The tubulin code: Molecular components,readout mechanisms,and functions

Microtubules are cytoskeletal filaments that are dynamically assembled from α/β-tubulin heterodimers. The primary sequence and structure of the tubulin proteins and, consequently, the properties and architecture of microtubules are highly conserved in eukaryotes. Despite this conservation, tubulin is subject to heterogeneity that is generated in two ways: by the expression of different tubulin isotypes and by posttranslational modifications (PTMs). Identifying the mechanisms that generate and control tubulin heterogeneity and how this heterogeneity affects microtubule function are long-standing goals in the field. Recent work on tubulin PTMs has shed light on how these modifications could contribute to a “tubulin code” that coordinates the complex functions of microtubules in cells.

Introduction

Microtubules are key elements of the eukaryotic cytoskeleton that dynamically assemble from heterodimers of α- and β-tubulin. The structure of microtubules, as well as the protein sequences of α- and β-tubulin, is highly conserved in evolution, and consequently, microtubules look alike in almost all species. Despite the high level of conservation, microtubules adapt to a large variety of cellular functions. This adaptation can be mediated by a large panel of microtubule-associated proteins (MAPs), including molecular motors, as well as by mechanisms that directly modify the microtubules, thus either changing their biophysical properties or attracting subsets of MAPs that convey specific functions to the modified microtubules. Two different mechanism can generate microtubule diversity: the expression of different α- and β-tubulin genes, referred to as tubulin isotypes, and the generation of posttranslational modifications (PTMs) on α- and β-tubulin (Figs. 1 and ​and2).2). Although known for several decades, deciphering how tubulin heterogeneity controls microtubule functions is still largely unchartered. This review summarizes the current advances in the field and discusses new concepts arising.Open in a separate windowFigure 1.Tubulin heterogeneity generated by PTMs. (A) Schematic representation of the distribution of different PTMs of tubulin on the α/β-tubulin dimer with respect to their position in the microtubule lattice. Acetylation (Ac), phosphorylation (P), and polyamination (Am) are found within the tubulin bodies that assemble into the microtubule lattice, whereas polyglutamylation, polyglycylation, detyrosination, and C-terminal deglutamylation take place within the C-terminal tubulin tails that project away from the lattice surface. The tubulin dimer represents TubA1A and TubB2B (Fig. 2), and modification sites for polyglutamylation and polyglycylation have been randomly chosen. (B) Chemical structure of the branched peptide formed by polyglutamylation and polyglycylation, using the γ-carboxyl groups of the modified glutamate residues as acceptor sites for the isopeptide bonds. Note that in the case of polyglutamylation, the elongation of the side chains generates classical peptide bonds (Redeker et al., 1991).Open in a separate windowFigure 2.Heterogeneity of C-terminal tails of tubulin isotypes and their PTMs. The amino acid sequences of all tubulin genes found in the human genome are indicated, starting at the last amino acid of the folded tubulin bodies. Amino acids are represented in single-letter codes and color coded according to their biochemical properties. Known sites for polyglutamylation are indicated (Eddé et al., 1990; Alexander et al., 1991; Rüdiger et al., 1992). Potential modification sites (all glutamate residues) are indicated. Known C-terminal truncation reactions of α/β-tubulin (tub) are indicated. The C-terminal tails of the yeast Saccharomyces cerevisiae are shown to illustrate the phylogenetic diversity of these domains.

Tubulin isotypes

The cloning of the first tubulin genes in the late 1970’s (Cleveland et al., 1978) revealed the existence of multiple genes coding for α- or β-tubulin (Ludueña and Banerjee, 2008) that generate subtle differences in their amino acid sequences, particularly in the C-terminal tails (Fig. 2). It was assumed that tubulin isotypes, as they were named, assemble into discrete microtubule species that carry out unique functions. This conclusion was reinforced by the observation that some isotypes are specifically expressed in specialized cells and tissues and that isotype expression changes during development (Lewis et al., 1985; Denoulet et al., 1986). These high expectations were mitigated by a subsequent study showing that all tubulin isotypes freely copolymerize into heterogeneous microtubules (Lewis et al., 1987). To date, only highly specialized microtubules, such as ciliary axonemes (Renthal et al., 1993; Raff et al., 2008), neuronal microtubules (Denoulet et al., 1986; Joshi and Cleveland, 1989), and microtubules of the marginal band of platelets (Wang et al., 1986; Schwer et al., 2001) are known to depend on some specific (β) tubulin isotypes, whereas the function of most other microtubules appears to be independent of their isotype composition.More recently, a large number of mutations in single tubulin isotypes have been linked to deleterious neurodevelopmental disorders (Keays et al., 2007; Fallet-Bianco et al., 2008; Tischfield et al., 2010; Cederquist et al., 2012; Niwa et al., 2013). Mutations of a single tubulin isotype could lead to an imbalance in the levels of tubulins as a result of a lack of incorporation of mutant isoforms into the microtubule lattice or to incorporation that perturbs the architecture or dynamics of the microtubules. The analysis of tubulin disease mutations is starting to reveal how subtle alterations of the microtubule cytoskeleton can lead to functional aberrations in cells and organisms and might provide novel insights into the roles of tubulin isotypes that have so far been considered redundant.

Tubulin PTMs

Tubulin is subject to a large range of PTMs (Fig. 1), from well-known ones, such as acetylation or phosphorylation, to others that have so far mostly been found on tubulin. Detyrosination/tyrosination, polyglutamylation, and polyglycylation, for instance, might have evolved to specifically regulate tubulin and microtubule functions, in particular in cilia and flagella, as their evolution is closely linked to these organelles. The strong link between those modifications and tubulin evolution has led to the perception that they are tubulin PTMs; however, apart from detyrosination/tyrosination, most of them have other substrates (Regnard et al., 2000; Xie et al., 2007; van Dijk et al., 2008; Rogowski et al., 2009).

Tubulin acetylation.

Tubulin acetylation was discovered on lysine 40 (K40; Fig. 1 A) of flagellar α-tubulin in Chlamydomonas reinhardtii (L’Hernault and Rosenbaum, 1985) and is generally enriched on stable microtubules in cells. Considering that K40 acetylation per se has no effect on the ultrastructure of microtubules (Howes et al., 2014), it is rather unlikely that it directly stabilizes microtubules. As a result of its localization at the inner face of microtubules (Soppina et al., 2012), K40 acetylation might rather affect the binding of microtubule inner proteins, a poorly characterized family of proteins (Nicastro et al., 2011; Linck et al., 2014). Functional experiments in cells have further suggested that K40 acetylation regulates intracellular transport by regulating the traffic of kinesin motors (Reed et al., 2006; Dompierre et al., 2007). These observations could so far not be confirmed by biophysical measurements in vitro (Walter et al., 2012; Kaul et al., 2014), suggesting that in cells, K40 acetylation might affect intracellular traffic by indirect mechanisms.Enzymes involved in K40 acetylation are HDAC6 (histone deacetylase family member 6; Hubbert et al., 2002) and Sirt2 (sirtuin type 2; North et al., 2003). Initial functional studies used overexpression, depletion, or chemical inhibition of these enzymes. These studies should be discussed with care, as both HDAC6 and Sirt2 deacetylate other substrates and have deacetylase-independent functions and chemical inhibition of HDAC6 is not entirely selective for this enzyme (Valenzuela-Fernández et al., 2008). In contrast, acetyl transferase α-Tat1 (or Mec-17; Akella et al., 2010; Shida et al., 2010) specifically acetylates α-tubulin K40 (Fig. 3), thus providing a more specific tool to investigate the functions of K40 acetylation. Knockout mice of α-Tat1 are completely void of K40-acetylated tubulin; however, they show only slight phenotypic aberrations, for instance, in their sperm flagellum (Kalebic et al., 2013). A more detailed analysis of α-Tat1 knockout mice demonstrated that absence of K40 acetylation leads to reduced contact inhibition in proliferating cells (Aguilar et al., 2014). In migrating cells, α-Tat1 is targeted to microtubules at the leading edge by clathrin-coated pits, resulting in locally restricted acetylation of those microtubules (Montagnac et al., 2013). A recent structural study of α-Tat1 demonstrated that the low catalytic rate of this enzyme, together with its localization inside the microtubules, caused acetylation to accumulate selectively in stable, long-lived microtubules (Szyk et al., 2014), thus explaining the link between this PTM and stable microtubules in cells. However, the direct cellular function of K40 acetylation on microtubules is still unclear.Open in a separate windowFigure 3.Enzymes involved in PTM of tubulin. Schematic representation of known enzymes (mammalian enzymes are shown) involved in the generation and removal of PTMs shown in Fig. 1. Note that some enzymes still remain unknown, and some modifications are irreversible. (*CCP5 preferentially removes branching points [Rogowski et al., 2010]; however, the enzyme can also hydrolyze linear glutamate chains [Berezniuk et al., 2013]).Recent discoveries have brought up the possibility that tubulin could be subject to multiple acetylation events. A whole-acetylome study identified >10 novel sites on α- and β-tubulin (Choudhary et al., 2009); however, none of these sites have been confirmed. Another acetylation event has been described at lysine 252 (K252) of β-tubulin. This modification is catalyzed by the acetyltransferase San (Fig. 3) and might regulate the assembly efficiency of microtubules as a result of its localization at the polymerization interface (Chu et al., 2011).

Tubulin detyrosination.

Most α-tubulin genes in different species encode a C-terminal tyrosine residue (Fig. 2; Valenzuela et al., 1981). This tyrosine can be enzymatically removed (Hallak et al., 1977) and religated (Fig. 3; Arce et al., 1975). Mapping of tyrosinated and detyrosinated microtubules in cells using specific antibodies (Gundersen et al., 1984; Geuens et al., 1986; Cambray-Deakin and Burgoyne, 1987a) revealed that subsets of interphase and mitotic spindle microtubules are detyrosinated (Gundersen and Bulinski, 1986). As detyrosination was mostly found on stable and long-lived microtubules, especially in neurons (Cambray-Deakin and Burgoyne, 1987b; Robson and Burgoyne, 1989; Brown et al., 1993), it was assumed that this modification promotes microtubule stability (Gundersen et al., 1987; Sherwin et al., 1987). Although a direct stabilization of the microtubule lattice was considered unlikely (Khawaja et al., 1988), it was found more recently that detyrosination protects cellular microtubules from the depolymerizing activity of kinesin-13–type motor proteins, such as KIF2 or MCAK, thus increasing their longevity (Peris et al., 2009; Sirajuddin et al., 2014).Besides kinesin-13 motors, plus end–tracking proteins with cytoskeleton-associated protein glycine-rich (CAP-Gly) domains, such as CLIP170 or p150/glued, specifically interact with tyrosinated microtubules (Peris et al., 2006; Bieling et al., 2008) via this domain (Honnappa et al., 2006). In contrast, kinesin-1 moves preferentially on detyrosinated microtubules tracks in cells (Liao and Gundersen, 1998; Kreitzer et al., 1999; Konishi and Setou, 2009). The effect of detyrosination on kinesin-1 motor behavior was recently measured in vitro, and a small but significant increase in the landing rate and processivity of the motor has been found (Kaul et al., 2014). Such subtle changes in the motor behavior could, in conjunction with other factors, such as regulatory MAPs associated with cargo transport complexes (Barlan et al., 2013), lead to a preferential use of detyrosinated microtubules by kinesin-1 in cells.Despite the early biochemical characterization of a detyrosinating activity, the carboxypeptidase catalyzing detyrosination of α-tubulin has yet to be identified (Hallak et al., 1977; Argaraña et al., 1978, 1980). In contrast, the reverse enzyme, tubulin tyrosine ligase (TTL; Fig. 3; Raybin and Flavin, 1975; Deanin and Gordon, 1976; Argaraña et al., 1980), has been purified (Schröder et al., 1985) and cloned (Ersfeld et al., 1993). TTL modifies nonpolymerized tubulin dimers exclusively. This selectivity is determined by the binding interface between the TTL and tubulin dimers (Szyk et al., 2011, 2013; Prota et al., 2013). In contrast, the so far unidentified detyrosinase acts preferentially on polymerized microtubules (Kumar and Flavin, 1981; Arce and Barra, 1983), thus modifying a select population of microtubules within cells (Gundersen et al., 1987).In most organisms, only one unique gene for TTL exists. Consequently, TTL knockout mice show a huge accumulation of detyrosinated and particularly Δ2-tubulin (see next section). TTL knockout mice die before birth (Erck et al., 2005) with major developmental defects in the nervous system that might be related to aberrant neuronal differentiation (Marcos et al., 2009). TTL is strictly tubulin specific (Prota et al., 2013), indicating that all observed defects in TTL knockout mice are directly related to the deregulation of the microtubule cytoskeleton.

Δ2-tubulin and further C-terminal modification.

A biochemical study of brain tubulin revealed that ∼35% of α-tubulin cannot be retyrosinated (Paturle et al., 1989) because of the lack of the penultimate C-terminal glutamate residue of the primary protein sequence (Fig. 2; Paturle-Lafanechère et al., 1991). This so-called Δ2-tubulin (for two C-terminal amino acids missing) cannot undergo retyrosination as a result of structural constraints within TTL (Prota et al., 2013) and thus is considered an irreversible PTM.Δ2-tubulin accumulates in long-lived microtubules of differentiated neurons, axonemes of cilia and flagella, and also in cellular microtubules that have been artificially stabilized, for instance, with taxol (Paturle-Lafanechère et al., 1994). The generation of Δ2-tubulin requires previous detyrosination of α-tubulin; thus, the levels of this PTM are indirectly regulated by the detyrosination/retyrosination cycle. This mechanistic link is particularly apparent in the TTL knockout mice, which show massive accumulation of Δ2-tubulin in all tested tissues (Erck et al., 2005). Loss of TTL and the subsequent increase of Δ2-tubulin levels were also linked to tumor growth and might contribute to the aggressiveness of the tumors by an as-yet-unknown mechanism (Lafanechère et al., 1998; Mialhe et al., 2001). To date, no specific biochemical role of Δ2-tubulin has been determined; thus, one possibility is that the modification simply locks tubulin in the detyrosinated state.The enzymes responsible for Δ2-tubulin generation are members of a family of cytosolic carboxypeptidases (CCPs; Fig. 3; Kalinina et al., 2007; Rodriguez de la Vega et al., 2007), and most of them also remove polyglutamylation from tubulin (see next section; Rogowski et al., 2010). These enzymes are also able to generate Δ3-tubulin (Fig. 1 A; Berezniuk et al., 2012), indicating that further degradation of the tubulin C-terminal tails are possible; however, the functional significance of this event is unknown.

Polyglutamylation.

Polyglutamylation is a PTM that occurs when secondary glutamate side chains are formed on γ-carboxyl groups of glutamate residues in a protein (Fig. 1, A and B). The modification was first discovered on α- and β-tubulin from the brain (Eddé et al., 1990; Alexander et al., 1991; Rüdiger et al., 1992; Mary et al., 1994) as well as on axonemal tubulin from different species (Mary et al., 1996, 1997); however, it is not restricted to tubulin (Regnard et al., 2000; van Dijk et al., 2008). Using a glutamylation-specific antibody, GT335 (Wolff et al., 1992), it was observed that tubulin glutamylation increases during neuronal differentiation (Audebert et al., 1993, 1994) and that axonemes of cilia and flagella (Fouquet et al., 1994), as well as centrioles of mammalian centrosomes (Bobinnec et al., 1998), are extensively glutamylated.Enzymes catalyzing polyglutamylation belong to the TTL-like (TTLL) family (Regnard et al., 2003; Janke et al., 2005). In mammals, nine glutamylases exist, each of them showing intrinsic preferences for modifying either α- or β-tubulin as well as for initiating or elongating glutamate chains (Fig. 3; van Dijk et al., 2007). Two of the six well-characterized TTLL glutamylases also modify nontubulin substrates (van Dijk et al., 2008).Knockout or depletion of glutamylating enzymes in different model organisms revealed an evolutionarily conserved role of glutamylation in cilia and flagella. In motile cilia, glutamylation regulates beating behavior (Janke et al., 2005; Pathak et al., 2007; Ikegami et al., 2010) via the regulation of flagellar dynein motors (Kubo et al., 2010; Suryavanshi et al., 2010). Despite the expression of multiple glutamylases in ciliated cells and tissues, depletion or knockout of single enzymes often lead to ciliary defects, particularly in motile cilia (Ikegami et al., 2010; Vogel et al., 2010; Bosch Grau et al., 2013; Lee et al., 2013), suggesting essential and nonredundant regulatory functions of these enzymes in cilia.Despite the enrichment of polyglutamylation in neuronal microtubules (Audebert et al., 1993, 1994), knockout of TTLL1, the major polyglutamylase in brain (Janke et al., 2005), did not show obvious neuronal defects in mice (Ikegami et al., 2010; Vogel et al., 2010). This suggests a tolerance of neuronal microtubules to variations in polyglutamylation.Deglutamylases, the enzymes that reverse polyglutamylation, were identified within a novel family of CCPs (Kimura et al., 2010; Rogowski et al., 2010). So far, three out of six mammalian CCPs have been shown to cleave C-terminal glutamate residues, thus catalyzing both the reversal of polyglutamylation and the removal of gene-encoded glutamates from the C termini of proteins (Fig. 3). The hydrolysis of gene-encoded glutamate residues is not restricted to tubulin, in which it generates Δ2- and Δ3-tubulin, but has also been reported for other proteins such as myosin light chain kinase (Rusconi et al., 1997; Rogowski et al., 2010). One enzyme of the CCP family, CCP5, preferentially removes branching points generated by glutamylation, thus allowing the complete reversal of the polyglutamylation modification (Kimura et al., 2010; Rogowski et al., 2010). However, CCP5 can also hydrolyze C-terminal glutamate residues from linear peptide chains similar to other members of the CCP family (Berezniuk et al., 2013).CCP1 is mutated in a well-established mouse model for neurodegeneration, the pcd (Purkinje cell degeneration) mouse (Mullen et al., 1976; Greer and Shepherd, 1982; Fernandez-Gonzalez et al., 2002). The absence of a key deglutamylase leads to strong hyperglutamylation in brain regions that undergo degeneration, such as the cerebellum and the olfactory bulb (Rogowski et al., 2010). When glutamylation levels were rebalanced by depletion or knockout of the major brain polyglutamylase TTLL1 (Rogowski et al., 2010; Berezniuk et al., 2012), Purkinje cells survived. Although the molecular mechanisms of hyperglutamylation-induced degeneration remain to be elucidated, perturbation of neuronal transport, as well as changes in the dynamics and stability of microtubules, is expected to be induced by hyperglutamylation. Increased polyglutamylation levels have been shown to affect kinesin-1–mediated transport in cultured neurons (Maas et al., 2009), and the turnover of microtubules can also be regulated by polyglutamylation via the activation of microtubule-severing enzymes such as spastin (Lacroix et al., 2010).Subtle differences in polyglutamylation can be seen on diverse microtubules in different cell types. The functions of these modifications remain to be studied; however, its wide distribution strengthens the idea that it could be involved in fine-tuning a range of microtubule functions.

Polyglycylation.

Tubulin polyglycylation or glycylation, like polyglutamylation, generates side chains of glycine residues within the C-terminal tails of α- and β-tubulin (Fig. 1, A and B). The modification sites of glycylation are considered to be principally the same as for glutamylation, and indeed, both PTMs have been shown to be interdependent in cells (Rogowski et al., 2009; Wloga et al., 2009). Initially discovered on Paramecium tetraurelia tubulin (Redeker et al., 1994), glycylation has been extensively studied using two antibodies, TAP952 and AXO49 (Bressac et al., 1995; Levilliers et al., 1995; Bré et al., 1996). In contrast to polyglutamylation, glycylation is restricted to cilia and flagella in most organisms analyzed so far.Glycylating enzymes are also members of the TTLL family, and homologues of these enzymes have so far been found in all organisms with proven glycylation of ciliary axonemes (Rogowski et al., 2009; Wloga et al., 2009). In mammals, initiating (TTLL3 and TTLL8) and elongating (TTLL10) glycylases work together to generate polyglycylation (Fig. 3). In contrast, the two TTLL3 orthologues from Drosophila melanogaster can both initiate and elongate glycine side chains (Rogowski et al., 2009).In mice, motile ependymal cilia in brain ventricles acquire monoglycylation upon maturation, whereas polyglycylation is observed only after several weeks (Bosch Grau et al., 2013). Sperm flagella, in contrast, acquire long glycine chains much faster, suggesting that the extent of polyglycylation could correlate with the length of the axonemes (Rogowski et al., 2009). Depletion of glycylases in mice (ependymal cilia; Bosch Grau et al., 2013), zebrafish (Wloga et al., 2009; Pathak et al., 2011), Tetrahymena thermophila (Wloga et al., 2009), and D. melanogaster (Rogowski et al., 2009) consistently led to ciliary disassembly or severe ciliary defects. How glycylation regulates microtubule functions remains unknown; however, the observation that glycylation-depleted axonemes disassemble after initial assembly (Rogowski et al., 2009; Bosch Grau et al., 2013) suggests a role of this PTM in stabilizing axonemal microtubules. Strikingly, human TTLL10 is enzymatically inactive; thus, humans have lost the ability to elongate glycine side chains (Rogowski et al., 2009). This suggests that the elongation of the glycine side chains is not an essential aspect of the function of this otherwise critical tubulin PTM.

Other tubulin PTMs.

Several other PTMs have been found on tubulin. Early studies identified tubulin phosphorylation (Eipper, 1974; Gard and Kirschner, 1985; Díaz-Nido et al., 1990); however, no specific functions were found. The perhaps best-studied phosphorylation event on tubulin takes place at serine S172 of β-tubulin (Fig. 1 A), is catalyzed by the Cdk1 (Fig. 3), and might regulate microtubule dynamics during cell division (Fourest-Lieuvin et al., 2006; Caudron et al., 2010). Tubulin can be also modified by the spleen tyrosine kinase Syk (Fig. 3; Peters et al., 1996), which might play a role in immune cells (Faruki et al., 2000; Sulimenko et al., 2006) and cell division (Zyss et al., 2005; Sulimenko et al., 2006).Polyamination has recently been discovered on brain tubulin (Song et al., 2013), after having been overlooked for many years as a result of the low solubility of polyaminated tubulin. Among several glutamine residues of α- and β-tubulin that can be polyaminated, Q15 of β-tubulin is considered the primary modification site (Fig. 1 A). Polyamination is catalyzed by transglutaminases (Fig. 3), which modify free tubulin as well as microtubules in an irreversible manner, and most likely contribute to the stabilization of microtubules (Song et al., 2013).Tubulin was also reported to be palmitoylated (Caron, 1997; Ozols and Caron, 1997; Caron et al., 2001), ubiquitinated (Ren et al., 2003; Huang et al., 2009; Xu et al., 2010), glycosylated (Walgren et al., 2003; Ji et al., 2011), arginylated (Wong et al., 2007), methylated (Xiao et al., 2010), and sumoylated (Rosas-Acosta et al., 2005). These PTMs have mostly been reported without follow-up studies, and some of them are only found in specific cell types or organisms and/or under specific metabolic conditions. Further studies will be necessary to gain insights into their potential roles for the regulation of the microtubule cytoskeleton.

Current advances and future perspectives

The molecular heterogeneity of microtubules, generated by the expression of different tubulin isotypes and by the PTM of tubulin has fascinated the scientific community for ∼40 years. Although many important advances have been made in the past decade, the dissection of the molecular mechanisms and a comprehensive understanding of the biological functions of tubulin isotypes and PTMs will be a challenging field of research in the near future.

Direct measurements of the impact of tubulin heterogeneity.

The most direct and reliable type of experiments to determine the impact of tubulin heterogeneity on microtubule behavior are in vitro measurements with purified proteins. However, most biophysical work on microtubules has been performed with tubulin purified from bovine, ovine, or porcine brains, which can be obtained in large quantities and with a high degree of purity and activity (Vallee, 1986; Castoldi and Popov, 2003). Brain tubulin is a mixture of different tubulin isotypes and is heavily posttranslationally modified and thus inept for investigating the functions of tubulin heterogeneity (Denoulet et al., 1986; Cambray-Deakin and Burgoyne, 1987b; Paturle et al., 1989; Eddé et al., 1990). Thus, pure, recombinant tubulin will be essential to dissect the roles of different tubulin isoforms and PTMs.Attempts to produce recombinant, functional α- and β-tubulin in bacteria have failed so far (Yaffe et al., 1988), most likely because of the absence of the extensive tubulin-specific folding machinery (Yaffe et al., 1992; Gao et al., 1993; Tian et al., 1996; Vainberg et al., 1998) in prokaryotes. An alternative source of tubulin with less isotype heterogeneity and with almost no PTMs is endogenous tubulin from cell lines such as HeLa, which in the past has been purified using a range of biochemical procedures (Bulinski and Borisy, 1979; Weatherbee et al., 1980; Farrell, 1982; Newton et al., 2002; Fourest-Lieuvin, 2006). Such tubulin can be further modified with tubulin-modifying enzymes, such as polyglutamylases, either by expressing those enzymes in the cells before tubulin purification (Lacroix and Janke, 2011) or in vitro with purified enzymes (Vemu et al., 2014). Despite some technical limitations of these methods, HeLa tubulin modified in cells has been successfully used in an in vitro study on the role of polyglutamylation in microtubule severing (Lacroix et al., 2010).Naturally occurring variants of tubulin isotypes and PTMs can be purified from different organisms, organs, or cell types, but obviously, only some combinations of tubulin isotypes and PTMs can be obtained by this approach. The recent development of an affinity purification method using the microtubule-binding TOG (tumor overexpressed gene) domain of yeast Stu2p has brought a new twist to this approach, as it allows purifying small amounts of tubulin from any cell type or tissue (Widlund et al., 2012).The absence of tubulin heterogeneity in yeast has made budding and fission yeast potential expression systems for recombinant, PTM-free tubulin (Katsuki et al., 2009; Drummond et al., 2011; Johnson et al., 2011). However, the expression of mammalian tubulin in this system has remained impossible. This problem was then partially circumvented by expressing tubulin chimeras that consist of a yeast tubulin body fused to mammalian C-terminal tubulin tails, thus mimicking different tubulin isotypes (Sirajuddin et al., 2014). Moreover, detyrosination can be generated by deleting the key C-terminal residue from endogenous or chimeric α-tubulin (Badin-Larçon et al., 2004), and polyglutamylation is generated by chemically coupling glutamate side chains to specifically engineered tubulin chimeras (Sirajuddin et al., 2014). These approaches allowed the first direct measurements of the impact of tubulin isotypes and PTMs on the behavior of molecular motors in vitro (Sirajuddin et al., 2014) and the analysis of the effects of tubulin heterogeneity on microtubule behavior and interactions inside the yeast cell (Badin-Larçon et al., 2004; Aiken et al., 2014).Currently, the most promising development has been the successful purification of fully functional recombinant tubulin from the baculovirus expression system (Minoura et al., 2013). Using this system, defined α/β-tubulin dimers can be obtained using two different epitope tags on α- and β-tubulin, respectively. Although these epitope tags are essential for separating recombinant from the endogenous tubulin, they could also affect tubulin assembly or microtubule–MAP interactions. Thus, future developments should focus on eliminating these tags.Current efforts have brought the possibility of producing recombinant tubulin into reach. Further improvement and standardization of these methods will certainly provide a breakthrough in understanding the mechanisms by which tubulin heterogeneity contributes to microtubule functions.

