predate the Greek and Roman descriptions (Wang 2007). Following the Renaissance, when herbals describing medicinal plants proliferated during the age of exploration, descriptions and drawings of “double flowers” also multiplied, not only for roses, for example, the double-flowered rose depicted on page 93 in Dodoens’ Florum, et Coronariarum Odoratarumque Nonnullarum Herbarum Historia (Dodoens 1568), but for other species as well. Gerard (Gerard et al. 1597) depicted no fewer than 15 species with double-flowered variants, including in the Brassicaceae (Fig. 1A), the latter being included in some earlier publications (L’Obel 1581; Dodoens 1583). Centuries later, double flowers of Aquilegia vulgaris were examined in greater detail with the finding that some stamens are converted into petals with nectaries, with petals from one whorl neatly fitting into those of another (Moquin-Tandon 1841).

While double-flowered blossoms are conspicuous, other less striking flower variants were also described centuries ago (Fig. 1B). For example, an apple tree (Malus) in M. Aux's orchard in Saint-Valery-sur-Somme that was “similar in every way to the common apple tree in terms of leaves and flower arrangement, differs in a very remarkable way in the structure of its flowers and fruits. A tomentose peduncle supports a flower composed of a calyx with ten folioles fused at the base, arranged in two alternating rows, the inner ones slightly shorter. The corolla and stamens are absent; the styles, fourteen in number, slightly hairy at the base, are surmounted by a very vigorous oblique stigma” (Tillette de Clermont-Tonnere 1825). Usually in apple flowers “the calyx is simple, with five leaflets; here, it is clearly double, with the inner part similar to the outer part, persisting like it and cannot be assimilated to a corolla. In the former, there are five styles and stigmas; in ours, there are fourteen. Ordinary apples have only five compartments; these have fourteen in two parallel rows, representing two apples fused end to end” (Tillette de Clermont-Tonnere 1825). This phenotype was maintained after grafting onto a normal apple tree, but unfortunately, the phenotype of progeny derived from crosses between mutant flowers and wild-type stamens was not reported (Tillette de Clermont-Tonnere 1825). A similar, albeit less severe, phenotype was noted in a Rhododendron variant, known as Misome-sho in Japanese and described in 1692, that exhibits a sepaloid corolla and altered stamen development (Cheon et al. 2018).

It is clear from the descriptions of the Malus and Aquilegia flowers that it was recognized that there had been a homeotic transformation of one organ type into another (Tillette de Clermont-Tonnere 1825; Moquin-Tandon 1841), although the term “homeotic” was not coined until later. For example, Moquin-Tandon's (1841, p. 386) interpretation of the mutant Malus flowers, “It is possible that, due to the absence of stamens, the androecium has transformed into a row of carpels that are fused to the ordinary carpels. In that case, the phenomenon in question would fall into the category of organic metamorphoses”. Later, Masters, in his Vegetable Teratology, uses the Malus example to illustrate his concept of metamorphy (Masters 1869). Bateson subsequently replaced Masters’ metamorphy with the term we know today, homeosis (Bateson 1894).

While anatomical and morphological descriptions of “normal” flower development and variations thereof abounded (see Moquin-Tandon 1841; Payer 1857; Masters 1869; Penzig 1890-4; Wordsell 1915-6), the connection to heritability was not often made. Perhaps some of the reluctance stemmed from observations such as made by Darwin: “Such flowers as the carnation, common tulip, and hyacinth, which are believed to be descended, each from a single wild form, present innumerable varieties, differing almost exclusively in the size, form, and colour of the flowers. These and some other anciently cultivated plants which have been long propagated by offsets, pipings, bulbs, & c., become so excessively variable, that almost each new plant raised from seed forms a new variety, ‘all of which to describe particularly,’ as old Gerarde wrote in 1597, ‘were to roll Sisyphus's stone, or to number the sands‘” (Darwin 1876). However, heritability is alluded to occasionally. Darwin himself described a number of heritable floral variants, including a poppy in which the stamens are converted into pistils [de Candolle quoted in (Darwin 1876)], and several authors noted the inheritance of a number of peloric floral varieties, including that of Antirrhinum majus (Naudin 1865; Godron 1874; Darwin 1876). And much earlier, Parkinson noted that while double-flowered plants never bore seed, if double-flowered plants are desired, one should collect plants from single-flowered plants in the vicinity of the double-flowered plant, and sometimes a proportion of these would be double-flowered even though the original single-flowered parents were indistinguishable (Parkinson 1629). Following the rediscovery of Mendel in the early 20th century, it was rapidly determined that a Matthiola double-flower phenotype was recessive with respect to the single-flower phenotype (Bateson et al. 1905), explaining why double flowers, which are entirely male and female sterile, only arise from single-flowered seed stocks, apparently at random, according to breeders in the days before genetics. Furthermore, as noted by some early breeders, single-flowered plants bearing flowers with extra petals may indicate plants that we now know as heterozygotes (Bateson et al. 1905).

For a majority of the 20th century, genetics and developmental biology of plants continued in distinct trajectories despite having merged in the study of animals in the mid-20th century (Meyerowitz et al. 1989; Prunet and Meyerowitz 2016). Thus, early models of flower development typically employed hormonal or biophysical signals or forces, rather than genes, that would influence floral organ identity. Perhaps the earliest model was proposed by Goethe where he viewed floral organs as modified leaves, with their modification from leaves into floral organs directed by ascending sap that became more refined as it made its way to the apex (Goethe 1790). Abnormal organs could be viewed as a failure of refinement of the sap. Other models relied on similar relay mechanisms such that one type of floral organ would induce the next inner set of organs. In some models the mechanism was proposed to be biomechanical (Green 1988), while in others, genetic programs were invoked (Heslop-Harrison 1963). However, these models failed to explain existing floral varieties where organ types were absent or in inappropriate positions. In contrast, models based on concentric rings of hormonal activity, avoiding the inter-whorl communication, were better equipped to explain known floral mutant phenotypes but invoked shifts in the positions of organ primordia rather than changes in the concentric rings of hormonal activity (Brieger 1935; Holder 1979).

