An electron microscopy journey in the study of microtubule structure and dynamics

Eva Nogales is the winner of the 2015 Dorothy Crowfoot Hodgkin Award.Abstract

Structural characterization of microtubules has been the realm of three‐dimensional electron microscopy and thus has evolved hand in hand with the progress of this technique, from the initial 3D reconstructions of stained tubulin assemblies, and the first atomic model of tubulin by electron crystallography of 2D sheets of protofilaments, to the ever more detailed cryoelectron microscopy structures of frozen‐hydrated microtubules. Most recently, hybrid helical and single particle image processing techniques, and the latest detector technology, have lead to atomic models built directly into the density maps of microtubules in different functional states, shading new light into the critical process of microtubule dynamic instability.

Introduction: Structural Studies of Tubulin

As central players in the life of the eukaryotic cells, microtubules (MTs) have fascinated cell biologists for decades. The purification of tubulin and the reconstitution of MT polymerization made possible the in vitro study of the assembly properties of these unique polymers. Early structural characterization showed the cylindrical character to which MTs owe their name, and how they are made of protofilaments, typically thirteen, that run parallel to the cylindrical axis.1, 2

What distinguishes MTs from other biological polymers is the phenomenon of dynamic instability, a property first described by Mitchison and Kirshner.3 MTs switch stochastically between phases of slow growth and rapid shrinkage, a metastable behavior powered by the energy of GTP hydrolysis (recently this behavior has also been shown for some bacterial actin homologs, which use ATP hydrolysis as an energy source).4 Many years of biochemical characterization by many laboratories provided critical information concerning the link between nucleotide state and MT polymerization.

The MT assembly unit is the αβ‐tubulin heterodimer. Each monomer binds one nucleotide. But while the nucleotide in α‐tubulin is nonexchangeable (N‐site nucleotide) and nonhydrolizable, the one in β‐tubulin is exchangeable in the unassembled dimer (E‐site nucleotide), but becomes hydrolyzed and nonexchangeable within the MT lattice (Fig. 1). Given the homology between α‐ and β‐tubulin, the expectation was that the N‐ and E‐sites would be in equivalent regions of the α‐ and β‐tubulin structures. Biochemical studies had also shown that: (1) it is the GTP‐bound form of tubulin (i.e., with GTP bound at the E‐site) that is capable of efficient incorporation into elongating microtubules, (2) that hydrolysis of that E‐site nucleotide occurs in the MT lattice following assembly, and (3) that this hydrolysis is required for MT depolymerization (e.g., the slowly hydrolyzable GTP analog GMPCPP promotes assembly, but inhibits disassembly5). A mechanistic understanding of these tubulin properties, linking nucleotide and assembly/disassembly, required a structural knowledge of tubulin that was not easy to obtain. Like actin, polymerization of tubulin precluded its crystallization. While the former was finally crystallized as an inhibited complex of actin and DNase I,6 the first high‐resolution structure of αβ‐tubulin was produced with a completely different approach: electron crystallography.

Figure 1.

Schematic of nucleotide binding, exchange and hydrolysis in tubulin, and its coupling to MT assembly. Exchange of GDP (orange) for GTP (magenta) at the E‐site in β‐tubulin (blue) happens in the unpolymerized dimer (left). The active, GTP‐bound tubulin dimer adds to a growing MT (right). Interaction of the incoming α‐tubulin (green) with the E‐site nucleotide at the plus end of a MT (with β‐tubulin exposed) results in GTP hydrolysis. The MT cartoon (bottom right) shows an oversimplified representation of a GTP cap as it first grows by tubulin addition and then shrinks by polymerization‐coupled GTP hydrolysis (here β‐tubulin that is bound to GTP is shown in red and that bound to GDP is shown in blue). Cryo‐EM density map (EMDB‐6349) and atomic model (PDB: 3JAK) for an EB3‐decorated MT bound to GTPγS. α‐tubulin, β‐tubulin and EB3 are colored green, blue, and orange, respectively.