Complexity of tubulin—understanding the regulatory principles.

The diversity of tubulin genes (isotypes) and the complexity of tubulin PTMs have led to the proposal of the term “tubulin code” (Verhey and Gaertig, 2007; Wehenkel and Janke, 2014), in analogy to the previously coined histone code (Jenuwein and Allis, 2001). Tubulin molecules consist of a highly structured and thus evolutionarily conserved tubulin body and the unstructured and less conserved C-terminal tails (Nogales et al., 1998). As PTMs and sequence variations within the tubulin body are expected to affect the conserved tubulin fold and therefore the properties of the microtubule lattice, they are not likely to be involved in generating the tubulin code. In contrast, modulations of the C-terminal tails could encode signals on the microtubule surface without perturbing basic microtubule functions and properties (Figs. 1 A and ​and4).4). Indeed, the highest degree of gene-encoded diversity (Fig. 2) and the highest density and complexity of PTMs (Fig. 1) are found within these tail domains.Open in a separate windowFigure 4.Molecular components of the tubulin code. Schematic representation of potential coding elements that could generate specific signals for the tubulin code. (A) The length of the C-terminal tails of different tubulin isotypes differ significantly (Fig. 2) and could have an impact on the interactions between microtubules and MAPs. (B) Tubulin C-terminal tails are rich in charged amino acid residues. The distribution of these residues and local densities of charges could influence the electrostatic interactions with the tails and the readers. (C) Although each glutamate residue within the C-terminal tails could be considered a potential modification site, only some sites have been found highly occupied in tubulin purifications from native sources. This indicates selectivity of the modification reactions, which can participate in the generation of specific modification patterns (see D). Modification sites might be distinguished by their neighboring amino acid residues, which could create specific modification epitopes. (D) As a result of the large number of modification sites and the variability of side chains, a large variety of modification patterns could be generated within a single C-terminal tail of tubulin. (E) Modification patterns as shown in D can be distinct between α- and β-tubulin. These modification patterns could be differentially distributed at the surface of the microtubule lattice, thus generating a higher-order patterning. Tub, tubulin. For color coding, see Fig. 2.Considering the number of tubulin isotypes plus all potential combinations of PTMs (e.g., each glutamate residue within the C-terminal tubulin tail could be modified by either polyglutamylation or polyglycylation, each of them generating side chains of different lengths; Fig. 4), the number of distinct signals generated by the potential tubulin code would be huge. However, as many of these potential signals represent chemical structures that are similar and might not be reliably distinguished by readout mechanisms, it is possible that the tubulin code generates probabilistic signals. In this scenario, biochemically similar modifications would have similar functional readouts, and marginal differences between those signals would only bias biological processes but not determine them. This stands in contrast to the concept of the histone code, in which precise patterns of different PTMs on the histone proteins encode distinct biological signals.The concept of probabilistic signaling is already inscribed in the machinery that generates the tubulin code. Polyglutamylases and polyglycylases from the TTLL family have preferential activities for either α- or β-tubulin and for generating different lengths of the branched glutamate or glycine chains. Although under conditions of low enzyme concentrations, as found in most cells and tissues, the enzymes seem to selectively generate their preferential type of PTM, higher enzyme concentrations induce a more promiscuous behavior, leading, for instance, to a loss of selectivity for α- or β-tubulin (van Dijk et al., 2007). Similarly, the modifying enzymes might prefer certain modification sites within the C-terminal tails of tubulin but might be equally able to modify other sites, which could be locally regulated in cells. For example, β-tubulin isotypes isolated from mammalian brain were initially found to be glutamylated on single residues (Alexander et al., 1991; Rüdiger et al., 1992), which in the light of the comparably low sensitivity of mass spectrometry at the time might rather indicate a preferential than a unique modification of these sites. Nevertheless, the neuron-specific polyglutamylase for β-tubulin TTLL7 (Ikegami et al., 2006) can incorporate glutamate onto many more modification sites of β-tubulin in vitro (Mukai et al., 2009), which clearly indicates that not all of the possible modification events take place under physiological conditions.Several examples supporting a probabilistic signaling mode of the tubulin code are found in the recent literature. In T. thermophila, a ciliate without tubulin isotype diversity (Gaertig et al., 1993) but with a huge repertoire of tubulin PTMs and tubulin-modifying enzymes (Janke et al., 2005), tubulin can be easily mutagenized to experimentally eliminate sites for PTMs. Mutagenesis of the most commonly occupied glutamylation/glycylation sites within the β-tubulin tails did not generate a clear decrease of glycylation levels nor did it cause obvious phenotypic alterations. This indicates that the modifying enzymes can deviate toward alternative modification sites and that similar PTMs on different sites can compensate the functions of the mutated site. However, when all of the key modification sites were mutated, glycylation became prominently decreased, which led to severe phenotypes, including lethality (Xia et al., 2000). Most strikingly, these phenotypes could be recovered by replacing the C-terminal tail of α-tubulin with the nonmutated β-tubulin tail. This α–β-tubulin chimera became overglycylated and functionally compensated for the absence of modification sites on β-tubulin. The conclusion of this study is that PTM- and isotype-generated signals can fulfill a biological function within a certain range of tolerance.But how efficient is such compensation? The answer can be found in a variety of already described deletion mutants for tubulin-modifying enzymes in different model organisms. Most single-gene knockouts for TTLL genes (glutamylases or glycylases) did not result in prominent phenotypic alterations in mice, even for enzymes that are ubiquitously expressed. Only some highly specialized microtubule structures show functional aberrations upon the deletion of a single enzyme. These “tips of the iceberg” are usually the motile cilia and sperm flagella, which carry very high levels of polyglutamylation and polyglycylation (Bré et al., 1996; Kann et al., 1998; Rogowski et al., 2009). It thus appears that some microtubules are essentially dependent on the generation of specific PTM patterns, whereas others can tolerate changes and appear to function normally. How “normal” these functions are remains to be investigated in future studies. It is possible that defects are subtle and thus overlooked but could become functionally important under specific conditions.A tubulin code also requires readout mechanisms. The most likely “readers” of the tubulin code are MAPs and molecular motors. Considering the probabilistic signaling hypothesis, the expected effects of the signals would be in most cases rather gradual changes, for instance, to fine-tune molecular motor traffic and/or to bias motors toward defined microtubule tracks but not to obliterate motor activity or MAP binding to microtubules. An in vitro study using recombinant tubulin chimeras purified from yeast confirmed this notion (Sirajuddin et al., 2014). By analyzing which elements of the tubulin code can regulate the velocity and processivity of the molecular motors kinesin and dynein, these researchers found that the C-terminal tails of α- and β-tubulin differentially influence the kinetic parameters of the tested motors; however, the modulation was rather modest. One of their striking observations was that a single lysine residue, present in the C-terminal tails of two β-tubulin isotypes (Figs. 2 and ​and4),4), significantly affected motor traffic and that this effect can be counterbalanced by polyglutamylation. These observations are the first in vitro evidence for the interdependence of different elements of the tubulin code and provide another indication for its probabilistic mode of signaling.

Future directions.

One of the greatest technological challenges to understanding the function of the tubulin code is to detect and interpret subtle and complex regulatory events generated by this code. It will thus be instrumental to further develop tools to better distinguish graded changes in PTM levels on microtubules in cells and tissues (Magiera and Janke, 2013) and to reliably measure subtle modulations of microtubule behavior in reconstituted systems.The current advances in the field and especially the availability of whole-organism models, as well as first insights into the pathological role of tubulin mutations (Tischfield et al., 2011), are about to transform our way of thinking about the regulation of microtubule cytoskeleton. Tubulin heterogeneity generates complex probabilistic signals that cannot be clearly attributed to single biological functions in most cases and that are not essential for most cellular processes. Nevertheless, it has been conserved throughout evolution of eukaryotes and can hardly be dismissed as not important. To understand the functional implications of these processes, we might be forced to reconsider how we define biologically important events and how we measure events that might encode probabilistic signals. The answers to these questions could provide novel insights into how complex systems, such as cells and organisms, are sustained throughout difficult and challenging life cycles, resist to environmental stress and diseases, and have the flexibility needed to succeed in evolution.     

Cell biology in neuroscience: Death of developing neurons: New insights and implications for connectivity

The concept that target tissues determine the survival of neurons has inspired much of the thinking on neuronal development in vertebrates, not least because it is supported by decades of research on nerve growth factor (NGF) in the peripheral nervous system (PNS). Recent discoveries now help to understand why only some developing neurons selectively depend on NGF. They also indicate that the survival of most neurons in the central nervous system (CNS) is not simply regulated by single growth factors like in the PNS. Additionally, components of the cell death machinery have begun to be recognized as regulators of selective axonal degeneration and synaptic function, thus playing a critical role in wiring up the nervous system.

Why do so many neurons die during development?

Programmed cell death occurs throughout life, as cell turnover is part of homeostasis and maintenance in most organs and tissues. The situation in the nervous system is principally different, as the vast majority of neurons undergo their last round of cell division early in development. Soon after exiting the cell cycle, neurons start elongating axons to innervate their targets. It is during this period that they are highly susceptible to undergo programmed cell death: a large percentage, as much as 50% in several ganglia in the peripheral nervous system (PNS) as well as in various central nervous system (CNS) areas, is eliminated around the time that connections are being made with other cells. Later in development, the propensity of neurons to initiate apoptosis progressively decreases. The likelihood for a neuron to undergo apoptosis seems to be determined by a tightly regulated apoptotic machinery (summarized in Fig. 1). Therefore, modulation of the expression levels or the activity of components of this apoptotic balance changes the sensitivity to death-promoting cues, allowing temporal restriction of cell death.Open in a separate windowFigure 1.Core components of the apoptotic machinery. The likelihood that a neuron undergoes apoptosis is determined by the interplay of the tightly interlinked apoptotic machinery, many components of which are highly conserved between species. The critical, and often terminal, step in programmed cell death is the proteolytic activation of the executor caspases (such as caspase 3, 6, 7) by the initiator caspases (i.e., caspase 8, 9, and 10; Riedl and Salvesen, 2007). In mammalian cells, initiation of the executor caspases is regulated by two distinct protein cascades: the intrinsic pathway, also known as the mitochondrial pathway, and the extrinsic pathway. The intrinsic pathway integrates a number of intra- and extracellular signal modalities, such as redox state (for example, the reactive oxygen species; Franklin, 2011), DNA damage (Sperka et al., 2012), ER stress (Puthalakath et al., 2007) and growth factor deprivation (Deckwerth et al., 1998; Putcha et al., 2003; Bredesen et al., 2005), or activation of the p75NTR neurotrophin receptor by pro-neurotrophins (Nykjaer et al., 2005). The stressors converge onto pro- and anti-apoptotic members of the Bcl-2 protein family (for example: BCL-2, BCL-X_l, BAX, and tBID; Youle and Strasser, 2008). These proteins regulate the release of cytochrome c from mitochondria, which activates the initiator caspase 9 through Apaf1 (Riedl and Salvesen, 2007). The extrinsic pathway links activation of ligand-bound death receptors (such as Fas/CD95 and TNFR) to the initiator caspase 8 and 10, through formation of the death-inducing signaling complex (DISC; LeBlanc and Ashkenazi, 2003; Peter and Krammer, 2003). Together with additional regulatory elements (including the Inhibitors of apoptosis proteins [IAP]; Vaux and Silke, 2005) and cFLIP (Scaffidi et al., 1999; Wang et al., 2005), the apoptotic machinery forms a balance that determines the propensity of the neuron to undergo apoptosis.Programmed cell death eliminates many neurons during development, even in organisms comprised of only few cells, such as Caenorhabditis elegans. As neurons and their targets are initially separated, it is possible that the initial generation of an overabundance of neurons is simply part of a mechanism to ensure that distal targets are adequately innervated (Oppenheim, 1991; Conradt, 2009; Chen et al., 2013). In various tissues other than the nervous system, programmed cell death is used to eliminate cells that are no longer needed, defective, or harmful to the function of the organism. However, there is strong evidence that the elimination of superfluous neurons in the developing nervous system is not essential. For example, early work in C. elegans revealed that preventing programmed cell death does not result in significant behavioral alterations (Ellis and Horvitz, 1986). In the C57BL/6 mouse strain, deletion of the executor caspases 3 and 7 (Fig. 1) has a remarkably limited neuropathological and morphological impact in the CNS (Leonard et al., 2002; Lakhani et al., 2006) compared with the 129X1/SvJ strain, in which deletion of these caspases causes neurodevelopmental defects (Leonard et al., 2002). Similar conclusions were reached by blocking the Bcl-2–associated X protein (BAX)–dependent pathway in many neuronal populations, including motoneurons (Buss et al., 2006a). A recent study in the developing retina showed that in mice lacking the central apoptotic regulator BAX, the normal mosaic distribution of intrinsically photosensitive retinal ganglion cells (ipRGCs) was perturbed (Chen et al., 2013). Although this abnormal distribution is dispensable for the intrinsic photosensitivity of the ipRGCs, it is required for establishing proper connections to other neurons in the retina, which is necessary for rod/cone photo-entrainment (Chen et al., 2013). Even though this finding highlights a physiological role for programmed cell death in the CNS, the functional consequences remain rather underwhelming in the face of a process that eliminates such large numbers of neurons (Purves and Lichtman, 1984; Oppenheim, 1991; Miller, 1995; Gohlke et al., 2004). It thus appears that apoptotic removal of the surplus neurons generated during development mainly serves the purpose to optimize the size of the nervous system to be minimal, but sufficient.

A molecular substrate for the neurotrophic theory

Quantitatively, programmed cell death of neurons in the PNS and CNS is most dramatic when neurons start contacting the cells they innervate. Because experimental manipulations such as target excision typically lead to the death of essentially all innervating neurons (Oppenheim, 1991), the concept emerged that the fate of developing neurons is regulated by their targets. This notion is often referred to as the “neurotrophic theory” (Hamburger et al., 1981; Purves and Lichtman, 1984; Oppenheim, 1991), but it is important to realize that it evolved in the absence of direct mechanistic or molecular support (Purves, 1988). Originally described as a diffusible agent promoting nerve growth, the eponymous NGF later provided a strong and very appealing molecular foundation for this theory (Korsching and Thoenen, 1983; Edwards et al., 1989; Hamburger, 1992). The tyrosine kinase receptor tropomyosin receptor kinase A (TrkA), which was initially identified as an oncogene (Martin-Zanca et al., 1986), was fortuitously discovered to be the critical receptor necessary for the prevention of neuronal cell death by NGF (Klein et al., 1991). Both the remarkable expression pattern of TrkA in NGF-dependent neurons and the onset of its expression during development (Martin-Zanca et al., 1990) provided further additional support for the neurotrophic theory. However, for a surprisingly long time, the question was not asked as to why only specific populations of neurons strictly depend on NGF for survival, while others do not. Indeed, it was only recently shown that TrkA causes cell death of neurons by virtue of its mere expression, and that this death-inducing activity is prevented by addition of NGF (Nikoletopoulou et al., 2010). These findings thus indicate that TrkA acts as a “dependence receptor,” a concept introduced after observations that various cell types die when receptors are expressed in the absence of their cognate ligands (Bredesen et al., 2005; Tauszig-Delamasure et al., 2007). Accordingly, embryonic mouse sympathetic or sensory neurons survive in the absence of NGF when TrkA is deleted (Nikoletopoulou et al., 2010). The closely related neurotrophin receptor TrkC also acts as a dependence receptor (Tauszig-Delamasure et al., 2007; Nikoletopoulou et al., 2010). Here, it is interesting to note a series of older, convergent results indicating that deletion of neurotrophin-3 (NT3), the TrkC ligand, leads to a significantly larger loss of sensory and sympathetic neurons in the PNS than the deletion of TrkC (Tessarollo et al., 1997). This phenotypic discrepancy fits well with the idea that inactivation of the ligand of a dependence receptor is expected to yield a more profound phenotype than inactivation of the receptor itself (Tauszig-Delamasure et al., 2007). How TrkA and TrkC induce apoptosis remains to be fully elucidated. It seems that proteolysis is involved, either of TrkC itself (Tauszig-Delamasure et al., 2007), as was suggested for other dependence receptors (Bredesen et al., 2005), or by the proteolysis of the neurotrophin receptor p75NTR, which associates with both TrkA and TrkC (Fig. 2; Nikoletopoulou et al., 2010). Surprisingly, although TrkA and TrkC cause cell death, the structurally related TrkB receptor does not (Nikoletopoulou et al., 2010), a difference that appears to be accounted for by their differential localization in the cell membrane. TrkA and TrkC colocalize with p75NTR in lipid rafts, whereas TrkB, which also associates with p75NTR (Bibel et al., 1999), is excluded from lipid rafts (Fig. 2; unpublished data). Interestingly, the transmembrane domains of TrkA and TrkC are closely related, and differ clearly from that of TrkB. It turns out that a chimeric protein of TrkB with the transmembrane domain of TrkA causes cell death, which can be prevented by the addition of the TrkB ligand brain-derived neurotrophic factor (BDNF; unpublished data). The suggestion that the lipid raft localization of TrkA and TrkC is important for their death-inducing function is in line with a number of reports indicating that certain apoptotic proteins preferentially localize in lipid rafts in the plasma membrane. After activation of the extrinsic apoptosis pathway, translocation of the activated receptors to lipid rafts in the membrane is required for assembling the death-inducing signaling complex (DISC; Davis et al., 2007; Song et al., 2007). Indeed, regulators of the extrinsic pathway (e.g., cFLIP; Fig. 1) prevent this translocation, explaining how they attenuate cell death induction (Song et al., 2007). Similarly, the localization of the dependence receptor DCC (deleted in colorectal cancer) in lipid rafts is a prerequisite for its pro-apoptotic activity in absence of its ligand, Netrin-1 (Furne et al., 2006).Open in a separate windowFigure 2.TrkA and TrkC as dependence receptors: mode of action and contrast with TrkB. All Trk receptors associate with the pan-neurotrophin receptor p75NTR (Bibel et al., 1999). A critical step in the induction of apoptosis by TrkA is the release of the intracellular death domain of p75NTR by the protease γ-secretase (Nikoletopoulou et al., 2010), which is localized in lipid rafts (Urano et al., 2005). Our membrane fractionation studies indicate that while TrkA and TrkC associate with p75NTR in lipid rafts, TrkB associated with p75NTR is excluded from this membrane domain (unpublished data). The 24–amino acid transmembrane domain of the Trk receptors may be responsible for this differential localization (see text).Despite the fact that TrkB does not act as a dependence receptor, its activation by BDNF is required for the survival of several populations of cranial sensory neurons (Ernfors et al., 1995; Liu et al., 1995). It appears that other death-inducing receptors predispose these neurons to be eliminated, such as p75NTR, which is expressed at high levels in some of these ganglia, or TrkC in vestibular neurons (Stenqvist et al., 2005). This latter case is of special interest, as NT3 is known not to be required for the survival of these neurons (Stenqvist et al., 2005). In addition to inducing apoptosis in the absence of their ligand, TrkA and TrkC have long been recognized to have a pro-survival function similar to TrkB, as can be inferred from the loss of specific populations of peripheral sensory neurons in mutants lacking these receptors (Klein et al., 1994; Smeyne et al., 1994).

Cell death in the CNS

Although TrkA is primarily expressed in peripheral sympathetic and sensory neurons, it is also found in a small population of cholinergic neurons in the basal forebrain (Sobreviela et al., 1994), a proportion of which requires NGF for survival (Hartikka and Hefti, 1988; Crowley et al., 1994; Müller et al., 2012). Selective deletion of TrkA was recently shown not to cause the death of these neurons (Sanchez-Ortiz et al., 2012). This supports the notion that TrkA acts as a dependence receptor for this small population of CNS neurons, like for peripheral sensory and sympathetic neurons. TrkA activation by NGF is essential for the maturation, projections, and function of these neurons (Sanchez-Ortiz et al., 2012), as was previously described for sensory neurons in the PNS as well (Patel et al., 2000).Whether or not receptors other than TrkA act as dependence receptors in the CNS is an important open question, particularly because TrkB, which is expressed highly by most CNS neurons, does not act as a dependence receptor (Nikoletopoulou et al., 2010). In retrospect, the structural similarities between TrkA and TrkB, just like those between NGF and BDNF (Barde, 1989), have substantially misled the field by suggesting that BDNF would act in the CNS like NGF in the PNS. Adding to the confusion were early findings showing that BDNF supports the growth of spinal cord motoneurons in vitro or in vivo after axotomy (Oppenheim et al., 1992; Sendtner et al., 1992; Yan et al., 1992). However, in the absence of lesion, deletion of BDNF does not lead to significant losses of neurons in the developing or adult CNS (Ernfors et al., 1994a; Jones et al., 1994; Rauskolb et al., 2010), unlike the case in some populations of PNS neurons. The poor correlation of the role of BDNF in CNS development and in axotomy and in vitro experiments is surprising, especially because the role of NGF in vivo could in essence be recapitulated by in vitro experiments. Although the reasons for this discrepancy are not fully understood, the strong up-regulation of death-inducing molecules such as p75NTR after axotomy (Ernfors et al., 1989) may be a part of the explanation. At present, most of the growth factors promoting the survival of PNS neurons fail to show significant survival properties for developing neurons in the CNS, as for example was shown for NT3 (Ernfors et al., 1994b; Fariñas et al., 1994), glial cell line–derived neurotrophic factor (GDNF; Henderson et al., 1994), ciliary neurotrophic factor (CNTF; DeChiara et al., 1995), and several others.In the developing CNS, neuronal activity and neurotransmitter input seem to play a more significant role than single growth factors in regulating neuronal survival. In particular, it has been known for a long time that blocking synaptic transmission at the neuromuscular junction has a pro-survival effect on spinal cord motoneurons (Pittman and Oppenheim, 1978; Oppenheim et al., 2008). By contrast, surgical denervation of afferent connections leads to increased apoptosis of postsynaptic neurons (Okado and Oppenheim, 1984), whereas inhibiting glycinergic and GABAergic synaptic transmission has both pro- and anti-apoptotic effects on motoneurons (Banks et al., 2005). Throughout the developing brain, blocking glutamate-mediated synaptic transmission involving NMDA receptors markedly increases normally occurring neuronal death (Ikonomidou et al., 1999; Heck et al., 2008). The mechanism involves a reduction of neuronal expression of anti-apoptotic proteins, such as B-cell lymphoma 2 (BCL-2; Hansen et al., 2004). Conversely, a limited increase in neuronal activity leads to down-regulation of the pro-apoptotic genes BAX and caspase 9 (Léveillé et al., 2010), thereby reducing the propensity of these cells to initiate programmed cell death (Hardingham et al., 2002). In addition to directly modulating the expression of apoptotic proteins, neuronal activity affects the expression of several secreted growth factors, such as BDNF (Hardingham et al., 2002; Hansen et al., 2004) and GDNF (Léveillé et al., 2010). So, even though BDNF is not a major survival factor in the developing CNS, it appears to be critical for activity-dependent neuroprotection (Tremblay et al., 1999). A recent publication revealed that certain populations of neurons in the CNS do not follow the predictions of the neurotrophic theory and showed that apoptosis of cortical inhibitory neurons is independent of cues present in the developing cerebral cortex (Southwell et al., 2012). This study indicates that programmed cell death of a large proportion of interneurons in the CNS is regulated by intrinsic mechanisms that are largely resistant to the presence or absence of extrinsic cues (Dekkers and Barde, 2013).Taken together, even though the extent of naturally occurring cell death in the different regions of the CNS is not nearly as well characterized as in the PNS, let alone quantified, it appears that its regulation may significantly differ. Although single secreted neurotrophic factors seem to be largely dispensable for survival, neuronal activity and other intrinsic mechanisms drive the propensity of the neurons in the CNS to undergo apoptosis. An important open question in this context is a possible involvement of non-neuronal cells, such as glial cells (see Corty and Freeman, in this issue).