Genetics

But I (J.L.B.) did not know most of the above as I peered through the dissecting microscope in a basement annex of the Kerckhoff building at Caltech in early 1990 when I was a PhD student in Elliot Meyerowitz's laboratory. I was searching among Arabidopsis plants for a triple mutant in a family segregating for apetala2-1, agamous-1, and pistillata-1. While its phenotype could be predicted based on already constructed double mutant combinations (Bowman et al. 1989), there was a thrill of observation and contemplating that you have knowledge no one else in the world possessed (Fig. 1C). It was late it the evening, and all members of Elliot's lab had left, and so I sauntered down the hallway to share the discovery with my mate, Joe Minor, who was working on sea urchins in the neighboring laboratory.

These observations were the culmination of experiments begun in the early 1980s, initially by Maarten Koornneef, who was constructing an Arabidopsis genetic map and who isolated several mutants affecting floral architecture (Koornneef et al. 1983). Elliot obtained these stocks from Maarten and, subsequently, Bob Pruitt, a previous PhD student in Elliot's laboratory, and Elliot thought the mutants exhibited homeotic phenotypes. When I joined the laboratory in late 1986, I continued the construction of multiple mutants, a process initiated by Bob and Elliot. At first, we had solitary mutant alleles of each of 5 loci [APETALA1 (AP1), APETALA2 (AP2), APEPALA3 (AP3), PISTILLATA (PI), AGAMOUS (AG)]. These were chosen based on their homeotic phenotypes, although AP1 was initially left aside due to its broader role that we now know as determining floral meristem (FM) identity. The chosen loci reflected an early decision to focus on the roles in organ identity, ignoring the effects of the mutations on organ number (e.g. organ initiation or fusion in the case of carpels) and determinacy, even though it was clear the chosen genes might have functions additional to those directly involved in organ identity. In January 1988, David Smyth joined the laboratory as a sabbatical visitor, and we set about looking at wild-type flower development using scanning electron microscopy (Smyth et al. 1990). Given the size of Arabidopsis flowers, this technique was critical to interpreting mutant phenotypes, although it was not without its challenges (see Bowman 2013). David also undertook a mutant screen to isolate additional mutant alleles of the above genes as well as identify other loci (Smyth 2023). The additional alleles were key to estimating whether alleles were likely to be null or hypomorphic, influencing the construction of the model. One early incarnation was drawn on the laboratory fridge door, where we debated whether the A/C boundary was sharp or fuzzy—and based on strong ap2 alleles (Komaki et al. 1988; Kunst et al. 1989; Bowman et al. 1991), we later drew the boundary sharp. In comparison to older models, the idea of concentric rings of activity was a common theme, only now each concentric ring was defined by specific gene activities (Bowman et al. 1989).

Simply stated, the ABC model postulates 3 distinct gene activities (A, B, and C), each of which are present in 2 adjacent whorls of the Arabidopsis flower, act alone and in combination to specify the 4 types of floral organ (Fig. 2, A to D) (Bowman et al. 1991). A second postulate was that the A and C activities are mutually antagonistic, a prerequisite to be consistent with the single mutant phenotypes (Fig. 2, E to I). The model immediately suggests phenotypes for most double and triple mutants (Fig. 2, J to M). While most double mutant combinations were additive, the key double mutant was ap2 ag, where both class A and C are missing, leaving only B present in whorls 2 and 3, and no homeotic function in whorls 1 and 4, both situations not found in any whorl in wild-type flowers (Fig. 2L). The ap2 ag phenotype could be described as carpelloid sepals in the first and fourth whorls and stamenoid petals in the second and third whorls, both being new organ types not observed in wild type (Bowman et al. 1989, 1991). Two conclusions can be drawn from the observed phenotype. First, agamous mutations have phenotypic effects in the outer whorls in an ap2 background and vice versa, indicating that the 2 gene activities are in fact mutually antagonistic. Second, that the organs in the middle 2 whorls are neither wild-type petals nor wild-type stamens indicates that B class alone in Arabidopsis does promote one or the other. Intriguingly, the cells of these organs are also intermediate between wild-type petals and wild-type stamen cells and thus formed a new cell type. This is in contrast to cases where we observed mosaic organs with distinct boundaries between known cell types in single or multiple mutant combinations (Bowman et al. 1989, 1991), suggesting that floral organ identities were determined largely cell autonomously. One mutant allele of APETALA3, ap3-1, and all alleles of APETALA2 were found to be temperature sensitive, allowing the timing of gene activity to be elucidated. Consistent with a role in organ identity specification, gene function was required during the time of organ initiation and early development, with the sensitive period of stamens being quite abbreviated while that of petals was more prolonged (Bowman et al. 1989).

Concomitant with our work in Arabidopsis, 2 groups working with Antirrhinum (Enrico Coen's group at the John Innes Centre in Norwich and Suzanna Schwarz-Sommer's group at the Max Planck Institute in Cologne) identified mutants with similar phenotypes and independently came up with a BC model of flower development (Carpenter and Coen 1990; Schwarz-Sommer et al. 1990), the major difference being the lack of Antirrhinum A function genes. While it might be difficult to fathom among the current generation of young geneticists, the ABC/BC models were constructed entirely with genetics (no molecular biology or genomics) and could have been formulated decades or even a century before [see (Bowman et al. 2012)]—we even named one room in Elliot's laboratory where histology was performed “the 19th-century room.” Thus, one question was, why did it take so long? And a second question, why did multiple groups converge on the same methodology and largely similar conclusions almost simultaneously? Such convergence is a common theme in science and must indicate that the time was ripe for discovery, either via a technical or conceptual advance, the latter being key in the present case. I was personally influenced by Ed Lewis, whose laboratory was upstairs and whose seminal work in Drosophila on the identification and characterization of the bithorax complex led to a model whereby multiple genes act in combination to specify the identity of the thoracic and abdominal segments of the fly body (Lewis 1978)—I recall with both fondness and horror that one of my first experiences as a graduate student was to present this paper in a journal club with Ed in the audience, who “gently” corrected me when I went astray. Thus, it did not seem such a large jump to think about a similar model for the concentric whorls of the flower.