With the purification of tubulin came in vitro reconstitution of its assembly. Interestingly, it was soon made clear that although tubulin had an exceptional capacity to self‐assemble, the biochemical conditions dictated the type of polymer obtained in the test tube. While certain conditions reproduced the cylindrical, microtubule assembly visualized in cross sections within preparations of embedded, sectioned cells, other produced aberrant assembly forms. Among these aberrant polymers were two‐dimensional sheets formed in the presence of zinc. Linda Amos and Tim Baker, using 2D electron crystallography methodology that was being developed by Richard Henderson and co‐workers at the LMB–MRC,7 showed that 3D reconstruction of negatively stained tubulin zinc‐sheets consisted of the antiparallel arrangement of protofilaments, while microtubules are formed by protofilaments that associate in parallel.8 Whilst parallel association gives rise to a close cylinder, the antiparallel one gives rise to a planar structure. In both cases the resolution was at the time limited to about 20–30 Å. A higher resolution structure of microtubules was proposed using X‐ray fiber diffraction of oriented gels of MT by Beese and Cohen a few years later.9

With the advent of cryoelectron microscopy, it became possible to carry out studies of fully hydrated MT, thus warranting better sample preservation. Structures of microtubules, both alone and decorated with the kinesin motor domain, were being obtained using different three‐dimensional reconstructions schemes to deal with the pseudo‐helical nature of the most common MT structures.10, 11, 12 It was at this time that the laboratory of Ken Downing at LBNL took on the task of obtaining an atomic model of assembled tubulin using electron crystallography of improved, zinc‐induced 2D tubulin sheets.13, 14 The antiparallel arrangement and optimal growth conditions, together with stabilization by binding to the anticancer agent taxol, allow for the growth of large 2D arrangement of protofilaments (∼.2 × 2 µm) that were used to collect both images and electron diffraction patterns. After an initial study at 6.5 Å resolution,15 which included the localization of the taxol binding site within the sheets, the first electron crystallographic atomic model of tubulin was initially reported in 199816 and later refined using additional electron diffraction data.17 Amos and Lowe18 published alongside the X‐ray crystal structure of FtsZ, the bacterial tubulin homolog. The comparison that followed showed that both structures contained a conserved N‐terminal, nucleotide‐binding domain that corresponds closely to the Rossmann fold of dinucleotide‐binding proteins like GAPDH, followed by a central helix (H7) and a smaller intermediate domain with structural homology to chorismate mutase19 (the C‐terminal region, alpha helical in tubulin, is not conserved in FtsZ). The remarkable similarity between the structures, in spite of the low sequence identity, which is mostly limited to nucleotide‐binding loops, allowed to deduce the likely interactions between FtsZ subunits along a filament to be similar to those along a tubulin protofilament.

Most importantly, the structure of the protofilament provided by the electron crystallographic study served to understand the structural basis of nucleotide exchange and hydrolysis that are at the heart of microtubule dynamic instability. The nucleotides were shown to be part of the longitudinal contact between tubulin subunits. The N‐site nucleotide in α‐tubulin (always GTP) is buried at the monomer–monomer interface within the dimer (Fig. 1, left), where it would be trapped during dimer biogenesis. The equivalent position for the E‐site is at the dimer‐dimer interface, which will be exposed in unpolymerized αβ‐tubulin, but becomes buried upon assembly (Fig. 1). Most importantly, the regions on α‐tubulin contacting the E‐site nucleotide at the longitudinal interface include a residue that when mutated in FtsZ dramatically reduce GTP hydrolysis20 and whose mutation is lethal in yeast.21 This structural arrangement for the E‐site nucleotide links the processes of polymerization and hydrolysis, concomitant with the burial of the E‐site GDP within the lattice (Fig. 1, right).22