The apoptotic machinery as a regulator of connectivity

Activation of the executor caspases has been most studied in cell bodies and typically results in the demise of the entire cell (Williams et al., 2006). However, recent evidence shows that caspases are also activated locally in neuronal processes and branches destined to be eliminated, for example in axons overshooting their targets that are subsequently pruned back to establish the precise adult connectivity (Finn et al., 2000; Raff et al., 2002; Luo and O’Leary, 2005; Buss et al., 2006b). Initially, axonal degeneration and axon pruning were thought to be independent of caspases (Finn et al., 2000; Raff et al., 2002). Later work in Drosophila melanogaster (Kuo et al., 2006; Williams et al., 2006) and in mammalian neurons (Plachta et al., 2007; Nikolaev et al., 2009; Vohra et al., 2010) demonstrated that interfering with the apoptotic balance or the executor caspases can prevent or at least delay axonal degeneration. Simon et al. (2012) have found that a caspase 9 to caspase 3 cascade is crucial for axonal degeneration induced by NGF withdrawal, with caspase 6 activation playing a significant but subsidiary role. Upstream of the caspases, BCL-2 family members such as BAX and BCL-X_l are required (Nikolaev et al., 2009; Vohra et al., 2010). It is conceivable that the failure of ipRGCs in BAX-deficient mice to form appropriate connections to other cells in the retina (Chen et al., 2013) may be in part attributable to defective axonal degeneration. Surprisingly, Apaf1 appears not to be involved in this process (Cusack et al., 2013), suggesting that axon degeneration depends on the concerted activation of the intrinsic initiator complex in a different way from apoptosis.Strikingly, a series of recent studies showed that several caspases and components of the intrinsic pathway also affect normal synaptic physiology in adulthood (Fig. 3, A–D). Here, pro-apoptotic proteins are predominantly involved in weakening the synapses, whereas the anti-apoptotic proteins have been mainly associated with synaptic strengthening (Fig. 3 B). In particular, caspase 3 promotes long-term depression (LTD), a stimulation paradigm that results in a period of decreased synaptic transmission (Li et al., 2010), and also prevents long-term potentiation (LTP), the converse situation leading to strengthened synaptic transmission (Jo et al., 2011). Likewise, the proapoptotic BCL-2 family members BAX and BAD stimulate LTD (Jiao and Li, 2011). By contrast, the anti-apoptotic protein BCL-X_l increases synapse numbers and strength (H. Li et al., 2008), and the inhibitor of apoptosis protein (IAP) family member survivin was reported to be involved in LTP in the hippocampus (Iscru et al., 2013) and in activity-dependent gene regulation (O’Riordan et al., 2008).Open in a separate windowFigure 3.Canonical and noncanonical functions of the apoptotic machinery. (A) The apoptotic machinery is not only involved in eliminating cells destined to die, but is also a central player in refining neuronal connectivity, by regulating synaptic transmission and by generating the adult connectivity through axon pruning (Luo and O’Leary, 2005; Hyman and Yuan, 2012). But how the canonical and noncanonical roles of the apoptotic machinery are interlinked and spatially restricted is not well understood. (B) In the adult nervous system, the pro-apoptotic proteins BAX, caspase 9, and caspase 3 promote weakening of synapses (long-term depression [LTD]; Li et al., 2010; Jiao and Li, 2011; Jo et al., 2011), while the anti-apoptotic proteins Bcl-X_l and the IAP survivin promote synaptic strengthening (long-term potentiation [LTP]; Li et al., 2008a; Iscru et al., 2013). It is unclear how the activation of these pathways is restricted to a single synapse, but a recent review suggested that the proteasomal degradation of activated caspases may prevent their diffusion (Hyman and Yuan, 2012). (C) Caspase activation is now known to be required for axon pruning during development to generate the adult refined connectivity (Luo and O’Leary, 2005; Simon et al., 2012). Different pathways are activated depending on the stimulus leading to degeneration. Growth factor deprivation during development leads to activation the executor caspases 3 and 6 (Simon et al., 2012) through the intrinsic apoptotic pathway, although its core protein Apaf1 does not seem to be required for this process (Cusack et al., 2013). On the other hand, a traumatic injury leads to reduced influx of NMNAT2 into the axon, which negatively affects the stability and function of mitochondria and leads to an increased calcium concentration (Wang et al., 2012). The effector caspase, caspase 6, is dispensable for this form of axonal degeneration (Vohra et al., 2010; Simon et al., 2012). Regulatory proteins such as the IAPs and also the proteasome seem to play a role in limiting the extent of activation to the degenerating part of the axon (Wang et al., 2012; Cusack et al., 2013; Unsain et al., 2013). (D) Simplified schematic of the main pro- and anti-apoptotic components. DISC, death-induced signaling complex. IAP, inhibitor of apoptosis protein. See Fig. 1 for details.These findings indicate that the apoptotic machinery acts at different levels in the cell, ranging from driving sub-lethal degradation of a compartment (Fig. 3 C) and attenuating synaptic transmission at the neuronal network level (Fig. 3 B) to destroying the entire cell during development or in disease (Fig. 3 D). How the cell spatially restricts the extent of activation of the apoptotic machinery is yet unclear. For example, elimination of the somata of developing neurons after neurotrophin deprivation is preceded by axonal degeneration, but not all instances of axonal degeneration lead to the death of the neuron (Campenot, 1977; Raff et al., 2002). Local regulation of caspase activation by IAPs is well established as a means for ensuring the elimination of neuronal processes in D. melanogaster (Kuo et al., 2006; Williams et al., 2006). Recent findings suggest a similar role for IAP in mammalian neurons, where it limits caspase activation to the degenerating axon (Fig. 3 C; Cusack et al., 2013; Unsain et al., 2013). The spontaneous mutation Wallerian degeneration slow (WldS; Lunn et al., 1989) has been instrumental to understand that trauma-induced axon degeneration is a regulated process different from, and independent of, cell body degeneration (Wang et al., 2012), but also distinct from axon pruning (Hoopfer et al., 2006). Work on the chimeric protein encoded by the WldS mutation also led to the identification of the protein NMNAT2 (nicotinamide mononucleotide adenylyltransferase 2) as a labile axon survival factor (Gilley and Coleman, 2010). How the WldS chimeric protein and NMNAT2 result in axon protection is unclear, but several lines of evidence seem to converge on local regulation of mitochondrial function and motility (Avery et al., 2012; Fang et al., 2012).Related to the spatial limiting of apoptotic activity is the question of how a local source of neurotrophins leads to the rescue of a developing peripheral neuron. When neurons encounter a source of neurotrophins, only the receptors close to the target will be activated, whereas the others, located further away, are not. The cell, therefore, needs to integrate a pro-survival signal from the activated receptors, and death-inducing signals from the nonactivated dependence receptors. The continued signaling of activated neurotrophin receptors that are retrogradely transported to the soma (Grimes et al., 1996; Howe et al., 2001; Wu et al., 2001; Harrington et al., 2011) likely play a role in counteracting the pro-apoptotic signaling proximal to the source of neurotrophins. It will be interesting to investigate whether similar mechanisms play a role in axon pruning and traumatic axon degeneration as well.

Programmed cell death in the adult brain

Most of the nervous system becomes post-mitotic early in development. In rodents, two brain areas retain the capacity to generate new neurons in the adult: the sub-ventricular zone, which generates neurons that migrate toward the olfactory bulb, and the sub-granular zone of the dentate gyrus of the hippocampus, where neurons are generated that integrate locally. Similar to what is observed during embryonic development, these adult-generated neurons are produced in excess, and a large fraction undergoes apoptosis when contacting its designated targets (Petreanu and Alvarez-Buylla, 2002; Kempermann et al., 2003; Ninkovic et al., 2007). Preventing apoptosis of adult-generated neurons in the olfactory bulb only has limited functional consequences (Kim et al., 2007), whereas a similar maneuver in the dentate gyrus does lead to impaired performance in memory tasks (Kim et al., 2009). Why superfluous hippocampal neurons would need to be eliminated for proper function is a matter of speculation, but may be linked with the fact that these are excitatory projection neurons, whereas in the olfactory bulb only axon-less inhibitory granule cells are integrated. The extent of survival in both these areas critically depends on the activity of the neuronal network in which these newly born neurons have to integrate (Petreanu and Alvarez-Buylla, 2002; Kempermann et al., 2006; Ninkovic et al., 2007). In this context, BDNF, the expression level of which is well known to be regulated by network activity, supports the survival of young adult–generated neurons and possibly even stimulates the proliferation of neural progenitors (Y. Li et al., 2008; Waterhouse et al., 2012). Interestingly, in young adult mouse mutants that exhibit spontaneous epileptic seizures, significantly higher levels of BDNF have been measured (Lavebratt et al., 2006; Heyden et al., 2011). Concomitantly, the entire hippocampal formation is considerably enlarged by as much as 40% (Lavebratt et al., 2006; Angenstein et al., 2007), which in turn is dependent on the epileptic seizures (Lavebratt et al., 2006). Whether or not there is a causal relationship between increased BDNF levels and hippocampal volume remains to be established.

Conclusion

Now that is has become clear that action of the apoptotic machinery can be limited spatially and temporally, several questions need to be addressed: how do neurons integrate intrinsic and extrinsic pro- and anti-apoptotic signals; and how they are spatially restricted to allow degradation of a dendrite or axon, or modulation of synaptic transmission? Another important issue is the regulation of cell death by intrinsic mechanisms in the central nervous system of vertebrates, not least because programmed cell death is observed in the CNS in a number of neurodegenerative diseases (Vila and Przedborski, 2003). Indeed, several of the central apoptotic components discussed here are also involved in these disorders (Hyman and Yuan, 2012). New insights in the regulation of programmed cell death in the developing nervous system may therefore continue to help to better understand the pathophysiological mechanisms of neurodegenerative disorders.     

AUXIN RESPONSE FACTOR17 Is Essential for Pollen Wall Pattern Formation in Arabidopsis

In angiosperms, pollen wall pattern formation is determined by primexine deposition on the microspores. Here, we show that AUXIN RESPONSE FACTOR17 (ARF17) is essential for primexine formation and pollen development in Arabidopsis (Arabidopsis thaliana). The arf17 mutant exhibited a male-sterile phenotype with normal vegetative growth. ARF17 was expressed in microsporocytes and microgametophytes from meiosis to the bicellular microspore stage. Transmission electron microscopy analysis showed that primexine was absent in the arf17 mutant, which leads to pollen wall-patterning defects and pollen degradation. Callose deposition was also significantly reduced in the arf17 mutant, and the expression of CALLOSE SYNTHASE5 (CalS5), the major gene for callose biosynthesis, was approximately 10% that of the wild type. Chromatin immunoprecipitation and electrophoretic mobility shift assays showed that ARF17 can directly bind to the CalS5 promoter. As indicated by the expression of DR5-driven green fluorescent protein, which is an synthetic auxin response reporter, auxin signaling appeared to be specifically impaired in arf17 anthers. Taken together, our results suggest that ARF17 is essential for pollen wall patterning in Arabidopsis by modulating primexine formation at least partially through direct regulation of CalS5 gene expression.In angiosperms, the pollen wall is the most complex plant cell wall. It consists of the inner wall, the intine, and the outer wall, the exine. The exine is further divided into sexine and nexine layers. The sculptured sexine includes three major parts: baculum, tectum, and tryphine (Heslop-Harrison, 1971; Piffanelli et al., 1998; Ariizumi and Toriyama, 2011; Fig. 1A). Production of a functional pollen wall requires the precise spatial and temporal cooperation of gametophytic and sporophytic tissues and metabolic events (Blackmore et al., 2007). The intine layer is controlled gametophytically, while the exine is regulated sporophytically. The sporophytic tapetum cells provide material for pollen wall formation, while primexine determines pollen wall patterning (Heslop-Harrison, 1968).Open in a separate windowFigure 1.Schematic representation of the pollen wall and primexine development. A, The innermost layer adjacent to the plasma membrane is the intine. The bacula (Ba), tectum (Te), and tryphine (T) make up the sexine layer. The nexine is located between the intine and the sexine layers. The exine includes the nexine and sexine layers. B, Primexine (Pr) appears between callose (Cl) and plasma membrane (Pm) at the early tetrad stage (left panel). Subsequently, the plasma membrane becomes undulated (middle panel) and sporopollenin deposits on the peak of the undulated plasma membrane to form bacula and tectum (right panel).After meiosis, four microspores were encased in callose to form a tetrad. Subsequently, the primexine develops between the callose layer and the microspore membrane (Fig. 1B), and the microspore plasma membrane becomes undulated (Fig. 1B; Fitzgerald and Knox, 1995; Southworth and Jernstedt, 1995). Sporopollenin precursors then accumulate on the peak of the undulated microspore membrane to form the bacula and tectum (Fig. 1B; Fitzgerald and Knox, 1995). After callose degradation, individual microspores are released from the tetrad, and the bacula and tectum continue to grow into exine with further sporopollenin deposition (Fitzgerald and Knox, 1995; Blackmore et al., 2007).The callose has been reported to affect primexine deposition and pollen wall pattern formation. The peripheral callose layer, secreted by the microsporocyte, acts as the mold for primexine (Waterkeyn and Bienfait, 1970; Heslop-Harrison, 1971). CALLOSE SYNTHASE5 (CalS5) is the major enzyme responsible for the biosynthesis of the callose peripheral of the tetrad (Dong et al., 2005; Nishikawa et al., 2005). Mutation of Cals5 and abnormal CalS5 pre-mRNA splicing resulted in defective peripheral callose deposition and primexine formation (Dong et al., 2005; Nishikawa et al., 2005; Huang et al., 2013). Besides CalS5, four membrane-associated proteins have also been reported to be involved in primexine formation: DEFECTIVE EXINE FORMATION1 (DEX1; Paxson-Sowders et al., 1997, 2001), NO EXINE FORMATION1 (NEF1; Ariizumi et al., 2004), RUPTURED POLLEN GRAIN1 (RPG1; Guan et al., 2008; Sun et al., 2013), and NO PRIMEXINE AND PLASMA MEMBRANE UNDULATION (NPU; Chang et al., 2012). Mutation of DEX1 results in delayed primexine formation (Paxson-Sowders et al., 2001). The primexine in nef1 is coarse compared with the wild type (Ariizumi et al., 2004). The loss-of-function rpg1 shows reduced primexine deposition (Guan et al., 2008; Sun et al., 2013), while the npu mutant does not deposit any primexine (Chang et al., 2012). Recently, it was reported that Arabidopsis (Arabidopsis thaliana) CYCLIN-DEPENDENT KINASE G1 (CDKG1) associates with the spliceosome to regulate the CalS5 pre-mRNA splicing for pollen wall formation (Huang et al., 2013). Clearly, disrupted primexine deposition leads to aberrant pollen wall patterning and ruptured pollen grains in these mutants.The plant hormone auxin has multiple roles in plant reproductive development (Aloni et al., 2006; Sundberg and Østergaard, 2009). Knocking out the two auxin biosynthesis genes, YUC2 and YUC6, caused an essentially sterile phenotype in Arabidopsis (Cheng et al., 2006). Auxin transport is essential for anther development; defects in auxin flow in anther filaments resulted in abnormal pollen mitosis and pollen development (Feng et al., 2006). Ding et al. (2012) showed that the endoplasmic reticulum-localized auxin transporter PIN8 regulates auxin homeostasis and male gametophyte development in Arabidopsis. Evidence for the localization, biosynthesis, and transport of auxin indicates that auxin regulates anther dehiscence, pollen maturation, and filament elongation during late anther development (Cecchetti et al., 2004, 2008). The role of auxin in pollen wall development has not been reported.The auxin signaling pathway requires the auxin response factor (ARF) family proteins (Quint and Gray, 2006; Guilfoyle and Hagen, 2007; Mockaitis and Estelle, 2008; Vanneste and Friml, 2009). ARF proteins can either activate or repress the expression of target genes by directly binding to auxin response elements (AuxRE; TGTCTC/GAGACA) in the promoters (Ulmasov et al., 1999; Tiwari et al., 2003). The Arabidopsis ARF family contains 23 members. A subgroup in the ARF family, ARF10, ARF16, and ARF17, are targets of miRNA160 (Okushima et al., 2005b; Wang et al., 2005). Plants expressing miR160-resistant ARF17 exhibited pleiotropic developmental defects, including abnormal stamen structure and reduced fertility (Mallory et al., 2005). This indicates a potential role for ARF17 in plant fertility, although the detailed function remains unknown. In addition, ARF17 was also proposed to negatively regulate adventitious root formation (Sorin et al., 2005; Gutierrez et al., 2009), although an ARF17 knockout mutant was not reported and its phenotype is unknown.In this work, we isolated and characterized a loss-of-function mutant of ARF17. Results from cytological observations suggest that ARF17 controls callose biosynthesis and primexine deposition. Consistent with this, the ARF17 protein is highly abundant in microsporocytes and tetrads. Furthermore, we demonstrate that the ARF17 protein is able to bind the promoter region of CalS5. Our results suggest that ARF17 regulates pollen wall pattern formation in Arabidopsis.     

Variability and Constancy in Cellular Growth of Arabidopsis Sepals

Growth of tissues is highly reproducible; yet, growth of individual cells in a tissue is highly variable, and neighboring cells can grow at different rates. We analyzed the growth of epidermal cell lineages in the Arabidopsis (Arabidopsis thaliana) sepal to determine how the growth curves of individual cell lineages relate to one another in a developing tissue. To identify underlying growth trends, we developed a continuous displacement field to predict spatially averaged growth rates. We showed that this displacement field accurately describes the growth of sepal cell lineages and reveals underlying trends within the variability of in vivo cellular growth. We found that the tissue, individual cell lineages, and cell walls all exhibit growth rates that are initially low, accelerate to a maximum, and decrease again. Accordingly, these growth curves can be represented by sigmoid functions. We examined the relationships among the cell lineage growth curves and surprisingly found that all lineages reach the same maximum growth rate relative to their size. However, the cell lineages are not synchronized; each cell lineage reaches this same maximum relative growth rate but at different times. The heterogeneity in observed growth results from shifting the same underlying sigmoid curve in time and scaling by size. Thus, despite the variability in growth observed in our study and others, individual cell lineages in the developing sepal follow similarly shaped growth curves.Cells undergo multiple rounds of growth and division to create reproducible tissues. In some plant tissues, such as expanding cotyledons, reproducibility can occur on a cellular level during specific intervals of development, where cotyledon cells exhibit uniform cellular growth (Zhang et al., 2011). However, several studies on cell division and growth in other developing plant tissues have demonstrated that plant cells exhibit considerable cell-to-cell variability during development (Meyer and Roeder, 2014). For example, in both the Arabidopsis (Arabidopsis thaliana) meristem and leaf epidermis, cells show spatiotemporal variation in individual cell growth rates (GRs; Asl et al., 2011; Elsner et al., 2012; Kierzkowski et al., 2012; Uyttewaal et al., 2012). Furthermore, cell divisions have been observed with marked randomness in their timing and orientation (Roeder et al., 2010; Besson and Dumais, 2011; Roeder, 2012). In this study, we identify a hidden, underlying pattern in the seemingly random GR (Box 1) of cells during the formation of sepals in Arabidopsis.Open in a separate windowBox 1.Definitions of GR terms. (For details on the calculations, see “Materials and Methods.”)Plant cell growth is defined as an increase in cell size due to an irreversible expansion of the cell wall. Neighboring cells physically accommodate one another during plant growth because their cell walls are glued together with a pectin-rich middle lamella, which prevents cell mobility. The cell wall is a thin, stiff layer composed of a polymer matrix including cellulose, hemicellulose, and pectin (Somerville et al., 2004; Cosgrove, 2005). Plant cells change their size and shape by modifying their turgor pressure and/or the mechanical properties of their walls, such as elasticity, plasticity, and extensibility. Growing plant cells exert forces on their neighbors through their walls, and cell wall stresses created by these forces feed back to alter the growth anisotropy (Hamant et al., 2008; Sampathkumar et al., 2014). Although these feedbacks can coordinate growth, they may also amplify differences in growth between neighboring cells (Uyttewaal et al., 2012).Two competing computational models have proposed explanations of the cellular heterogeneity observed in growing tissues by making different assumptions about how cells grow. In the first, it is assumed that relative growth rates (RGRs) of all cells are uniform in space and time, whereas variation in the timing of division causes the heterogeneity of cell sizes (Roeder et al., 2010). This model suggests that cell divisions cut the sepal into semiindependent cells, which grow uniformly within the expanding organ (Kaplan and Hagemann, 1991). The second model postulates the reverse process: timing of cell division is uniform, but cellular growth is variable and depends on the size of the cell (Asl et al., 2011). This model suggests that cells are autonomous. Currently, there is biological evidence for both models. Variability in cell division timing is observed in sepals and meristems, whereas variability in cellular GRs has been observed in leaves and meristem cells (Reddy et al., 2004; Roeder et al., 2010; Asl et al., 2011; Elsner et al., 2012; Kierzkowski et al., 2012; Uyttewaal et al., 2012). Thus, the debate on how the growth of individual cells within an organ relates to one another remains unresolved.The identification of underlying patterns in noisy cellular growth processes is challenging. Technical difficulties include the capability for cellular-resolution imaging of the tissue at sufficiently small time intervals. Previous studies (Zhang et al., 2011; Elsner et al., 2012; Kierzkowski et al., 2012) did not image and track individual cells, or they had a coarse time resolution, with 11- to 48-h intervals between images, which may have hidden important temporal dynamics. We studied growing cells in the Arabidopsis sepal, which allows for live imaging with cellular resolution at 6-h intervals (Roeder et al., 2010). The sepal is the leaf-like outermost floral organ of Arabidopsis (Fig. 1) with four sepals of stereotypical size produced per flower. Its accessibility for live imaging makes the sepal an excellent system for studying organogenesis (Roeder et al., 2010, 2011, 2012; Qu et al., 2014). Sepals exhibit high cellular variability in the timing of division and endoreduplication, an alternative cell cycle in which a cell replicates its DNA but fails to divide (Roeder et al., 2010). Furthermore, quantifying cell growth in sepals may shed light on growth mechanisms of other plant organs, such as leaves (Poethig and Sussex, 1985; Roeder et al., 2010).Open in a separate windowFigure 1.Diverse sizes of Arabidopsis sepal cells. A, Four sepals (s) are the outermost green leaf-like floral organs in Arabidopsis. B and C, Scanning electron micrographs of a mature Arabidopsis sepal show that the outer epidermal cells have a wide range of sizes. Asterisks mark some of the largest cells (giant cells) that can span 1/4 the length of the sepal. Scale = 100 µm.Another key challenge in analyzing cellular growth is the identification of trends in noisy data. Inaccuracies in data acquisition, such as segmentation errors, and noisy growth of individual cells can hide meaningful spatiotemporal trends in growth. GRs measured over longer time intervals will have reduced noise, but they may also obscure important temporal dynamics. Alternatively, previous studies have examined growth of the whole organ or its subregions to avoid cellular noise (De Veylder et al., 2001; Mündermann et al., 2005; Rolland-Lagan et al., 2005, 2014; Kuchen et al., 2012; Remmler and Rolland-Lagan, 2012). However, precise cellular patterns are not resolved. In our study, we use cellular resolution data to define spatially averaged kinematics while keeping the full temporal resolution to identify course-grained spatial trends in the dynamics of cellular growth (Box 1).We analyze the relationships among the growth of individual cell lineages in a developing Arabidopsis sepal by live imaging and computational analyses. We have developed continuous low-order displacement fields to represent the spatially averaged kinematics of the sepal (Box 1). We find that the growth of the tissue surface area, cell lineage area, and wall length follows S curves, suggesting that their GRs vary over time. Additionally, we find that there is a linear correlation between the maximum GR (i.e. size increase per hour) and the size of the cell. We furthermore find that each sepal cell lineage reaches the same maximum RGR (i.e. GR divided by size). However, each cell reaches the maximum RGR at a different time during its development, generating the observed heterogeneity. Thus, we find underlying similarities in the growth curves of sepal cells.     

Aneuploidy in health,disease, and aging

Synapse formation is a highly regulated process that requires the coordination of many cell biological events. Decades of research have identified a long list of molecular components involved in assembling a functioning synapse. Yet how the various steps, from transporting synaptic components to adhering synaptic partners and assembling the synaptic structure, are regulated and precisely executed during development and maintenance is still unclear. With the improvement of imaging and molecular tools, recent work in vertebrate and invertebrate systems has provided important insight into various aspects of presynaptic development, maintenance, and trans-synaptic signals, thereby increasing our understanding of how extrinsic organizers and intracellular mechanisms contribute to presynapse formation.Chemical synapses are highly specialized, asymmetric intercellular junction structures that are the basic units of neuronal communication. Proper development of synapses determines appropriate connectivity for the assembly of functional neuronal circuits. Synaptic circuits arise during development through a series of intricate steps (Waites et al., 2005; McAllister, 2007; Jin and Garner, 2008). First, spatiotemporal cues guide axons through complex cellular environments to contact their appropriate postsynaptic targets. At their destination, synapse formation is specified and initiated through adhesive interactions between synaptic partner cells or by local diffusible signaling molecules. Stabilization of intercellular contacts and assembly into functional synapses involves cytoskeletal rearrangements, aggregation, and insertion of pre- and postsynaptic components at nascent synaptic sites. Maturation and modulation of these newly formed synapses can then occur by altering the organization or composition of synaptic proteins and post-translational modifications to achieve its required physiological responsiveness (Budnik, 1996; Lee and Sheng, 2000). Conversely, retraction of contacts and elimination of inappropriate synaptic proteins help to refine the neuronal circuitry (Goda and Davis, 2003; Sanes and Yamagata, 2009).Over the last decade, new insights have furthered our understanding of synapse development through the identification of new molecular players and by advanced imaging technology that has allowed for high-resolution inspection of the dynamics and relative positions of synaptic proteins. This review will highlight recent results on the development of presynaptic specializations, and the roles of trans-synaptic organizers, intracellular synaptic proteins, and the cytoskeleton during the formation and maintenance of synapses.

