KNOX genes: versatile regulators of plant development and diversity

The maize knotted1 (kn1) gene was isolated two decades ago through transposon tagging in a mutant with striking ‘knotted’ leaves, and revealed that the predicted gene product encoded a member of the homeodomain superfamily of transcriptional regulators (Fig. 1) (Vollbrecht et al., 1991). This discovery generated considerable excitement among developmental biologists, not least because the isolation of animal homeobox genes a few years earlier had revolutionized our understanding of the molecular basis of metazoan development and evolution. The cloning of the first member of a KNOX gene family in maize led to a similar explosion of new research in plant development (Hake et al., 1995). Today, a substantial body of information exists on the function of KNOX proteins in both model and non-model plants. These studies have revealed parallels with the mechanistic action of animal TALE (see Glossary, Box 1) homeodomain proteins, and have helped us to understand how these proteins influence plant development and to unravel key aspects of the logic that underpins cell fate allocation and tissue differentiation in the ‘green branch’ of the tree of life.

The discovery of KNOX genes provided the first molecular insights into the function of the shoot apical meristem (SAM; see Glossary, Box 1). The nuclear expression of KN1 protein in the maize shoot is confined to meristem cells and is excluded from leaf founder cells, thus providing the earliest marker for meristem versus lateral organ (see Glossary, Box 1) cell fate (Fig. 1C) (Smith et al., 1992). The SAM is established at the shoot pole during embryogenesis and harbours a stem cell population that allows continued organogenesis throughout the life of a plant (Fig. 1A). This continuous development contrasts with how animals develop and allows for plasticity in plant form, such that plants (which are sessile organisms) can readily modify their development in response to environmental cues. KN1 activity is required to prevent pluripotent cells in the maize SAM from adopting differentiated cell fates, as does the related protein SHOOT MERISTEMLESS (STM) in Arabidopsis, such that stm and kn1 loss-of-function mutants fail to establish and maintain a SAM (Fig. 1D; Fig. 3A,B) (Long et al., 1996; Vollbrecht et al., 2000).

KNOX genes comprise a small family of TALE homeobox genes that are found in all green plant lineages (Table 1) and fall into two subclasses on the basis of sequence similarity within the homeodomain, intron position, expression pattern and phylogenetic analysis (Kerstetter et al., 1994; Mukherjee et al., 2009). Class I KNOX genes are most similar to kn1 and are expressed in overlapping domains within the SAMs of both monocot and eudicot plants (see Glossary, Box 1) (Hake et al., 2004; Jackson et al., 1994). Class II KNOX genes form a monophyletic group that is distinct from Class I genes, and show diverse expression patterns and few known functions (Zhong et al., 2008). We therefore focus this review on Class I KNOX genes.

Although the first plant homeobox gene was cloned over 20 years ago, only recently are we beginning to understand how KNOX genes function in diverse developmental contexts, and how these functions relate to developmental transitions during land plant evolution. In this review, we discuss the functions of Class I KNOX proteins during development and the advances in understanding of KNOX gene regulatory networks, including upstream regulators, protein partners and downstream effectors of KNOX function. We highlight the discovery of a novel class of KNOX genes that lack a homeobox and insights into KN1 protein trafficking in plants (Box 2). We also discuss recent work in the unicellular green alga Chlamydomonas reinhardtii that suggests ancestral KNOX genes controlled diploid development and diversified in land plants in order to control multicellular body plans (Lee et al., 2008). Finally, we consider how evolutionary changes in KNOX gene regulation may have influenced plant diversity.

The Arabidopsis genome contains four Class I KNOX genes: STM, BREVIPEDICELLUS (BP), Kn1-like in Arabidopsis thaliana2 (KNAT2) and KNAT6 (Table 2). These genes participate in various developmental processes during the plant life cycle; however, their expression in specific domains of the SAM maintains the activity of the SAM and of its lateral organ and stem boundaries throughout development (Fig. 2). STM is the first KNOX gene expressed during early embryogenesis and its expression marks the entire SAM (Long et al., 1996). KNAT6 is expressed in the embryonic SAM once bilateral symmetry is established and later marks the SAM boundaries (Belles-Boix et al., 2006). BP shares aspects of both these gene expression patterns in the SAM during post-embryonic development (its embryonic expression marks the hypocotyl), whereas KNAT2 is expressed during embryogenesis and marks the base of the SAM (Byrne et al., 2002; Dockx et al., 1995). Genetic redundancy masks the contribution of KNAT6 and BP, which act, in addition to STM, to maintain SAM activity and organ separation (Belles-Boix et al., 2006; Byrne et al., 2002).

