The tubulin code in microtubule dynamics and information encoding

Summary

Microtubules are non-covalent mesoscale polymers central to the eukaryotic cytoskeleton. Microtubule structure, dynamics and mechanics are modulated by a cell’s choice of tubulin isoforms and posttranslational modifications, a “tubulin code” which is thought to support the diverse morphology and dynamics of microtubule arrays across various cell types, cell cycle and developmental stages. We give a brief historical overview of research into tubulin diversity and highlight recent progress towards uncovering the mechanistic underpinnings of the tubulin code. As a large number of essential pathways converge upon the microtubule cytoskeleton, understanding how cells utilize tubulin diversity is crucial to understanding cellular physiology and disease.

Introduction

Microtubules are one of the four non-covalent polymers that constitute the cytoskeleton, an integrated filamentous protein network that gives cells their shape, facilitates their movement, organizes their cytoplasm and builds the spindle needed to segregate genetic material. Microtubules are highly dynamic and oscillate stochastically between periods of growth and depolymerization, a property known as “dynamic instability” (Figure 1; (Horio, 1986; Mitchison, 1984; Walker, 1988)). The microtubule building block is the αβ-tubulin heterodimer. It assembles head-to-tail to form a polar protofilament with α-tubulin exposed on one end and β-tubulin at the other (reviewed in (Nogales, 2001)). Protofilaments assemble laterally to form a hollow cylinder (Figure 1). Most eukaryotic cells make microtubules with 13 protofilaments (Tilney et al., 1973). At the resolution of the optical microscope, all microtubules look the same, but this is deceptive. Microtubules in cells are highly complex, composed of multiple tubulin isoforms and functionalized with diverse evolutionarily conserved posttranslational modifications. This tubulin diversity alters intrinsic microtubule properties such as their dynamics and mechanics, as well as the recruitment and activity of motors and microtubule-associated proteins (MAPs). This variation has been interpreted as a “tubulin code” by which cells regulate microtubule diversity to achieve specific, physiological goals (Janke and Bulinski, 2011; Verhey and Gaertig, 2007; Yu et al., 2015). How cells write and read the tubulin code is largely unknown, but the advent of new tools for in vitro and in vivo tubulin manipulation, and the identification of an extensive list of tubulin-modifying enzymes is now bringing this fundamental problem into focus.

Figure 1. Microtubule structure and dynamics.

Microtubules are non-covalent cylindrical polymers assembled from 13 protofilaments formed by the head-to-tail addition of αβ-heterodimers. They grow by the addition of GTP-tubulin at their ends. Once incorporated into the microtubule, the GTP is hydrolyzed to GDP. Microtubules cycle between periods of growth and depolymerization. The transition from growth to depolymerization is called catastrophe. The transition from depolymerization to growth is called rescue. Microtubule dynamics are characterized by four parameters: rate of polymerization and depolymerization and frequency of catastrophe and rescue events.

The genetic and chemical complexity of tubulin

Humans have nine isoforms of both α- and β-tubulin. Certain isoforms, like α1A (TUBA1A) and βI (TUBB) are ubiquitous. Others have cell- or tissue-specific expression and are important for the function of specialized microtubule arrays in sperm, platelets and neurons (reviewed in (Ludueña, 2013) and (Breuss et al., 2017)). Early evidence for isoform-specific tubulin function, from studies in Drosophila showed that βIV-tubulin, the dominant β-isoform in flagella and cilia, is important for specifying correct axonemal structure. A sequence motif conserved in this isoform is important for this activity (Kemphues et al., 1979; Nielsen et al., 2001). The first mechanosensation screen revealed that the mec-12 and mec-7 α- and β-tubulin genes are required for the formation of 15-protofilament microtubules found in the mechanosensory neurons of Caenorhabditis elegans (Fukushige et al., 1999; Savage et al., 1989), providing further support for the specialized functions of tubulin isoforms. We now know that mutations in various tubulin isoforms are associated with a broad spectrum of human pathologies (Hussey and Fritz‐Laylin, 2019) ranging from bleeding disorders (Kunishima et al., 2014) and susceptibility to pathogens, (Salinas et al., 2014) to neurodevelopmental and neurodegenerative disorders (reviewed in (Breuss et al., 2017)). The molecular mechanisms behind these physiological responses remain largely unknown and will provide fertile ground for investigations for a new generation of cell biologists.

Tubulin is further diversified through many chemically distinct posttranslational modifications: acetylation of α- and β-tubulin, methylation of α-tubulin, phosphorylation of α- and β-tubulin, palmytoylation of α-tubulin, polyamination of α- and β-tubulin, detyrosination of α-tubulin, removal of the two C-terminal glutamates in α-tubulin to generate Δ2 tubulin, glutamylation and glycylation of α or β-tubulin (Figure 2; reviewed in (Janke and Bulinski, 2011; Yu et al., 2015)).

Figure 2. Sequence variability among tubulin isoforms concentrates on the intrinsically disordered C-terminal tails and at polymerization interfaces.

(A) Molecular surface of three protofilaments from a microtubule showing lattice contacts of the central αβ-tubulin dimer within a microtubule. α- and β-tubulin colored as in Figure 1. Sequences from Homo sapiens α- and β-tubulin isoforms were aligned and conservative sequence substitutions mapped in magenta on the atomic model (PDB ID: 6dpu). The M-loop position is highlighted by a yellow ellipse.

(B) Molecular surface of the αβ-tubulin dimer colored as in (A) illustrating the sequence variability on the C-terminal tails and lateral (between tubulin heterodimers in neighboring protofilaments) and longitudinal (between tubulin heterodimers within the same protofilament) polymerization interfaces.