Molecules

The mid-1980s were a time of revolutionary insight into the molecular bases of animal development, with the discovery that genes of the bithorax complex and Antennapedia encoded paralogous proteins now known as homeobox proteins (Garber et al. 1983; McGinnis et al. 1984b; Scott and Weiner 1984; Karch et al. 1985) and that homologous genes were widespread in animals (McGinnis et al. 1984a). It was immediately speculated that homologs might exist in plants and these might also encode homeotic functions, despite the independent evolution of multicellular plants and animals from a single-celled ancestor. While homeobox genes were subsequently identified in plants (Vollbrecht et al. 1991), the first ABC gene cloned was DEFICIENS, a B class gene in Antirrhinum, and rather than being similar to homeobox proteins, it had sequence similarity with the known transcription factors MCM1 of yeast and SRF of mammals (Sommer et al. 1990). Some months later, the cloning of the Arabidopsis C class gene, AGAMOUS, was reported, and this too was homologous with MCM1, DEF, and SRF (Yanofsky et al. 1990), giving rise to the moniker of “MADS”-box for the putative DNA binding sequence of the gene family (Schwarz-Sommer et al. 1990). Given that 2 of the ABC genes encoded MADS-box genes, it was immediately speculated that other ABC genes may also encode homologous proteins. While the first genes cloned, DEFICIENS and AGAMOUS, were identified via transposon/T-DNA tagging approaches, cloning of subsequent genes took advantage of homology-based approaches. Although one could not scan genome sequences at the time, low stringency Southern blots and library screens provided a fruitful approach to clone paralogs (Fig. 3A). Thus, the remaining B and C class genes in both Arabidopsis and Antirrhinum were quickly identified, either by homology to MADS-box genes in general or via assumption of orthology between the 2 species (Jack et al. 1992; Tröbner et al. 1992; Bradley et al. 1993; Goto and Meyerowitz 1994). However, in retrospect the Antirrhinum C-class gene PLENA turned out to be a ortholog of Arabidopsis SHATTERPROOF1/2 genes, which are sister paralogs of AGAMOUS (Davies et al. 1999). Later, the Arabidopsis A class gene APETALA1 was also cloned based on MADS-box homology (Mandel et al. 1992b). The Antirrhinum AP1 ortholog, SQUAMOSA, was not recognized as an A class gene but rather was proposed to play a role in flower meristem identity (Huijser et al. 1992), a theme we will return to shortly. It was also soon recognized that plant genomes harbor a large contingent of MADS-box genes that extended well beyond the ABC genes [e.g. (Ma et al. 1991; Pnueli et al. 1991)], and these would subsequently be targets of future reverse genetic approaches (Fig. 3B).

The cloning of the genes enabled the examination of the ABC model from a molecular perspective, with expression patterns of B and C class genes largely congruent with mutant phenotypes (Yanofsky et al. 1990; Jack et al. 1992; Schwarz-Sommer et al. 1992; Tröbner et al. 1992; Bradley et al. 1993; Goto and Meyerowitz 1994). Critically, in Arabidopsis, the expression pattern of AGAMOUS expanded into all floral whorls in the A-class mutant apetala2, supporting the hypothesized antagonism between these 2 classes (Drews et al. 1991). Having the cloned genes in hand also allowed further tests by creating gain-of-function alleles in Arabidopsis (disappointingly, Antirrhinum is still difficult to transform). Thus, it was noted that ectopic expression of the Brassica napus AGAMOUS gene throughout the flower was sufficient to transform the identity of the outer whorls of tobacco flowers into carpels and stamens, respectively, upholding one tenet of the model and also providing additional support for universal ABC gene activity across eudicots (Mandel et al. 1992a). Likewise, ectopic expression of both APETALA3 and PISTILLATA was sufficient to produce homeotic transformations in the first and fourth whorls (Krizek and Meyerowitz 1996) (Fig. 3C). Given that APETALA3 and PISTILLATA were required for B-class activity [they act as an obligate heterodimer (Schwarz-Sommer et al. 1992; Goto and Meyerowitz 1994; Davies et al. 1996; Riechmann et al. 1996)], one surprise was that ectopic expression of PISTILLATA alone partially transformed sepals into petals and ectopic expression of APETALA3 alone partially transformed carpels into stamens. These observations were reconciled by careful examination of gene expression patterns, with APETALA3 expressed during early sepal development and PISTILLATA expression detected in the fourth whorl. Thus, where the 2 gene expression patterns overlap determines B-class activity, and it is possible that the “sloppy” expression patterns could facilitate evolution of perianth phenotypes. Notably, where B- or C-class genes were expressed throughout the entire plant, the leaves were curled and small, but no transformation of leaves into floral organs was observed (Mandel et al. 1992a; Krizek and Meyerowitz 1996).

Converting leaves into floral organs

A major advance at the end of beginning of this century was the discovery of E-class MADS box function. Pivotal was the continued investigation of MADS-box paralogs in Arabidopsis and the discovery that a triple mutant of 3 paralogs resulted in a flower with all sepals (Fig. 3D), implying that this class of gene was required for B- and C-class function (Pelaz et al. 2000). Inclusion of mutations in a fourth paralog resulted in flowers with all leaf-like organs, indicating their requirement for all 4 floral organ types (Ditta et al. 2004). The genes, named SEPALLATA, were expressed through the flower and did not appear to provide any organ identity specification activity on their own because their ectopic expression alone is not sufficient to convert leaves to floral organ identity (Honma and Goto 2001); for this reason, I argued that this should be called F-class, for Flower. Remarkably, ectopic expression of ABC genes along with SEPALLATA genes was sufficient to convert leaves into floral organs (Honma and Goto 2001; Pelaz et al. 2001). Thus, the MADS-box genes are necessary for the specification of floral organ identity and are sufficient (along with E-class) to convert leaves into floral organs, supporting ideas of the evolution of floral organs from leaf-like ancestors (Fig. 3, E and F). It was noted earlier that DEFICIENS, GLOBOSA, and SQUAMOSA proteins could form a ternary complex (Egea-Cortines et al. 1999), and ternary and quaternary complexes of Arabidopsis proteins including ABC and SEPALLATA MADS-box proteins (e.g. B + C + SEP and A + B + SEP) were observed, prompting the hypothesis that these complexes form the molecular basis of the ABC model (Honma and Goto 2001). These observations were summarized in the “quartet” model wherein MADS-box proteins act as “dimers of dimers’ in their binding to regulatory sequences of target genes (Theissen and Saedler 2001) (see “Good things come in fours” section below).