Axonal transport of synaptic vesicle and active zone proteins

After cell fate determination and morphogenesis, neurons continue to differentiate by entering the phase of synapse formation. Most synaptic material required for this process is synthesized in the cell body of neurons and transported to synapses by microtubule (MT)-based molecular motors (Fig. 1). MTs are intrinsically polarized filaments with a plus and a minus end (Fig. 1 B). MT-based molecular motors use this polarity to transport cargoes to specific cellular locations. Examination of MTs by electron microscopy in dissociated cultured neurons showed that the organizations of MTs is different in axon and dendrite (Baas et al., 1988, 2006). In axons, all microtubules have their minus ends oriented toward the cell body and their plus ends extend distally. On the contrary, the MT polarity in dendrites is mixed. Recent studies tracking the movement of end-binding MT-capping proteins confirmed these results in vivo. Specifically, axonal MTs are uniformly organized with their plus ends pointing distally in all organisms. Dendrites of vertebrate neurons show more plus end–out MTs in vivo, whereas flies and worms have more minus end–out MTs in dendrites (Stepanova et al., 2003; Rolls et al., 2007; Stone et al., 2008).Open in a separate windowFigure 1.Regulatory steps during polarized motor-based transport of synaptic material. (A) At the Golgi apparatus, synaptic proteins have to be sorted into appropriate vesicles. These vesicles and other cargo such as mitochondria get loaded onto specific motor proteins. (B) Establishment of proper microtubule polarity along the axon determines anterograde and retrograde trafficking by plus end– and minus end–directed motor proteins such as kinesins and dynein. (C) At the appropriate destination, motor-cargo unloading occurs in a regulated fashion to achieve the appropriate distribution of synaptic boutons. At synapses, synaptic vesicle precursors give rise to mature synaptic vesicles. Proteins required for the SV cycle and trans-synaptic adhesion coalesce into the active zone (AZ) underneath the plasma membrane juxtaposed against the postsynaptic membrane.Does the difference in microtubule organization and polarity help to segregate synaptic cargoes between axons and dendrites? Recent studies have started to identify some molecules that create these differences in MT polarity in different neuronal subcellular compartments and show how disruption of their function affects synapse formation. For example, a recent paper showed that kinesin-1 is required to establish the predominantly minus end–out organization in the dendrites of Caenorhabditis elegans motor neurons (Yan et al., 2013). In kinesin-1/unc-116 mutants, dendrites adopt the axon-like MT polarity causing presynaptic cargoes to mislocalize into dendrites (Seeger and Rice, 2010; Yan et al., 2013). Similarly, loss of the MT-binding CRMP protein UNC-33 or the actin–spectrin adaptor protein ankyrin/UNC-44 in worms also results in MT polarity defects, which also results in ectopic localization of synaptic vesicles and active zone proteins into dendrites (Maniar et al., 2012). These results support the idea that MT polarity ensures the faithful targeting of presynaptic components to the axon. However, another way motors can distinguish between axons and dendrites is through MT-associated proteins (MAPs). In a recent study, Banker and colleagues showed that plus end–orienting kinesins can differentiate axon and dendrite, likely due to specific MT-binding proteins in these compartments (Huang and Banker, 2012).The direct regulation of motor activity by MTs or synaptic vesicle–associated proteins is likely to contribute to the trafficking of synaptic cargoes. Doublecortin, a MAP, binds to kinesin-3/KIF1A to affect the trafficking of the synaptic vesicle protein, synaptobrevin, in hippocampal neurons by altering the affinity of ADP-bound KIF1A to MTs (Liu et al., 2012). The Rab3 guanine nucleotide exchange factor, DENN/MADD, functions as an adaptor between kinesin-3 and GTP-Rab3–containing synaptic vesicles to promote the trafficking of synaptic vesicles in the axon (Niwa et al., 2008).Precise regulation of motor-based transport ensures that synaptic cargoes are delivered to and maintained at synapses. Several recent studies have provided evidence that two postmitotic cyclin-dependent kinases are important regulators of anterograde and retrograde trafficking of presynaptic cargoes. The kinase CDK-5 is required in many aspects of nervous system function. In the context of presynaptic development and function, CDK-5 has been shown to regulate the transport of synaptic vesicles and dense core vesicles, which contain neuropeptides, by inhibiting a dynein-mediated pathway that mobilizes presynaptic components to the somatodendritic compartments in C. elegans neurons (Ou et al., 2010; Goodwin et al., 2012). A paralogue of CDK-5, the PCT-1 kinase acts in a partially redundant pathway to prevent the mislocalization of presynaptic material to dendrites. In animals lacking both kinases or their activators, synaptic cargoes completely mislocalize to the dendrites, leaving an “empty” axon (Ou et al., 2010). Vertebrate CDK-5 also plays profound roles in the regulation of synaptic vesicle pools by modifying Ca²⁺ channels. Genetic ablation or pharmacological inhibition of CDK-5 increases the pool of synaptic vesicles that are docked at the active zone, termed the readily releasable pool, and potentiates synaptic function (Kim and Ryan, 2010, 2013). These results suggest that CDK-5 and its paralogue control local and global vesicle pools. Regulation of the exchange between these pools can affect membrane trafficking at presynaptic terminals as well as the overall polarity of neurons.To form synapses at defined locations, cargoes not only need to know how to “get on” the transport system but also need to know where to precisely “get off” at their destination (Fig. 1 C). Loss of a conserved small G-protein of the Arf-like family, ARL-8, in C. elegans, resulted in premature exit of synaptic cargoes during transport and showed ectopic aggregations of synaptic vesicles in the proximal axon. This causes a reduction in the number but an increase in the size of synapses (Klassen et al., 2010). ARL-8 localizes to both stable and trafficking synaptic vesicles and promotes trafficking by increasing kinesin-3 activity and suppressing aggregation-induced stoppage of synaptic cargoes along the axon (Wu et al., 2013). Hence, the balance between motor activity and aggregation propensity of trafficking cargoes may determine the number, size, and location of presynaptic terminals. Interestingly, the small GTPase Rab3, which normally associates with synaptic vesicles, has recently been shown to affect the distribution of active zone proteins at fly neuromuscular junction (NMJ) synapses, further suggesting that the trafficking of synaptic vesicles and formation of active zones are linked (Graf et al., 2009).Besides synaptic material, another major organelle cargo that is often present at the presynaptic terminal is mitochondria. The Milton–Miro complex functions as an adaptor between kinesin-1 and mitochondria to support axonal transport of mitochondria. Interestingly, the coupling of the Milton–Miro complex to kinesin is regulated by Ca²⁺ (Macaskill et al., 2009; Wang and Schwarz, 2009), providing a mechanism for neuronal activity controlling transport of mitochondria along the axon.Previous studies have suggested that components of the presynaptic active zone are transported in a preassembled form by Piccolo-Bassoon transport vesicles (PTVs) that may contain multiple components required to build a synapse (Zhai et al., 2001; Shapira et al., 2003). Recent studies found that Golgi-derived PTVs contain many active zone proteins including Piccolo, Bassoon, RIM1α, and ELKS2/CAST, but lack another active zone component, Munc-13, which may exit the Golgi on separate vesicles (Maas et al., 2012). Packing of various active zone components that have the propensity to self-assemble into separate vesicles may contribute a way to control synaptogenesis. This is interesting in light of the finding that Munc-13 can function as a protein scaffold for Bassoon and ELKS2 (Wang et al., 2009). The link between trafficking of synaptic vesicle and active zone components is not well understood. In vivo time-lapse imaging of synaptic vesicle and active zone trafficking showed that these components, possibly in the form of dense core vesicles, could be trafficked together in C. elegans neurons, suggestive of prepackaged presynaptic material during transport (Wu et al., 2013). Taken together, axonal transport of synaptic components is a necessary step for synapse formation and maintenance. The regulation of MTs, molecular motors, and synaptic cargoes ensure the targeting of appropriate proteins to synapses.

Role of the actin cytoskeleton in presynaptic assembly

Although MT-mediated transport is critical for long-range trafficking, actin-based mechanisms often organize local protein complexes in subcellular domains. A large body of work has described the role of the actin cytoskeleton in postsynaptic structure and function (Schubert and Dotti, 2007; Hotulainen and Hoogenraad, 2010). We will focus on more recent work that has highlighted the importance of the actin cytoskeleton in presynaptic formation.F-actin is required for presynaptic assembly during the early stages of synaptogenesis. Depolymerization of F-actin in young hippocampal neuronal cultures results in a reduction in the size and number of synapses. This effect was not seen with older cultures when synapses are more mature (Zhang and Benson, 2001). This observation correlates with an increase in both pre- and postsynaptic F-actin levels in newly formed synapses compared with mature synapses (Zhang and Benson, 2002).F-actin has been implicated in many steps of synapse assembly and function (Fig. 2; Cingolani and Goda, 2008). One of the roles that has been proposed for F-actin is to act as a scaffold for other presynaptic proteins (Sankaranarayanan et al., 2003). A recent study identified an F-actin–binding active zone molecule Neurabin/NAB-1 that is recruited by a presynaptic F-actin network (Chia et al., 2012). In addition, knockdown of Rac/Cdc42 GTPase exchange factor β-Pix resulted in a decrease in actin at synapses with a concomitant loss of synaptic vesicle clustering (Sun and Bamji, 2011). These studies demonstrate that F-actin at presynaptic sites can recruit and stabilize presynaptic components.Open in a separate windowFigure 2.Assembling the presynaptic active zone. Scaffolding proteins including Liprin, SYD-1, ELKS, Neurabin, Piccolo, and Bassoon form the dense protein network in the presynaptic cytomatrix that facilitates synaptic vesicle docking and fusion. The presynaptic F-actin networks are required for presynaptic assembly and maintenance.Studies of Drosophila NMJs have found that the presynaptic spectrin–actin cytoskeleton is important for synapse stability. Loss of presynaptic spectrin led to retraction of synapses (Pielage et al., 2005). Intriguingly, loss of postsynaptic spectrin increased the total number of the active zone specializations, termed T-bars, and affected the size and distribution of presynaptic sites. Thus, the spectrin cytoskeleton can impose a trans-synaptic influence on synapse development (Pielage et al., 2006).Given the importance of F-actin at synapses, it is crucial to understand the signaling pathways that instruct F-actin organization. Multiple studies have shown that signaling from synaptic cell adhesion molecules can lead to cytoskeletal rearrangements at synapses. Adhesion of hippocampal neurons to syndecan-2–coated beads is sufficient to induce F-actin clustering and downstream formation of presynaptic boutons (Lucido et al., 2009). In mice, the adhesion molecule L1CAM may bind to spectrin–actin adaptor ankyrin to mediate GABAergic synapse formation (Guan and Maness, 2010). Another adhesion molecule of the immunoglobulin superfamily SYG-1 in C. elegans has also been shown to be necessary and sufficient to recruit F-actin to synapses (Chia et al., 2012). In a recent study, secreted bone morphogenetic protein (BMP) can signal in a retrograde fashion to regulate Rac-GEF Trio expression in presynaptic neurons, which is important for controlling synaptic growth (Ball et al., 2010).Interestingly, presynaptic active zone proteins can also affect F-actin assembly (Fig. 2). Knockdown of Piccolo reduced activity-dependent assembly of F-actin at synapses and enhanced dispersion of Synapsin1a and synaptic vesicles in hippocampal neurons. Loss of Piccolo also resulted in a loss of Profilin 2, a regulator of actin polymerization (Waites et al., 2011).Various studies have begun to shed light on the actin regulators required for synaptic F-actin establishment and maintenance. Diaphanous, a formin-related gene that associates with barbed ends of F-actin, was found to function downstream of presynaptic receptor Dlar at fly NMJs. Spectrin–actin capping protein, Adducin, is enriched at presynaptic sites and is required to prevent synapse retraction and elimination (Bednarek and Caroni, 2011; Pielage et al., 2011). Activators of the Arp2/3 complex, WASP and WAVE, have also been implicated in the regulation of F-actin at synapses (Coyle et al., 2004; Stavoe et al., 2012; Zhao et al., 2013). This diversity of F-actin modulators suggests that there are probably different F-actin structures at different stages of development or even in subcellular domains within the synapse. This is supported by observations that F-actin can localize with synaptic vesicles, at the active zone and in the perisynaptic region (Bloom et al., 2003; Sankaranarayanan et al., 2003; Waites et al., 2011; Chia et al., 2012). Thus, much remains to be done in our understanding how distinct F-actin structures are formed and regulated to mediate various processes during synapse assembly and maintenance.

Assembly of the molecular network at presynaptic terminals

Although F-actin might help to initiate the presynaptic assembly process, many other ensuing molecular interactions are required to form the mature presynaptic apparatus (Fig. 2). The presynaptic active zone is comprised of a framework of scaffolding proteins that function as protein-binding hubs for other presynaptic components. Piccolo and Bassoon are important vertebrate multidomain proteins that traditionally have been widely used as active zone markers. Recent electrophysiology data on Piccolo mutant and Bassoon knockdown neurons showed that these molecules are dispensable for synaptic transmission but affect synaptic vesicle clustering (Mukherjee et al., 2010). Furthermore, Piccolo and Bassoon were found to be required for maintaining synapse integrity by regulating ubiquitination and degradation of presynaptic components (Waites et al., 2013).Forward genetic approaches in worms and flies have made important contributions to our understanding of the presynaptic cytomatrix. Studies have found that two active zone scaffolding molecules, SYD-1 and Liprin-α/SYD-2, are required for proper synapse formation (Zhen and Jin, 1999; Patel et al., 2006; Astigarraga et al., 2010; Owald et al., 2010; Stigloher et al., 2011). Interestingly, at fly NMJs, SYD-1 is necessary for clustering presynaptic neurexin that in turn clusters postsynaptic neuroligin (Owald et al., 2012). The presynaptic assembly function of SYD-1 and SYD-2 appears to be conserved because mutation analysis of mammalian SYD-1 and knockdown of Liprin-α both caused defects in presynaptic development and function (Spangler et al., 2013; Wentzel et al., 2013). In flies, the active zone T-bar structure is comprised of ERC/CAST family protein bruchpilot (brp) as the major active zone organizing protein (Fouquet et al., 2009). Brp is not only present at the active zone but also plays important scaffolding roles in localizing Ca²⁺ channels. In C. elegans, the Brp homologue ELKS-1 is also localized to the active zone; however, the importance of ELKS-1 during development of synapses was only revealed in sensitized genetic backgrounds (Dai et al., 2006; Patel and Shen, 2009), suggesting that there are likely redundant molecular pathways for presynaptic assembly. In the vertebrate system, loss of one of the three ELKS genes, surprisingly, caused an increase in the inhibitory synaptic transmission (Kaeser et al., 2009). Besides Brp, Rab3-interacting molecule (RIM) binding protein (RBP) was found to be important for active zone structural integrity in flies. Using super-resolution microscopy, RBP was found to surround Ca²⁺ channels at T-bars and loss of RBP resulted in defective Ca²⁺ channel clustering and reduced evoked neurotransmitter release (Liu et al., 2011).Assembly of the presynaptic active zone is subjected to several layers of regulation. The assembly process is balanced by inhibitory mechanisms that control the number and size of synapses. Loss of the E3 ubiquitin ligase Highwire/RPM-1 results in an increased number of synaptic boutons in flies and multiple active zones in worms (Wan et al., 2000; Zhen et al., 2000). Working together with F-box protein FSN-1, RPM-1 down-regulates the DLK MAP kinase signaling pathway (Liao et al., 2004; Nakata et al., 2005; Yan et al., 2009). Another E3 ubiquitin ligase, the SKP complex, has been shown to eliminate transient synapses during development in worms (Ding et al., 2007). Therefore, ubiquitin-mediated mechanisms play important roles in controlling the presynaptic assembly program.Other inhibitory mechanisms include SRPK79D, a serine–arginine protein kinase discovered in flies that represses T-bar formation (Johnson et al., 2009). In the mutant, the T-bar component Brp is ectopically accumulated in the axonal shaft. Regulator of synaptogenesis, RSY-1, limits the extent of presynaptic assembly by directly binding to active zone scaffold molecule Liprin-α/SYD-2 and SYD-1 (Patel and Shen, 2009). In addition, Liprin-α/SYD-2 may inhibit its own activity via intramolecular interactions (Taru and Jin, 2011; Chia et al., 2013).Taken together, the presynaptic assembly process driven by scaffolding molecules is controlled by complex inhibitory mechanisms to achieve the appropriate extent of aggregation in the process of synapse formation.

Trans-synaptic signals orchestrate pre- and postsynaptic formation

Coordinated pre- and postsynaptic development requires the precise apposition of presynaptic components to postsynaptic specializations. It is conceivable that signals from pre and postsynaptic sides functioning across the synaptic cleft coordinate synaptic differentiation reciprocally. Although a vast assortment of factors have been identified as synaptic organizers, the fact that genetic ablation of some synaptic organizers in vivo fails to elicit dramatic synaptic defects suggests the incomplete view of the trans-synaptic signaling. Moreover, the underlying mechanisms and the cross talk of these signaling pathways are still unclear. In recent years, an emerging body of literature has begun to shed light on trans-synaptic signaling and the importance of environmental cues in synapse formation.

Adhesion proteins instruct synaptic differentiation

A large body of literature suggests that trans-synaptic interactions between synaptic adhesion molecules function bi-directionally for synapse formation and maturation (Fig. 3). Neurexin–neuroligin is the first pair to be shown to induce pre- and postsynapse formation (Scheiffele et al., 2000; Graf et al., 2004; Chih et al., 2005; Nam and Chen, 2005; Chubykin et al., 2007). Recent in vitro studies have unveiled more components interacting with neurexin or neuroligin in specific synaptic differentiation events (Fig. 3, B and C). In early developmental stages, a secreted synaptic organizer, thrombospondin 1 (TSP1, see next section) increases the speed of synaptogenesis through neuroligin 1 (Xu et al., 2010). At excitatory synapses, a retrograde signaling controls synaptic vesicle clustering, neurotransmitter release, and presynaptic maturation by cooperation of neuroligin and N-cadherin (Wittenmayer et al., 2009; Stan et al., 2010; Aiga et al., 2011). A leucine-rich repeat transmembrane (LRRTM) protein family was also identified as an organizer of the function of excitatory synapses through interactions with neurexin (Linhoff et al., 2009). Further studies showed that binding of LRRTMs and neuroligins to neurexin acts redundantly to maintain excitatory synapses by preventing activity and Ca²⁺-dependent synapse elimination during early development, while performing divergent functions upon synapse maturation (de Wit et al., 2009; Ko et al., 2009, 2011; Soler-Llavina et al., 2011).Open in a separate windowFigure 3.Adhesive trans-synaptic signalings orchestrate excitatory and inhibitory synaptic assembly. Multiple pairs of trans-synaptic adhesion molecules organize synaptic differentiation and function on both pre- and postsynaptic sites. Note that different adhesion molecules are used at excitatory and inhibitory synapses. LPH1, latrophilin 1; α-DG, α-dystroglycan; β-DG, β-dystroglycan; S-SCAM, synaptic scaffolding molecule; Lasso, LPH1-associated synaptic surface organizer; IL-1RAcp, interleukin-1 receptor accessory protein.The function of neurexin and neuroligin in mediating synaptic differentiation has also been shown at Drosophila NMJs and mammalian CNS. In mammalian, although neither compound knockout of three neurexins nor two individual neuroligin knockout mice display severe defects in the number or morphology of synapses (Missler et al., 2003), the deletion of either neurexin or neuroligin affects the neurotransmitter release and in turn impairs the relevant behavior (Zhang et al., 2005; Blundell et al., 2009, 2010; Etherton et al., 2009; Jedlicka et al., 2011). Neurexin loss of function in fly leads to reduced number and defective morphology of synaptic boutons and active zones from early developmental stages (Li et al., 2007; Chen et al., 2010). In contrast, deletion of either neuroligin 1 or 2 causes NMJ defects and alternations of active zones only in the larval stage, indicating that they function mainly in the expansion of NMJs during development (Banovic et al., 2010; Sun et al., 2011). These abnormalities further impair synaptic transmission at the NMJs (Li et al., 2007; Banovic et al., 2010; Chen et al., 2010; Sun et al., 2011). Moreover, these phenotypes are enhanced when the Teneurin family of adhesion molecules is deleted, suggestive of functional redundancy between adhesion molecules (Mosca et al., 2012). Recently, it has been reported that an active zone protein, SYD-1, is required for the formation and function of the neurexin–neuroligin complex in flies (Fig. 2; Owald et al., 2012), providing an example of how trans-synaptic neurexin–neuroligin signaling orchestrates synaptic assembly bi-directionally. Interestingly, at postsynaptic sites, the NMDA receptor activity-triggered Ca²⁺-dependent cleavage of neuroligin 1 was found to destabilize presynaptic neurexin, reduce presynaptic release probability, and depress synaptic transmission (Peixoto et al., 2012). This observation raises a possibility that neurexin and neuroligin could fine-tune synaptogenesis both positively and negatively.Although Drosophila neuroligin and neurexin mutants share many phenotypes in synaptic differentiation, there are some unique features for each mutant, suggesting that they play distinct roles. For example, some aspects of synaptic specificity are achieved by different pairs of neurexin–neuroligin interactions. Neuroligin 1 promotes the growth and differentiation of excitatory synapses by binding to PSD-95, whose amount balances the ratio of excitatory-to-inhibitory synaptic specializations (Prange et al., 2004; Banovic et al., 2010). Neuroligin 2, on the contrary, binds to a scaffold protein gephyrin at inhibitory synapses, instructing inhibitory postsynaptic assembly (Fig. 3, B and C; Poulopoulos et al., 2009). Different isoforms of neurexin also contribute to the differentiation of excitatory and inhibitory synapses (Fig. 3, B and C; Chih et al., 2006; Graf et al., 2006; Kang et al., 2008).Other novel trans-synaptic interactions have also been identified to organize synaptic differentiation (Fig. 3, B and C). For example, Netrin-G ligand 3 (NGL-3), localized at postsynaptic region, induces excitatory synaptic differentiation by interacting with the receptor tyrosine phosphatase LAR family proteins, including PTPδ and PTPσ (Woo et al., 2009; Kwon et al., 2010). PTPδ can also trans-interact with Slitrk3 and IL-1 receptor accessory protein (IL-1RAcP) to promote presynaptic formation (Takahashi et al., 2012; Yoshida et al., 2012). Molecules that function in other neuronal developmental processes have also been shown to regulate synaptic differentiation. Farp1, essential for the dynamics of dendritic filopodia, regulates postsynaptic development and triggers a retrograde signal promoting active zone assembly by binding to SynCAM 1 (Cheadle and Biederer, 2012). Teneurins, instructing synaptic partner selection in fly olfactory system (Hong et al., 2012), act in synaptogenesis through trans-synaptic interaction at NMJs (Mosca et al., 2012). Another splice variant of a postsynaptic Teneurin-2 in rat, Lasso, binding with presynaptic Latrophilin 1 (LPH1), induces presynaptic Ca²⁺ signals and regulates synaptic function (Silva et al., 2011). Neural activity is also involved in controlling the growth of the presynapse. Conditioning or BDNF application induces presynaptic bouton development via an ephrin-B–dependent manner (Li et al., 2011), suggesting the role of EphB/ephrin-B signaling in activity-dependent synaptic modification.