The architecture of bp mutants differs markedly from wild-type plants owing to their reduced height, irregularly shortened internodes and reduced apical dominance (see Glossary, Box 1; Fig. 3C,D) (Byrne et al., 2003; Douglas et al., 2002; Smith and Hake, 2003; Venglat et al., 2002). Plant architecture consists of repeating modules produced at the SAM called phytomers, each containing an internode, leaf and axillary meristem (Fig. 1A). In Arabidopsis inflorescences, growth of the leaf is suppressed such that only the floral meristem develops. BP expression at the boundary of the inflorescence SAM might, therefore, regulate cell allocation between floral primordia and internodes, explaining why the flowers are positioned aberrantly along bp mutant stems. Defective internode patterning is also observed in rice plants in which the related gene Oryza sativa homeobox15 (OSH15) is mutated (Sato et al., 1999). However, the flower stalks, or pedicels, are specifically shortened and curved downwards in bp mutants, owing to defects in cell division, cell elongation and cell differentiation (Douglas et al., 2002; Venglat et al., 2002), and it remains unclear which aspects of BP function are shared between this tissue and the SAM. KNAT6 and KNAT2 are part of a segmental chromosomal duplication in Arabidopsis, and single mutations in these genes do not affect shoot development (Belles-Boix et al., 2006; Byrne et al., 2002). However, KNAT6 and KNAT2 gene expression is expanded in bp mutants, and mutations in each of these genes show antagonistic genetic interactions with bp mutants, suggesting that the role of BP in inflorescence development is mediated at least in part by repression of KNAT6 and KNAT2 (Fig. 2) (Ragni et al., 2008).

In model organisms with simple leaves (see Glossary, Box 1), such as Arabidopsis, maize and tobacco, KNOX gene expression is confined to the shoot meristem and stem, and leaf development is dramatically altered by the ectopic expression of KNOX genes in leaves (Lincoln et al., 1994; Sinha et al., 1993; Vollbrecht et al., 1991). For example, transgenic expression of any Class I KNOX gene ectopically in the leaves of Arabidopsis can dramatically alter the simple leaf margin to produce a highly lobed shape (Fig. 3E,F) (Shani et al., 2009). These studies

suggested that ectopic KNOX gene expression within a simple leaf development programme produced differential growth at the leaf margin, owing to the creation of novel ‘meristem-leaf’ boundaries. The ectopic initiation of shoot meristems in transgenic tobacco and Arabidopsis leaves supported this idea (Chuck et al., 1996; Sinha et al., 1993). Strikingly, this correlation between the expression of KNOX genes in the leaf and a more complex leaf shape was shown to hold for tomato and many other plant species with naturally compound leaves (also known as dissected leaves; see Glossary, Box 1) (Bharathan et al., 2002; Hareven et al., 1996). It is now clear that KNOX gene expression is reactivated following leaf initiation in order to facilitate leaflet formation in both tomato and the Arabidopsis relative Cardamine hirsuta (Hay and Tsiantis, 2006; Shani et al., 2009). Gain of KNOX function in these species where KNOX genes are part of a compound leaf development programme can produce a striking reiteration of leaflets upon leaflets (see Fig. 3G,H) (Hareven et al., 1996; Hay and Tsiantis, 2006). Therefore, Class I KNOX genes have specific functions in compound leaf development that are distinct from their ability to induce shoot meristem formation.

KNOX proteins interact with another group of TALE proteins, the BEL1-like homeodomain family (BELL or BLH, see Table 2), in a highly connected, complex network that determines not only high-affinity KNOX target selection but also their subcellular localization (Bellaoui et al., 2001; Bhatt et al., 2004; Cole et al., 2006; Hackbusch et al., 2005; Smith et al., 2002). These interactions depend on the KNOX MEINOX domain, and each homeodomain of the two proteins binds to target DNA as a protein heterodimer (Box 3). Such regulatory interactions are reminiscent of those between different TALE proteins in animals, indicating that they might have an ancient origin. In Drosophila, for example, nuclear translocation of the MEIS-class TALE protein Homothorax depends on its interaction with the PBC-class TALE protein Extradenticle (see Glossary, Box 1) (Rieckhof et al., 1997). MEIS proteins in animals and KNOX proteins in plants share a MEINOX domain, which mediates their TALE protein interactions (Burglin, 1997). However, plants do not have easily recognizable PBC proteins, and this role of selectively interacting with KNOX proteins to regulate their nuclear translocation and target affinity is fulfilled by BELL proteins (Bellaoui et al., 2001).

This protein interaction network also includes Arabidopsis thaliana OVATE family proteins (AtOFP), which negatively control the activity of TALE protein dimers by causing their relocalization from the nucleus, where they are functional, to the cytoplasmic space (Box 3) (Hackbusch et al., 2005). Genetic evidence for AtOFP regulation of KNOX-BELL activity in Arabidopsis embryo sac development came recently from an analysis of a gain-of-function BLH1 allele called eostre (Pagnussat et al., 2007). The eostre phenotype depends on the Class II KNOX gene KNAT3 and is partially phenocopied by loss of AtOFP5 function, suggesting that repression of KNAT3-BLH1 dimer activity by AtOFP5 is essential for embryo sac development.

TALE protein interactions are selective between specific members of the KNOX and BELL protein families, and these interactions are required for high-affinity DNA binding (Smith et al., 2002). Different combinations of KNOX/BELL transcription factors may, therefore, regulate different downstream genes. Genetic analyses in Arabidopsis indicate that the formation of different heterodimers of STM or BP with the BELL protein

BELLRINGER (BLR) plays a role in maintaining the SAM and in patterning the inflorescence and fruit (Fig. 2) (Byrne et al., 2003; Roeder et al., 2003; Smith and Hake, 2003). Additional STM-BELL dimers are likely to promote SAM activity, as triple mutants of the BELL genes BLR, POUND-FOOLISH, ARABIDOPSIS THALIANA HOMEOBOX1 are shootless (Rutjens et al., 2009). This phenocopying of stm mutants could indicate that these BELL proteins regulate the nuclear localization of STM in a redundant manner (Box 3).