Two years after the electron crystallographic structure of tubulin in a polymerized state of straight protofilaments was obtained (Fig. 2, left), the X‐ray crystal structure of tubulin bound to a MT depolymerizer, in a “curved”, inhibited state, was reported by Knossow and coworkers23 (Fig. 2, right). This structure, and a number of others to follow at improved resolution, used a strategy that parallels that used for the crystallization of actin: to crystallize tubulin by inhibiting its self‐assembly. While the electron crystallographic structure of straight, polymerized tubulin has served as a surrogate for the structure of tubulin within the MT (see following section), those from X‐ray crystallography have served as a surrogate for the peels of curved protofilaments observed at depolymerizing MT ends (Fig. 2), or for αβ‐tubulin dimers in an unassembled state (Fig. 2, center). Later studies complemented our structural view of tubulin polymerization with the cryo‐EM description, at intermediate resolution, of two polymer structures proposed to mimic intermediates in the assembly and disassembly of microtubules24 that illustrated the conformational consequences of the nucleotide state and how they relate to longitudinal and lateral assembly.25

Figure 2.

Crystallographic structures of tubulin and their correspondence to MT assembly states. The electron crystallographic structure of tubulin in straight protofilaments, from zinc‐induced sheets (left), has served as a surrogate for the structure of assembled tubulin within the MT, while the X‐ray crystallography structures of “curved” tubulin dimers bound to MT depolymerizing proteins (right) has served as a surrogate for the low energy state of protofilaments peels at the end of depolymerizing MTs. The center cartoon shows a depolymerizing MT end.

MT Structure: Improving the Resolution of MT Cryo‐EM Reconstructions

While the electron crystallographic structure of the protofilament defined the details of the longitudinal contacts between αβ‐tubulin dimers, information on how protofilaments are arranged in the MT, and which structural elements are involved in lateral contacts between protofilaments, required the use of hybrid methodology that combined the cryo‐EM structure of the MT, with the atomic model of the protofilament.26 Ron Milligan and Ivan Rayment27 had pioneered such hybrid methodology a few years earlier in the study of the actomyosin complex. The highest resolution structure of frozen‐hydrated MTs at the time, 20 Å, was from Milligan himself [Fig. 3(A), top]. The electron crystallographic structure of the protofilament could be docked with significant precision into the cryo‐EM map of the microtubule (Fig. 3(A), top), indicating that the structure of tubulin dimers and their longitudinal interactions were preserved between the MT and the zinc‐induced sheet.26 The model of the full MT that came from such analysis identified a loop, therefore named M‐loop, within the intermediate domain of both α and β‐tubulin, as a key element forming the lateral interface in the physiologically relevant polymer. It placed the C‐terminal H11 and H12 helices and the following disordered tail of both subunits on the outside “crest” of the protofilament, where kinesin motor domains had been shown to interact. α:Lys40, the only site of major post‐translational modification in tubulin (i.e., acetylation) not located on the disordered C‐terminal tail, faced the lumen of the microtubule. The electron crystallographic structure had been obtained in the presence of the anticancer drug Taxol, a microtubule stabilizer that also stabilized the zinc‐induced tubulin sheets. This drug binds within β‐tubulin near the M‐loop, occupying a pocket that in α‐tubulin is filled by an insertion in one of the loops within the intermediate domain.16 The MT model also placed this pocket on the luminal surface of the microtubule.

Figure 3.

Cryo‐EM reconstructions of MTs at increasing resolutions. (A) End‐on views of a fragment from three different MT reconstructions, from Nogales et al.²⁶, Li et al.²⁸, and Zhang et al.⁴⁰ (from top to bottom). The electron crystallographic atomic model (1JFF) of one or several protofilaments are fitted into the density (top two panels) or an atomic model refined directly into the density map (bottom). (B) Outside view of the 3.5 Å map of MTs from Zhang et al.⁴⁰ An interactive view is available in the electronic version of the article.