Secreted molecules organize synapse differentiation

In addition to adhesion molecules, some secreted molecules also serve as synaptic organizers (Fig. 4). For example, the motor neuron–derived ligand agrin, which was the first identified secreted organizing molecule for postsynaptic differentiation, activates MuSK, a postsynaptic receptor tyrosine kinase, to regulate NMJ specialization (Glass et al., 1996; Zhou et al., 1999). Recently, a low-density lipoprotein receptor–related protein, LRP4, was identified as the co-receptor of agrin, forming a complex with MuSK and mediating MuSK signaling (Kim et al., 2008; Zhang et al., 2008). Several Wnts appears to act together with agrin to activate the LRP4–MuSK receptor complex to promote postsynaptic differentiation (Jing et al., 2009; Zhang et al., 2012). LRP4 also acts as a direct retrograde signal, functioning independently of MuSK for presynaptic differentiation (Yumoto et al., 2012), demonstrating that LRP4 acts as a bi-directional synaptic organizer (Fig. 4, left).Open in a separate windowFigure 4.Secreted trans-synaptic signaling at NMJs and CNS synapses. (Left) At Drosophila neuromuscular junctions (NMJs), Wnts are secreted from presynaptic terminals in association with Evi in the form of exosomes. In vertebrate NMJs, Wnt binds to the Agrin–LRP4–MuSK complex to regulate synapse formation. (Right) At CNS synapses, glia-derived thrombospondins (TSPs) and presynaptic neuron–derived cerebellin (Cbln) organize synapse differentiation and formation bi-directionally through binding to GluD2 and an isoform of neurexin (S4+) on the postsynaptic and presynaptic membranes, respectively. LTCC, L-type Ca²⁺ channel complex; AChR, acetyl choline receptor.Wnt is another well-characterized signaling molecule regulating many developmental processes including synaptic differentiation bi-directionally. Wnt regulates synaptic assembly both positively and negatively. For example, Wnt3 collaborates with agrin to promote the clustering of acetyl choline receptor (AChR) at the vertebrate NMJs (Henriquez et al., 2008), while Wnt3a inhibits AChR aggregation through β-catenin signaling (Wang et al., 2008). In the C. elegans NMJ, a Wnt molecule, CWN-2, stimulates the delivery and insertion of AchR to the postsynaptic membrane through the activation of a Frizzled–CAM-1 receptor complex (Jensen et al., 2012). Local Wnt gradient can suppress synapse formation in both C. elegans and Drosophila (Inaki et al., 2007; Klassen and Shen, 2007). Interestingly, in these contexts, Wnts are secreted from nonneuronal or nonsynaptic partner cells, suggesting that environmental factors can shape synaptic connections. Wnt can also be secreted from presynaptic neurons. A recent study demonstrated the trans-synaptic transmission of Wnt by exosome-like vesicles containing the Wnt-binding protein Evi at Drosophila NMJs (Fig. 4, left; Korkut et al., 2009; Koles et al., 2012). Presynaptic vesicular release of Evi is required for the secretion of Wnt. Intriguingly, different Wnt ligands regulate synapse formation in distinct cellular contexts. Wnt3a promotes excitatory synaptic assembly through CaMKII, whereas Wnt5a mediates inhibitory synapse formation by stabilizing GABA_A receptors (Cuitino et al., 2010; Ciani et al., 2011). This functional diversity indicates that different Wnts, receptors, and downstream pathways, as well as cell-specific contexts dictate the action of extracellular cues. Another conserved secreted molecule, netrin/UNC-6, can also pattern synapses by either promoting or inhibiting synapse formation (Colón-Ramos et al., 2007; Poon et al., 2008). Because Wnt and netrin often exist in gradients, these observations suggest that the localization of synapses can be specified by the gradient of extrinsic cues.In mammalian, several glia-derived cues have been shown to play important roles in regulating synapse formation or elimination. Thrombospondins (TSPs) are trans-synaptic organizers secreted from immature astrocytes (Christopherson et al., 2005). Both in vitro and in vivo data demonstrate the capacity of TSPs to increase synapse number, promote the localization of synaptic molecules, and refine the pre- and postsynaptic alignment (Christopherson et al., 2005; Eroglu et al., 2009). Recently, two transmembrane molecules were uncovered in mediating TSP-induced synaptogenesis (Fig. 4, right). Neuroligin 1 interacts with TSP1 with its extracellular domain mediating the acceleration of synaptogenesis in hippocampal neurons (Xu et al., 2010). α2δ-1, a subunit of the L-type Ca²⁺ channel complex (LTCC), was also identified as the postsynaptic receptor of TSP in excitatory CNS neurons (Eroglu et al., 2009). Interaction between TSP and α2δ–1 triggers the conformational changes and sequentially recruits synaptic scaffolding molecules and initiates synapse formation (Eroglu et al., 2009). Interestingly, TSP-induced synapses, although structurally normal and presynaptically active, are postsynaptically silent due to the lack of AMPA receptors (Christopherson et al., 2005), indicating the existence of other glia-derived signals involved in synapse formation. In fact, in cultured hippocampal neurons, a glia-derived neurotrophic factor GDNF enhances the pre- and postsynaptic adhesion by triggering the trans-homophilic interaction of its receptors GFRα1 localized at both pre- and postsynaptic sites (Ledda et al., 2007). Several other glia-derived factors have been shown to play critical roles in synaptogenesis. Astrocytes secrete extracellular molecules hevin and SPARC to regulate synapse formation in vitro and in vivo (Kucukdereli et al., 2011). Astrocytes also express a transmembrane adhesion protein, protocadherin-γ, serving as a local cue to promote synapse formation (Garrett and Weiner, 2009). TGF-β secreted from the NMJ glia acts together with the muscle-derived TGF-β to control synaptic growth (Fuentes-Medel et al., 2012). In a similar fashion, secretion of BDNF by vestibular supporting cells is required for synapse formation between hair cells and sensory organs (Gómez-Casati et al., 2010).Another important synaptic organizer is cerebellin (Cbln), a presynapse-derived complement protein, C1q-like family protein. In cbln1-null mice the number of parallel fibers (PF)–Purkinje synapses is dramatically reduced; the postsynaptic densities in the remaining synapses are larger than the apposite active zones (Hirai et al., 2005). Cbln was also found to regulate synaptic plasticity, as cbln1-null mice show impaired long-term depression in cerebellum (Hirai et al., 2005). These defects precisely resemble those in mice lacking a putative glutamate receptor, GluD2 (Kashiwabuchi et al., 1995; Kurihara et al., 1997), suggesting that Cbln1 and GluD2 function in synaptic differentiation through a common pathway. Interestingly, the C-terminal domain and N-terminal domains of GluD2 are indispensable for cerebella LTD and PF–Purkinje synaptic morphology, respectively (Kohda et al., 2007; Uemura et al., 2007; Kakegawa et al., 2008, 2009). Further studies suggested that Cbln1 directly binds to the N-terminal domain of GluD2 and recruits postsynaptic proteins by clustering GluD2 (Matsuda et al., 2010). Neurexin was recently reported as the presynaptic receptor of Cbln in promoting synaptogenesis (Uemura et al., 2010), which reinforces the understanding of Cbln-mediated trans-synaptic signaling: Cbln serves as a bi-directional synapse organizer by linking presynaptic neurexin and postsynaptic GluD2 (Fig. 4, right).Besides being required for synapse formation at early stages, genetic ablation of GluD2 in adult cerebellum leads to loss of PF–Purkinje synapses (Takeuchi et al., 2005), indicating that Cbln1–GluD2 signaling is also important for the maintenance of PF–Purkinje synapses. Chronic stimulation of neural activity decreases Cbln1 expression and diminishes the number of PF–Purkinje synapses (Iijima et al., 2009), suggesting the importance of Cbln1–GluD2 signaling for synaptic plasticity and homeostasis.Cbln subfamily proteins are widely expressed throughout the brain (Miura et al., 2006), suggesting that their synaptogenic roles may be wide spread in other regions of the brain. Cbln2 and 4 are also secreted proteins, whereas Cbln3 is retained in the cellular endomembrane system (Iijima et al., 2007). Cbln1 and 2, interacting with an isoform of presynaptic neurexin, induce synaptogenesis (Joo et al., 2011; Matsuda and Yuzaki, 2011). Notably, the cortical synapses induced by neurexin–Cbln signaling are preferentially inhibitory (Joo et al., 2011), distinguishing the effects of Cbln from neuroligin. GluD1 was recently found to be the postsynaptic receptor of Cbln1 and 2 in cortical neurons, mediating the differentiation of inhibitory presynapses (Yasumura et al., 2012). On the other side, Cbln4 selectively binds to the netrin receptor DCC in a netrin-displaceable manner (Fig. 4, right), suggesting a potential function of Cbln4 through DCC signaling pathway (Iijima et al., 2007). Intriguingly, C1q, although sharing similar structure with Cbln, serves an opposite role by regulating the synapse elimination: C1q released from retinal ganglion cells refines the retinogeniculate connections by eliminating unneeded synapses (Stevens et al., 2007).

Concluding remarks

Synapse development is regulated in multiple steps. Research over the last few years have uncovered many regulatory mechanisms on how trafficking of synaptic material is regulated and how scaffold proteins act with cytoskeleton networks and trans-synaptic signaling to instruct the synapse formation. Nevertheless, our understanding of the cellular and molecular mechanisms regulating synapse development is still incomplete. For example, how is the direction, speed, and amount of synaptic material being transported specified? How is a synapse’s size determined? How is synapse type and strength specified through adhesive and secreted trans-synaptic signaling? How do the redundant synapse-inducing pathways interact with each other? Given the rapidly emerging improvements of technologies, especially super-resolution microscopy and high-throughput genomics and proteomics, the synapse development field will likely rapidly evolve in the near future.     

Cell biology in neuroscience: Cellular and molecular mechanisms underlying axon formation,growth, and branching

Proper brain wiring during development is pivotal for adult brain function. Neurons display a high degree of polarization both morphologically and functionally, and this polarization requires the segregation of mRNA, proteins, and lipids into the axonal or somatodendritic domains. Recent discoveries have provided insight into many aspects of the cell biology of axonal development including axon specification during neuronal polarization, axon growth, and terminal axon branching during synaptogenesis.

Introduction

Axon development can be divided into three main steps: (1) axon specification during neuronal polarization, (2) axon growth and guidance, and (3) axon branching and presynaptic differentiation (Fig. 1; Barnes and Polleux, 2009; Donahoo and Richards, 2009). These three steps are exemplified during neocortical development in the mouse: upon neurogenesis, newly born neurons engage long-range migration and polarize (Fig. 1, A and B) by adopting a bipolar morphology with a leading and a trailing process (Fig. 1 C). During migration (approximately from embryonic day [E]11 to E18 in the mouse cortex), the trailing process becomes the axon and extends rapidly while being guided to its final destination (lasts until around postnatal day [P]7 in mouse corticofugal axons with distant targets like the spinal cord; Fig. 1, D–F). Finally, upon reaching its target area, extensive axonal branching occurs during the formation of presynaptic contacts with specific postsynaptic partners (during the second and third postnatal week in the mouse cortex; Fig. 1, G–I). Disruption of any of these steps is thought to lead to various neurodevelopmental disorders ranging from mental retardation and infantile epilepsy to autism spectrum disorders (Zoghbi and Bear, 2012). This review will provide an overview of some of the cellular and molecular mechanisms underlying axon specification, growth, and branching.Open in a separate windowFigure 1.Axon specification, growth, and branching during mouse cortical development. Three stages of the development of callosal axons of cortical pyramidal neurons from the superficial layers 2/3 of the somatosensory cortex in the mouse visualized using long-term in utero cortical electroporation. For this class of model axons, development can be divided in three main stages: (1) neurogenesis and axon specification, occurring mostly at embryonic ages (A–C); (2) axon growth/guidance during the first postnatal week (D–F); and (3) axon branching and synapse formation until approximately the end of the third postnatal week (G–I). A, D, and G show coronal sections of mouse cortex at the indicated ages after in utero cortical electroporation of a GFP-coding plasmid at E15.5 in superficial neuron precursors in one brain hemisphere only (GFP signal in inverted color, dotted line indicates the limits of the brain). B, E, and H are a schematic representation of the main morphological changes observed in callosally projecting axons (red) at the corresponding ages. C shows the typical bipolar morphology of a migrating neuron emitting a trailing process (TP) and a leading process (LP) that will ultimately become the axon and dendrite, respectively. F and I show typical axon projections of layer 2/3 neurons located in the primary somatosensory area at P8 and P21, respectively. Neurons and axons in C, F, and I are visualized by GFP expression (inverted color). Image in C is modified from Barnes et al. (2007) with permission from Elsevier. Images in D, F, G, and I are reprinted from Courchet et al. (2013) with permission from Elsevier.

Neuronal polarization and axon specification

Neuronal polarization is the process of breaking symmetry in the newly born cell to create the asymmetry inherent to the formation of the axonal and somatodendritic compartments (Dotti and Banker, 1987). The mechanisms underlying this process have been studied extensively in vitro and more recently in vivo, but the exact sequence of events has remained elusive (Neukirchen and Bradke, 2011) partly because it is studied in various neuronal cell types that might not use the same extrinsic/intrinsic mechanisms to polarize. It is highly likely that at least three factors underlie neuronal polarization: extracellular cues, intracellular signaling cascades, and subcellular organelle localization. The partition-defective proteins (PARs) are a highly conserved family of proteins including two dyads (Par3/Par6 adaptor proteins and the Par4/Par1 serine/threonine kinases) that are required for polarization and axon formation (Shi et al., 2003, 2004; Barnes et al., 2007; Shelly et al., 2007; Chen et al., 2013), while many other intracellular signaling molecules also support axon formation (Oliva et al., 2006; Rašin et al., 2007; Barnes and Polleux, 2009; Shelly et al., 2010; Cheng et al., 2011; Hand and Polleux, 2011; Cheng and Poo, 2012; Gärtner et al., 2012). These intracellular signaling pathways are influenced by localized extracellular cues that instruct which neurite becomes the axon by either directly promoting axon extension or repressing axon growth in favor of dendritic growth (Adler et al., 2006; Yi et al., 2010; Randlett et al., 2011b; Shelly et al., 2011).The role of organelle subcellular localization during neuronal polarization is a more controversial issue. Initially, the orientation of organelles, including the Golgi complex, centrosomes, mitochondria, and endosomes, was shown to correlate with the neurite that becomes the axon in vitro (Bradke and Dotti, 1997; de Anda et al., 2005, 2010) and in vivo (de Anda et al., 2010). However, more recent studies suggest that the positioning of the centrosome is not necessary for neuronal polarization (Distel et al., 2010; Nguyen et al., 2011). Centrosome localization is likely constrained by microtubule organization within the cell, and therefore the centrosome position within the cell changes dynamically during different stages of polarization (Sakakibara et al., 2013). The question of how the interplay between extracellular cues, intracellular signaling, and organelle localization lead to polarization has pushed the field to perform more extensive in vivo imaging studies as in vitro systems/models have a difficult time recapitulating the complex environment and rely on neurons that were previously polarized in vivo.Like other epithelial cells, neural progenitors present a high degree of polarization along the apico-basal axis (Götz and Huttner, 2005). One of the major questions still needing to be addressed is how, or if, newly born mammalian neurons inherit some level of asymmetry from their parent progenitors (Barnes and Polleux, 2009). Recent studies have attempted to answer this question in vivo but have found that the answer might vary in each neuronal subtype. Retinal ganglion cells (RGCs), retinal bipolar neurons, and tegmental hindbrain nuclei neurons seem to inherit the apical/basolateral polarity from their progenitors (Morgan et al., 2006; Zolessi et al., 2006; Distel et al., 2010; Randlett et al., 2011a). In cortical neurons, hippocampal neurons, and cerebellar granule neurons, this relationship is unclear, in part because newly born cortical neurons first exhibit a multipolar morphology with dynamic neurites emerging from the cell body before adopting a bipolar morphology, suggesting they may not retain a predisposed parental polarity (Hand et al., 2005; Barnes et al., 2007). Other factors also suggest that different neuronal subtypes use different mechanisms during polarization. One such factor is the position where neurons specify their axon relative to the original apical/basolateral axis of their progenitors. As an example, cortical neurons in the mouse brain protrude an axon from the membrane facing the original apical surface toward the ventricular zone (Hand et al., 2005; Barnes et al., 2007; Shelly et al., 2007), whereas zebrafish RGCs form their axon from the membrane on the basolateral side (Zolessi et al., 2006; Randlett et al., 2011b). Another significant difference between cortical neurons and RGCs is related to the timing of axogenesis and dendrogenesis. RGCs tend to form their axons and dendrites at the same time during migration (Zolessi et al., 2006; Randlett et al., 2011b). However, cortical neurons form a long axon during migration before significant dendrite arborization takes place. These differences in the regulation of polarization and sequence of axon versus dendrite outgrowth may be linked to the localization of extracellular cues relative to the immature neuron during polarization (Yi et al., 2010).

Neuronal polarization, cytoskeletal dynamics, and polarized transport

What exactly makes the axonal compartment distinct from the somatodendritic domain? This can most easily be illustrated by focusing on the cytoskeleton that forms the framework of the developing axon. The cytoskeleton is composed of microtubules, actin filaments, and intermediate filaments (also called neurofilaments) along with their associated binding partners. Microtubules are composed of α- and β-tubulin subunits that polymerize to form a long filament intrinsically polarized by the addition of tubulin subunits to only one side of the growing filament called the plus end, while on the opposite side depolymerization occurs. It was discovered more than thirty years ago that the axon of a neuron contains a very uniform distribution of microtubules with the plus end facing away from the cell body (Heidemann et al., 1981). Through the years this observation was confirmed in many neuron cell types, and it was determined that dendrites do not have this uniform plus-end out network of microtubules (Fig. 2; Baas et al., 1988). Dendrites appear to have a complex array of microtubule orientations that may vary between species and/or neuronal subtypes. Current research shows that proximal dendrites are composed of mainly minus-end out microtubules, whereas more distal dendrites transition from an equal distribution of minus-end out and plus-end out microtubules to mainly plus-end out microtubules (Stone et al., 2008; Yin et al., 2011; Ori-McKenney et al., 2012). The orientation of microtubules matters greatly because it determines the relative contribution of microtubule-dependent motor proteins (kinesins and dyneins), which are the main motor proteins carrying various cargoes within cells and in particular are responsible for long-range transport in very large cells such as neurons. Dynein (a minus end–directed microtubule motor) is known to be responsible for both the transport of microtubules away from the cell body and for the uniform polarity of microtubules in the axon (Ahmad et al., 1998; Zheng et al., 2008). Recently, it was discovered that kinesin-1 (a plus end–directed microtubule motor) is required for the minus-end out orientation of microtubules in the dendrites of Caenorhabditis elegans DA9 neurons through selective transport of plus-end out microtubule fragments out of the dendrite (Yan et al., 2013). Another hallmark that differentiates the axonal and somatodendritic compartments is the microtubule-associated proteins (MAPs) that decorate microtubules to regulate their bundling and stability (Hirokawa et al., 2010). Microtubules in the axon are mainly decorated by Tau and MAP1B, whereas microtubules in the dendrites are labeled by proteins of the MAP2a-c family. The role of Tau in axon elongation remains controversial because early reports (Harada et al., 1994; Tint et al., 1998; Dawson et al., 2001) of Tau knockout alone suggested that axons were unaffected, but this apparent lack of phenotype might originate from the functional redundancy between MAPs as Tau/MAP1b double knockout mice show clear axon growth defects (Takei et al., 2000).Open in a separate windowFigure 2.Polarity maintenance and trafficking of somatodendritic and axonal proteins. Neurons are polarized into two main compartments: the somatodendritic domain and the axon. These domains are characterized by the underlying cytoskeleton and their unique protein signatures. The axonal cytoskeleton is defined by its uniform microtubule orientation where each microtubule is oriented with its plus end away from the cell body, while the dendrites contain a mixture of microtubules oriented in both directions. The proximal axon is characterized by a structure known as the axon initial segment (AIS, see inset). This highly ordered structure creates a diffusion barrier between the axonal compartment and the rest of the cell. F-actin is responsible for the cytoplasmic barrier, while sodium channels anchored by Ankyrin G form the basis of the plasma membrane barrier. Tau is retained in the axon by a microtubule-based filter at the AIS. Molecular motors (including kinesin, dynein, and myosin) then use the underlying cytoskeleton to restrict cargo transport to either the axon (such as Cntn1 and L1) or the dendrites (such as PSD95, AMPARs, and NMDARs).The dynamics of actin polymerization into actin filaments (F-actin) also play an important role in defining the axonal compartment, and contain an intrinsic polarity based on the polymerization of the free G-actin subunits (Hirokawa et al., 2010). Beyond the well-documented early role of F-actin dynamics in neurite outgrowth, multiple groups have shown that the disruption of actin polymerization allows dendritically localized proteins to incorrectly enter the axonal compartment (Winckler et al., 1999; Lewis et al., 2009; Song et al., 2009). The existence of a “diffusion barrier” in the proximal part of newly formed axons (Song et al., 2009) was long suspected. One of the current hypotheses is that a dense F-actin meshwork creates a cytoplasmic diffusion barrier shortly after polarization, which in part separates the axonal compartment from the neuronal cell body (Fig. 2, inset). Based on functional analysis and electron microscopy analysis, this “F-actin–based filter” is oriented so that the plus ends point toward the cell body while the minus ends point into the axon (Lewis et al., 2009, 2011; Watanabe et al., 2012). Two recent papers show via high resolution imaging techniques that indeed the axon has a unique F-actin network that is not found in dendrites (Watanabe et al., 2012; Xu et al., 2013). The development of this F-actin meshwork appears to directly precede the formation of the axon initial segment (AIS; Song et al., 2009; Galiano et al., 2012). An intra-axonal diffusion barrier, composed of Spectrins and Ankyrin B, defines the eventual position of the AIS. This boundary excludes Ankyrin G, which instead clusters in the most proximal part of the axon close to the cell body, where the AIS will form (Galiano et al., 2012). Ankyrin G is required for AIS formation and maintenance, and its loss causes the axon to start forming protrusions resembling dendritic spines (Hedstrom et al., 2008). Microtubules also play an important role at the AIS, as recent evidence suggests that Tau is retained in the axon through a microtubule-based diffusion barrier independently of the F-actin based filter (X. Li et al., 2011). The AIS is important in the formation of a plasma membrane barrier between the axonal and somatodendritic domains and its disruption affects both neuronal polarity and function because it is critical for clustering of voltage-dependent sodium channels and action potential initiation (Rasband, 2010).One of the critical cellular mechanisms underlying neuronal polarization is the polarized transport of various cargoes in axons and dendrites. Transport of proteins and various organelles is performed by the microtubule-dependent motor proteins kinesin and dynein (Hirokawa et al., 2010). Studies from many laboratories have demonstrated that kinesin motors can carry cargo to both the axonal and dendritic compartments (Burack et al., 2000; Nakata and Hirokawa, 2003). The mechanism for how selection occurs is not completely understood, but it probably incorporates both the affinity of the kinesin head for microtubules and the cargo bound to the motor protein (Nakata and Hirokawa, 2003; Song et al., 2009; Jenkins et al., 2012). In the axon, dynein works to bring cargo and retrograde signals back to the cell body, whereas in the dendrites it is responsible for much of the transport from the soma into the dendrites (Zheng et al., 2008; Kapitein et al., 2010; Harrington and Ginty, 2013). Additionally, the F-actin–dependent myosin motors can affect the polarized transport of cargos by using the F-actin–based cytoplasmic filter at the AIS to deny or facilitate entry of vesicles into the axon. Loss of the actin filter or myosin Va activity (a plus end–directed motor) allows dendritic cargos into the axon, whereas myosin VI (a minus end–directed motor) both removes axonal proteins from the dendritic surface and funnels vesicles containing axonal proteins through the actin filter at the AIS (Lewis et al., 2009, 2011; Al-Bassam et al., 2012). A current working hypothesis is that vesicles composed of multiple cargoes contain binding sites for each of these motors, and that through unknown mechanisms the activity of the motors can be differentially regulated to control the directionality of transport. An interesting example of how the interplay between different motors and cargo adaptors could lead to polarized transport was recently described for mitochondria (van Spronsen et al., 2013).

Axon growth

Microtubule dynamics regulate axon growth.

After axon specification, axon growth constitutes the second step of axonal development and is tightly linked to axon guidance toward the proper postsynaptic targets. Axon elongation by the growth cone is the product of two opposite forces (Fig. 3): slow axonal transport and the polymerization of microtubules providing a pushing force from the axon shaft, and the retrograde flow of actin providing a pulling force at the front of the growth cone (Letourneau et al., 1987; Suter and Miller, 2011). Although coordinated actin and microtubule dynamics are required for the proper function of the growth cone, it was reported that agents disrupting the actin cytoskeleton have limited consequences on axon elongation and are rather involved in axon guidance in vitro (Marsh and Letourneau, 1984; Ruthel and Hollenbeck, 2000) and in vivo (Bentley and Toroian-Raymond, 1986). Furthermore, local disruption of actin organization in the growth cone of minor neurites allows them to turn into axons (Bradke and Dotti, 1999; Kunda et al., 2001), indicating that the dense actin network present at the periphery of an immature neuronal cell body and in immature neurites may prevent microtubule protrusion and elongation necessary for axon specification.Open in a separate windowFigure 3.Cytoskeletal changes during axon elongation and branching. Representation of axon elongation and collateral branch formation in a cultured neuron. Axon growth is a discontinuous process, and collateral branches often originate from sites where the growth cone paused (gray dotted line), after it has resumed its progression. Other modalities of branch formation can occur through the formation of filopodia and lamellipodia. Red box shows a magnification of the main growth cone. Microtubules from the axon shaft spread into the central (C) zone. Some microtubules pass through the transition (T) zone, containing F-actin arcs, to explore filopodia from the peripheral (P) zone. Upon the proper stimulation by extracellular guidance cues or growth-promoting cues, microtubules are stabilized and invade the P-zone where they provide a pushing force, which, combined with the traction force from the actin treadmilling, provides the force required for growth cone extension. Green box shows the cytoskeletal changes occurring during collateral branch formation in the axon. Filopodia and lamellipodia are primarily F-actin–based protrusions that get invaded by microtubules, then elongate upon microtubule bundling. At later developmental stages, axon branches are stabilized or retracted (blue box) by mechanisms relying on the access to extracellular neurotrophins and/or neuronal activity and synapse formation.Contrary to actin, microtubule polymerization is required to sustain axon elongation and branching (Letourneau et al., 1987; Baas and Ahmad, 1993). Axonal proteins and cytoskeletal elements are transported along the axon through slow axonal transport by molecular motors (Yabe et al., 1999; Xia et al., 2003). It is still controversial whether tubulin and other cytoskeletal elements are transported in the axon as monomers and/or as polymers (Roy et al., 2000; Terada et al., 2000; Wang et al., 2000; Brown, 2003; Terada, 2003). Nonetheless, disruption of the slow transport of tubulin impairs the pushing force resulting from microtubule polymerization and impairs axon elongation (Suter and Miller, 2011). Therefore, it is not surprising that axon growth is affected in vitro and in vivo by disruption of plus-end microtubule-binding proteins such as APC (Shi et al., 2004; Zhou et al., 2004; Yokota et al., 2009; Chen et al., 2011) or EB1 and EB3 (Zhou et al., 2004; Jiménez-Mateos et al., 2005; Geraldo et al., 2008), microtubule-associated proteins such as MAP1B (Black et al., 1994; Takei et al., 2000; Dajas-Bailador et al., 2012; Tortosa et al., 2013), or proteins regulating microtubule severing and reorganization such as KIF2A (Homma et al., 2003), katanin, and spastin (Karabay et al., 2004; Yu et al., 2005; Wood et al., 2006; Butler et al., 2010).The contribution of microtubule dynamics to axon growth is not limited to growth cone dynamics but also involves axonal transport and polymerization along the axon shaft. Moreover, changing the balance between microtubule stabilization and depolymerization by local application of microtubule stabilizing agents is sufficient to instruct one neurite to grow and adopt an axon fate (Witte et al., 2008). Many kinase pathways converge on Tau and other axonal MAPs to regulate their function by phosphorylation (Morris et al., 2011). Among them, the MARK kinases regulate microtubule stability and axonal transport through phosphorylation of Tau (Drewes et al., 1997; Mandelkow et al., 2004). Interestingly, MARK-related kinases SAD-A/B control axon specification in part through phosphorylation of Tau (Barnes et al., 2007) and have very recently been linked to the growth and branching of the axons of sensory neurons (Lilley et al., 2013). Our work recently demonstrated that another family member related to MARKs and SAD kinases, called NUAK1, controls axon branching of mouse cortical neurons through the regulation of presynaptic mitochondria capture (Courchet et al., 2013). To what extent the regulation of Tau and other MAPs by the MARKs, SADs, and NUAK1 kinases contributes to axon elongation remains to be explored.