Establishing a boundary between KNOX-expressing and non-expressing cells in the SAM is crucial for initiating lateral organs, whereas KNOX gene expression outside the SAM disrupts lateral organ development, causing transformations of cell fate and organ shape (Hake et al., 2004). Establishing such a boundary in the SAM, therefore, requires the repression of KNOX gene expression in leaf founder cells, and maintaining this repression regulates leaf development in many plants, such as in maize and Arabidopsis. Negative regulators of KNOX genes have been identified, mostly through studying recessive mutants that phenocopy the effects of KNOX overexpression in the leaf (Fig. 4). These regulatory pathways all act to maintain the repression of KNOX genes during lateral organ development and suggest that as yet unidentified processes might control the repression of KNOX activity in founder cells.

The first upstream regulator of KNOX genes was identified from a recessive maize mutant rough sheath2 (rs2), which has leaves that resemble those of dominant Knox mutants (Schneeberger et al., 1998). rs2 encodes a myb-type protein that represses KNOX expression in the lateral organs of several plant species with both simple leaves (such as maize, Arabidopsis and snapdragon) and compound leaves (such as pea and the Arabidopsis relative C. hirsuta). Collectively, these are called ARP proteins [ASYMMETRIC LEAVES1 (AS1) in Arabidopsis, Rough sheath2 in maize, PHANTASTICA in snapdragon] (Byrne et al., 2000; Hay and Tsiantis, 2006; Tattersall et al., 2005; Timmermans et al., 1999; Tsiantis et al., 1999). In these plants, ARP and KNOX gene expression distinguishes leaf founder cell from meristem cell fate in the shoot apex. Although mutations in ARP genes cause ectopic KNOX expression in lateral organs, the boundary between KNOX-expressing and non-expressing cells in the SAM is maintained (Schneeberger et al., 1998). Expression of the KNOX genes BP, KNAT2 and KNAT6 is reactivated in leaf primordia of Arabidopsis as1 mutants, producing asymmetric leaves with lobed margins and occasional shoot meristems (Byrne et al., 2000; Ori et al., 2000). Transgenic expression of each of these KNOX genes is sufficient to elicit similar alterations in Arabidopsis leaf patterning (Shani et al., 2009), and loss of all

The mechanistic basis for this regulation in Arabidopsis is beginning to shed light on how ARP proteins maintain a KNOX repressive state. AS1 forms a heterodimer with the LATERAL ORGAN BOUNDARIES (LOB) domain protein ASYMMETRIC LEAVES2 (AS2) in the adaxial (see Glossary, Box 1) domain of the Arabidopsis leaf, and genetic epistasis indicates that these genes function together (Ori et al., 2000; Xu et al., 2003). The BTB-POZ domain BLADE-ON-PETIOLE proteins (BOP1 and BOP2) are active specifically at the adaxial base of lateral organs, where BOP1 directly activates AS2 transcription (Jun et al., 2010). This AS1-AS2 protein dimer directly binds the promoters of BP and KNAT2, possibly as a repressive chromatin complex through recruitment of the histone chaperone histone regulatory protein A (HIRA) (Fig. 4) (Guo et al., 2008; Phelps-Durr et al., 2005).

Consistent with an epigenetic mode of KNOX gene repression, a second pathway confining KNOX activity to meristems requires the polycomb-group proteins CURLY LEAF (CLF), SWINGER (SWN) and FERTILISATION INDEPENDENT ENDOSPERM (Katz et al., 2004; Schubert et al., 2006). The SET domains of CLF and SWN have histone methyltransferase activity, which facilitates the maintenance of silenced target gene expression states through cell divisions. Recent work has shown how the combined actions of two polycomb repressive complexes (PRCs) repress KNOX transcription (Xu and Shen, 2008). CLF-containing PRC2 marks KNOX genes by trimethylation of histone H3 at lysine 27 (H3K27me3) (Schubert et al., 2006). A PRC1-like complex, which contains the chromodomain protein LIKE HETEROCHROMATIN PROTEIN1 and two RING domain proteins, then binds to these methylated histones, resulting in the transcriptional repression of KNOX genes (Fig. 4) (Xu and Shen, 2008).

YABBY (YAB) proteins define a seed plant-specific family that promotes lamina outgrowth (see Glossary, Box 1) and represses KNOX expression in Arabidopsis leaves (Fig. 4). YAB gene expression is confined to lateral organs, and the cumulative loss of YAB function results in the progressive loss of lamina growth and in the reactivation of STM, BP and KNAT2 expression in fil;yab3 leaves (Kumaran et al., 2002). These leaves produce ectopic shoot meristems, associated with KNOX expression; however, similar to as1 mutants, the repression of KNOX genes is maintained in leaf founder cells. Genetic evidence does suggest, however, that regulating this boundary of KNOX-expressing and non-expressing cells in the SAM involves antagonism between meristem determinants, such as STM, and lateral organ determinants such as YAB, AS1 and AS2 genes. Genetic interactions result in the suppression of stm mutants by fil;yab3 or by as1 or as2, indicating that cells that lack STM in the SAM boundary region are consumed by lateral organ production in the absence of YAB, AS1 or AS2 function (Byrne et al., 2000; Kumaran et al., 2002). Although it is not yet clear what component of YAB activity is mediated by KNOX repression, these observations strengthen the notion that restricting KNOX expression to the SAM is a prerequisite for normal leaf development in Arabidopsis.