Improving the resolution of MT cryo‐EM maps required the use of high‐end instruments and the merging of large data sets. A major breakthrough came with the structure at better than 10 Å resolution of Downing and Li28 [Fig. 3(A), middle]. This structure allowed a better definition of the lateral contacts between protofilaments and confirmed that no major rearrangements existed between the tubulin structure in MTs and the zinc‐induced sheets. Ten years later, numerous cryo‐EM studies of MTs, alone or decorated with a number of motors or other MT‐associated factors, had provided a richness of biological information, but none of these MT structures had broken the 8 Å resolution barrier [e.g., Refs. ( 29, 30, 31, 32, 33)]. One possible limitation was the quality and quantity of data used in each reconstruction. Another problem affecting undecorated MTs is the inability to distinguish α‐tubulin from β‐tubulin at low resolution, as their structures are very similar, and the fact that the most common microtubule lattices do not have true helical symmetry. To address the first limitation, we collected images on a 300 kV FEI Titan electron microscope, which clearly showed signal to 5 Å resolution, and collected large data sets. To address the second, we decorated the microtubules with kinesin, which binds the MT lattice once per αβ‐tubulin dimer and thus provide a low‐resolution signal to drive the alignment of tubulin dimers. We then employed a modified IHRSR34 (iterative helical real space refinement) image processing protocol during which MT structures with different protofilament numbers are separated and reconstructed. Finally, we developed and implemented a reconstruction scheme that took advantage of the pseudo‐helical symmetry of the microtubule, while still accounting for the presence of the so call “seam”, in which lateral contacts are heterotypic only along one protofilament interface. As a result of our efforts, we produced record resolution structures at the time for microtubules bound to GMPCPP (4.7 Å), GDP (4.9 Å), and GDP + Taxol (5.5 Å).35 At resolutions of ∼5 Å, direct atomic modeling into the cryo‐EM maps was not possible. Instead, we used each density map in conjunction with Rosetta modeling36 to define an ensemble of 20 low‐energy structural conformers, and used their average to describe each MT state. This analysis allowed us to describe the differences between the three states analyzed, providing insight into the effects of GTP hydrolysis on microtubule structure. Hydrolysis results in a compression of the dimer–dimer longitudinal interface along the filament axis and a conformational change in α‐tubulin where the C‐terminal region moves with respect to the N‐terminal, nucleotide binding domain and the β subunit. Interestingly, binding of Taxol counteracts most of the effects of GTP hydrolysis.35

Findings from the Recent Atomic Structures of Microtubules

The studies described above used data collected on film, at the time the best detection media for the 300 keV electrons used during imaging. While CCDs performance, especially at this voltage, is suboptimal,37, 38 new detector technology has recently dramatically improved data collection and is revolutionizing the cryo‐EM field.39

In our most recent studies of MT structure, we have used a direct electron detector and improved data processing strategies to obtain cryo‐EM reconstructions of MTs at 3.5 Å or better resolution [Interactive Fig. 3(A), bottom, (B), and Fig. 1].40 At this resolution, side‐chain densities can be seen for most residues, with the noticeable exception of those from acidic residues (Glu and Asp), presumably due to radiation damage during electron exposure.41 In addition to MTs in GMPCPP and GDP states, we used copolymerization with excess human EB3, a plus‐end tracking protein (based on the work of Surrey and co‐workers42), in order to obtain the structure of EB3 bound to its preferred substrate, GTPγS‐MT. The structures of kinesin‐decorated and kinesin‐free GTPγS‐MT (both at ∼3.5 Å resolution) were very similar, and merging the two datasets improved the resolution to 3.3 Å. This result indicated that the presence of kinesin has no significant effect on the structure of EB3‐bound MTs.

Unexpectedly, the GTPγS‐MT structure displays a compacted lattice (Fig. 4(A)), thus making it more similar to the GDP state.40 The detail of our structures and atomic models allowed us to see that during this compaction there is an “anchor point” across the longitudinal interface, contacting the E‐site and near the MT surface, that does not move [Fig. 4(A), purple circle]. This unchanging contact involves hydrophobic interactions between the H8 helix and T7 loop of α‐tubulin and the H11′ helix of β‐tubulin [Fig. 4(B)]. As the lattice compaction brings adjacent dimers along a protofilament closer together around this unchanging anchor point, there is a concomitant conformational rearrangement of every α‐tubulin. The H8 helix (which contains the catalytic residue Glu 254) and the T7 loop are among the structural elements in α‐tubulin undergoing the most significant changes [e.g., Fig. 4(A), purple arrow], which are directly coupled to the movement of the next tubulin dimer.