Tubulin posttranslational modifications vary widely between cell types and are developmentally and spatially regulated (reviewed in (Yu et al., 2015)). Many studies have now documented the selective transport of various cargo on subsets of microtubules with specific modifications. For example, lysosomes switch between tyrosinated and acetylated microtubules as they move from the periphery to the cell interior (Guardia et al., 2016). Vesicles in epithelial cells exit the Trans-Golgi network along polyglutamylated microtubules functionalized with septin-2 fibers and aligned with the apicobasal cell axis (Spiliotis et al., 2008). In flagella, cargo movement is efficiently segregated for anterograde and retrograde transport by moving on the adjoining tracks of the microtubule doublet which are differentially modified (Stepanek and Pigino, 2016). Consistent with their involvement in basic cell biological processes, perturbations in tubulin modifications and the enzymes that introduce them are associated with cancers, neurodevelopmental and neurodegenerative disorders (reviewed in (Garnham and Roll‐Mecak, 2012; Magiera et al., 2018b)). The molecular mechanisms that give rise to the formation and readout of microtubule modification patterns in cells are largely unknown.

Historical Overview

Despite knowledge of the chemical complexity of tubulin and its importance for cell physiology dating as far back as the 1970s, our understanding of the cellular and developmental mechanisms behind this diversity and its effects on microtubule functions is still in its infancy. Why do we know so little? Arguably, this is due to (1) the historical development of the microtubule field which emphasized other aspects of tubulin biology after the initial realization of the diversity of microtubule structures and tubulin heterogeneity; (2) the challenges in studying the tubulin code because of the lack of an in vitro reconstitution system, antibodies for tubulin isoforms and tools for the specific and quantitative detection of tubulin modifications in vivo, and the difficulty with fluorescently tagging tubulin in vivo without adverse effects.

The isolation of tubulin (Borisy and Taylor, 1967; Shelanski and Taylor, 1967) ushered in a new biochemical era in microtubule research. Biochemical studies soon revealed that organisms have multiple tubulin forms (Feit et al., 1971; Witman et al., 1972) and that some are posttranslationally modified (Arce et al., 1975). This tubulin heterogeneity gave rise to the “multi-tubulin hypothesis” (Fulton and Simpson, 1976) which postulated that different tubulin isotypes were segregated in cells and required for the biogenesis and function of specialized microtubule structures. The large tubulin diversity became evident with the identification of multigene tubulin families in the 1980s (Cleveland et al., 1980; Lopata et al., 1983; Sullivan and Cleveland, 1986). A chemically diverse array of posttranslational modifications was identified in subsequent decades. Certain tubulin isoforms were found to be ubiquitous while other exhibited cell-type specific expression (reviewed in (Ludueña, 2013)). Likewise, tubulin posttranslational modifications were stereotyped and distributed asymmetrically in various cells (reviewed in (Janke and Bulinski, 2011; Yu et al., 2015)). These results generated a lot of excitement. Landmark studies documented the increase in tubulin modifications during cellular differentiation (Gundersen et al., 1989) and the different susceptibility to drug or temperature-induced depolymerization of cellular microtubules with different posttranslational modifications (Kreis, 1987; Schulze et al., 1987; Schulze and Kirschner, 1987; Webster et al., 1987). However, because many different tubulin isotypes could co-assemble both in vitro and in cells and because the few loss-of-function studies attempted failed to give easily interpretable phenotypes, there was no unifying theory of the roles of tubulin isoforms. As a result, the multi-tubulin hypothesis lost its sheen and most of the field directed its attention elsewhere.

The decades after the identification of tubulin isoforms saw an explosion in the discovery of factors that interact with tubulin and microtubules to build morphologically and dynamically diverse microtubule structures in cells: proteins that stabilize microtubules such as tau (Cleveland et al., 1977), motors that transport cargo (Brady, 1985; Vale et al., 1985a) or end-depolymerize microtubules (Walczak et al., 1996), proteins that localize to microtubule plus ends and modulate their dynamics (Rickard and Kreis, 1990) and enzymes that sever microtubules (Vale, 1991). Crucial to understanding the mechanism of these newly discovered molecular players was the introduction of in vitro reconstitution assays for motility (Vale et al., 1985b) and microtubule dynamics (Horio, 1986; Kinoshita et al., 2001; Walker, 1988). However, throughout all of this exciting time of building the microtubule cytoskeleton from the ground-up, tubulin itself was left out, its chemical and genetic diversity largely ignored and swept under the rug. This was in large part because many tubulin modifying enzymes remained unknown, making the manipulation of modifications difficult, and because there was no biochemical platform to investigate the consequences of tubulin diversity.

Tubulin diversity has returned into focus largely for three reasons: (1) identification of many of the enzymes that posttranslationally modify tubulin (Aillaud et al., 2017; Akella et al., 2010; Janke et al., 2005; Nieuwenhuis et al., 2017; Rogowski et al., 2009; Rogowski et al., 2010; van Dijk et al., 2007), (2) the development of biochemical tools for the production of naïve and recombinant tubulin (Minoura et al., 2013; Ti et al., 2018; Vemu et al., 2016; Widlund et al., 2012), and (3) the generation of tubulin with quantitatively defined posttranslational modifications (Valenstein and Roll-Mecak, 2016; Vemu et al., 2014) for in vitro reconstitution. The “multi-tubulin hypothesis” has resurfaced as the more expansive “tubulin code” hypothesis which postulates that tubulin isoforms and their chemically diverse posttranslational modifications regulate microtubule structure and dynamics and function as cellular GPS signals recognized and interpreted by cellular effectors for localized microtubule-based functional outputs (Verhey and Gaertig, 2007). Here we concentrate on recent advances that have returned the genetic and chemical complexity of tubulin to the fore of the microtubule field.