That the ABC triple mutant produced carpelloid leaves while the organs of the sepallata1234 quadruple were more leaf-like suggested that additional MADS-box genes active in the ABC triple mutant background contributed to carpel differentiation. Two AGAMOUS-related paralogs, SHATTERPROOF1/2, were found to be ectopically expressed in the apetala2 agamous background and shown to be responsible for AGAMOUS-independent carpel development (Savidge et al. 1995; Pinyopich et al. 2003). Thus, flowers of the pentuple mutant, apetala2 pistillata agamous shatterproof1/2 (Fig. 3G), consist entirely of leaf-like organs (Lee et al. 2005a).

A molecular explanation for the indeterminacy of agamous flowers was provided by the observation that WUSCHEL, which acts to maintain stem cells in the flower meristem, is repressed directly by AGAMOUS (Sun et al. 2009; Liu et al. 2011). The repression of WUSCHEL may also explain other determinacy effects observed when carpels develop in inappropriate whorls, such as the reduced numbers of organs in genotypes containing apetala2 null alleles. Finally, that carpels promote fusion and determinacy is supported by reduced determinacy in genotypes, where carpel identity is reduced by introduction SPATULA or CRABS CLAW mutant alleles (Alvarez and Smyth 1999).

The curious case of A-class and subfunctionalization of B- and C-classes

When venturing away from Arabidopsis, genetic manipulations did not always yield the phenotypes expected if function was strictly conserved with Arabidopsis ABC orthologs. Diving into the complex histories of plant genomes revealed that gene/genome duplications, followed by neo- and subfunctionalization or loss, are the source of many variations observed.

The conservation, or even the existence, of A-class has been debated, with the most common hypothesis being that A-class is a continuation of FM identity (e.g. Causier et al. 2010), as is also observed in Arabidopsis. In Arabidopsis, A-class was defined as having 2 functions in the perianth—the repression of C-class and an organ identity role (Bowman et al. 1991). That there exists a distinct A-class activity is supported by the observation that organs of the middle whorls of apetala2 agamous flowers are petal/stamen hybrids, and thus, B-class plus SEPALLATA does not specify petals. Furthermore, sepals are not leaves (or bracts), and thus their identity must be specified by some activity, even if it overlaps with the activity that promotes FM identity.

A-class activity was proposed to be composed of the MADS-box gene APETALA1 and APETALA2 (Irish and Sussex 1990; Bowman et al. 1991, 1993), encoding an AP2-type transcription factor (Jofuku et al. 1994). In null apetala1 alleles sepals are converted into bract-like organs (Fig. 3H), and petals are usually missing (Irish and Sussex 1990; Bowman et al. 1993). However, in hypomorphic apetala1 alleles (Fig. 3I) carpelloid tissue can be formed on the first whorl organs and the petals may be partially transformed into stamens (Bowman et al. 1993), a phenotype often neglected in later reviews of the ABC model. A role for APETALA1 in organ identity could also be masked by the action of CAULIFLOWER, a closely related paralog (Kempin et al. 1995): the double mutant ap1 cal produces inflorescence meristems instead of flower ones (Fig. 3J), indicating both genes have a role in FM identity and precluding observation of potential organ identity defects (Bowman et al. 1993). Notably, the sepallata1234 quadruple mutant does not appear to have defects in FM specification, although SEPALLATA4 may have redundant roles with APETALA1 in this regard (Ditta et al. 2004).

Consistent with mutual antagonism and A and C classes in Arabidopsis, AGAMOUS mRNA is detected in all 4 whorls of apetala2 mutants (Drews et al. 1991), and APETALA1 mRNA expands to the inner whorls in agamous flowers (Gustafson-Brown et al. 1994). However, AGAMOUS mRNA also accumulates in the inner whorls of agamous flowers (Gustafson-Brown et al. 1994), and APETALA2 mRNA can be found at some levels in all 4 whorls of wild-type flowers (Jofuku et al. 1994), indicating that their mutual antagonism is complex. The negative regulation of APETALA2 by miR172 provides one mechanism to restrict AP2 activity post-transcriptionally (Aukerman and Sakai 2003; Chen 2004). Expression of miR172 in the central whorls during early flower development appears to provide quantitative regulation of AP2 activity, with the balance of AGAMOUS and APETALA2 activity, especially in the third whorl, proposed to influence organ identity (Wollmann et al. 2010).

Outside of Arabidopsis, findings in asterids demonstrated that floral homeotic phenotypes are incompatible with a straightforward transfer of Arabidopsis A ortholog function (Cartolano et al. 2007; Morel et al. 2017). The AP2 family comprises APETALA2-type and TOE-type clades (Wang et al. 2016). Petunia AP2A/ROB1 and Antirrhinum LIP1/2 genes are APETALA2 orthologs, but unlike APETALA2, these are unable to repress the C-function. Instead, it is another miRNA, called BLIND in Petunia and FIS in Antirrhinum, that tackles this task by targeting NF-YA transcription factors. BLIND is helped by BLIND ENHANCER (BEN), and only the double mutant blind ben mimics the expected A class phenotype (Morel et al. 2017). BEN is a TOE-type AP2 whose members control flowering time, not organ identity, in Arabidopsis. BEN was also shown to work with the 3 Petunia AP2-type genes called ROBs (Repressor Of B function) so that BEN and ROBs together repress B function (but not C-function) in the sepal whorl. To summarize, in Petunia A function relies on a combination of a related AP2 gene and an unrelated microRNA (BEN + BLIND); reciprocally, the APETALA2 orthologs (ROB genes) do not repress C function but rather team up with BEN to repress B function.