Figure 4.

Changes in MT structure with GTP hydrolysis. (A) Comparison of the Cα traces of two consecutive tubulin dimers around the E‐site nucleotide between the GMPCPP‐K state (gold) and EB3‐GTPγS state (green and blue for α‐ and β‐tubulin, respectively) showing the compaction at the interdimer interface. The purple circle marks the position of the anchor point. The catalytic residue, E254, and D251 are indicated. The region of helix H8 between these residues (purple arrow) is distorted with GTP hydrolysis. (D) Detailed view of the hydrophobic interactions at the anchor point (Modified from Zhang et al.⁴⁰).

The high resolution of the most recent MT structures allows, for the first time, the atomic description of the lateral contacts between protofilaments (Fig. 5). Our analysis has now showed that these lateral contacts do not change significantly with nucleotide state, with only small changes seen for MTs with different protofilament numbers. The lateral interface is limited to a single point of contact, with exquisite shape complementarity that resembles a lock‐and‐key configuration. The M‐loop of both α‐ and β‐tubulin positions a strategic aromatic residue, H283 in α‐tubulin and Y283 in β‐tubulin, to function as the “key” that inserts into a complementary “lock” formed by the H2–S3 and H1′–S2 loops on the subunit across the interface [Fig. 5(A), right]. Additionally, residue K60 in the H1′–S2 loop, which is conserved between α‐ and β‐tubulin, is fully extended and further “locking” the position of the aromatic residue Y/H283.

Figure 5.

Lateral contacts between protofilaments. (A) Homotypic interactions as seen in the symmetrized reconstruction. Left, cryo‐EM map seen from the MT lumen. Right, detailed view of the interaction between β‐tubulins. (B) Heterotypic interactions at the seam as seen in the C1 asymmetric reconstruction. Left, cryo‐EM map seen from the MT lumen. Right, detailed view of the interaction between β‐tubulin and α‐tubulin, compared to the homotypic interaction between α‐tubulins (pink). In all panels, blue maps, and models correspond to β‐tubulin and green maps and models correspond to α‐tubulin (Modified from Zhang et al.⁴⁰).

The use of a direct electron detector and our improved accuracy in determining the seam location allowed us to obtain asymmetric (C1) reconstructions (in which pseudo‐helical symmetry is not applied) at 4.4 Å or better resolution. By merging three EB3‐bound states to generate an “EB consensus” map at a resolution of 3.9 Å, we were able to visualized, with unprecedented detail, the seam, the special lateral contact involving heterotypic contacts, that is thought to be involved in MT closure. Interestingly, the seam contact is similar to that seen at a nonseam location [Fig. 5(B)], likely due to the high degree of conservation between α‐ and β‐tubulin for the amino acids involved in the lateral contacts. However, due to differences in the position of the adjacent H2‐S3 loop between α‐ and β‐tubulin [Fig. 5(B)], there are small adjustments that optimize contacts while avoiding steric clashes between adjacent protofilaments at the seam. As a result, the position of the two protofilaments involved in seam contacts deviates from the cylindrical shape for the rest of the tube,40 indicating that this site may be a week point for MT disassembly.

Concluding Remarks

Our molecular image of MTs has become sharper and sharper over that last two decades, as the quality and quantity of cryo‐EM images used for a 3D reconstruction has increased, and the image processing has become more sophisticated and adjusted to suit the special character of MTs. Today, structures of MTs at a resolution that allow side chain visualization and direct atomic modeling of MT structure and structural changes, is providing us with the mechanistic details underlying the critical process of dynamic instability. Cryo‐EM should also soon provide us with the atomic details of the many interactions MTs establish with associated factors to carry out and regulate their numerous and essential cellular functions. Furthermore, cryo‐EM may provide complementary information to X‐ray crystallographic structures of antimitotic agents bound to tubulin for which a description of potential allosteric changes within the microtubule lattice may be a critical part of their mode of action.