The tubulin problem

Why do we still understand so little about the effects of tubulin diversity, either genetically encoded or arising from posttranslational modifications? One reason is that tubulin did not succumb to standard molecular biology approaches until recently. While most other fields take recombinant tools for granted, tubulin has resisted attempts at recombinant production. One of the main difficulties is that the generation of microtubule-assembly-competent αβ-tubulin dimers requires a complex collection of chaperones and downstream factors (Lewis et al., 1997). Moreover, eukaryotic cells regulate overall tubulin levels tightly (Burke et al., 1989) and tubulin isoforms copolymerize. Because of a lack of a recombinant source, the field resorted to purifying bulk tubulin from mammalian brain, as tubulin is highly abundant in this tissue. This is problematic, however, because tubulin from this source contains multiple isoforms and is abundantly and diversely modified. In fact, most of the tubulin in the adult nervous system is heavily posttranslationally modified, with glutamylation being most abundant (Wolff et al., 1992; Wolff et al., 1982). Tens of different α– and β-tubulin species are present in these preparations (Vemu et al., 2017; Wolff et al., 1982). Tubulin purification from brain entails repeated cycles of temperature and GTP-driven polymerization and depolymerization (Weisenberg, 1972). This procedure has a critical unintended consequence for deciphering the roles of tubulin diversity: the topographical information originally encoded in the patterns of tubulin isoforms and modifications in cells is lost by tissue homogenization; modifications present in the original cellular microtubules become incorporated in random mosaic patterns when the tubulin is repolymerized in vitro. Thus, microtubules assembled from this tubulin are comprised of a randomized mixture of isoforms bearing multiple, chemically distinct and quantitatively varied posttranslational modifications. As a result, the substrates used to study motors and MAPs have jumbled codes; all the spatial information encoded in them by the cell has been irretrievably scrambled. This tubulin purification procedure also discards tubulin subpopulations that do not depolymerize robustly when subjected to cold, as microtubules with certain posttranslational modifications do (Brady et al., 1984). Lastly, this source also limits investigations to a fixed repertoire of tubulin isoforms found in neuronal tissue. To decipher the tubulin code, novel biochemical and biophysical methods were required for in vitro reconstitution to probe the functions of tubulin diversity.

Initial attempts to express tubulin in bacteria and then refold it using chaperones from reticulate lysate gave very small yields (Shah et al., 2001). Yeast tubulin could be purified, albeit at low yields (Barnes et al., 1992; Kilmartin, 1981), and offered the hope of purifying engineered variants (Gupta Jr et al., 2002). Since then, S. cerevisiae has been used to overexpress and purify tubulin mutants suitable for structural studies (Ayaz et al., 2012; Johnson et al., 2011) or mechanistic work into polymer dynamics (Geyer et al., 2015). More recently, it became possible to express and purify recombinant, engineered, and isotypically pure human tubulin (Minoura et al., 2013; Ti et al., 2018; Vemu et al., 2016).

The second major reason for the limited progress in understanding the code is that in vivo tubulin is chemically complex, and modifications are posttranslational. Thus, making a certain type of tubulin requires knowing, understanding and harnessing the properties of the enzymes involved in modifying it. The ability to make recombinant or unmodified tubulin coupled with the knowledge garnered from recent biochemical and structural characterization of tubulin modifying enzymes (Garnham et al., 2015; Garnham et al., 2017; Li et al., 2019; Liao et al., 2019; Mukai et al., 2009; Prota et al., 2013; Shida et al., 2010; Szyk et al., 2011; Szyk et al., 2014; Wang et al., 2019) allows now the generation of tubulin with quantitatively defined posttranslational modifications to be used in deciphering the direct effects of the tubulin code on microtubule behavior and the factors that interact with microtubules.

Tubulin isoforms specialize and diversify microtubules

Multicellular organisms contain multiple tubulin genes. Each tubulin protomer consists of a compactly folded “body” and an intrinsically disordered, negatively charged C-terminal tail (Nogales et al., 1998). The tubulin body participates in α-tubulin-β-tubulin intradimer and interdimer (microtubule lattice) interactions. The C-terminal tails (two per dimer) decorate the microtubule exterior (Figure 1), where they participate in interactions with molecular motors and MAPs. Since the tails are evolutionarily unconstrained by the need to preserve dimer folding and functional polymerization (microtubule lattice) interfaces, they are hotspots for tubulin sequence variation. The remaining sequence variation concentrates at lateral and longitudinal polymerization interfaces where it can modulate microtubule structure and dynamics (Figure 3). A hotspot for sequence variation is around the M-loop, which participates in inter-protofilament interactions (Figures 3A and B). The M-loop conformation is modulated by microtubule-targeting agents such as taxol (Kellogg et al., 2017), one of the most widely used anti-cancer chemotherapy drugs. The response of the different tubulin isoforms to microtubule targeting agents is of great interest because drug-resistant tumors change their tubulin isoform expression profiles (reviewed in (Gan et al., 2010)).

Figure 3. αβ-tubulin dimers are functionalized with a diverse arrays of chemically distinct posttranslational modifications.

Schematic of a neuron (microtubules, blue; actin, pink) with a close-up of a microtubule decorated with diverse posttranslational modifications. Microtubule colored as in Figure 1. α-tubulin tyrosination, orange; monoglutamaylation, light red; polyglutamylation red; monoglycylation, light green; polyglycylation, green; acetylation or methylation on α-tubulin Lys40, magenta; phosphorylation on Ser172, yellow; polyamination, purple. Microtubule cross-section shows the luminal position of Lys40. Boxes show the chemical structure of selected posttranslational modifications. We note that the linkage of the polyglutamate chain is still not clear with studies reporting both α-linked glutamate chains (Redeker et al., 1991), shown here (where the elongating glutamate forms a peptide bond with the branch glutamate residue) and γ-linked chains (where the elongating glutamate forms an isopeptide bond with the branch glutamate residue (Wolff et al., 1994)).