Another mechanistic link between KNOX downregulation and sites of lateral organ initiation involves repression of KNOX expression by concentration maxima of the phytohormone auxin (see Glossary, Box 1) in the periphery of the SAM (Fig. 4). These auxin maxima are established by action of the PINFORMED1 (PIN1) auxin efflux transporter in Arabidopsis (Benkova et al., 2003; Reinhardt et al., 2003). PIN1 and KNOX transcripts are expressed in mutually exclusive domains in the SAM, and compromising PIN1 activity or auxin signaling results in ectopic expression of the KNOX gene BP in Arabidopsis leaves (Hay et al., 2006). Loss of BP function can partially rescue the loss of lateral organs in pin1 mutants, suggesting that failure to repress BP expression in the boundary of the SAM in pin1 mutants could antagonize lateral organ formation (Hay et al., 2006). Furthermore, maize apices grown in the presence of an auxin transport inhibitor fail to repress KNOX genes or to initiate leaves at the shoot apex (Scanlon, 2003). Thus, it will be important to understand whether repression of KNOX gene expression in leaf founder cells is a direct read-out of auxin signaling. One possibility is that AUXIN RESPONSE FACTORs (ARFs) repress KNOX transcription because ectopic KNOX expression causes floral organ defects in arf6 arf8 double mutants (Tabata et al., 2010).

Despite mounting evidence for antagonism between KNOX and auxin activities, the proportion of auxin action in lateral organ formation that depends on KNOX repression at the SAM boundary is unclear. The boundary regulator CUP-SHAPED COTYLEDON2 (CUC2) initially activates STM expression and later overlaps with STM in a region of repressed growth and low auxin activity at the SAM boundary (Fig. 4) (Heisler et al., 2005). Failure to initiate cotyledons in pin1;pinoid embryos depends on STM and CUC1,2 function, and eliminating STM activity in pin1;pid;stm triple mutants recovers cotyledons (Furutani et al., 2004). The similarity of these genetic interactions between KNOX and auxin transport-related genes in the embryo, vegetative and reproductive shoots indicates that antagonism between auxin and KNOX genes might operate in multiple contexts throughout Arabidopsis development to promote organ initiation and the associated elaboration of organ boundaries (Fig. 2) (Furutani et al., 2004; Hay et al., 2006). A role for combined regulation of KNOX and auxin activities in boundary delimitation was also suggested by analysis of the boundary-expressed LOB domain gene JAGGED LATERAL ORGANS (JLO), which is sufficient to activate STM and BP expression and repress PIN auxin efflux transporters (Fig. 4) (Borghi et al., 2007).

This progress highlights two avenues of future research. First, isolation of higher-order KNOX-repressive complexes and analysis of their dynamic nature during development and the cell cycle should help to explain how these repressive pathways are integrated. Second, the precise phenotypic relevance of KNOX repression by different pathways will need to be resolved in order to understand which KNOX-repressive pathways have a clear contribution to cell-fate decisions and growth regulation. For example, although the mechanistic basis of PRC-mediated KNOX silencing is being clarified, it is unclear to what extent the mutant phenotypes of such chromatin repressors are attributable to misregulation of KNOX versus other target genes.

The few downstream targets of KNOX transcription factors identified thus far are beginning to shed light on how these proteins influence the balance between differentiation and replenishment of cells in the SAM. KNOX targets modulate the abundance of gibberellins (GAs) and cytokinins (CKs) (see Glossary, Box 1), and of the secondary cell wall polymer lignin (Fig. 4). CKs activate cell division, whereas GAs promote cell elongation, which is a cell differentiation process – as is lignin deposition – and KNOX proteins regulate the balance between GAs and CKs in meristem and leaf founder cells. In the SAM, KNOX proteins raise CK levels by activating the transcription of ISOPENTENYL TRANSFEREASE7 (IPT7) and lower GA levels by directly inhibiting the biosynthetic gene GA 20-oxidase1 (GA20ox1) and activating the catabolic gene GA 2-oxidase1 (GA2ox1) (Bolduc and Hake, 2009; Jasinski et al., 2005; Sakamoto et al., 2001; Yanai et al., 2005). Genetic analyses demonstrate that this high CK:low GA ratio is important for KNOX function – to prevent cell differentiation and thus maintain pluripotent cell fate in the SAM. For example, stm mutant phenotypes are suppressed by elevating CK biosynthesis and are enhanced by either elevating GA activity or by reducing CK biosynthesis, and ultimately the combination of elevated GAs and reduced CKs is sufficient to phenocopy stm mutant phenotypes in wild-type Arabidopsis (Jasinski et al., 2005; Yanai et al., 2005). Conversely, the absence of KNOX expression in lateral organ founder cells likely correlates with a low CK:high GA ratio and with the competence to differentiate (Jasinski et al., 2005; Yanai et al., 2005).