Compared to A class, the molecular agents fulfilling B and C functions appear broadly conserved across angiosperms. This difference in extent of conservation may be related to the antiquity of the functions, with B and C evolving at the dawn of seed plants, while A-function appears to be an angiosperm novelty (Wellmer et al. 2014). Nonetheless, variations involving related paralogs exist. For example, the B gene APETALA3 is only 1 of 2 old twins: a duplication created APETALA3 and TM6 somewhere near the base of eudicots (Kramer et al. 1998; Kim et al. 2004), but Arabidopsis and Antirrhinum have both lost their TM6 ortholog. So the original floral development genetic models only tell part of the story, and B function can be more complicated in other species. In Petunia the APETALA3 ortholog (PhDEF) and PhTM6 redundantly control stamen identity, but petals are specified by PhDEF only. Phtm6 mutants have a wild-type phenotype, whereas in a Phdef mutant petals are replaced by sepals but the stamens are normal (Vandenbussche et al. 2004; Rijpkema et al. 2006). A similar logic applies to the PISTILLATA lineage, which represents the other leg of the B function in Arabidopsis: the Petunia genome hosts 2 PI orthologs acting redundantly to specify petal and stamen identity (Vandenbussche et al. 2004), with one paralog having acquired an additional role controlling stamen to petal tube fusion. Hence, the expected B phenotype is only obtained when both paralogs are defective. Such duplications also play important roles for petal elaboration (see the “Beyond eudicots” section below), impacting not only organ identity specification but morphogenesis as well (Kramer et al. 2007; Sharma and Kramer 2013; Hsu et al. 2015, 2021).

Evolution within the MADS-box gene family also affected how C-function is executed. Duplications yielded 3 genes in the “C-lineage”: AGAMOUS, SHATTERPROOF1/2 (PLENA), and SEEDSTICK, with the first duplication producing SEEDSTICK occurring in the ancestral seed plant and the other occurring within eudicots [e.g. (Carvalho et al. 2018)]. In Petunia both AGAMOUS and SHATTERPROOF1/2 orthologs must be inactivated to produce typical C-class mutant phenotype (Kapoor et al. 2002; Heijmans et al. 2012). In Arabidopsis, AGAMOUS fulfils the C function with SHATTERPROOF1/2 controlling aspects of carpel and fruit development (Liljegren et al. 2000). In contrast, PLENA executes the C function in snapdragon, while the AGAMOUS ortholog, FARINELLI, only plays a secondary role during stamen development (Bradley et al. 1993; Davies et al. 1999). The third paralog of the clade, SEEDSTICK, directs ovule development (Pinyopich et al. 2003) and is sometimes referred to as D class that is unrelated to floral organ identity. However, even SEEDSTICK alone can promote aspects of carpel development in the absence AGAMOUS (Favaro et al. 2003), and perhaps its role in ovule development reflects an ancestral function of this clade since ovules predate flowers.

Regarding E-class, in rice (Ohmori et al. 2009; Li et al. 2010), Petunia (Rijpkema et al. 2009), orchid (Hsu et al. 2015, 2021), and maize (Thompson et al. 2009) SEPALLATA orthologs and also their close paralogs, the AGL6 genes, can all deliver E function, adding another layer of redundancy and complexity to the ABC story (for review, see Dreni and Zhang 2016).

To accommodate these variations, it was proposed that the model be referred to as the (A)B(C) model (Causier et al. 2010; Theissen et al. 2016) with (A) accounting for A and E activities and (C) combining C and D roles. However, now that we have a broader evolutionary perspective of gene evolution within the ABCE clades (Fig. 3B) combined with functional analyses (above and later), most described variation within core eudicots is derived from vagaries in subfunctionalization of paralogous genes rather than any fundamental differences. It remains to be seen whether exceptions will occur in species with “noncanonical” perianths, such ANA-grade and magnoliid angiosperms (Wellmer et al. 2014).

Paddling upstream and downstream

As all big discoveries, the formulation of the ABC model immediately raised difficult questions: what regulatory processes generate the spatio-temporal expression patterns central to the homeotic function of ABC genes? How can the combinations uncovered by genetic analyses transpose into molecular mechanisms allowing different protein complexes to activate distinct developmental programs? How do the regulatory activities of TF-encoding ABC genes translate into morphogenic effects that shape the different types of floral organs, specifying a range of often organ-specific cell fates? Although we still lack a high-resolution picture, the last 20 years of floral development research have added many pieces to this incomplete puzzle. Our understanding of the complex networks behind the characteristic ABC expression domains has been synthesized regularly over the last 2 decades (Zahn et al. 2006; Causier et al. 2009; Pelayo et al. 2021). In the interest of brevity, we only summarize here the main messages and highlight recent breakthroughs toward solving long-lasting riddles.

Swimming upstream

Early work rapidly established that regulation of ABC genes expression is a multi-layered process integrating inputs from hormonal signaling pathways with the activity of transcription factors, histone modifying enzymes, and chromatin remodeling complexes and roles for post-transcriptional mechanisms often involving mobiles noncoding RNAs and RNA processing factors (e.g. Cheng et al. 2003; Chen 2004; Zhao et al. 2007; Yamaguchi et al. 2009; Wu et al. 2012, 2018; Jung et al. 2014; Sacharowski et al. 2015; Poyatos-Pertíñez 2016; Morel et al. 2017; Lee et al. 2022; Lin et al. 2023). These investigations exposed a 3-step waltz regulating ABC genes: upstream factors initiate transcription across broad territories (step 1), followed by refinement (step 2), and then maintenance (step 3) of precise expression domains. Step 2 and 3 rely extensively on cross regulatory interactions and feedback loops between floral homeotic genes in a manner reminiscent of the processes establishing discreet stripes of pair-rule gene expression in flies: combined inputs of maternal determinants and GAP genes first trigger expression of pair-rule genes in broad territories, which then get refined through cross regulations between the pair-rule genes themselves (Edgar et al. 1989; Pankratz and Jäckle 1990) (Fig. 4). Flower patterning follows a similar logic: the broad expression domains of ABC genes in nascent floral meristems largely overlap at first, but cross inhibition between A and C later renders their expression mutually exclusive while positive feedback loops between APETALA3 and PISTILLATA enhance and eventually restrict any significant B genes transcription to the subset of meristematic cells that initially expressed both (Fig. 4). Additionally, B-class genes may act to directly repress genes required for carpel development, providing another layer of cross-regulation (Wuest et al. 2012).