Both α– and β-tubulin protomers bind to GTP. The GTP site on α-tubulin is at the αβ intradimer interface and not accessible for exchange. The GTP bound to β-tubulin, is hydrolyzed to GDP when tubulin incorporates into the microtubule lattice, and exchanges back to GTP on the depolymerized tubulin dimer (reviewed in (Desai and Mitchison, 1997)). Amino acids in the structural core and GTP-binding sites of the tubulin dimer are largely invariant (Figure 3D), suggesting either a unique solution to the requirement for tubulin to act as its own GTPase activating protein during tubulin polymerization, or that the need for the different nucleotide states of tubulin to interact with multiple partners has limited evolutionary drift (i.e., “the principle of continuity” (Orgel, 1968)). Interestingly, there is at present no evidence for a tubulin GTP exchange factor.

Remarkably, we do not know the exact stoichiometry of tubulin isoforms in most cell types, or even how α- and β-tubulin isoforms pair with each other. Are all combinations equally favorable? Or are there preferred pairings of α and β isoforms? Some evidence exists for tubulin sorting in vivo (Hoyle et al., 2001), as well as for α- and β-tubulin isoform coevolution (Nielsen et al., 2010). More recent studies identified differences in dimerization affinity between α–β-tubulin protomer pairs from different sources (Montecinos-Franjola et al., 2018), but so far, our understanding of the rules for tubulin heterodimerization is rudimentary. We also have limited understanding of the transcriptional, translational and degradation programs that regulate tubulin isoform levels throughout the lifetime of an organism. Early work showed that an increase in soluble tubulin triggers tubulin mRNAs degradation, providing a quick feedback mechanism to achieve tubulin homeostasis, a process termed “tubulin autoregulation” (Cleveland et al., 1981; Pachter et al., 1987). Tubulin targeting agents as well as nutrient deprivation regulate tubulin mRNA levels (Gasic et al., 2019). Recent work finally shed light on the molecular mechanism for this regulation by identifying tetratricopeptide protein 5 (TTC5) as the factor that recognizes the amino terminus of the nascent tubulin peptide on the translating ribosome, targeting it for degradation (Lin et al., 2020).

Early studies supported the idea that microtubule dynamics can be changed in cells by varying the relative levels of tubulin isoforms (Lu and Luduena, 1994; Panda et al., 1994). Recent studies using recombinant tubulin have shown that, indeed, tubulin isoforms have different dynamic parameters (Pamula et al., 2016; Vemu et al., 2017; Vemu et al., 2016) and that titrating tubulin isoform composition can proportionally tune microtubule dynamics and elicit changes of magnitude comparable to those induced by MAPs. For example, microtubules assembled from non-neuronal specific isoforms βI, II, or IVb polymerize faster and are less dynamic than those containing neuronal βIII tubulin (Pamula et al., 2016; Vemu et al., 2017), indicating that changes in tubulin isoform expression during neurogenesis directly impact microtubule turnover in addition to possibly recruiting isoform-specific regulators. Interestingly, βIII tubulin is only a minor isoform in the brain of Antarctic fish which live in temperatures as low as −1.8 C° (Detrich et al., 2000), compared to mammals where it comprises 25% of total tubulin (Banerjee et al., 1992). Tubulin polymerization is finely tuned to organism temperature. Mammalian tubulin depolymerizes rapidly at lower temperatures. Thus, temperature homeostasis of microtubule function could be achieved through differential expression of tubulin isoforms triggered by changes in environmental temperatures. This promises to be an interesting area of future exploration.

Structures of microtubules composed of isotypically pure α1AβIII tubulin revealed structural changes at lateral polymerization interfaces which are a hotspots of sequence variability between tubulin isoforms, and thus can modulate microtubule dynamics parameters (Vemu et al., 2016). More recently, structural studies of C. elegans microtubules also revealed the structural plasticity of lateral polymerization interfaces and proposed that the structuring of these lateral contacts is rate-limiting in microtubule growth (Chaaban et al., 2018). Interestingly, early structure-based sequence analysis of Antarctic fish tubulin isoforms revealed key mutations at lateral interfaces when compared to tubulins from species that live at warmer temperatures (Detrich et al., 2000). Thus, tuning of lateral interfaces could be a general evolutionary mechanism for biochemical adaptation of the tubulin dimer for physiologically optimal dynamic parameters.

The ability to make recombinant tubulin or isolate it from varied sources opens the way to the complete structural and functional characterization of tubulin isoforms. The coming years will likely bring a comprehensive understanding of the basic biophysical properties of tubulin isoforms from a variety of organisms: their polymerization properties, their effect on microtubule structure, dynamics and mechanics and how these relate to cell physiology. The spatial organization of tubulin isoforms is largely unexplored: do tubulin isoforms incorporate into patterns in microtubule arrays and are such patterns used by cells to recruit specific cellular effectors? Such patterns could be generated through different rates of incorporation and relative expression levels of tubulin isoforms or could result from sophisticated regulatory mechanisms such as local translation and degradation. The microtubule plus-end-tracking protein adenomatous polyposis coli (APC) associates with the 3’ UTR of βIIb (TUBB2B) to control its local translation at the growth cone periphery where it incorporates into microtubules (Preitner et al., 2014). Unlike βI and βIII, βII-tubulin localizes to neurites and not cell bodies (Guo et al., 2011). Thus, local translation can support local remodeling of microtubule isoform composition. Enrichment of certain tubulin isoforms at other subcellular locations such as cilia, cell cortex and centrosome has been also documented (reviewed in (Ludueña, 2013)). The mechanism for this selective incorporation of tubulin isoforms at subcellular locations is an interesting area of exploration.