Where does axon elongation take place?

Growth cone progression and guidance constitute the main driver of axonal growth during development, but this process is unlikely to account for the totality of axon elongation. This is especially true after the axon has reached its final target but the axon shaft keeps growing in proportion to the rest of the body. One mechanism that may contribute to this “interstitial” form of axon elongation during brain/body size increase (see Fig. 1 for an example during postnatal cortex growth) is axon stretching, a process that can induce axon elongation in vitro (Smith et al., 2001; Pfister et al., 2004; Loverde et al., 2011) and in vivo (Abe et al., 2004). Aside from extreme stretching performed through the application of external forces, stretching could also contribute to the natural elongation of the axon in response to the tension resulting from growth cone progression (Suter and Miller, 2011).Axon elongation requires the addition of new lipids, proteins, cytoskeleton elements, and organelles along the axon. Where does the synthesis and incorporation of new components take place? Polysaccharides and cholesterol synthesis mostly occur in the cell body; however, lipid synthesis and/or incorporation can occur along the axon as well (Posse De Chaves et al., 2000; Hayashi et al., 2004). The growth cone is also a site of endocytosis, membrane recycling, and exocytosis (Kamiguchi and Yoshihara, 2001; Winckler and Yap, 2011; Nakazawa et al., 2012). One of the best studied examples of endocytosis and its role in axon growth and neuronal survival is the retrograde trafficking of TrkA receptor by target-derived nerve growth factor (NGF) in the peripheral nervous system (Harrington and Ginty, 2013).

Axon branching and presynaptic differentiation

Where do axon branches form?

The last step of axon development is terminal branching, which allows a single axon to connect to a broad set of postsynaptic targets. Collateral branches are formed along the axon through two distinct mechanisms: the first modality of branching is through splitting or bifurcation of the growth cone, which is linked to axon guidance and to the capacity of one single neuron to reach two targets that are far apart with one single axon. Growth cone splitting is observed in vivo in various neuron types including cortical neurons (Sato et al., 1994; Bastmeyer and O’Leary, 1996; Dent et al., 1999; Tang and Kalil, 2005), sympathetic neurons (Letourneau et al., 1986), motorneurons (Matheson and Levine, 1999), sensory neurons (Ma and Tessier-Lavigne, 2007), or mushroom body neurons in Drosophila (Wang et al., 2002). The second modality, known as interstitial branching, occurs through the formation of collateral branches directly along the axon shaft. Contrary to growth cone splitting, interstitial branching serves the purpose of raising axon coverage locally in order to define their “presynaptic territory”, and may contribute to increased network connectivity (Portera-Cailliau et al., 2005). Although both mechanisms can occur simultaneously in the same neuron, the relative importance of splitting versus interstitial branching is highly divergent from one neuron type to the other (Bastmeyer and O’Leary, 1996; Matheson and Levine, 1999; Portera-Cailliau et al., 2005).In culture, the axon grows in a non-continuous fashion with frequent growth cone pausing. Time-lapse imaging of sensorimotor neurons revealed that interstitial branching often occurs at the site where the growth cone paused, shortly after it has continued its course (Szebenyi et al., 1998). Accordingly, treatments with neurotrophins that slow the growth cone correlate with increased axon branching (Szebenyi et al., 1998). This suggests that growth cone pausing leaves a “mark” along the axon shaft that might predetermine future sites of branching (Kalil et al., 2000). However, local applications of neurotrophins shows that aside from growth cone pause sites the axon shaft remains competent to form collateral branches upon stimulation by extracellular factors (Gallo and Letourneau, 1998; Szebenyi et al., 2001), through the formation of filopodia or lamellipodia. Similar observations in vivo revealed that cortical axons are highly dynamic during development and form multiple filopodia that are the origin of collateral branches (Bastmeyer and O’Leary, 1996). Lamellipodia can be observed as motile, F-actin–dependent “waves” along the axon in vitro (Ruthel and Banker, 1998) and in vivo (Flynn et al., 2009). Moreover, disruption of microtubule organization impairs lamellipodia formation along the axon and is correlated with decreased axon branching (Dent and Kalil, 2001; Tint et al., 2009).

Cytoskeleton dynamics and axon branch formation.

Regardless of what type of protrusion gives rise to a branch, cytoskeletal reorganization in the nascent branch generally follows a similar sequence (Fig. 3): initially F-actin filament reorganization gives rise to a protrusion (filopodia, lamellipodia), followed by microtubule invasion of this otherwise transient structure to consolidate it, before the mature branch starts elongating through microtubule bundling (Gallo, 2011). Actin filaments accumulate along the axon and form “patches” that serve as nucleators for axon protrusions such as filopodia and lamellipodia (Korobova and Svitkina, 2008; Mingorance-Le Meur and O’Connor, 2009; Ketschek and Gallo, 2010). The mRNA for β-actin and regulators of actin polymerization such as Wave1 or Cortactin accumulate along the axons of sensory neurons and form hot-spots of local translation that are associated to NGF-dependent branching (Spillane et al., 2012; Donnelly et al., 2013). Subsequently, microtubules in the axon shaft undergo fragmentation at branch points as a prelude to branch invasion by microtubules (Yu et al., 1994, 2008; Gallo and Letourneau, 1998; Dent et al., 1999; Hu et al., 2012), a process that may disrupt transport locally to help trap molecules and organelles at branch points. Moreover, severed microtubules are transported into branches, a process required for branch stabilization (Gallo and Letourneau, 1999; Ahmad et al., 2006; Qiang et al., 2010; Hu et al., 2012). Interestingly, it is clear that, just like growth cone–mediated axon elongation, F-actin and microtubule reorganization events are interconnected to sustain axon branching (Dent and Kalil, 2001). As an example, microtubule-severing enzymes can also contribute to actin nucleation and filopodia formation (Hu et al., 2012).

Is axon branching linked to axon elongation?

Like in the growth cone, cytoskeleton reorganization constitutes the backbone of branch formation. It is therefore not surprising that many manipulations of the cytoskeleton affect both axon elongation and branch formation (Homma et al., 2003; Chen et al., 2011). Moreover, conditions that primarily disrupt axon elongation could secondarily disrupt branching by impairing the ability of the nascent branch to grow. However, axon elongation and axon branching can be considered as two separate phenomena and can be operationally separated because conditions disrupting one do not systematically affect the other. As an example, the microtubule-severing proteins katanin and spastin have differential consequences on axon elongation (primarily dependent upon katanin function) and branching (mostly spastin mediated; Qiang et al., 2010), taxol treatment (which stabilizes microtubules) affects axon elongation but not branching (Gallo and Letourneau, 1999), and disruption of TrkA endocytosis by knock-down of Unc51-like kinase (ULK1/2) proteins has opposite effects on axon elongation and branching (Zhou et al., 2007). In vivo, superficial layer cortical neurons initially go through a phase of elongation through the corpus callosum without branching (see Fig. 1), then stop elongating and form collateral branches in the contralateral cortex (Mizuno et al., 2007; Wang et al., 2007). It is conceivable that even before myelination, axons are actively prevented from branching at places and stages when they elongate (for example in the white matter of the neocortex) where they tend to be highly fasciculated. The identities of the molecules that inhibit interstitial branching along the axon shaft are currently unknown.

Regulation of axon branching by activity.

Immature neurons display spontaneous activity in the form of calcium waves (Gu et al., 1994; Gomez and Spitzer, 1999; Gomez et al., 2001) and spontaneous vesicular release long before they have completed axon development, which suggested a role for early neuronal activity in axon development and guidance (Catalano and Shatz, 1998). Cell-autonomous silencing of neurons in vivo by transfection of the hyperpolarizing inward-rectifying potassium channel Kir2.1 in olfactory neurons (Yu et al., 2004), in RGCs (Hua et al., 2005) or in cortical pyramidal neurons (Mizuno et al., 2007; Wang et al., 2007), or in vitro through infusion of tetrodotoxin (which blocks action potentials generation) in co-cultures of thalamo-cortical projecting neurons (Uesaka et al., 2007) results in a decrease in terminal axon branching, indicating that synaptic activity is required for axons to fully develop their branching pattern. Moreover, inhibition of synaptic release by expression of tetanus toxin light chain (TeTN-LC; Wang et al., 2007) also abolished terminal axon branching, suggesting that the formation of functional presynaptic release sites is required cell autonomously to control terminal axon branching. However, one potential limitation of the experiments involving TeTN-LC is that it blocks most VAMP-mediated vesicular release (with the exception of VAMP7, also called tetanus toxin–independent VAMP, or TI-VAMP). Therefore TeTN-LC action may not be limited to blocking synaptic vesicle release, but could also inhibit peptide release through vesicles containing neurotrophins for example, or other important trophic factors required for axon branching. More recent experiments through silencing of postsynaptic targets revealed that branching of callosal or thalamocortical axons is also dependent upon the activity of the postsynaptic targets (Mizuno et al., 2010; Yamada et al., 2010), albeit activity of the presynaptic neuron is required earlier during the branching process than activity of the postsynaptic targets (Mizuno et al., 2010). Activity is also required in some neurons at the phase of axon elongation through a feedback loop involving the activity-dependent up-regulation of guidance molecules (Mire et al., 2012).How much does spontaneous or evoked neuronal activity contribute to branching? Reduction of neuronal activity through hyperpolarization induced by overexpression of Kir2.1 significantly reduces axon branching without completely eliminating it (Hua et al., 2005; Mizuno et al., 2007; Wang et al., 2007). Activity seems to serve as a competitive regulator of axon branching with regard to its neighbors because silencing of neighboring axons restores normal branching (Hua et al., 2005). Interestingly, neuronal activity induces neurotrophin expression locally, suggesting that activity can contribute to branching partly through activation of activity-independent branching mechanisms (Calinescu et al., 2011).Neuronal activity can regulate branching through modification of the actin cytoskeleton via RhoA activation (Ohnami et al., 2008), and mRNA accumulates at presynaptic sites, indicating a correlation between local translation and synaptic activity (Lyles et al., 2006; Taylor et al., 2013). Neuronal activity is associated with changes in intracellular Ca²⁺ signaling, which has been shown to play a deterministic function in axon growth (Gomez and Spitzer, 1999). Calcium signaling activates the Ca²⁺/calmodulin-dependent kinases (CAMKs) that are known to regulate axon branching in vitro (Wayman et al., 2004; Ageta-Ishihara et al., 2009) and in vivo (Ageta-Ishihara et al., 2009).

Stabilization and refinement of the axonal arborization.

Axon branches are often formed in excess during development, then later refined to select for specific neural circuits (Luo and O’Leary, 2005). Long-range axon branch retraction has long been observed in layer V cortical neurons that initially project to the midbrain, hindbrain, and spinal cord (O’Leary and Terashima, 1988; Bastmeyer and O’Leary, 1996). At later stages, pyramidal neurons from the primary visual cortex will retract their spinal projection through axon pruning, whereas pyramidal neurons from the primary motor cortex will stabilize this projection but retract their axonal branches growing toward visual targets such as the superior colliculus. The molecular mechanisms controlling this area-specific pattern of axon branch pruning are still poorly understood, but seem to involve extracellular cues such as semaphorins and Rac1-dependent signaling (Bagri et al., 2003; Low et al., 2008; Riccomagno et al., 2012). Another example is the well-characterized refinement of retino-geniculate axons during the selective elimination of binocular input of RGC axon synapses onto relay neurons in the dorsal lateral geniculate nucleus (Muir-Robinson et al., 2002). Interestingly, some axons use caspase-dependent pathways locally to induce the selective retraction of axon branches during the process of pruning (Nikolaev et al., 2009; Simon et al., 2012).Circuit refinement and selective branch retraction can be observed in vivo at the level of the neuromuscular junction where individual branches of motor axons are eliminated asynchronously (Keller-Peck et al., 2001). In the developing CNS, neurotrophin-induced branch retraction can also be observed in a context of competition between neighboring axons (Singh et al., 2008). One other way of stabilizing axon branches is through the formation of synapses with postsynaptic targets. In the visual system, the initial axon arbor is refined to establish ocular dominance through activity-dependent retraction of less active branches (Ruthazer et al., 2003). Time-lapse imaging of RGC axons in zebrafish or in Xenopus tadpole revealed that the formation of presynaptic sites occurs concomitantly to axon branching, and branches that form presynaptic structures are less likely to retract (Meyer and Smith, 2006; Ruthazer et al., 2006). The stabilization of axon branches through formation of synaptic contacts parallels with the stabilization of dendritic branches through synapse formation and stabilization (Niell et al., 2004; J. Li et al., 2011). The role of presynaptic bouton formation goes beyond the stabilization of axonal branches because in vivo, new axon branches can emerge from existing presynaptic terminals (Alsina et al., 2001; Javaherian and Cline, 2005; Panzer et al., 2006).In conclusion, axon growth and branching can be regulated by both activity-dependent and activity-independent mechanisms during development. However, for mammalian CNS axons, much more work is needed to define (1) the precise molecular mechanisms underlying axon branching; (2) the cellular and molecular mechanisms regulating the key transition between axon growth and branching when axons start forming presynaptic contacts with their postsynaptic partners; and (3) the mechanisms regulating axon pruning during synapse elimination.     

Rab GTPases and membrane identity: Causal or inconsequential?

Rab GTPases are highly conserved components of vesicle trafficking pathways that help to ensure the fusion of a vesicle with a specific target organelle membrane. Specific regulatory pathways promote kinetic proofreading of membrane surfaces by Rab GTPases, and permit accumulation of active Rabs only at the required sites. Emerging evidence indicates that Rab activation and inactivation are under complex feedback control, suggesting that ultrasensitivity and bistability, principles established for other cellular regulatory networks, may also apply to Rab regulation. Such systems can promote the rapid membrane accumulation and removal of Rabs to create time-limited membrane domains with a unique composition, and can explain how Rabs define the identity of vesicle and organelle membranes.

Rab GTPases regulate membrane tethering and vesicle fusion

Eukaryotic cells are defined in part by their complex membrane organelles. This organization permits the coexistence of different chemical environments within the same cell. For example, the endoplasmic reticulum (ER) is a neutral pH, reducing environment containing chaperones conducive to protein folding and the formation of disulfide bonds, whereas the lysosomes are ∼pH 5 and contain catabolic enzymes maximally active at acidic pH. Though valuable, this organization requires some form of active transport machinery for the exchange of material between these compartments because large hydrophilic molecules such as proteins cannot easily cross membranes. This transfer of molecules between compartments is achieved by vesicular transport systems that use cytosolic coat protein complexes to select small regions of membrane and shape these into defined 40–80-nm-diameter transport vesicles (Bonifacino and Glick, 2004; Faini et al., 2013). Vesicle coats contain binding sites for specific transport sequences, and thus only transfer a subset of proteins into the vesicle. Once produced, these vesicles have to identify, tether to, and then fuse with a specific target organelle (Zerial and McBride, 2001). Research over many years has defined small transmembrane proteins (SNAREs) and a set of accessory factors as the minimal machinery for membrane fusion (McNew et al., 2000; Shi et al., 2012). Tethering is a less well-defined event involving the Rab GTPases and effector protein complexes, typically large extended molecules thought to bridge the space between two approaching membranes (Gillingham and Munro, 2003).Rab GTPases were first linked to vesicle transport by groundbreaking genetic screens for mutants defective in protein secretion (Novick et al., 1980; Salminen and Novick, 1987). Sec4, Rab8 in humans, was found to function in the terminal step of the secretory pathway, delivery of Golgi-derived transport vesicles to the cell surface (Salminen and Novick, 1987; Goud et al., 1988). Ypt1, Rab1 in humans, was then shown to regulate secretion at the Golgi apparatus (Segev et al., 1988; Bacon et al., 1989). These findings led to an influential model for Rab function in which the cycle of GTPase activation and inactivation is coupled to recognition events in vesicle docking (Bourne, 1988). Consistent with the idea that they control vesicle targeting, work in mammalian cells then showed that there is a large family of highly conserved Rab GTPases, each with a specific subcellular localization (Chavrier et al., 1990). A series of seminal studies has since provided direct evidence that Rab1 and Rab5 promote membrane fusion (Gorvel et al., 1991; Segev, 1991) by regulating the activation and engagement of SNAREs (Lian et al., 1994; Søgaard et al., 1994), as a consequence of recruiting tethering factors to membrane surfaces (Segev, 1991; Sapperstein et al., 1996; Cao et al., 1998; Christoforidis et al., 1999; McBride et al., 1999; Allan et al., 2000; Shorter et al., 2002). Similar findings were also made for the Rab Ypt7, which functions in vacuole docking in yeast (Price et al., 2000; Ungermann et al., 2000), a system that allows direct visualization of docked or tethered intermediates due to the large size of the membrane structures (Wang et al., 2002).The evidence that Rabs function upstream of SNARE protein in vesicle trafficking pathways has led to the notion that Rabs help to define the identity of vesicle and organelle membranes (Pfeffer, 2001; Zerial and McBride, 2001). This is best exemplified by the early endocytic pathway, where the identity of early and late endosomes is thought to be determined by Rab5 and Rab7, respectively (Rink et al., 2005). However, in most other cases it remains unclear if this is a causal relationship, where the Rab directly defines the identity of the membrane rather than acting as an upstream regulator of vesicle targeting before the SNARE-mediated membrane fusion event. In addition to Rabs, GTPases of the Arf/Arl family and specific phosphoinositide lipids have also been proposed to act in specifying membrane identity (Munro, 2002; Di Paolo and De Camilli, 2006). It therefore seems likely that no single factor can explain how membrane identity is achieved in vesicle transport, and that Rabs, phosphoinositides, and other factors act in concert.

Rab GEFs provide the minimal machinery for targeting and activation

Despite the progress in defining Rab function, the claim that Rab GTPases define organelle identity therefore remains premature due to crucial unanswered questions. In particular, the issue of how Rabs are targeted to specific organelles, or even restricted to subdomains of these organelles, has remained problematic. Initial work using chimeric GTPases suggested that the variable C-terminal region of the different Rabs provided a targeting mechanism (Chavrier et al., 1991). However, subsequent work indicated that this failed to provide a general mechanism to explain specific Rab targeting, and that multiple regions of the Rab including C-terminal prenylation contribute to membrane recruitment (Ali et al., 2004). Emerging evidence based on the improved understanding of the family of Rab guanine nucleotide exchange factors (GEFs) now provides an alternative view for Rab activation at specific membrane surfaces. Mechanistic details of how Rab GEFs activate Rabs have been discussed elsewhere (Barr and Lambright, 2010), and are not directly relevant for this discussion so won’t be detailed further. Two studies now show that Rab GEFs can provide the minimal machinery needed to target a Rab to a specific membrane within the cell (Gerondopoulos et al., 2012; Blümer et al., 2013). In both cases, Rab GEFs were fused to mitochondrial outer membrane targeting sequences, and the effects on different Rabs observed. Using this strategy it was possible to specifically target Rab1, Rab5, Rab8, Rab35, and Rab32/38 to mitochondria with biochemically defined cognate GEFs (Gerondopoulos et al., 2012; Blümer et al., 2013). Mutants that either reduced the nucleotide exchange activity of the GEF or the target GTPase gave a correspondingly reduced rate of Rab targeting (Blümer et al., 2013). Alone this does not provide a full explanation for Rab targeting; for this an understanding of the interaction of prenylated Rabs with the chaperone GDI (guanine nucleotide displacement inhibitor) is needed. Structural and biophysical analysis of the Ypt1–GDI complex has revealed two components of this interaction relevant for Rab targeting (Pylypenko et al., 2006). Domain I of GDI interacts with the switch II region of Ypt1 only when this is in the GDP-bound inactive form. The doubly prenylated C terminus of Ypt1 occupies a hydrophobic cavity created by domain II of GDI. Simulation of this system and direct biophysical measurements suggests that in the absence of other factors GDI will rapidly deliver Rabs to and extract them from a lipid bilayer (Pylypenko et al., 2006; Wu et al., 2010). These ideas can be combined into a simple model for Rab activation at specific membrane surfaces (Fig. 1 A). In simple terms this model is a form of molecular speed-dating in which the Rab spends a short time sampling each membrane surface it encounters before finally meeting its cognate GEF partner, triggering a period of longer residence at that site (Fig. 1 A). In this model, GEF-mediated nucleotide exchange renders the Rab resistant to extraction by GDI, and thus drives accumulation of the active GTP-bound form of the Rab. This active Rab can then recruit effector proteins to the membrane surface and promote the desired recognition event. Such a system is analogous to the rapid proofreading of amino-acyl tRNAs during protein synthesis by the ribosome (Ibba and Söll, 1999). All amino-acyl tRNAs can enter the so-called acceptor site, but only if stable codon recognition occurs is the peptidyltransferase reaction initiated, otherwise the tRNA is rejected (Steitz, 2008). The two-stage kinetic proofreading of membrane surfaces by Rabs may similarly increase fidelity at little overall cost to the rate of vesicular traffic.Open in a separate windowFigure 1.The Rab activation and inactivation cycle. (A) Prenylated Rabs (black wavy lines) are bound by the chaperone GDI in the cytosol. Partitioning of the prenylated tail moiety between the hydrophobic pocket in GDI and the membrane bilayer allows Rabs to rapidly and reversibly sample membrane surfaces. When the GDP-bound inactive Rab encounters a cognate GEF nucleotide exchange occurs. This GTP-bound active Rab species does not interact with GDI and can therefore accumulate on the membrane surface, where it may further recruit effector proteins with specific biological functions. This cycle is reset when a GTP-bound Rab encounters a GAP (GTPase-activating protein) and the bound GTP is hydrolyzed to generate GDP and inorganic phosphate. (B) Additional specification of membrane domains within complex organelles, such as tubular domains of endosomes, or the fenestrated rims and different cisternae of the Golgi apparatus, may involve membrane receptors for Rabs (shown as light blue, dark blue, and green boxes). This could either involve (a) sequestration of the active Rab to a subdomain defined by the membrane receptor, or (b) direction of GDI unloading of an inactive Rab to specific sites on the organelle membrane also defined by a membrane receptor. Accumulation of a Rab at a specific site may be favored by GAPs opposing Rab activation at unwanted sites (Haas et al., 2007).Although this minimal Rab-targeting system does not require any additional factors, it is important to mention that this does not mean such factors do not exist. A family of membrane proteins with prenylated Rab-binding activity that can promote dissociation of some prenylated Rabs from GDI and favor retention of the GDP-bound form of the Rab downstream of membrane delivery by GDI has been identified (Dirac-Svejstrup et al., 1997; Martincic et al., 1997; Hutt et al., 2000; Sivars et al., 2003). These may therefore favor Rab activation, although recent data has suggested that such factors are not generally essential (Blümer et al., 2013). Intriguingly, other evidence links this family of proteins to factors involved in shaping subdomains of the ER and to the Golgi apparatus (Yang et al., 1998; Calero et al., 2001; Chen et al., 2004; Voeltz et al., 2006), perhaps suggesting that they may play roles in defining at which subdomain of an organelle an active Rab is enriched (Fig. 1 B).In addition to these regulatory factors, covalent modification can also be used to modulate the Rab activation cycle. Phosphorylation of Rab1 and Rab4 in mitosis alters the fraction of these GTPases that can associate with membranes (Bailly et al., 1991; van der Sluijs et al., 1992), although the exact mechanisms remain unclear. Furthermore, emerging evidence indicates that one Rab in yeast, Ypt11, is controlled by a phosphorylation-dependent mechanism regulating its activation and abundance (Lewandowska et al., 2013). A number of bacterial pathogens also encode enzymes that directly modify Rab GTPases and as a consequence alter the Rab regulatory cycle. During Legionella infection, Rab1 is modulated by a cycle of adenylylation and de-adenylylation by DrrA and SidA, respectively, and this modification of the conserved tyrosine residue in the switch II renders the protein constitutively active (Müller et al., 2010; Neunuebel et al., 2011; Tan and Luo, 2011). DrrA also has a GEF domain and can therefore directly activate and trap Rab1 in an active form independent of other cellular factors (Schoebel et al., 2009). A second bacterial protein, AnkX, mediates phosphocholination of an adjacent serine within the switch II region (Mukherjee et al., 2011; Campanacci et al., 2013). Pathogens such as Legionella use this covalent modification of Rabs to modulate their localization and activation (Stein et al., 2012). Although cellular enzymes that carry out related modification of Rabs are currently unknown, it would be premature to dismiss the possibility of their existence and use by cells to similarly control Rab activation and inactivation at specific sites.