Cadastral genes help define and maintain sharp boundaries between the expression domains of floral homeotic genes: for example, SUPERMAN and RABBIT EARS encode C2H2 zinc finger repressors that turn off APETALA3 in whorl 4 (Sakai et al. 1995; Prunet et al. 2017) and AGAMOUS in whorl 2 (Krizek et al. 2006; Huang et al. 2012), respectively. In superman mutants, it appears as if the flower meristem is “stuck” in the production of third whorls (Fig. 3, K and L) (Bowman et al. 1992; Baum et al. 2001). These boundaries are no-mans-land for ABC gene expression that keep organ identity apart but also physical borders characterized by a lack of growth, allowing morphogenesis to proceed independently in each whorl to produce discreet organ types.

Many regulators of ABC genes are not exclusive agents of floral organs but instead participate in the control of other developmental processes. A key finding was the discovery that molecular players inducing floral transition directly participate in the activation of floral homeotic selectors. One of those elements, LEAFY (LFY), stands out because it can activate the transcription of A, B, C, and E genes by binding directly to their regulatory regions (Parcy et al. 1998; Busch et al. 1999; Wagner et al. 1999; Honma and Goto 2001; Lohmann et al. 2001; Lamb et al. 2002; Benlloch et al. 2011; Moyroud et al. 2011; Winter et al. 2011; Rieu et al. 2023). Because LFY is a central regulator of the reproductive transition, these results elegantly connect induction of flowering to specification of floral organ identity but also yield an initial paradox: how can a regulator expressed throughout the floral buds activate different floral homeotic genes in spatially distinct territories? With its uniform expression across floral meristems, LFY seemed at first ill-suited to such task.

Floating downstream

Another key question is how combinations of ABC proteins lead to floral organ differentiation. One approach to identifying downstream targets was to isolate mutants affecting only a single organ type, e.g. carpels (Alvarez and Smyth 1999). For example, CRABS CLAW mutants alter carpel development and whereas some effects are independent of AGAMOUS activity, others seem to rely on AGAMOUS (Alvarez and Smyth 1999). Indeed, when molecular approaches were available, CRABS CLAW was found to be a direct downstream target of AGAMOUS (Gómez-Mena et al. 2005; Ó’Maoiléidigh et al. 2013), with requisite dual CArG boxes (MADS-box binding site) for binding of MADS-quartets required for its expression (Lee et al. 2005a). CRABS CLAW's role in carpel development is conserved across angiosperms, with the rice ortholog, DROOPING LEAF, indispensable for carpel specification (Yamaguchi et al. 2004).

CRABS CLAW was also found to be required for nectary development (Bowman and Smyth 1999). Arabidopsis nectaries are normally found at the base of stamens and appear to be a third whorl organ independent of the identity of the floral organ occupying that whorl (Baum et al. 2001; Lee et al. 2005a). Such a role for CRABS CLAW was found to be widespread in eudicots regardless of nectary position within the flower, or even elsewhere in the plant, leading to speculation that nectaries came under CRABS CLAW control when they became associated with reproductive organs within the flower (Lee et al. 2005b). However, my (J.L.B.) foray into flower development was coming to an end as we discovered CRABS CLAW was a member of a small gene family, the YABBY gene family (Bowman and Smyth 1999; Siegfried et al. 1999), and enhancer mutagenesis revealed a role for this family and another, the KANADI gene family, in establishing organ polarity not only in the floral organs but also leaves (Eshed et al. 1999, 2001).

Disappointingly, 20 years later our understanding of what lays downstream of ABC remains surprisingly fragmentary: genome-wide binding approaches (SELEX-seq, ChIP-seq, and Dap-seq) have successfully identified in vivo binding sites of ABC regulators associated with genes controlling cell differentiation, division, and expansion, participating in hormonal signaling, or coding for other TFs and even chromatin remodeling and assembly factors (Kaufmann et al. 2009, 2010b; Yant et al. 2010; Wuest et al. 2012; Ó’Maoiléidigh et al. 2013, Pajoro et al. 2014; Chen et al. 2018; Lai et al. 2020). However, picking out bona fide regulatory cis-elements from inactive binding sites remains challenging and although many ABC direct targets have been predicted, only a relatively small number have been validated. Despite this, remarkable progress has been made uncovering the mode of action of the floral quartets [recently reviewed in (Goslin et al. 2023; Käppel et al. 2023)] and in particular toward understanding how closely related MADS proteins with very similar DNA binding properties identify targets accurately and selectively among thousands possible targets. Teamwork is key and ABC proteins achieve specificity by following the metaphor of the Swiss cheese model: many leaky layers, which individually would confer very imperfect selectivity, can forge a highly specific system when compounded with each other.

ABC through the evolutionary lens

When one examines flower architecture across the quarter-million species of angiosperms, a common theme is organ order. Although flowers differ enormously in numbers, shapes, positions and even numbers of whorls of organs, the organ order (sepals, petals, stamens, carpels) is nearly constant (although missing organ types is a common theme). That similar phenotypes were observed in Arabidopsis, a rosid, and Antirrhinum, an asterid, immediately suggested that their common ancestor (that of core eudicots) and perhaps all angiosperms utilized a common genetic program, consistent with a singular origin of flowers (Coen and Meyerowitz 1991). Yet could a universal and relatively simple model underpin the diversity of floral forms seen across the angiosperms? And what is the evolutionary origin of the floral network? To account for various floral morphologies, modifications of ABC model by shifting/sliding or fading boundaries have been proposed (van Tunen et al. 1993; Bowman 1997; Kramer et al. 2003; Buzgo et al. 2004), and we consider only a few variations here.

Variations on a theme

One early line of investigation concerned how the ABC model might apply to grass flowers, which have a highly derived perianth architecture. Maize silky1 flowers exhibited transformations of stamens into carpels and lodicules into palea/lemma-like organs—a canonical B-class mutant phenotype—and subsequent cloning of the gene revealed silky1 to encode an APETALA3 ortholog (Ambrose et al. 2000). Thus, lodicules were interpreted to be petal homologs and the palea/lemma to be sepal homologs (Fig. 5A). C-function was also found to be conserved, although it was split across 2 paralogs (Mena et al. 1996). Similar results were found in rice (Kyozuka et al. 2000), indicating the ABC model applies to grass flowers, despite their divergent perianths. A minor variation of the ABC model was proposed for tulip flowers based on a B-class mutant in which both outer whorls of tepals (perianth organs not differentiated into sepals and petals) were transformed into sepal-like organs not normally found in tulips (Fig. 5B), and stamens were transformed into carpels (van Tunen et al. 1993), with gene expression patterns of ABC orthologs supporting this hypothesis (Kanno et al. 2003).