A characterization of tubulin isoform patterns in cells and tissues has been hampered by a limited repertoire of isoform specific antibodies and difficulties with systematic tagging of tubulin without altering the tubulin isoform balance in the organism (e.g., the adverse effects of GFP tagging on microtubule function (Jenkins et al., 2017)). Recent progress in developing better tags for live-cell imaging and electron microscopy will greatly aid these efforts. The renewed interest in the tubulin code, coupled with advances in recombinant tubulin production, should also catalyze the isolation of high-affinity affimers and antibodies specific for different tubulin isoforms and suitable for high-resolution imaging to obtain comprehensive maps of tubulin isoforms in microtubules from different cell types and organisms.

Tubulin modifications: effects on microtubule structure, mechanics and dynamics

The posttranslational modifications most strongly associated with tubulin involve the covalent addition of amino acids with distinctly different physio-chemical properties (glutamate, glycine and tyrosine). Glutamates and glycines are added to internal glutamates in the intrinsically disordered C-terminal tails of α- or β-tubulin (Figure 2). Tyrosine is added at the carboxy-end of α-tubulin (Figure 2), after being first removed posttranslationally as part of the detyrosination/tyrosination cycle. Most of the tubulin purified from brain has three to six added glutamates, and as many as eleven and seven (on α- and β-tubulin, respectively) (Redeker, 2010) and is also detyrosinated. Tubulin in axonemes has as many as 21 added glutamates (Geimer et al., 1997; Schneider et al., 1998). Larger numbers of posttranslationally added glutamates are likely found but are not detected by mass spectrometry because of the challenge in ionizing highly-charged species. Axonemal tubulin is also abundantly glycylated with one to 40 and two to 26 glycines detected on α– and β-tails, respectively (Wall et al., 2016). Thus, the posttranslationally added amino acid chains can be longer than the genetically-encoded tubulin tails, which are ten to 15 residues long and thus can significantly affect the access and recruitment of cellular factors to the microtubule surface. In vivo, glutamylated as well as detyrosinated microtubules turn over slower and are more resistant to cold or drug induced depolymerization (Schulze and Kirschner, 1987; Webster et al., 1987). Glycylation is primarily found in axonemes where it is important for their long-term stability (Bosch Grau et al., 2013). It is not yet clear whether the different behaviors of these microtubule subpopulations in cells reflect different intrinsic polymer properties or are a result of specific interactions with cellular effectors. This question will only be answered by measuring the effects of detyrosination, glutamylation or glycylation on microtubules in vitro with well-defined tubulin preparations.

In addition to the flexible tails, residues in the tubulin body are also modified through acetylation, methylation, phosphorylation, palmytoylation, sumoylation, phosphorylation and polyamination (Figure 2, reviewed in (Yu et al., 2015)). Several of these modifications affect microtubule properties. Acetylation at Lys40 in the microtubule lumen weakens interactions between protofilaments leading to an increase in resistance against buckling (Xu et al., 2017). In the absence of acetylation, microtubules in C. elegans mechanosensory neurons splay apart and have frequent protofilament number changes (Cueva et al., 2012; Topalidou et al., 2012). Acetylated α-tubulin is also essential for touch sensation in mice and Drosophila (Morley et al., 2016; Yan et al., 2018). Acetylation of β-tubulin Lys 252 impairs tubulin polymerization, likely by affecting the α-β intradimer interface and promoting a conformation that disfavors incorporation into microtubules (Chu et al., 2011). Phosphorylation of β-tubulin Ser172 (Figure 2) inhibits tubulin dimer incorporation into microtubules (Fourest-Lieuvin et al., 2006; Ori-McKenney et al., 2016). Several other phosphorylation sites were identified in α and β-tubulin (Liu et al., 2015), however their effects are unknown. Polyamination (Figure 2), primarily on residues at polymerization interfaces renders microtubules resistant to cold-induced depolymerization (Song et al., 2013).

Tubulin code readers: effects of tubulin diversity on microtubule regulators

The microtubule surface, as seen by cellular effectors, is conformationally and chemically complex and not the compact, uniform surface that we have been accustomed to seeing in textbooks (Figure 2). In addition to the changes in electrostatic character introduced by modifications, the flexibility and increased search space of the tubulin tails modified with posttranslationally added amino acids also increase the odds of productive interactions with rigid-body binding partners such as the ATPase domains of molecular motors. The velocity and processivity of kinesin-1, 2 and 3 are all upregulated by polyglutamylation (Sirajuddin, 2014)(Figure 4A, B). Kinesin-1 processivity on βIII microtubules is three-fold lower than on microtubules assembled from βII tubulin because of a positively charged residue in the βIII-tail. A negative charge increase through glutamylation increases run length, restoring it to that on βII microtubules (Sirajuddin, 2014). This illustrates the potential combinatorial effects of tubulin isoform diversity and posttranslational modifications. A striking example of microtubule track specialization is found in flagella where kinesin-II powered intraflagellar trains move on the B-tubule enriched in detyrosination, glutamylation and glycylation while dynein-II powered IFT move on the mainly tyrosinated, unmodified A-tubule modified (Figure 4C; (Johnson, 1998; Lechtreck and Geimer, 2000; Multigner et al., 1996; Stepanek and Pigino, 2016).

Figure 4. The tubulin code regulates interactions with microtubule effectors.

(A) Schematic of a neuron with microtubule and posttranslational modifications colored as in Figure 3. Insets show in clockwise direction, spastin recruited preferentially to glutamylated microtubules, CLIP170 and dynein-dynactin recruited to the plus ends of microtubules enriched in tyrosination, and dynein/dynactin-initiated motility at tyrosinated microtubule sections through the Cap-Gly domain in p150/dynactin; once initiated, motility is independent of the microtubule tyrosination status.