Evidence for Rab activation on vesicle and target membrane surfaces

Based on the model and discussion so far it seems obvious that Rabs accumulate on the same membrane as their cognate GEF. Indeed, there is evidence that Rab1 may be activated and recruit the p115 tethering factor during the COP II vesicle formation stage of ER-to-Golgi transport (Allan et al., 2000). This would have the advantage that identity would be created at an early stage in vesicle biogenesis, and the vesicle could therefore be tethered to the Golgi before completion of the vesicle, thus increasing targeting efficiency. However, there is also evidence that Rab activation can occur at the target membrane and not only on a vesicle surface. Careful analysis of cell-free ER–Golgi transport assays revealed that Ypt1–Rab1 is not always required on the vesicle fraction, but is essential on the target Golgi membranes (Cao and Barlowe, 2000). Furthermore, a Ypt1 mutant with reduced nucleotide hydrolysis (which prevents its recycling from the Golgi compartment; Richardson et al., 1998), or Golgi membrane-anchored forms of Ypt1 (Cao and Barlowe, 2000) both support apparently normal ER–Golgi transport and cell growth. Subsequently, it was found that the COP II coat required to form ER–Golgi transport vesicles is the membrane receptor for the Ypt1–Rab1 GEF TRAPP (transport protein particle; Jones et al., 2000; Wang et al., 2000; Cai et al., 2007), indicating that Rab1 activation may occur on the coated vesicle. This raises questions about how the cytosolic Rab–GDI complex can access the membrane surface of a still-coated vesicle. However, because the COP II coat has an open lattice structure (Faini et al., 2013), it may be possible in this case for Ypt1–Rab1 to approach the membrane and insert. A further possibility is that COP II vesicles recruit TRAPP and promote the activation of Ypt1–Rab1 at the adjacent Golgi membranes to signal that an ER-derived vesicle is in close proximity (Fig. 2 A). This Golgi pool of activated Rab would then recruit effector proteins such as Uso1/p115 that trap and tether the incoming vesicle by directly engaging with vesicle SNAREs (Cao et al., 1998; Shorter et al., 2002).Open in a separate windowFigure 2.Recruitment mechanisms for Rab GEFs. Rab GEFs can be divided into two groups according to the mechanism of membrane recruitment. (A) Discrete coat protein complexes (green) recruit the first group. For example, COP II recruits the Rab1 GEF TRAPP to ER-Golgi vesicles, while clathrin-AP2 recruits DENND1A, the Rab35 GEF, to endocytic sites at the cell surface. In the case of TRAPP, biochemical and genetic data suggest that Rab1 can be activated on the target membrane, before vesicle tethering and SNARE-mediated fusion. (B) The larger second group of Rab GEFs is recruited by Rab GTPases either alone or in combination with a second factor (Rabs/factors listed next to arrow). For example, the GEF Sec2 is recruited to late-Golgi vesicles trafficking to the bud in yeast by the activated Rab Ypt31/32 and phosphatidylinositol 4-phosphate (PI4P), where it activates the Rab Sec4 (Rab8 in humans). The Rabex5–rabaptin complex, which is a Rab5 GEF, interacts with activated Rab4 or Rab5 and ubiquitylated cargo proteins on endocytic vesicles and early endosomes. A number of other GEFs (some additional examples shown) have been found to interact with active Rabs. Whether or not these represent the sole mode of membrane interaction for these GEFs is not defined at this time. PM, plasma membrane. (C) In situations where the GEF for a second Rab in the pathway is an effector for the first, a cascade can develop, where Rab-A promotes the recruitment of GEF-B for this second Rab-B.

Rab GEF targeting and regulation

The mechanism of GEF targeting is of crucial importance for understanding how Rabs are activated at a particular membrane site. At present, two different solutions to the problem depending on the GEF are known. First, as already mentioned, is vesicle coat–dependent GEF targeting (Fig. 2 A). Three examples are known at present: COP II recruitment of the Rab1 GEF TRAPP-I (Cai et al., 2007), and clathrin-adaptor protein complex 2–dependent recruitment of either the Rab35 GEF DENND1A (Allaire et al., 2010; Yoshimura et al., 2010) or the Rab5 GEF RME-6 (Sato et al., 2005; Semerdjieva et al., 2008) during endocytic transport from the plasma membrane. In the latter cases the exact nature of the membrane on which the target Rab is activated is unclear, but it is tempting to speculate that like COP II, the coated vesicle promotes Rab activation on the target organelle to signal the presence of an incoming vesicle to be tethered. The second larger group of GEFs comprises those known to interact with active Rab GTPases (Fig. 2 B). The first of these Rab GEF effectors defined was the Rabex-5–rabaptin complex, which is both a Rab5 exchange factor and effector for Rab4 and Rab5 (Horiuchi et al., 1997). Rabex-5 also binds to ubiquitin via a specific domain and this is important for regulating its recruitment to early endosomes (Lee et al., 2006; Mattera et al., 2006; Mattera and Bonifacino, 2008) where it activates Rab5.Specific phosphatidylinositols play a key role in defining membrane identity (Di Paolo and De Camilli, 2006)), and this is in part due to a role in recruitment or regulation of Rab exchange factors. Sec2, the exchange factor for Sec4–Rab8, is recruited to post-Golgi vesicles by a combination of the Rab Ypt32 and phosphatidylinositol 4-phosphate generated by Pik1 (Ortiz et al., 2002; Sciorra et al., 2005; Mizuno-Yamasaki et al., 2010). Similarly, in mammalian cells the Rab GEF Sec2–Rabin8 is recruited by the Ypt31/32 orthologue Rab11 (Knödler et al., 2010), and phosphatidylinositol 4-phosphate generated by the Pik1 orthologue Fwd is important for Rab11 regulation in Drosophila (Polevoy et al., 2009). Although less is known about the targeting of other Rab GEFs, the clear theme is that many are effectors for a Rab other than the one they activate (Fig. 2 B). The Ric1–Rgp1 complex is a GEF for Rab6 and effector for Rab33B at the Golgi (Pusapati et al., 2012) and the Rab21 GEF VARP is an effector for Rab32/38 (Zhang et al., 2006; Tamura et al., 2009). Additionally, a GEF for Rab32/38 is an effector for Rab9 (Kloer et al., 2010; Gerondopoulos et al., 2012), and the DENND5A Rab39 GEF is an effector for Rab6 (Recacha et al., 2009; Yoshimura et al., 2010). In addition to these canonical trafficking functions there are specialized examples that indicate there is some plasticity to both GEF targeting and specificity. The Ypt1 GEF TRAPP exists in an alternate form (TRAPP-II) with additional subunits that promote late-Golgi targeting and may create additional GEF activity toward Ypt31/32 (Morozova et al., 2006). Interestingly, in higher eukaryotes there is evidence that TRAPP-II may regulate the Ypt31/32 orthologues Rab11 in male meiotic cytokinesis in flies (Robinett et al., 2009) and Rab-A in plant cell polarization and division (Qi et al., 2011), respectively. TRS85 in another alternate TRAPP complex (TRAPP-III) promotes localization to the forming autophagosome and activates Rab1 during autophagy (Lynch-Day et al., 2010).The counterpart to this interlinked network of Rab activation is an equally complex set of interactions between Rabs and Rab GAPs. The GAP Gyp1 is an effector for Ypt32 and promotes GTP hydrolysis by Ypt1 in budding yeast (Rivera-Molina and Novick, 2009). In the absence of Gyp1, Ypt1 spreads into the later compartments of the secretory pathway that should be occupied by Ypt32 (Rivera-Molina and Novick, 2009). Interestingly, one of the cellular GAPs for Ypt1–Rab1 is a transmembrane protein of the ER that may prevent Rab1 activity from spreading earlier in the pathway to the ER rather than act to terminate Rab1 activity at the Golgi (Haas et al., 2007; Sklan et al., 2007). Similarly, two related proteins, RUTBC1 and RUTBC2, bind to active Rab9 and are GAPs for Rab32 and Rab36, respectively (Nottingham et al., 2011, 2012).Together, these findings have led to the general idea that the order of trafficking events in a pathway can potentially be defined by a series of Rabs acting as a cascade (Fig. 2 C). In such models one Rab triggers the next in the pathway by recruiting its cognate GEF, and then feedback develops as a GTPase-activating protein (GAP) is recruited to terminate the action of the previous Rab in the series (Mizuno-Yamasaki et al., 2012; Pfeffer, 2013). In part, this simply passes the problem on because we are then left with the question of how the previous Rab in the pathway or a cofactor for recruitment such as phosphatidylinositol 4-phosphate or ubiquitin is localized and generated only when required. In the case of the secretory pathway the ER provides a defined starting point where activation of Rab1–Ypt1 will inevitably result in a defined and correctly timed wave of Rab activation through the secretory pathway. However, a note of caution is needed when considering these ideas because far more support from experimental data looking at the biochemical properties of these systems both in vitro and in vivo is required to come to any definitive conclusions.

Ultrasensitive Rab activation switches

One of the key tenets of the membrane identity hypothesis is that Rabs should rapidly and accurately establish membrane identity and then be lost once the membrane recognition event is over. Although biochemical data on Rab GEFs clearly indicate these molecules generally have sufficiently high specificity to ensure activation of only one Rab or a set of closely related Rabs (Delprato et al., 2004; Yoshimura et al., 2010; Gerondopoulos et al., 2012), how rapid switch-like accumulation is ensured is less obvious. Similar issues exist for termination of the Rab cycle by Rab GAPs. As already mentioned, Rab cascade models give part of the solution to this problem, and provide features that can ensure vectorial flow in a membrane traffic pathway (Mizuno-Yamasaki et al., 2012; Pfeffer, 2013). However, they do not fully explain how switch-like transitions and defined compartmental boundaries are achieved (Del Conte-Zerial et al., 2008). A possible solution to this problem comes from studies on the regulation of other complex biological systems, exemplified by control of cell cycle transitions (Tyson et al., 2001). Rather than displaying the expected Michaelis-Menten kinetics (Fig. 3 A), Rab cycles may yield properties of ultrasensitivity (Goldbeter and Koshland, 1981, 1984). This would appear to be a valid proposal if the Rab cycle is treated as being analogous to a covalent modification (Rab and Rab-modified, for GDP and GTP forms, respectively) and because GEF activity is generally assumed to be limiting (Blümer et al., 2013). In such a situation, inputs activating the GEF, for example membrane recruitment requiring multiple or binding of an activator, would be amplified and give rise to very large changes in the amount of activated Rab (Fig. 3 A). When combined with feedback loops, this can create a bistable switch between two states as shown for cell cycle transitions (Novak and Tyson, 1993; Pomerening et al., 2003). In the case of GTPase regulation, as the input controlling the GEF increases then the system transitions to a Rab-active state that remains stable over a wide range of GEF activity. GAP activation could then trigger exit from this state. This is also useful for providing a potential explanation for the timing properties of a Rab cascade. Ultrasensitivity and bistability are therefore likely to be useful concepts when explaining the behavior of Rabs, especially when considering complex interlinked cycles (Fig. 3 B) because they avoid the futile cycles where GAPs and GEFs fight one another and thus don’t do any useful work.Open in a separate windowFigure 3.Ultrasensitivity and bistability in Rab regulatory networks. (A) A simplified schematic of a Rab activation cycle is shown treating GDP–GTP exchange as equivalent to a covalent modification cycle such as phosphorylation. Because the reaction can only occur at a membrane surface, membrane recruitment factors are treated as activating inputs. Assuming no feedback and normal first-order reaction kinetics, Rab recruitment would be expected to follow Michaelis-Menten behavior. In cases where substrate is saturating and the reaction becomes zero-order, Goldbeter and Koshland (1984) have shown that product formation becomes more sensitive to enzyme concentration. In this case, generation of GTP-bound Rab becomes ultrasensitive to GEF concentration at the membrane surface. If additional positive feedback controls exist as shown in the bottom panel, then bistability may develop. In this case a rapid switch-like transition in Rab activity develops as Rab GEF concentration increases. Once in the active state the system becomes less dependent on continued high GEF activity. (B) A model for an interlinked Rab cascade is shown. The GEF for Rab-B is an effector for activated Rab-A, while the GAP for Rab-A is regulated by Rab-B. An example of this latter situation is provided by the Ypt1–Yp32 system discussed in the main text and shown in the bottom panel, where a Ypt1 GAP Gyp1 is an effector for Ypt32 (Rivera-Molina and Novick, 2009) and inhibits Ypt1. This coupling of the two cycles can result in coupled ultrasensitive switch-like transitions or bistability.A groundbreaking study in this area has applied these ideas to the conversion of Rab5-positive early endosomes to Rab7-positive late endosomes and lysosomes (Del Conte-Zerial et al., 2008). This analysis has provided strong evidence that positive and negative feedback loops in this system mediated by Rab GEFs and GAPs result in bistability in the form of a cut-out switch, so that Rab5 accumulation is followed by an abrupt transition at which Rab5 is rapidly lost and Rab7 accumulates (Del Conte-Zerial et al., 2008). Underpinning this is a biochemical network in which the Mon1–Ccz1 Rab7 GEF complex displaces Rabex-5, thus breaking the positive feedback loop to Rab5 activation (Poteryaev et al., 2010) and simultaneously promoting recruitment and activation of Rab7 (Nordmann et al., 2010; Gerondopoulos et al., 2012). Although there are only few studies where these ideas have been considered, they can be experimentally tested and are likely to be of increasing importance in membrane traffic regulation.

Origins of Rab GTPase control systems

One of the most difficult questions in membrane trafficking relates to the origins of complex internal membrane systems in eukaryotes. Analysis of Rab GTPases themselves suggests a pattern of evolution of Rabs consistent with the evolution of a core set of membrane organelles of the endocytic and secretory pathways (Diekmann et al., 2011; Klöpper et al., 2012). Yet, this provides little insight into how membrane organelles initially arose. Recent data on the structure of Rab GTPase regulators and coat protein complexes has identified common features with GTPase regulators in other systems including prokaryotes (Kinch and Grishin, 2006; Zhang et al., 2012; Levine et al., 2013). The conserved Longin–Roadblock fold has emerged as a structural feature of the large family of DENN-domain Rab GEFs in human cells (Yoshimura et al., 2010; Wu et al., 2011; Levine et al., 2013). Intriguingly, related domains are also present in the signal sequence receptor involved in protein translocation into the ER, vesicle coat protein complexes, and the MglA GTPase–MglB bacterial cell polarity regulator (Sun et al., 2007; Miertzschke et al., 2011; Levine et al., 2013). Although far from conclusive, these findings provide important pointers to the development of GTPase control systems, and more generally the early origins of membrane traffic pathways in eukaryotes from membrane-associated GTPases and their effector proteins.Are Rabs alone capable of triggering the pathways defining membrane identity? Multiple lines of evidence show Rab GTPases are clearly important and far from inconsequential regulators of vesicle traffic; however, further evidence is required before we should conclude that they are causal regulators of vesicle or organelle membrane identity. Neither of the studies using strategies to modulate the cellular localization of Rab GEFs reported that the mitochondria altered their identity or were converted into an endosome or Golgi because of the mistargeted Rabs (Gerondopoulos et al., 2012; Blümer et al., 2013). The picture emerging is therefore one in which Rabs cannot program membrane identity alone and must work in concert with other factors. Defining and reconstituting the systems needed to create membrane identity is therefore a major goal for membrane traffic research.     

Cell biology in development: The cell biology of planar cell polarity

Planar cell polarity (PCP) refers to the coordinated alignment of cell polarity across the tissue plane. Key to the establishment of PCP is asymmetric partitioning of cortical PCP components and intercellular communication to coordinate polarity between neighboring cells. Recent progress has been made toward understanding how protein transport, endocytosis, and intercellular interactions contribute to asymmetric PCP protein localization. Additionally, the functions of gradients and mechanical forces as global cues that bias PCP orientation are beginning to be elucidated. Together, these findings are shedding light on how global cues integrate with local cell interactions to organize cellular polarity at the tissue level.The collective alignment of cell polarity across the tissue plane is a phenomenon known as planar cell polarity (PCP). Exemplified by the uniform orientation of bristles covering the insect epidermis or of the hairs covering the mammalian body surface (Fig. 1 A), PCP patterns can align over thousands, even billions of cells. This phenomenon is controlled by the so-called PCP pathway, which integrates global directional cues to produce locally polarized cell behaviors. There has been a recent surge in interest in PCP after discoveries that various processes such as vertebrate gastrulation, mammalian ear patterning and hearing, and neural tube closure all require a conserved set of PCP genes (Heisenberg et al., 2000; Tada and Smith, 2000; Wallingford et al., 2000; Kibar et al., 2001; Murdoch et al., 2001; Curtin et al., 2003; Montcouquiol et al., 2003; Copley et al., 2013). Since that time, the PCP pathway has been found to coordinate cell behaviors in numerous diverse settings including polarized ciliary beating in the trachea and brain ventricles (Tissir et al., 2010; Vladar et al., 2012), oriented cell divisions (Gong et al., 2004; Baena-López et al., 2005; Ségalen et al., 2010; Mao et al., 2011), lung branching (Yates et al., 2010), and hair follicle alignment (Guo et al., 2004; Devenport and Fuchs, 2008), to name a few (Fig. 1). Genetic disruptions to PCP cause severe developmental abnormalities in vertebrates, notably neural tube defects, left/right patterning defects, and ciliopathies, which highlights the essential requirement for PCP in development (Kibar et al., 2001; Murdoch et al., 2001; Curtin et al., 2003; Wang et al., 2006a,b; Kim et al., 2010; Song et al., 2010).Open in a separate windowFigure 1.Planar cell polarity and the core PCP components. (A and B) The Drosophila wing blade and mammalian epidermis illustrate the phenomenon of PCP. In both cases, hairs point in a single direction along the tissue axis, where they align locally with neighboring hairs and globally across the tissue. Whereas Drosophila wing hairs are produced by single cells, mammalian hairs emerge from multicellular hair follicles, which orient as a unit. A conserved PCP pathway controls the collective alignment of both types of structures. (C) Core PCP components localize to the plasma membrane and asymmetrically segregate along the epithelial plane as indicated.Like many types of cell polarity, the establishment of PCP involves (1) a global orienting cue, (2) asymmetric segregation of dedicated polarity proteins, and (3) translation of polarity information into polarized outputs. But unlike other types of cell polarity, the PCP mechanisms we currently understand involve coupling between adjacent cells, allowing for the alignment of polarity over many cell distances.First described in insects and then genetically dissected in Drosophila melanogaster, PCP was long confined to the realm of experimental embryology and genetics until the discovery that the protein products of several PCP genes were localized asymmetrically within the cell, thrusting PCP into the domain of cell biology (for review see Strutt and Strutt, 2009). The challenge to understanding PCP on a molecular level is that long-range PCP is, in essence, an in vivo phenomenon that is difficult to recapitulate in a tissue culture dish. However, recent advances in imaging technology combined with increasingly sophisticated genetic tools are helping us to decipher the in vivo cell biology of PCP. In this review, I highlight some of the recent advances made toward understanding the cell biology underlying the establishment of coordinated polarized cell behaviors. For clarity, I limit discussions of PCP phenomena that meet the definition of PCP proposed by Goodrich and Strutt (2011): namely, that “cell–cell communication causes two or more cells to adopt coordinated polarity” in a process that is mechanistically “dependent upon planar polarity proteins.” Other aligned cellular patterns or examples of noncanonical Wnt signaling, sometimes described as “Wnt/PCP” signaling, will not be discussed.

PCP components and molecular asymmetries

Two molecular systems control PCP behavior, the “core” and the “Fat–Dachsous (Ft–Ds)” PCP pathways. A key feature of both is the asymmetric distribution of their constituents (Fig. 2). The core PCP pathway is composed of the multipass transmembrane proteins Frizzled (Fz), Van Gogh (Vang; also known as Strabismus/Stbm), and Flamingo (Fmi; also known as Starry night/Stan), and the cytosolic components Dishevelled (Dsh), Prickle (Pk), and Diego (Dgo). On one edge of the cell reside Fz, Dsh, and Dgo, and on the opposite side lie Vang and Pk (Figs. 1 C and 2 B; Axelrod, 2001; Strutt, 2001; Feiguin et al., 2001; Tree et al., 2002; Bastock et al., 2003). The atypical cadherin, Fmi, resides on both sides, where it forms homodimers between neighboring cells (Usui et al., 1999; Shimada et al., 2001). These molecular asymmetries are observed in sensory hair cells of the vertebrate inner ear (Wang et al., 2005, 2006a,b; Montcouquiol et al., 2006; Deans et al., 2007; Song et al., 2010), the mammalian epidermis (Devenport and Fuchs, 2008; Devenport et al., 2011), brain ventricles (Tissir et al., 2010), and trachea (Vladar et al., 2012). Mutations in core PCP components lead to a loss or randomization of polarity and misalignment of cellular structures along the tissues axis.Open in a separate windowFigure 2.Asymmetric localization of PCP components. (A) PCP asymmetry develops progressively from an initially uniform distribution of core PCP proteins. Fz, Dsh, and Dgo (red) localize to the distal/posterior edge, whereas Vang and Pk (turquoise) localize to the proximal/anterior side. Fmi (dark blue) localizes to both sides, where it forms homodimers between neighboring cells. (B) Feedback interactions between core PCP components. A Fz–Fmi complex interacts preferentially with a Vang–Fmi complex between cells, whereas proximal and distal complexes antagonize one another within the cell. (C) Ds and Fj are expressed in opposing gradients in the Drosophila wing blade. Fj positively modulates Ft activity, leading to a gradient of Ft activity across the wing (not depicted). (D) Graded expression of Ds and Fj leads to asymmetric cellular localization of Ds and Ft, which form heterodimers between adjacent cells. Dachs, a downstream component of the Ft–Ds pathway, also localizes asymmetrically in association with Ds.The Ft–Ds pathway includes the large protocadherins Ft and Ds and the Golgi resident transmembrane kinase, Four-jointed (Fj; for review see Matis and Axelrod, 2013; Thomas and Strutt, 2012). Similar to the core system, Ft–Ds also displays molecular asymmetries in flies. Ds and its ligand Ft accumulate on opposite cell edges, where they form intercellular heterophilic interactions (Fig. 2 D; Matakatsu and Blair 2004; Ambegaonkar et al., 2012; Brittle et al., 2012). Unlike the core components, Ds and Fj are expressed in complementary gradients in the Drosophila eye and developing wing, which contribute to the cellular asymmetries of Ds and Ft (Fig. 2, C and D). Whether Ft–Ds–Fj gradients and asymmetries are conserved in vertebrate systems has yet to be determined.

Segregation of cortical polarity proteins: Shaking hands with the enemy

The asymmetric segregation of Fz–Dsh–Fmi and Vang–Pk–Fmi complexes to opposite sides of the cell relies on their mutual exclusion intracellularly and their preferential binding between neighboring cells (Fig. 2 B; for review see Strutt and Strutt, 2009). There is mutual interdependence among core PCP components for their asymmetric localization. Depletion of any one core PCP component results in a loss of asymmetry of all the others. In addition, PCP asymmetry develops progressively from initially uniform distributions (Fig. 2 A). Thus, PCP asymmetry can be thought of not as a simple hierarchy of interactions, but the result of feedback amplification of an initial directional bias.

Intercellular interactions.

PCP requires cell–cell communication, mediated by the transmembrane components of the core system, where it is thought that Fz–Fmi on one cell interacts with Vang–Fmi on its neighbor. These interactions are best understood in the Drosophila wing blade, where PCP controls the alignment of wing hairs along the proximal–distal axis (Figs. 1 A and 2, A and B). In the wing, Vang–Pk localize to the proximal face of each cell, whereas Fz–Dsh–Dgo localize distally adjacent to the wing hair (Figs. 1 C and 2, A and B; Axelrod, 2001; Strutt, 2001; Tree et al., 2002; Bastock et al., 2003; Das et al., 2004). By generating mutant clones and examining PCP localization at the clone border, the intercellular interactions between neighboring cells can be assessed in vivo. For example, when Fz is lacking within a clone, leaving only Vang–Fmi available at cell junctions, then Fz–Fmi in adjacent wild-type cells is recruited to the clone border (Chen et al., 2008). Vang mutant clones produce a similar effect, but in this case the excess Fz recruits Vang to clone borders (Bastock et al., 2003). What mediates these intercellular asymmetric interactions? One possibility is that Vang and Fz interact directly, and in vitro binding assays between the Fz extracellular domain and Vang suggest that this mechanism is possible (Wu and Mlodzik, 2008). However, mutants of Fz or Vang lacking their extracellular domains can still recruit one another between cells, which suggests that something else must bridge the two proteins (Chen et al., 2008). The seven-pass transmembrane cadherin, Fmi, likely performs this function. Fmi is essential for the junctional recruitment of Fz and Vang, and Fmi homodimers appear to be functionally asymmetric (Chen et al., 2008; Strutt and Strutt, 2008; Struhl et al., 2012). Clonal overexpression of Fmi preferentially recruits Fz to the clone border, even in the absence of Vang, which suggests that excess or unpaired Fmi is in a configuration that has higher affinity for Fmi–Fz than Fmi–Vang (Chen et al., 2008; Strutt and Strutt, 2008). Thus, Fmi may exist in two forms depending on whether it is paired with Fz or Vang, but the molecular basis for this difference is not known (Chen et al., 2008; Strutt and Strutt, 2008; Struhl et al., 2012).

Amplification of asymmetry.