One implication of the fact that modulating GA and CK pathways is largely sufficient to account for STM function in the SAM is that only a few target genes might mediate the function of this protein. This mirrors the scenario in Drosophila, in which HOX proteins have been shown to exert their function in some contexts by regulating only a few crucial targets (Lovegrove et al., 2006). Genome-wide target identification is required to evaluate this hypothesis and should help to reveal how distinct KNOX proteins exert their effects in diverse developmental contexts.

How these hormone signals are integrated with developmental processes operating in the SAM is less clear. For example, STM not only implements a high CK:low GA regime but also prevents cell differentiation by repressing leaf determinants such as AS1 and AS2 (Byrne et al., 2000; Byrne et al., 2002). The mechanistic basis for this repression is unlikely to be direct, so it will be interesting to know whether these different aspects of STM activity are independent or whether AS1 and AS2 respond to, and influence, hormone activity.

As land plants evolved from green algae, the dominant phase of the life cycle changed from a haploid gametophyte to a diploid sporophyte (see Glossary, Box 1; Fig. 5). In seed plants, the haploid stage of the life cycle is reduced to few-celled gametophytes that produce male and female gametes, while morphological diversity resides in the sporophyte. However, land plants evolved from a group of green algae in which the diploid stage of the life cycle comprised merely the unicellular zygote. Mechanisms that control multicellular development in land plants, therefore, either evolved from genes directing unicellular development in the sporophyte or were recruited from existing genetic toolkits that controlled multicellular morphology in the gametophyte. Evidence exists for the recruitment of gametophyte genes into the sporophyte (Menand et al., 2007); however, genetic analyses in green algae and basal land plants suggest that ancestral KNOX genes controlled diploid development and diversified in land plants to control multicellular body plans.

A leafy shoot develops in the gametophyte of the moss Physcomitrella patens – a model organism for basal land plants (shown in Fig. 5). Given the role of KNOX genes in shoot development, it was surprising to find that Class I and II KNOX genes are neither expressed nor functional in the moss gametophyte, acting instead during sporophyte development (Sakakibara et al., 2008; Singer and Ashton, 2007). Genetic control of the shoot system in haploid moss and in diploid flowering plants might, therefore, represent convergent evolution (see Glossary, Box 1). This suggests that the regulatory relationships between KNOX transcription factors and their downstream effectors in higher plants could be later innovations that were absent in basal land plant lineages. Indeed, P. patens KNOX deletion mutants show no alteration in IPT, GA20ox or GA2ox gene expression, suggesting that KNOX proteins in moss do not regulate these hormone pathways (Sakakibara et al., 2008). This contrasts with the biochemical activities of KNOX proteins, which might be conserved across land plants despite target gene divergence. For example, KNOX genes from both moss and the fern Ceratopteris richardii, where expression is also restricted to the sporophyte, elicit a similar spectrum of phenotypes to Arabidopsis KNOX genes when overexpressed in Arabidopsis (Sakakibara et al., 2008; Sano et al., 2005). These experiments highlight potential pitfalls for interpreting gene transfer experiments between different developmental contexts across very distant lineages if genetic tools are not available in both organisms to aid the understanding of endogenous gene function. Studies to identify KNOX target genes in moss will provide an interesting future direction for investigating the extent of divergence and conservation of the KNOX pathway, and how this relates to morphological transitions during land plant evolution.

If KNOX proteins have only diploid-specific functions, does this relate back to a history in regulating diploid development in algal ancestors? The discovery of interacting TALE proteins that are necessary and sufficient for zygote formation in the unicellular green alga C. reinhardtii suggests that this could be the case (shown in Fig. 5). Two haploid mating types in C. reinhardtii each contribute a different TALE protein, which interact and translocate to the nucleus in a diploid zygote – one is a Class II KNOX protein, the other a BELL protein (Lee et al., 2008). This BELL gene belongs to a lineage lost in land plants, despite an overall expansion of TALE genes during land plant evolution, which potentially allowed new interactions between BELL and KNOX proteins (see Table 2). An attractive correlation therefore arises between the diversification of KNOX/BELL genes and morphological diversity of plant sporophytes. For example, new TALE heterodimers in ancestral green plant lineages might have modified zygote behavior to allow a series of mitoses that produced a multicellular sporophyte. Exploiting the regulatory complexity created by multiple KNOX-BELL interactions could have facilitated the explosion of body plan complexity, which occurred as plants colonized terrestrial ecosystems. This hypothesis predicts that loss of TALE activity in land plants might condition zygotic defects, reflecting conservation of ancestral TALE gene functions to promote zygote formation. However, KNOX expression has not been detected in the Arabidopsis zygote – in fact, STM first acts at the globular stage of embryo development when the sporophyte comprises 32 cells (Long et al., 1996). Further studies are required, therefore, to map diversification of KNOX and BELL gene families to major morphological transitions during land plant evolution.