(B) Schematic of the mitotic spindle with microtubule and posttranslational modifications colored as in Figure 3. Inset shows preferential recruitment of chromosome-associated CENP-E to detyrosinated microtubules.

(C) Schematic of a flagellum with microtubules and posttranslational modifications colored as in Figure 3. Spheres denote cargo trains. Inset shows designated transport lanes on the microtubule doublet with kinesin anterograde trains running on the glutamylated, glycylated- and detyrosinated-enriched B-tubule and dynein retrograde trains running on the tyrosinated A-tubule.

Because glutamylation involves the addition of variable glutamate numbers, it can elicit graded responses. The microtubule severing activity of spastin is regulated by the number of glutamates on tubulin tails, providing precise, substrate driven control of microtubule stability (Valenstein and Roll-Mecak, 2016)(Figure 4A). Thus, tubulin modifications can be highly information-rich. Given the high abundance of glutamylated tubulin in neurons and axonemes, it is not surprising that perturbation of glutamylation levels leads to broad neuronal and ciliary phenotypes (reviewed in (Magiera et al., 2018b)). The effects of glycylation on motor or MAP behavior has so far not been investigated, largely because of a lack of a biochemical platform to generate microtubules with controlled glycylation levels.

The detyrosination/tyrosination cycle of α-tubulin plays important roles in many biological processes ranging from cell division to neuronal trafficking and cardiomyocyte beating. Early studies identified detyrosination as a signal for the recruitment of vimentin intermediate filaments by kinesin (Gurland and Gundersen, 1995; Kreitzer et al., 1999). Detyrosination also guides chromosomes towards the cell equator during mitosis by increasing the time the chromosome-associated CENP-E motor remains attached to microtubules under load (Barisic et al., 2015)(Figure 4B). Asymmetric detyrosination of microtubules in the female meiotic spindle, achieved through CDC42-dependent signaling from the cortex, drives non-Mendelian segregation of selfish elements (Akera et al., 2017) through a yet unknown molecular mechanism. The preferential interaction of the depolymerizing kinesin-13 MCAK with tyrosinated microtubules contributes to the higher turnover of tyrosinated microtubules in cells and an imbalance between detyrosination and tyrosination results in spindle positioning defects (Peris et al., 2009). The detyrosination/tyrosination cycle also affects microtubule dynamics by regulating the recruitment of Cap-Gly containing proteins such as CLIP170 to growing microtubule ends (Peris et al., 2006) (Figure 4A). CLIP170 promotes fast actin polymerization from microtubule ends via recruitment of the formin mDia1 (Henty-Ridilla et al., 2016). Thus, tyrosination dependent localization of CLIP170 could mediate communication between actin and microtubule arrays. Tyrosinated microtubules also recruit the dynein-dynactin complex through the Cap-Gly domain of the p150^Glued subunit of dynactin (McKenney et al., 2016; Nirschl et al., 2016; Peris et al., 2006). This interaction is used synergistically with CLIP170 to load retrograde vesicles in the distal axon, enriched in tyrosinated microtubules (Nirschl et al., 2016); Figure 4A). The tyrosination dependent recruitment of dynein-dynactin is also important for centrosome positioning (Barbosa et al., 2017).

Desmin intermediate filaments couple the sarcomere to the microtubule. The binding to the microtubule is strongly dependent on detyrosination. (Robison et al., 2016). Because microtubules provide mechanical resistance during cardiac contraction, an increase in detyrosination negatively regulates contractility. Strikingly, cardiomyocytes from failing hearts have heavily detyrosinated microtubules. Reduction in detyrosination reduces stiffness and partially restores contractile function (Chen et al., 2018), revealing a potential point of therapeutic intervention. Motor traffic can also be controlled indirectly by the tubulin code through its effects on microtubule organization. For example, Lys40 acetylation has not been shown so far to have a direct effect on motors (Kaul et al., 2014; Walter et al., 2012), but it promotes the formation of microtubule bundles on which kinesin-1 traffics preferentially (Balabanian et al., 2017; Tas et al., 2017). This preference is important for kinesin-1 to drive the entry of simian virus 40 into the cytosol (Ravindran et al, 2017).

Recent years have seen a growing number of examples of microtubule effectors regulated by individual posttranslational modifications. We will likely soon have a comprehensive characterization of the response to the tubulin code of major cytoskeletal regulators through targeted as well as proteomic approaches. The effects of the microtubule track on motor stalling forces have been largely unexplored, with the exception of CENP-E (Barisic et al., 2015) and this is an area of physiological importance since motors carry large cargo in constrained environments in cells. Multiple posttranslational modifications occur on the same tubulin dimer or on the same microtubule in vivo and it is not yet known whether tubulin modifications act combinatorially. What is the minimal mosaic block on a microtubule enriched in a particular modification that can elicit a response from a cellular effector? What are the effects of different modification types and levels on neighboring microtubules? The physiological output of the tubulin code can be highly sensitive to the concentration of the microtubule effectors and the spatial distribution of microtubules with different modifications. For example, spastin-mediated microtubule stability can be regulated in trans by neighboring microtubules depending on their glutamylation levels. Glutamylated microtubules sequester the enzyme and protect nearby unmodified microtubules from severing (Valenstein and Roll-Mecak, 2016). Thus, cell types with different concentrations of an effector will respond differently to the same changes in the levels of a tubulin modification. Here the microtubule field may learn from the analytical and system approaches in the cell signaling field.