Transient assays, such as virus-induced gene silencing, facilitated functional investigations in a wider range of systems. Combined with modern -omics approaches, these not only painted a more holistic view of how floral organ identity is specified by the ABC but also started to unveil the molecular bases accounting for morphological variation. Two cases of parallel evolution exemplify sub- and neofunctionalization of B-class genes in the evolution of flower novelty. [I (J.L.B.) once remarked that researchers in the field followed a similar path—they multiplied and diversified via sub- and neofunctionalization…] In the rununculid Aquilegia, diversification of APETALA3 orthologs results in 3 distinct combinations that promote petal, stamen, and staminode identity (Kramer et al. 2007; Sharma and Kramer 2013). In this case a novel organ, the staminode, is produced, whose ecological function is unknown, although they likely play a role in pollination (Fig. 5C). Another ranunculid, Thalictrum, exhibits expansion of B-class activity into the outer whorl that results in petaloid sepals in a parallel evolution of this phenotype with tulips (Martínez-Gómez et al. 2020).

Sub- and neofunctionalization of B-class is also observed in orchids, where there has been a diversification of APETALA3 orthologs into 4 distinct clades (Fig. 5D). In this case, the perianth organs are specified by a “perianth code,” with each distinct organ type having a different qualitative, and sometimes quantitative, suite of B and E activities that compete with each other, making individual tepals morphologically distinct from each other: sepal/petal vs. labellum (Mondragon-Palomino and Theissen 2011; Pan et al. 2011; Hsu et al. 2015). Such complexes also appear to have gained function beyond specifying organ identity: quartet components seem to vary spatially with exquisite precision so that subregions within an organ could harbor different multimers to sculpt and pattern a given organ section (Fig. 6)—for instance, interfering with the formation of specific quartets in moth orchids prevented curved edge formation and red anthocyanin spots production at the base of lateral sepal (Hsu et al. 2021).

Downregulation of the B-class gene GGLO1 in the Asteraceae family plant Gerbera hybrida resulted in transformation of second whorl organs into ones bearing pappus bristles, indicating that the pappus bristles of wild-type Asteraceae flowers represent sepals modified to facilitate seed dispersal (Yu et al. 1999) (Fig. 5E). Another proposed modification of the ABC model is one where the boundaries between the different classes are indistinct, that is, fading/fuzzy borders (Buzgo et al. 2004; Soltis et al. 2007), and which may apply to species where the distinction between sepals and petals or petals and stamens can be vague (Fig. 5F). Lastly, the only known angiosperm in which the organ order of the reproductive organs is transposed is Lacandonia schismatica (Fig. 5G), a saprophytic plant growing only in the Lacandon jungle of southern Mexico. In this species the B- and C-class gene expression patterns mirror organ identity (Alvarez-Buylla et al. 2010). Closely related to L. schismatica is the dioecious Triuris brevistylus (Espinosa-Matias et al. 2012), whose female flowers resemble those of L. schismatica except for the central stamens, providing a possible evolutionary pathway to the unique floral architecture of L. schismatica.

Another early avenue of excitement was whether the ABC genes would be involved in the evolution of monoecy or dioecy in angiosperms. Now that a number of dioecious angiosperms have been molecularly characterized, it seems clear that the ABC genes are not the sex determining genes [as speculated in (Bowman 1997)] but rather that the sex determination mechanisms may converge on changes in B-class gene activity [reviewed in (Montalvão et al. 2021; Zhang et al. 2022)]. One potential exception may be in a gymnosperm, Cycas, where GGM13 (which resides in the phylogenetic sister group to B-class genes) is located in the male determining region of the Y chromosome (Liu et al. 2022). Finally, it must be stressed that not all floral diversity originates from modification of floral MADS-box genes or their downstream targets. For example, change in floral ground plan, such as number and position of floral organ primordia appears largely independent of ABC gene activity (Smyth 2018).

ANA grade and basal eudicots

ABC homologs tend to exhibit broader expression patterns in early diverging angiosperms (Kim et al. 2005; Yoo et al. 2010; Li et al. 2015; Moschin et al. 2021). In vitro studies indicate those are capable of forming floral quartets, but protein-protein interactions are more promiscuous in ANA grade species and early diverging eudicots compared to core eudicots (Liu et al. 2010; Melzer et al. 2014; Li et al. 2015; Rümpler et al. 2018). Hence, the elaboration of the ABC complexes during flowering plant diversification may be best understood by drawing parallels with the formation of a neural network: during animal brain development a large number of neurons is produced (i.e. extensive duplication of MIKC^C-MADS TFs), then as many synapses as possible form between those neurones producing innumerable connections (i.e. almost all combinations of protein-protein interactions between promiscuous MADS-box genes are possible as expression patterns largely overlap). Finally, pruning ensues to selectively eliminate many connections [i.e. paralogous MADS are lost, protein-protein interaction become exclusive between those that remain (Melzer et al. 2014; Li et al. 2015), and expression patterns are sharpened]. Taken together these processes certainly contributed to sculpt the ABC model of eudicots as we know it but also likely introduced the robustness and canalization needed for reliable floral development.

Gymnosperms

While flowers are an angiosperm innovation, ABC-type MADS box genes predate the evolution of flowers (Tandre et al. 1995). Gymnosperm C-class gene expression was detected in both developing female and male cones (Tandre et al. 1998), whereas B-class gene expression is limited to male cones (Mouradov et al. 1999; Sundstrom et al. 1999), suggesting that the interplay between B and C components of the ABC model, key to sex determination, evolved in the ancestral seed plant (Fig. 5H). However, functional investigations in gymnosperms remain difficult, and progress has been limited to heterologous expression, in vitro biochemical assays, and characterization of a handful of natural variants (Carlsbecker et al. 2013). That gymnosperm genomes possess genes similar to A- or E-class angiosperm genes, some of which are expressed in cones (Carlsbecker et al. 2013; Gramzow et al. 2014), suggests that quartets of MADS-box proteins may act in gymnosperms as well (Theißen et al. 2016). Indeed, gymnosperm MIKC^C-type proteins can form quartet complexes, with C-homotetramers and BC-heterotetramers able to loop DNA and E proteins (AGL6) not essential to complex formation, at least in vitro (Wang et al. 2010). Whether BC and BCE complexes are both functionally relevant in vivo and/or are capable of activating different sets of genes cannot yet be tested. However, reconstruction and characterization of an ancestral version of BCE complexes suggested that E was necessary for quartet formation specifying male identity in the ancestral seed plant (Ruelens et al. 2017). Some regulatory players accounting for the spatio-temporal expression patterns typical of angiosperms ABC also predate flower evolution: B-gene promoters in Welwitschia can recruit LFY to high-affinity binding sites also present in regulatory regions of other conifer B-genes (Moyroud et al. 2017). LFY may already rely on its cofactor UFO to regulate gymnosperm B gene expression because the LFY-UFO connection is likely ancient, but functional UFO genes remain to be identified in nonflowering plants (Rieu et al. 2023).