Divergent Class I KNOX expression among species closely correlates with leaf shape such that in simple-leafed species, such as A. thaliana and maize, KNOX expression is confined to the meristem while KNOX proteins accumulate in compound leaves of many species (Bharathan et al., 2002). KNOX expression in leaves was independently recruited to control compound leaf development in multiple seed plant lineages, although legume species (including pea) are notable exceptions (Bharathan et al., 2002; Hofer et al., 2001) in which compound leaf development is independent of KNOX function. Ten years after KNOX overexpression was first shown to increase tomato leaf complexity, genetic evidence was provided that KNOX activity is necessary and sufficient for leaflet formation in C. hirsuta (Hay and Tsiantis, 2006). More recently, activity of the tomato Class I KNOX protein Tomato knotted 2 (Tkn2) was perturbed by means of translational fusion to a 12 amino acid EAR repressor motif of the Arabidopsis SUPERMAN gene termed SDRX, which confirmed a role for Tkn2 in tomato leaflet initiation (Shani et al., 2009).

These studies raise the question of what processes are responsible for KNOX genes being expressed in leaves of some species but not others. Parallel genetic studies in the closely related species Arabidopsis and C. hirsuta offer a useful approach for investigating the mechanistic basis for such evolutionary diversification in KNOX gene expression. KNOX promoter swaps between Arabidopsis and C. hirsuta showed that STM and BP promoters directed reporter expression in a pattern characteristic of the native species (Hay and Tsiantis, 2006). Thus, differences in the cis-regulation of KNOX genes contribute to species-specific leaf shapes. It will, therefore, be interesting to identify the precise KNOX gene regulatory elements responsible for leaf expression in C. hirsuta and other species. Phylogenetic foot-printing of STM regulatory sequences, sampled from species with both simple and dissected leaves, identified the K-box as a conserved regulatory element required for repression of STM in the simple leaves of A. thaliana and tobacco (Uchida et al., 2007). Notably, this defined the first cis-regulatory element to mediate KNOX repression. Conservation of this K-box between A. thaliana and C. hirsuta, however, makes the precise significance of this element for elaboration of divergent leaf morphologies unclear.

The importance of cis-regulatory changes in KNOX loci for transitions in leaf form follows the logic that mutations in the cis-regulatory regions of developmental patterning genes are likely to underlie most of phenotypic evolution (Carroll, 2005). Extending the comparative genetic approach used between C. hirsuta and Arabidopsis to tomato relatives, in which KNOX regulation of leaf form evolved independently, should provide a broader picture of cis-regulatory evolution and should allow the individual sequence elements that control KNOX expression and leaf form to be functionally dissected.

Changes in KNOX gene expression between species could also arise by diversification of repressors, such as the ARP proteins. However, AS1 expression and function is conserved between Arabidopsis and C. hirsuta, such that, while AS1 prevents KNOX expression in Arabidopsis leaves, its repressive function defines the correct domain of BP expression within the context of the C. hirsuta leaf (Hay and Tsiantis, 2006). Transgenic complementation of Arabidopsis as1 mutants by the gene SkARP1 from the lycophyte Selaginella kraussiana suggests that the ability to repress KNOX expression might be an ARP function that is ancestral to vascular plants (Harrison et al., 2005). Isolation of KNOX repressors from C. hirsuta and tomato should help clarify to what degree novel versus conserved repressors of KNOX genes were used during evolution to produce compound leaves. For example, the CLAUSA and TRIPINNATE gene products repress KNOX transcripts in the adaxial domain of tomato leaves and regulate leaflet number (Jasinski et al., 2007). Thus, it is tempting to speculate that many avenues exist via which evolutionary change in either the trans-regulatory landscape or cis-regulatory elements can reconfigure KNOX expression. Modularity in KNOX regulation is also likely to circumvent the potentially pleiotropic effects of KNOX gene expression in lateral organs. Thus, a challenge for the future is to understand how evolutionary tinkering with the KNOX gene regulatory network has produced morphological diversity in different lineages.

How does our understanding of KNOX gene function in meristem development relate to the formation and delimitation of leaflets in compound leaf development? Recent work suggests that this process reflects the function of different KNOX developmental modules that control primordium initiation, boundary formation and meristem maintenance in the context of both pluripotent cells of the SAM and in differentiating cells of the compound leaf.

Leaflet formation from the leaf margin in C. hirsuta and tomato unfolds through a series of events similar to that of leaf formation at the periphery of the SAM (Barkoulas et al., 2008; Koenig et al., 2009). For example, PIN1 action facilitates the generation of auxin maxima sequentially along the leaf margin at sites where KNOX expression is repressed. These auxin maxima recruit cell-division centres at the leaf margin, resulting in outgrowth of individual lateral leaflets in the context of repressed growth elsewhere on the rachis. On the one hand, KNOX overexpression causes ectopic auxin maxima within leaflets, resulting in ectopic leaflet formation; on the other hand, exogenous auxin application can repress KNOX expression in the leaf (Barkoulas et al., 2008). These data suggest the following sequence of events: KNOX expression facilitates the formation of auxin maxima, which feed back to repress KNOX expression, thus allowing leaflet outgrowth. Whether such a feedback between KNOX/PIN1/auxin operates both in a compound leaf and in the context of the SAM, and its mechanistic basis, remains an interesting subject for future research.