Pattern generation, maintenance and plasticity: the tubulin code writers and erasers

The in vivo chemical complexity of tubulin is generated by the collective action of tubulin modifying enzymes, the code writers and erases. These enzymes belong to diverse protein families ranging from acetyltransferases and kinases to transglutaminases, ATP-dependent ligases and carboxypeptidases. They have roles in cell cycle regulation, neuronal development and cilia function (reviewed in (Yu et al., 2015). Their mutation has been linked to cancers, neurodevelopmental and neurodegenerative disorders (reviewed in (Magiera et al., 2018b)).

The genetically encoded tyrosine on α-tubulin is removed by the vasohibins VASH1 and 2 (Aillaud et al., 2017; Nieuwenhuis et al., 2017), although these are likely not the only enzymes with this activity (Nieuwenhuis et al., 2017). Vasohibins are angiogenesis regulators; however, the role of detyrosination in this pathway is not yet known. Detyrosination is reversed by tubulin tyrosine ligase (TTL) (Ersfeld et al., 1993; Raybin and Flavin, 1977). TTL knockout mice die shortly after birth due to disorganized neuronal arrays (Erck et al., 2005). Higher levels of microtubule detyrosination are associated with tumor aggressiveness and poor prognosis, and TTL inactivation is common in carcinomas and sarcomas (Whipple et al., 2010). Interestingly, VASH2 is expressed in cancer and promotes tumor growth by stimulating angiogenesis (Sato, 2012). Thus, tumors upregulate detyrosination by dysregulation of both TTL and vasohibins.

While the first tubulin modification enzyme, TTL, was isolated in 1977 (Raybin and Flavin, 1977), many tubulin modification enzymes were identified in the last fifteen years. These include the enzymes responsible for the most abundant and evolutionarily widespread tubulin modifications, glutamylation (Janke et al., 2005; van Dijk et al., 2007) and glycylation (Wloga et al., 2009). Glutamylases and glycylases belong to the tubulin tyrosine ligase-like (TTLL) family and share a core structurally homologous to TTL (Garnham et al., 2015; Garnham et al., 2017), consistent with the fact that they all catalyze the RNA-independent ligation of amino acids. Mammals encode thirteen TTLLs. TTLL1, 2, 4, 5, 6, 7, 9, 11 and 13 are glutamylases, while TTLL3, 8, and 10 are glycylase (reviewed in (Yu et al., 2015)). So far, no enzymatic activity has been reported for TTLL12. TTLL enzymes are specialized either to initiate the glutamate or glycine chains by catalyzing the formation of a branch point formed by the addition of a glutamate or glycine to the γ-carboxyl group of an internal glutamate in the tubulin tail, or for elongating these chains once the branch point is formed (Figure 2). Recent studies using purified enzymes and well-defined tubulin substrates demonstrated these segregated activities for TTLL enzymes and defined the active site signatures that specialize TTLLs into glutamate and glycine elongases and initiases (Mahalingan et al., 2020). It is not yet clear whether the exact position of the modification on the tubulin tails is important, or just its presence. Tubulin isolated from brain tissue or axonemes has only a subset of the glutamates in the tails glutamylated or glycylated (Redeker et al., 1998; Wall et al., 2016) and modification enzymes (TTLL6, 7 and 3) have a preference for certain modification sites (Mahallingan et al., 2020; Garnham et al., 2017), indicating some substrate specificity. Intriguingly, experiments in Tetrahymena using tubulin mutants suggest that what is critical is that a tubulin tail carry the modification, even if it is grafted on the body of the other protomer (Xia et al., 2000). Studies in other systems that systematically mutate tubulin modification sites will be needed to gain a clearer understanding of the positional effects of modifications on the tubulin tails.

The action of TTLLs is reversed by the M14D subfamily of metallo-dependent carboxypeptidases, CCPs (Berezniuk et al., 2012; Kimura et al., 2010; Rogowski et al., 2010). Mammals express six CCPs (CCP1–6) with different substrate specificities. Overexpression experiments and use of modification-specific antibodies indicate that CCP1, 2, 3, 4, and 6 shorten polyglutamate chains and produce Δ2-tubulin, while CCP5 removes branching point glutamates (Rogowski et al., 2010). To date, no enzyme with tubulin deglycylating activity has been identified. Both TTLL and CCP enzymes are linked to defects in cilia function and neuronal development. In C. elegans, glutamylation by TTLL11 controls the architecture of the AB doublet in the cilia of sensory neurons, regulates the velocity of kinesin-2 OSM-3/KIF17 and kinesin-3 KLP-6 and is required for ciliary extracellular vesicle release (O’Hagan et al., 2017). TTLL11 loss is epistatic to CCP1 loss in this system. Excessive glutamylation in the mouse cerebellum due to CCP1 loss causes Purkinje cell degeneration and impairs axonal transport (Magiera et al., 2018a; Rogowski et al., 2010). The neurodegenerative phenotype can be rescued by downregulating the glutamylase TTLL1. CCP1 mutations are also linked to infantile-onset neurodegeneration (Shashi et al., 2018). Thus, perturbation of tubulin glutamylation is likely an important mechanism contributing to neurodegeneration.

Even though acetylation of α-tubulin Lys40 was described in 1986 (Maruta et al., 1986), it took more than twenty years to identify the acetyltransferase responsible for this modification, TAT (Akella et al., 2010; Shida et al., 2010). Tubulin is deacetylated by histone deacetylase 6 and NAD-dependent deacetylase sirtuin2, enzymes extensively studied in the context of histone modifications (reviewed in (Garnham and Roll‐Mecak, 2012)). Lys40 was also found recently to be methylated by the Set2 histone methyltransferase (Park et al., 2016), thus providing another common regulator between the tubulin and histone codes.