Thus, key components required to build a flower were already present in the ancestral seed plant, so that the floral ABC network likely derived from a minimal gene regulatory circuit that controlled reproductive organ identity in the last common ancestor of seed plants. In this view the origin of the flower primarily required the construction of an elaborate perianth surrounding the reproductive organs and the evolution of the carpel enclosing the ovules, with the subsequent development of a fruit enclosing the seeds. In this respect, the continued generation of MADS-box gene paralogs was instrumental in the evolution and diversification of angiosperm fruit morphologies (e.g. Gu et al. 1998; Ferrándiz and Fourquin 2014; Karlova et al. 2014; Schilling et al. 2018; Liu et al. 2020).

Origins

The origin of MIKC^C-MADS-box genes lies in a duplication in the ancestral land plant, and while most bryophyte species have effectively only a single MIKC^C-MADS-box gene (Bowman et al. 2017; Koshimizu et al. 2018), the seeds of ABCE origins can be seen in the ancestral euphyllophyte (Fig. 3B) (Henschel et al. 2002; Kofuji et al. 2003; Wellmer et al. 2014; Qiu et al. 2023). Land plant MIKC^C-MADS proteins can form quartets in vitro, with a duplication of sequence within the K domain critical for forming the interface required for tetramer formation (Rümpler et al. 2023). Hence, quartet emergence predates the MADS-boxes learning their ABCs. The ability to form tetramers combined with expansion and diversification of MADS-box genes in the ancestral seed plant, and more dramatically within the ancestral angiosperm, enhanced the combinatorial power of MIKC^C complexes (Rümpler et al. 2023) and may have contributed to the evolution of tissue and organ types during development. In the ancestral land plant, sex was determined during the haploid generation (Bowman 2022), but in seed plants the sporophyte largely controls gametophyte development, including sex expression. In seedless vascular plants, MIKC^C-MADS genes are broadly expressed, often in both generations [e.g. (Hasebe et al. 1998; Ambrose et al. 2021)]. One intriguing speculation is that MIKC^C-MADS genes were co-opted into a role in determining sporophyte sex during tracheophyte evolution as the sporophyte was assuming control over gametophyte differentiation. And peering deeper in time, given the role of MCM1 in the yeast life cycle, an idea worth exploring is whether land plant MADS-box genes have retained vestiges of a fundamental role in the eukaryotic life cycle (Herskowitz 1989), as have land plant homeodomain genes (Dierschke et al. 2021; Hisanaga et al. 2021).

Conclusion: a short and biased selection of floral riddles

Thirty years ago, we could not have predicted what a cottage industry the characterization of ABC MADS-box genes in various angiosperms would become, contributing to a deeper and broader understanding of evo-devo processes that have shaped floral architecture. Reciprocally, they also accentuate remaining gaps in our knowledge. For instance, a given quartet is expected to activate different sets of genes at different timepoints as organ primordia initiate, grow, and differentiate, but whether this is the case and how it is achieved remain obscure. Equally, to generate the diversity of floral forms genes regulated by the ABC genes is likely to differ somewhat between species. However, a systematic examination of ABC targets conservation between species is still lacking. Even if a full picture of those targets becomes available, this still does not explain how gene activity translates into morphogenetic effects: here, high-resolution live-imaging coupled with computational simulations and theoretical modeling [see e.g. (Refahi et al. 2021)] could be instrumental in understanding how homeotic regulators interact with gene regulatory networks that specify organ polarity, floral symmetry, and tissue layer identity to sculpt undifferentiated primordia into floral organs with characteristic shape, size, patterned cell types, and metabolic properties. Research into the origins of the ABC genes established that the gene regulatory networks controlling flower development stem from recycling of an ancient circuit at work in the reproductive structures of last common ancestor of seed plants. However, what rewiring occurred along the lineage leading to flowering plants so that discreet B and B + C expression domains could be established enclosed by a sterile perianth remains enigmatic.

These puzzles represent only a handful of the remaining questions and by no means constitute an exhaustive list, but we hope they will spark the curiosity of future flower enthusiasts to continue the successful adventure of the ABC.

Author contributions

Both authors contributed to the outline and preparation of this review. JLB was the primary author for the History, Genetics, Molecules and Converting leaves into floral organs sections; EM was the primary author of the Swimming upstream and Floating downstream sections; both authors contributed to the Curious case and A and subfunctionalization of B- and C-classes, and the ABC through the evolutionary lens sections.

Funding

Work in J.L.B.'s laboratory is funded by the Australian Research Council (DP200100225, DP210101423, CE200100015). E.M. thanks the Gatsby Charitable Foundation and the Isaac Newton/Wellcome Trust ISSF for funding.

Dive Curated Terms

The following phenotypic, genotypic, and functional terms are of significance to the work described in this paper:

• AGL24 Gramene: AT4G24540

• AGL24 Araport: AT4G24540

• SEP3 Gramene: AT1G24260

• SEP3 Araport: AT1G24260

• SVP Gramene: AT2G22540

• SVP Araport: AT2G22540

• CHR11 Gramene: AT3G06400

• CHR11 Araport: AT3G06400

• AGL6 Gramene: AT2G45650

• AGL6 Araport: AT2G45650

• AP2 Gramene: AT4G36920

• AP2 Araport: AT4G36920

• WUS Gramene: AT2G17950

• WUS Araport: AT2G17950