Intercellular Fz–Fmi and Vang–Fmi complexes can form between cells in any orientation, so how do they resolve into discrete and opposed asymmetric domains? One way is through clustering of Fz–Fmi and Vang–Fmi complexes of the same orientation, and the cytoplasmic PCP components are particularly important for this function. As PCP complexes grow increasingly asymmetric, they cluster into discrete puncta that are stably associated with the plasma membrane and are resistant to endocytosis (Strutt et al., 2011). FRAP analysis of Fz-containing puncta demonstrated that they are highly stable compared with diffuse Fz-GFP, and have limited lateral mobility within the membrane. In the absence of Dsh, Pk, or Dgo, the size, intensity, and stability of Fz-containing puncta are diminished (Strutt et al., 2011), whereas overexpression causes Fz accumulation and coalescence into larger puncta (Feiguin et al., 2001; Tree et al., 2002; Bastock et al., 2003). Although the precise mechanisms driving PCP puncta formation are not known, the cytoplasmic components do not affect endocytosis, which suggests that they contribute to puncta formation by clustering intercellular complexes (Strutt et al., 2011). Pk can interact homophilically (Jenny et al., 2003; Ayukawa et al., 2014), which might promote clustering of proximal Vang–Pk–Fmi complexes. It will also be interesting to determine whether the cytoskeleton is directly connected to PCP complexes to minimize lateral mobility within the membrane.A second mechanism contributing to PCP asymmetry is directed transport. Live imaging of fluorescently tagged PCP proteins in pupal wings showed that Fz- and Dsh-containing particles travel across the cell in a proximal-to-distal direction (Shimada et al., 2006; Matis et al., 2014; Olofsson et al., 2014). These particles most likely represent endosomes undergoing transcytosis, as they arise from the proximal cortex and are labeled by the endocytic tracer FM4-64. This mechanism could serve to amplify asymmetry or even provide the initial polarity bias by removing proximal Fz–Dsh–Fmi complexes and transporting them to the distal side. Directed PCP transport is mediated by an array of subapical, noncentrosomal microtubules (MTs) that align along the proximal–distal axis, with the plus ends oriented with a slight distal bias (Hannus et al., 2002; Shimada et al., 2006; Harumoto et al., 2010; Matis et al., 2014; Olofsson et al., 2014). Ft and Ds are required for proximal–distal MT alignment (Harumoto et al., 2010), which suggests that the Ft-Ds system may feed into the core PCP system by orienting cytoskeletal architecture to deliver Fz–Dsh–Fmi complexes to the distal edge of the cell.Directed transport of Vang-containing endosomes has not been reported in flies, but selective trafficking could target Vang to specific membrane domains. In mammalian cells, exit of the Vang homologue Vangl2 from the trans-Golgi network (TGN) requires Arfrp1 (an Arf-like GTPase) and the clathrin adaptor complex AP-1, neither of which are required for the transport of a mammalian Fz homologue Fz6 or Fmi/Celsr1, which suggests that the differential sorting of PCP complexes to opposite sides of the cell could initiate at TGN export (Guo et al., 2013). Whether newly synthesized Vang and Fz proteins are transported to opposing cell surfaces from the TGN has not yet been explored.Microtubule orientation also correlates with PCP asymmetry in mouse trachea epithelial cells, where PCP coordinates the alignment of motile cilia (Vladar et al., 2012). MTs are planar polarized with their plus ends oriented toward the Fz–Dvl domain, and disruption of MTs with nocodazole impairs core PCP localization. Similarly, MTs are needed to establish Pk asymmetry in gastrulating zebrafish embryos (Sepich et al., 2011). However, in the skin epithelium, MTs align perpendicular to the axis of PCP asymmetry (unpublished data). Thus, directed transport along MTs may not be required in all tissue types for the establishment of PCP asymmetry.

Negative regulation.

Repulsive interactions between Vang- and Fz-containing complexes may also contribute to the amplification of asymmetry, and cytoplasmic proteins have been proposed to perform this function. Pk and Dgo both bind to Dsh in vitro, interacting with the same domain on Dsh in a mutually exclusive manner (Jenny et al., 2005). In addition, overexpression of Pk can prevent Dsh translocation to the membrane (Tree et al., 2002; Carreira-Barbosa et al., 2003), which suggests that Pk binding to Dsh could displace it from the proximal side of the cell. On the distal side, Dgo binding to Dsh would prevent association with Pk, thus enhancing Dsh distal localization. This increase in Dsh and Pk asymmetry would then positively feed back by clustering the transmembrane components into stable membrane domains.Modulation of PCP protein levels by ubiquitin-mediated degradation also leads to feedback by restricting the amount of one PCP protein to antagonize another. In flies, regulation of Dsh by a Cullin-3-BTB E3 ubiquitin ligase complex limits its levels at cell junctions (Strutt et al., 2013a). Reduction of Cullin-3 leads to an increase in overall core PCP protein levels, a reduction of asymmetry, and defects in wing hair polarity, which is consistent with Dsh overexpression phenotypes (Strutt et al., 2013a). SkpA, a subunit of the SCF E3 ligase, regulates Pk levels by promoting its degradation in a Vang-dependent manner (Strutt et al., 2013b). In mice, Smurf E3 ligases ubiquitinate Pk and promote its local degradation by binding to phosphorylated Dvl2 (a mammalian homologue of Dsh; Narimatsu et al., 2009). Smurfs are required for Pk localization in the inner ear and floor plate, and their removal leads to defects in convergent extension (CE) and stereocilia alignment (Narimatsu et al., 2009). Thus, targeting Pk for degradation either balances total Pk protein levels or targets a specific pool of Pk for ubiquitination and proteasome degradation.

Tissue-level polarity cues: This way or that?

What provides the tissue-level polarity cue that biases core PCP asymmetry in one direction over another? This is perhaps the most fundamental, yet poorly understood, element of PCP. Current models propose that an upstream, graded cue provides an initial bias in PCP asymmetry by regulating the levels, localization, or activity of one or more of the core proteins. Gradients are attractive candidates for providing global polarity cues, as they can act across many cells and define the tissue boundaries over which polarity must be oriented.

Ft–Ds–Fj.

Unlike the core proteins, Ds and Fj are nonuniformly expressed in the Drosophila eye, wing, and abdominal segments, and as such, the Ft–Ds module has been proposed to provide a global polarity cue (Fig. 2, C and D; for review see Ma et al., 2003; Yang et al., 2002; Thomas and Strutt, 2012; Matis and Axelrod, 2013). Ft and Ds are heterodimeric cadherins, regulated by the Golgi kinase Fj (Ishikawa et al., 2008; Brittle et al., 2010; Simon et al., 2010). The complementary expression patterns of Ds and Fj are thought to give rise to asymmetric Ft and Ds protein localization, with Ft and Ds localizing to opposite sides of each cell (Fig. 2 D; Ambegaonkar et al., 2012; Bosveld et al., 2012; Brittle et al., 2012). Because Fj positively regulates the activity of Ft, a gradient of Ft activity is expressed across the wing complementary to that of its ligand, Ds (Simon et al., 2010). Mutations in the Ft–Ds system give rise to swirling wing hair patterns, and disrupt the global alignment of core PCP proteins, but not their asymmetric distributions.An appealing model for symmetry breaking in the early Drosophila wing is that cellular asymmetries of Ft–Ds polarize MT organization and promote the distal transport of Fz–Dsh–Fmi vesicles (Shimada et al., 2001, Harumoto et al., 2010; Matis and Axelrod, 2013; Matis et al., 2014). This would produce an initial bias in Fz–Dsh localization, which would then be amplified by feedback interactions. However, several pieces of evidence have prevented the model from gaining universal acceptance. First, Ds and Fj gradients are oriented in opposite directions with respect to the core PCP proteins in the wing compared with the eye and abdomen. This discrepancy has been rectified with the finding by two independent groups that cells interpret Ft–Ds–Fj gradients differently depending on which of two Pk isoforms is expressed (Ayukawa et al., 2014; Olofsson et al., 2014). Second, the Ft–Ds system can orient PCP independently of the core pathway, and thus the two systems orient polarity in parallel, as opposed to in a single, common pathway (Casal et al., 2006). Third, Ft–Ds mutations affect core PCP orientation only regionally in the wing, which suggests that, if Ft–Ds provides a global bias, other, redundant cues must also exist (Matakatsu and Blair, 2006; Matis et. al., 2014). Finally, the direction of Ft–Ds and core PCP asymmetry diverges late in wing development, where the two systems become completely uncoupled. Intriguingly, the extent of coupling depends on which isoform of Pk is expressed (Merkel et al., 2014). Perhaps the simplest explanation for Ft–Ds function is that it can both transmit polarity information independent of the core system and organize the cytoskeleton to provide an initial bias of core PCP asymmetry, but which mechanism predominates depends on the tissue and developmental stage.

Wnts.

Wnt proteins have long been considered attractive candidates to provide tissue-level polarity cues because Fz and Dsh are primary components of the Wnt–β-catenin signaling pathway. Wnts are secreted glycoproteins that bind to Fz and other receptors, and often display graded expression. In vertebrates, Wnts are clearly important regulators of PCP, but whether they act instructively or permissively is unclear. In zebrafish, Wnt5a and Wnt11 are required for CE movements during gastrulation, but uniform expression of Wnt11 rescues the mutant phenotype, which suggests that it is permissive rather than instructive (Heisenberg et al., 2000; Kilian et al., 2003). Wnt5a is expressed in a gradient along the axis of polarity in the mouse inner ear, where it interacts genetically with Vangl2 in cochlear hair cell orientation (Qian et al., 2007). In the mouse limb, Wnt5a and its atypical receptor Ror2 are required for limb elongation and the asymmetric localization of Vangl2 at the proximal face of converging and extending chondrocytes (Gao et al., 2011). Wnt5a is expressed in a distal-to-proximal gradient, which induces a gradient of Vangl2 phosphorylation. The functional consequences of Vangl2 phosphorylation are unknown but Vangl2 cellular asymmetry appears to be strongest distally, where Wnt5a and Vangl2 phosphorylation levels are highest (Gao et al., 2011).While several studies had argued against the involvement of Wnt proteins in Drosophila PCP (Lawrence et al., 2002; Chen et al., 2008), it was recently discovered that Wingless (Wg) and Wnt4a act redundantly to orient PCP in the wing, particularly near the wing margin (Wu et al., 2013). Misexpression of Wg or Wnt4a reorients wing hair polarity in a pattern reminiscent of Fz loss of function, which suggests that Wnt gradients may orient polarity by antagonizing Fz. Consistently, the ability of Fz and Vang to recruit one another between adjacent cells in culture was inhibited by the addition of Wg or Wnt4a, which suggests that Wnts could provide a polarizing cue by diminishing Fz–Vang interactions at the margin of the wing, where Wnt expression is highest (Wu et al., 2013). However, Wnt4a overexpression also reorients MT alignment, suggesting that Wnts may act as polarity cues thorough an effect on the cytoskeleton (Matis et. al., 2014). Alternatively, Sagner et al. (2012) suggest that Wg orients core PCP indirectly through its effects on wing patterning and growth. Although the evidence for Wnt gradients as global PCP cues is accumulating, the mechanisms by which they regulate core protein levels or activity remain to be elucidated.

Mechanical forces.

Anisotropic mechanical forces that accompany growth and morphogenesis can also provide global polarizing cues. During wing development, PCP reorients in response to extensive morphogenetic changes that elongate the wing along the proximal–distal axis. In early pupal wings, PCP aligns toward the wing margin and then reorients during wing elongation and contraction of the wing hinge (Aigouy et al., 2010). These morphogenetic changes have broad effects on cell behavior, inducing cell elongation, oriented divisions, and cell rearrangements with a concomitant reorientation of PCP. Severing the wing pouch from the hinge blocks cell flows and PCP reorientation, which suggests that the anisotropic tension from hinge contraction drives tissue flow and the reorientation of polarity (Aigouy et al., 2010). Although this model doesn’t explain what initially biases PCP, it does demonstrate how the morphogenetic processes that shape tissues can completely remodel global PCP alignment. This is an attractive model to explain how PCP aligns over very large tissues, like the mammalian skin, where hairs consistently reorient along regions of extensive tissue elongation such as the face, limbs, and ears.

Downstream effectors of PCP: Steering the wheel

If PCP is the cell’s compass, it is also the steering wheel, directing downstream, polarized cell behaviors in response to global directional cues. PCP can polarize a wide range of cell behaviors, which suggests that it can intersect with numerous downstream effectors. We focus here on three examples where the molecular mechanisms linking core PCP to their polarized outputs have recently been elucidated.

Distal positioning of wing hairs.

Each cell of the Drosophila wing blade emits a single actin-rich protrusion from its distal edge. The placement of the wing hair strongly correlates with the position of Fz–Dsh–Fmi, which suggests that core proteins may localize cytoskeletal regulators to distinct positions within the cell (Strutt and Warrington, 2008). On the proximal side, Vang recruits a group of proteins that negatively regulate actin prehair formation: Inturned, Fuzzy, and Fritz (Adler et al., 2004; Strutt and Warrington, 2008). These three proteins regulate Multiple Wing Hairs, a GTPase-binding/formin-homology 3 (GBD/FH3) domain protein thought to repress actin polymerization (Strutt and Warrington, 2008; Yan et al., 2008). This restricts actin nucleation to distal positions within the cell, and in the absence of Multiple Wing Hairs, ectopic actin bundles form across the apical surface (Wong and Adler, 1993). On the distal side, casein kinase 1 γ CK1g/gilamesh is required to further refine prehair nucleation to a single site through a parallel mechanism involving Rab11-dependent vesicle traffic to the site of prehair formation (Gault et al., 2012). Rho and Rho kinase (Drok) have also been implicated in wing hair formation, but their roles are difficult to dissect due to the numerous functions of Rho in cell shape and cell division (Winter et al., 2001; Yan et al., 2009).

Actomyosin contraction and convergent extension (CE).

CE was the first vertebrate process to be linked molecularly to PCP (Wallingford et al., 2000). During CE, mesenchymal cells elongate, form mediolateral-directed protrusions, and intercalate mediolaterally, narrowing the mediolateral axis while simultaneously lengthening the anterior–posterior (A-P) axis (Fig. 3 A; Keller, 2002). Mediolateral polarization, elongation, and intercalation are lost when core PCP components are disrupted, leading to a failure in CE (Tada and Smith, 2000; Wallingford et al., 2000; Goto and Keller, 2002; Jessen et al., 2002). While several PCP-dependent mechanisms have been proposed to mediate CE movements, two recent studies provide direct mechanistic links between asymmetrically localized core PCP components and CE behaviors. In neuroepithelial cells, PCP specifies the localization of myosin to the A-P faces of intercalating cells. Fmi/Celsr1 and Dvl recruit the formin DAAM1 to the A-P junction, which in turn binds and activates PDZ-RhoGEF. This likely activates RhoA and myosin contractility specifically at A-P junctions, resulting in medial-directed cell intercalation and neural plate bending (Nishimura et al., 2012). A similar mechanism was found to drive CE movements of mesenchymal cells during Xenopus gastrulation. In this case, Fritz and Dsh help to localize septins to mediolateral vertices, where they spatially restrict cortical actomyosin contractility and junctional shrinking to A-P cell edges, thus driving cell intercalation (Fig. 3 A; Kim et al., 2010; Shindo and Wallingford, 2014). Together these studies show how asymmetric PCP localization produces collectively polarized cell behaviors through spatial modulation of the cytoskeleton.Open in a separate windowFigure 3.Polarized cell behaviors controlled by PCP. (A) PCP drives convergent extension (CE). CE in vertebrates is driven by mediolateral intercalation, which narrows the mediolateral axis while simultaneously lengthening the A-P axis. Mediolateral intercalation is accompanied by cell polarization and elongation and the formation of mediolateral protrusions, all of which require core PCP function. Pk localizes anteriorly (Ciruna et. al., 2006; Yin et al., 2008), whereas Dsh localizes posteriorly (Yin et. al., 2008). In addition, PCP proteins recruit myosin to A-P cell borders, leading to actomyosin contractility and junctional shrinking. (B) Asymmetric cell division. Drosophila sensory organ precursors (SOPs) divide asymmetrically along the epithelial plane, giving rise to distinct anterior and posterior daughters. Spindle alignment along the A-P axis is PCP dependent. Dsh interacts with Mud/NuMA and the dynein complex posteriorly while Vang links Pins/LGN-Mud/NuMA-dynein on the anterior. This links astral MTs to the A-P cortex, bringing the spindle into register with the A-P axis. (C) Positioning of the kinocilium in the inner ear. The placement of kinocilium in sensory hair cells of the inner ear determines the position of V-shaped stereocilia bundles. Gαi and mPins/LGN localize on the abneural side on the hair cell, where they are required for abneural positioning the MT-based kinocilium. The collective alignment of kinocilia and stereocilia bundles across the epithelium requires the core PCP component Vangl2. Vangl2 (light green) localizes to the abneural side of supporting cells. Whether Fz (dark blue) associates on the opposite face is not yet clear (Ezan et. al., 2013).

Positioning of centrosomes and cilia.

PCP regulates the positioning of MT-based structures including the mitotic spindle and cilia. In Drosophila sensory organ precursors and early zebrafish embryos, PCP controls mitotic spindle orientation along the epithelial plane by interacting with the highly conserved spindle orientation complex, which links astral MTs to the cell cortex through Mud/NuMA-mediated recruitment of the dynein complex (Ségalen et al., 2010). To orient the spindle, posteriorly localized Dsh binds to Mud/NuMA, which recruits the dynein complex and astral MTs to the posterior cortex. On the anterior side, Pins/LGN recruits Mud/NuMA, bringing the spindle into A-P alignment (Fig. 3 B). Similarly, PCP was recently shown to interact with the spindle orientation machinery to position the kinocilium in nondividing cells of the inner ear (Ezan et al., 2013; Tarchini et al., 2013). In vestibular hair cells, Gαi and mPins/LGN localize to the abneural cortex, opposite Vangl2, where they are required for kinocilium positioning and subsequent alignment of stereocilia bundles (Fig. 3 C; Ezan et al., 2013). MT plus ends and dynein also show an abneural bias suggesting that Gαi-mPins/LGN induces pulling on MTs by a similar mechanism that orients the centrosome during spindle orientation. Vangl2 is required for the alignment of Gαi-Pins/LGN crescents between cells, coordinating kinocilia positioning and stereocilia polarity across the tissue (Fig. 3 C; Ezan et al., 2013). Thus the PCP pathway co-opts the spindle orientation machinery to specify not only the division plane but also cilia position in nondividing cells. As PCP is required for asymmetric cilia positioning in a wide range of cell types, including the node (Antic et al., 2010; Borovina et. al., 2010; Song et al., 2010; Hashimoto et. al., 2010), it will be interesting to determine whether this mechanism is conserved.

Concluding remarks

PCP is a fundamental and highly conserved process coordinating a vast number of polarized cell behaviors. While the number of functions ascribed to PCP continues to grow, an understanding of the mechanisms establishing PCP is still far from complete. The development of cellular asymmetry from uniform distributions is not well understood, and will benefit from recent advances in high-resolution, time-lapse imaging with photoconvertible fluorescent tags. Other important issues to resolve include deciphering the structural domains and biochemical interactions mediating intercellular communication, identifying the global cues that orient PCP especially in vertebrates, and deciphering the mechanisms by which complex multicellular structures, like lung branches and hair follicles, are oriented by the PCP machinery.     

Keratins and protein synthesis: the plot thickens

In addition to protecting epithelial cells from mechanical stress, keratins regulate cytoarchitecture, cell growth, proliferation, apoptosis, and organelle transport. In this issue, Vijayaraj et al. (2009. J. Cell Biol. doi:10.1083/jcb.200906094) expand our understanding of how keratin proteins participate in the regulation of protein synthesis through their analysis of mice lacking the entire type II keratin gene cluster.Keratins are members of the intermediate filament family. They form intricate cytoskeletal networks via the assembly and organization of 10–12-nm filaments, whose formation is initiated through coiled-coil interactions between a type I keratin (e.g., K18 and -19) and a type II keratin (e.g., K8; Kim and Coulombe, 2007). Most keratin proteins have been shown to contribute to the protection of cells and tissues from mechanical and nonmechanical stresses (Toivola et al., 2005; Kim and Coulombe, 2007). Whereas the large number of keratin genes (n = 54; Schweizer et al., 2006) and the heteropolymerization-based assembly of their protein products should in part serve the purpose of modulating the viscoelastic properties of keratin networks to assist various cellular needs, common sense dictates that there should be additional roles for these proteins. Not surprisingly, efforts over the last decade have implicated keratin proteins in several nontraditional functions, including cytoarchitecture, proliferation and growth, apoptosis, and organelle transport, to name a few (Toivola et al., 2005; Kim and Coulombe, 2007). Yet, the high homology between several keratin proteins along with their overlapping distribution in epithelia has limited researchers'' progress toward uncovering the full range of keratin function in vivo (Baribault et al., 1994; Tamai et al., 2000; McGowan et al., 2002; Kerns et al., 2007). In this issue, Vijayaraj et al. report on the ultimate bypass of redundancy by eliminating all keratin filaments via the generation of a mouse strain lacking all type II keratins (KtyII^−/− mice). The study of these mice, which are viable until embryonic day 9.5, led to the discovery of a novel mechanism through which keratin proteins regulate protein synthesis and cell growth (Kim et al., 2006, 2007; Galarneau et al., 2007). The authors'' findings also showcase the recent conceptual and technical advances of chromosome engineering in the mouse genome.For over a decade, the Cre-loxP site-specific recombination system has been a popular method to generate targeted conditional knockout embryonic stem (ES) cells and mice. Although recombination efficiency is inversely proportional to the distance between loxP sites, larger chromosomal rearrangements have been successfully engineered into mouse ES cells using Cre-loxP (Ramírez-Solis et al., 1995). Generating such targeting vectors is cumbersome using traditional cloning methods. This said, DNA recombineering eliminates many of the constraints of finding unique restriction enzyme sites in genomic DNA sequences (Liu et al., 2003). Also, an Sv129 bacterial artificial chromosome (BAC) library generated from AB2.2 ES cells makes it easier to obtain large genomic sequences or even target ES cells directly with loxP-containing BACs (Liu et al., 2003; Adams et al., 2005). Finally, the Mutagenic Insertion and Chromosome Engineering Resource (MICER), a library of ready-made targeting vectors spread throughout the mouse genome, is now available (Adams et al., 2004). Vijayaraj et al. (2009) used MICER vectors to remove the entire 0.68-Mb keratin type II cluster on mouse chromosome 15 (Fig. 1 A). Owing to the interdependency of type I and II keratins for 10-nm filament assembly (Fig. 1 B), the resulting KtyII^−/− mice represent the first successful elimination of all keratin filaments from an organism as complex as a mouse.Open in a separate windowFigure 1.Genome organization, assembly, and epithelial function of keratins. (A) Arrangement of keratin clusters in the mouse genome. Human keratin genes that have not been identified or annotated in the mouse genome are shown on the bottom side and marked with a question mark. The arrows mark the boundaries of the region deleted by Vijayaraj et al. (2009) on mouse chromosome 15. (B) Summary of the multistep pathway through which type I and II keratin protein monomers polymerize to form 10-nm filaments. The antiparallel docking of the lollipop-shaped coiled-coiled dimers along their lateral surfaces generates structurally apolar tetramers and accounts for the lack of polarity of assembled keratin intermediate filaments. For all steps in the pathway, the forward (assembly promoting) reaction is heavily favored in vitro (Kim and Coulombe, 2007). (C) Keratins influence the localization and function of many cellular components. As highlighted here, keratins interact with and modulate the mTOR pathway in several ways, both in skin keratinocytes and gut epithelial cells, and regulate the localization of microtubules, γ-tubulin, and GLUT transporters in polarized epithelia. Components are not drawn to scale in this schematic.KtyII^−/− embryos display severe growth retardation and die midgestation (Baribault et al., 1993; Hesse et al., 2000; Tamai et al., 2000). Smaller cell size has been observed previously in K17^−/− skin keratinocytes and K8^−/− liver hepatocytes, correlating with altered Akt/mammalian target of rapamycin (mTOR) signaling (Fig. 1 C) and a reduction in bulk protein synthesis (Kim et al., 2006; Galarneau et al., 2007). Although K17 appears to modulate the mTOR pathway through its physical interaction with 14-3-3–σ in keratinocytes (Fig. 1 C; Kim et al., 2006), the mechanism for how K8 influences protein synthesis in hepatocytes is less clear but appears to integrate responses to both insulin and integrin stimulation (Galarneau et al., 2007). Loss of K8 is also associated with an increase in Akt activity (Galarneau et al., 2007), which is contrary to the findings in the K17^−/− setting (Kim et al., 2006), calling into question whether the two settings use the same mechanism to modulate mTOR signaling. Vijayaraj et al. (2009) uncover yet another path through which keratins are able to influence protein synthesis. The authors find that loss of all keratin filaments causes mislocalization of GLUT transporters and disruption of glucose homeostasis through AMP kinase (AMPK) activation. In addition, the authors report that in the absence of the keratin network, AMPK phosphorylates Raptor, which then interacts with mTOR to repress protein synthesis and hamper cell growth (Fig. 1 C). These findings further the evidence for an important role of keratin proteins (or filaments) in the regulation of translation and epithelial cell growth. However, they also raise the question of whether keratins affect mTOR signaling via an as of yet unknown, common denominator or whether several mechanisms come together, perhaps in a cell type– and context-dependent fashion, to achieve the same downstream effect.Unlike actin and microtubules, keratin filaments are not believed to possess intrinsic polarity (Fig. 1 B). However, K8/K18 and/or K8/K19 filaments play a significant role in maintaining apicobasal compartmentalization in simple epithelial linings in both the small intestine (Ameen et al., 2001; Oriolo et al., 2007) and colon of adult mice (Toivola et al., 2004) and have also been implicated in organelle transport (Toivola et al., 2005; Kim and Coulombe, 2007). The mechanism or mechanisms accounting for this surprising influence of keratins on the establishment and maintenance of spatial order in epithelial cells are unknown. Ameen et al. (2001) and Oriolo et al. (2007) recently made a dent in this mystery by showing that K8/K18 filaments are necessary for the proper localization of γ-tubulin to the apical compartment in polarized epithelial cells, thereby participating in the organization of noncentrosomic microtubules (note: the interested reader should examine a recent study by Bocquet et al. [2009], which shows a role for neuronal intermediate filaments in tubulin polymerization in axons). Similar to previous observations made in K8^−/− mice (Ameen et al., 2001; Toivola et al., 2004), Vijayaraj et al. (2009) show that apical proteins, particularly GLUT1 and -3, are mislocalized in KtyII^−/− embryonic epithelia. However, in this instance, microtubule organization appears to be intact. Although the authors'' experimental findings again nicely demonstrate a role for keratin proteins in the establishment of polarity in simple epithelial settings, the underlying mechanism or mechanisms still need to be ascertained.The mouse model generated by Vijayaraj et al. (2009) has important implications for the field of keratin biology and intermediate filaments in general. It will allow researchers to address central questions about the contributions of keratins during development and tissue homeostasis unencumbered by the redundancy of properties and functions among members of this large family. The availability of tissue- or cell type–specific promoters makes it possible to express the Cre recombinase in specific epithelial settings, thereby promoting the elimination of keratins in a more restricted fashion. It will be interesting to see how the total loss of keratin filaments affects different tissues and subpopulations of cells, highlighting essential functions and perhaps uncovering previously unappreciated roles for keratins in complex cellular processes.