Tubulin phosphorylation is poorly characterized and many phosphorylation sites have no identified kinases so far. β-tubulin is phosphorylated on Ser172 by cyclin-dependent kinase 1 in mitotic cells (Fourest-Lieuvin et al., 2006) and by the down syndrome minibrain/DYRK1a kinase in neurons where it regulates dendritic arborization (Ori-McKenney et al., 2016). Phosphorylation of α-tubulin Ser165 by protein kinase C enhances motility by promoting microtubule growth and suppressing catastrophe (De et al., 2014). Polyamination is catalyzed by transglutaminases (Figure 2) Transglutaminases catalyze the polyamination of both soluble tubulin and tubulin incorporated in microtubules (Song et al., 2013). They have functions in neuronal differentiation and degeneration (Jeitner et al., 2009), but it is not clear whether these functions are executed only through the modification of tubulin or also other substrates.

For many of the cases examined so far, there is an asymmetry in the free tubulin dimer versus microtubule preference for the code writers and erases. Early experiments established that detyrosination occurs post-polymerization. Once the microtubule depolymerizes, the detyrosinated tubulin is converted to the tyrosinated form (Gundersen et al., 1987). The recently identified tyrosine carboxypeptidase VASH1/2 does indeed prefer the microtubule polymer (Nieuwenhuis et al., 2017), while TTL tyrosinates the tubulin dimer (Raybin and Flavin, 1975). TTLLs also prefer the polymer substrate (Garnham et al., 2015; Mukai et al., 2009), although this has not been experimentally demonstrated for all TTLLs. TAT, despite having to access the microtubule lumen to acetylate Lys40, also prefers the polymeric state of tubulin (Kormendi et al., 2012; Shida et al., 2010) by recognizing the luminal interface between protofilaments (Szyk et al., 2014). Recent structural and molecular dynamics studies indicate that acetylation reduces the conformational flexibility of the acetylation loop, and possibly affects lateral contacts between protofilaments (Eshun-Wilson et al., 2019). The eraser HDAC6 prefer the soluble tubulin dimer (Miyake et al., 2016). Thus, the tubulin code can be reset by tubulin depolymerization if the catalytic rates of the writers and erasers are so balanced. In addition, recent work showing the exchange of lattice tubulin dimers with the soluble tubulin pool as a result of buckling (Schaedel et al., 2015) or the action of severing enzymes (Vemu et al., 2018), raises the question of whether the modifications on the lattice can be edited without depolymerization since the tubulin dimers incorporated from the soluble tubulin pool will have a different history and thus different tubulin modifications than those on the pre-existing microtubule lattice. Such an editing mechanism would create islands within a microtubule that have different lattice structures, as newly incorporated tubulin is bound to GTP, while the old microtubule lattice contains GDP-tubulin (Aumeier et al., 2016; Vemu et al., 2018), but also different posttranslational modifications. Soluble tubulin, which would be incorporated at repair sites is mostly unmodified, while older microtubules accumulate glutamylation and acetylation, the former being a regulator of microtubule severing (Lacroix et al., 2010; Valenstein and Roll-Mecak, 2016).

The complex and diverse spatial and temporal microtubule modification patterns in cells are a function of the expression levels, subcellular distributions, substrate specificity (tubulin versus microtubules; α- versus β-tubulin) and kinetic parameters of tubulin posttranslational modification enzymes. In addition to these first order factors, pre-existing modifications as well as MAPs can influence the addition of new modifications since many MAPs interact with the microtubule C-terminal tails. For example, cilia and spindle-associated protein (CSAP) has been recently identified as a positive regulator of TTLL glutamylases (Bompard et al., 2018). Thus, the tubulin code likely coexists in an intimate relation with a MAP code. Moreover, the intersection between the lifetime of the microtubule, which can be influenced by MAPs, and the kinetic rate of the modification enzymes can generate microtubule arrays where microtubules with different modifications are in close juxtaposition even when the enzymes are broadly distributed. Once the unitary response to each individual modification is quantitatively characterized, deciphering this multilevel regulatory complexity will require in vitro reconstitution experiments with multiple modification enzymes and chemically defined tubulin substrates, coupled with optogenetic and chemical perturbations in cells that will allow quick inactivation or redistribution of modification enzymes to interrogate effects on pattern formation and functional outputs.

Concluding remarks and perspectives

More than 50 years after microtubules were characterized and named (Ledbetter and Porter, 1963; Slautterback, 1963) and tubulin was isolated (Borisy and Taylor, 1967; Shelanski and Taylor, 1967), we still have much to uncover about the physical principles that govern the dynamic behavior and mechanical properties of this cytoskeletal polymer and how these properties are modulated by isoform-specific sequence variations and posttranslational modifications. Deciphering the tubulin code is a perfect problem for the new interdisciplinary era in biology. It will require physical chemistry to understand non-covalent polymer properties and their modulation by strongly polyanionic intrinsically disordered regions (IDRs); forward genetics to discover signaling pathways that regulate tubulin diversity and reverse genetics to interrogate the contribution of tubulin diversity to cell autonomous, tissue and organism behavior. CRISPR technology in non-model organisms that have pushed their microtubule architectures to extremes for specialized functions promises to yield insights into the key evolutionary levers in tubulin sequence and modifications. Cell biology and proteomics coupled with new faster high-resolution light microscopy and high-resolution ultrastructural characterization using cryo-focused ion beam milling scanning electron microscopy (cryo-FIB SEM) will provide detailed cellular maps of microtubule modifications and isoform usage and how they change with stress, developmental stage or synaptic activity. Quantitative in vitro reconstitution will continue to be key to these efforts because cells have likely devised combinatorial, complex means to regulate microtubule behavior due to their centrality to the survival of the eukaryotic cell.