Genetic analysis in four distantly related eudicot species showed that CUC gene expression defines the distal boundary of leaflet initiation on the leaf rachis and is required for leaflet formation (Berger et al., 2009; Blein et al., 2008). CUC activity promotes KNOX expression within the leaf and vice versa, forming a positive-feedback loop that also functions during SAM and lateral organ boundary development (Blein et al., 2008; Takada et al., 2001).

KNOX activity ensures that low GA levels are maintained in the SAM, which favours meristem activity over cell differentiation. This KNOX-GA module also operates in tomato leaves where expression of the GA biosynthetic gene LeGA20ox1 is repressed in response to Tkn2 overexpression, and the effects of Tkn2 overexpression are suppressed by constitutive GA signaling (Hay et al., 2002; Jasinski et al., 2008). Thus, antagonism between KNOX and GA activities in the tomato leaf regulates the correct pattern of leaflet formation.

In each of these regulatory modules, KNOX activity appears to influence the duration and direction of growth, while repressing tissue differentiation. Therefore, detailed, quantitative descriptions of growth will be useful in order to understand the role of KNOX proteins in different contexts. This may help to tease apart the degree to which genetic interactions, described above, reflect the integration of KNOX activity in specific modules versus the indirect consequences of KNOX expression on cellular differentiation. KNOX modules probably regulate meristem activity across all seed plants, and the repeated deployment of these modules to regulate compound leaf morphology is likely to reflect evolutionary parallelism (see Scotland, 2010). In some instances, however, the regulation of different morphologies by KNOX modules may reflect underlying homologies between structures. For example, tuber development in potato is regulated in part by an antagonistic interaction between KNOX and GA, probably reflecting the fact that tubers are modified shoots (Chen et al., 2004).

The action of KNOX proteins in these growth-regulating modules also provides a unified framework with which to interpret other striking morphologies caused by novel patterns of KNOX expression. For example, ectopic expression of the kn1 orthologue in barley produces Hooded florets, a trait selected in some barley varieties in which the awn (see Glossary, Box 1) is transformed into reiterative inflorescence axes (Müller et al., 1995; Williams-Carrier et al., 1997). Similarly, ectopic KNOX expression in two snapdragon mutants Hirzina and Invaginata produces novel structures that resemble petal spurs (see Glossary, Box 1) (Golz et al., 2002). In both examples, ectopic KNOX expression leads to the formation of new tissue-organizing centres that can direct altered cell division and growth, which might reflect redeployment of KNOX modules in novel contexts. Whether selection has fixed evolutionary changes in KNOX expression that cause divergent morphologies in nature is an exciting issue that can be approached using population genetics methodologies empowered by an increasing wealth of genomic resources available in multiple taxa.

The diverse effects of ectopic KNOX expression in lateral organs also highlight the fact that KNOX gene action is highly context and dose dependent. For example, dominant Mouse ear (Me) and Curl (Cu) mutations cause aberrant transcription of the tomato Tkn2 gene, yet Me leaves vary between highly ramified vegetative leaves and bladeless reproductive leaves, while the growth of Cu leaves is severely arrested (Parnis et al., 1997). Variation in the expression domain of ARP repressors during development may influence these phenotypes; however, ARP expression has been reported to be both absent and present in the SAM of tomato so this issue requires further investigation (Kim et al., 2003; Pien et al., 2001).

Inducible KN1 expression in Arabidopsis has provided some insight into such phenotypic variation, as increased KN1 dose produces a qualitative increase in leaf phenotype from mild indentations at the leaf margin, to highly lobed margins, to strong growth arrest (Hay et al., 2003). This study also defined a window of competence during Arabidopsis leaf development in which only proliferating, but not differentiating, tissues could respond to KN1 activity. A recent study elegantly demonstrated the importance of developmental timing for the phenotypic output of KNOX overexpression in Arabidopsis and tomato leaves, which we further discuss in the next section (Shani et al., 2009). This work raises the issue of what the molecular basis of a differential competence to respond to KNOX expression could be.

Genetic analyses in Arabidopsis and tomato have identified two such mechanisms that influence the competence of shoot tissue to respond to KNOX activity. Studies in Arabidopsis suggest that KNOX-independent regulation of common target genes might modulate the competence of leaf tissue to respond to ectopic KNOX expression. For example, as1 and as2 mutant phenotypes are enhanced by mutations in PICKLE (PKL) and SERRATE (SE) genes, which encode a chromatin-remodeling protein and a miRNA biogenesis component, respectively [(Ori et al., 2000) and references therein]. KNOX genes are regulated normally in both pkl and se mutant leaves; however, expression of the KNOX target gene GA20ox1 is repressed in the leaves of both mutants, indicating that PKL, SE and KNOX activities converge on at least one common target gene (Grigg et al., 2005). In the case of SE, this convergence could reflect its repression of a small family of microRNA-targeted Class III HD-ZIP genes, which promote meristem activity in common with KNOX genes (Grigg et al., 2005; McConnell et al., 2001). For example, se mutants showed heightened responses to KNOX activity that reflected elevated levels of HD-ZIP III expression, and HD-ZIP III gain-of-function mutants had reduced GA20ox1 expression, thus mimicking the effects of KNOX overexpression (Grigg et al., 2005).

A second mechanism highlights the regulation of tissue differentiation as a key determinant for tissue response to KNOX gene expression. The Lanceolate (La) mutation in tomato produces miRNA-insensitive TEOSINTE BRANCHED1/CYCLOIDEA/PROLIFERATING CELL FACTOR 4 (TCP4) transcripts during leaf development, which cause precocious cell differentiation and prevent leaflet formation (Ori et al., 2007). KNOX expression is unaffected in La mutants, but KNOX gain-of-function effects are suppressed, suggesting that TCP-mediated control of cellular differentiation affects the competency of leaf tissue to respond to KNOX activity. Investigating the molecular underpinnings of this regulation and understanding how TCP and KNOX activities are coordinated during compound leaf development are interesting avenues for future research.

Almost 20 years ago, an influential review article described a conceptual model for maize leaf development based on the perturbations caused mainly by dominant KNOX mutants (Freeling, 1992). Freeling hypothesized that a leaf primordium progresses through a series of defined developmental stages to form distinct tissue types – sheath, ligule (see Glossary, Box 1), auricle and blade – along the proximal-to-distal axis of the maize leaf. He termed this developmental trajectory a maturation schedule. Ectopic expression of KNOX genes in dominant mutants caused tissue transformations that he interpreted as developmental delays, whereby cells failed to progress through subsequent phases of the maturation schedule (Muehlbauer et al., 1997). For example, ectopic kn1 expression along the lateral veins in Kn1 mutants retarded this tissue development such that ligule differentiated in place of blade (Fig. 1B). Freeling's view that ectopic KNOX activity influenced the temporal progression of a leaf cell through developmental states has provided a valuable framework for conceptualizing compound leaf development, where KNOX proteins are active in endogenous, rather than ectopic, contexts.

Leaf development can be divided into three stages termed leaf initiation (I), primary morphogenesis (PM) and secondary morphogenesis (SM) (Fig. 6). The analysis of transcriptome dynamics during Arabidopsis leaf maturation described a sequential schedule of developmental steps operating during these stages and correlated expression signatures with each stage (Efroni et al., 2008). Using the promoters of such stage-specific genes to drive KNOX expression in the tomato leaf suggested that KNOX proteins act to prolong PM, thus allowing leaflet initiation (Fig. 6A) (Shani et al., 2009). Cell populations that occupy distinct positions along the maturation schedule may respond differently to KNOX expression, partly owing to their proliferative state and partly owing to receiving different tissue maturation signals from surrounding cells as the primordium grows. For example, KNOX overexpression during PM, but not SM, in tomato caused reiterative leaflet initiation (Shani et al., 2009). Here, KNOX activity retained cells in a divisive state for longer periods of time, permitting prolonged tissue responses to signals that promote leaflet formation (Fig. 6B). Reduction of KNOX activity during I in tomato, by expressing a KNOX-SRDX gene fusion, prevented leaflet initiation (Shani et al., 2009). In this case, reduced KNOX activity resulted in faster cessation of cell division, abbreviating PM to result in precocious tissue maturation, as observed in C. hirsuta STM RNAi leaves (Fig. 6C) (Hay and Tsiantis, 2006). Leaflet initiation was also blocked for different reasons by KNOX overexpression in tomato during I (Shani et al., 2009). Here, KNOX activity retained cells in a state in which they are not yet competent to respond to leaflet-promoting signals.

It is tempting to speculate that KNOX proteins retard progression from divisional growth to differentiation programmes by elevating CK and repressing GA levels, hence ultimately influencing the timing at which cells exit the cell cycle (Fig. 6A). This idea is consistent with observations in tomato where elevated GA signaling antagonizes KNOX action and results in faster growth rates during PM and in consequent reductions in leaflet number (Jasinski et al., 2008). It will be important to understand how KNOX activity in compound leaves integrates into the pathways that are shared by all leaves, which control the progression of a plant along a maturation schedule, in order to disentangle the influence of KNOX action on cell fate, tissue maturation and the direction and duration of cellular growth.

Twenty years on from cloning the first KNOX gene, we have a robust framework within which to understand how KNOX proteins control plant development. However, important gaps in our knowledge remain. First, the identification of genome-wide targets of distinct Class I KNOX proteins and their co-factors is required to understand the basis of context-specific KNOX activity in different tissues. Second, the identification of those targets that are able to account for the bulk of function of individual proteins will require in depth genetic analyses to overcome redundancy. Third, the implementation of quantitative and dynamic phenotyping frameworks to analyse KNOX function will help to explain how these proteins control growth and differentiation. In this context, it will be useful to devise in vivo assays to quantify KNOX contributions to distinct developmental processes and to exploit computational models to understand the logic of increasingly complex genetic circuitries. Finally, the development of genetic tools in diverse plant taxa will be needed to fully understand the role of KNOX genes in morphological evolution. This, in turn, will help to elucidate how the balance of conservation versus divergence in a fundamental developmental pathway sculpts plant form.

We thank S. Hake, Y. Yasumura and C. Canales for kindly providing images, and R. Scotland and D. Bailey for discussions. M.T. received support for this work from the Gatsby Charitable Foundation and Biotechnology and Biological Sciences Research Council. M.T. is a recipient of a Royal Society Wolfson Merit Award and A.H. is a recipient of a Royal Society University Research Fellowship.