Mitotic spindles revisited – new insights from 3D electron microscopy

During mitosis, chromosomes are segregated with high precision to generate two identical daughter cells. The process of segregation is driven by a dynamic bipolar spindle apparatus (Helmke et al., 2013; McIntosh, 2017). Microtubules are the main building blocks of such spindles. They polymerize from heterodimers consisting of α- and β-tubulin and display a characteristic stochastic switch from slow growth to fast shrinkage, described as dynamic instability (Mitchison and Kirschner, 1984).

Microtubules show a distinct polarity with a relatively stable minus-end and a dynamic plus-end (Helmke et al., 2013). Most microtubule minus-ends are associated with the centrosome (Wu and Akhmanova, 2017). This non-membrane-bound organelle is the major site of microtubule nucleation in animal cells, although other sites of microtubule nucleation have been reported. In general, microtubules can also be formed within the spindle itself, a phenomenon called microtubule branching (Goshima et al., 2008; Petry et al., 2013), or nucleated around chromosomes as observed in Xenopus extracts (Heald et al., 1996). Microtubule nucleation at the mitotic centrosome, however, causes a distinct orientation of microtubules within the bipolar spindle, in that the microtubule minus-ends are located within the pericentriolar material (PCM) of the centrosome and the microtubule plus-ends are growing away from the centrosome. According to the direction of microtuble plus-end growth and interaction with a particular cellular target site, microtubules of bipolar spindles are grouped into different classes. A canonical view of mitotic spindle structure in metaphase (Fig. 1) shows the following classes of microtubules (MTs): astral-MTs (AMTs), kinetochore-MTs (KMTs), interdigitating-MTs (IMTs) and spindle MTs (SMTs). AMTs are those microtubules that grow away from centrosomes towards the cellular cortex, thus mainly playing a role in positioning the spindle apparatus (Grill et al., 2001). The plus-ends of KMTs are directly connected to the kinetochores (i.e. to specific centromeric microtubule-binding sites on the chromosomes) (Musacchio and Desai, 2017). IMTs are thought to interact with each other in the midzone of the spindle. This interaction is supposed to build a direct pole-to-pole connection through microtubules of opposite polarity (Mastronarde et al., 1993). SMTs are microtubules in the body of the spindles that are neither overlapping with other microtubules (IMTs), nor connecting to the kinetochore (KMTs) (Redemann et al., 2017). In addition, numerous microtubule-associated proteins and molecular motors function during spindle formation, positioning and chromosome segregation (Helmke et al., 2013; McIntosh et al., 2012).

This general scheme of spindle organization is used in many textbooks to describe the ‘basic organization’ of mitotic spindles. However, despite the evolutionary conservation of essential proteins and regulatory factors, there is a remarkable variability in the structure and organization of mitotic spindles between organisms and within cells from a single organism. In addition, the process of chromosome segregation appears to be variable, in that organisms show differences in the mechanism of anaphase (Scholey et al., 2016).

So, where does our current understanding of mitotic spindle structure and organization come from and what technology was used to increase this body of knowledge? Clearly, our understanding of mitosis is intimately linked to improvements in light and electron microscopy, as well as in advances in specimen preparation and image processing. Key steps in specimen preparation and imaging are briefly summarized in the following paragraphs.

The vast majority of information we currently have about mitotic spindles is derived from light microscopy. The first investigators to describe the process of mitosis were the Polish scientist Wacław Mayzel (Mayzel, 1875) and the German scientist Otto Bütschli (Bütschli, 1875, 1876). About 3 years later, the term ‘mitosis’, derived from the Greek word for ‘thread’, was coined by the German scientist Walther Flemming (Flemming, 1878, 1965). Flemming investigated cell division and used aniline dyes to stain cells and observe chromosome distribution in the fins and gills of salamanders (Fig. 2). Even though Flemming also observed living cells, the groundbreaking findings of mitotic chromosome segregation were made from analysis of fixed and stained samples, providing a static view about dynamic spindles. The hand-drawn images of mitotic spindles produced by Flemming provided the first detailed images of spindles. It is important to point out that chromosomes, because they were stained, were easy to identify. In contrast, the microtubule cytoskeleton could not be observed, and appeared as empty regions in the cytoplasm. Nevertheless, the formation of spindle fibers and astral arrays was proposed (Flemming, 1878). The nature and composition of such fibers and arrays, however, was unknown at the time and many ideas were postulated. For example, fibers and arrays were thought to crystallize out of the protoplasm around a spindle pole, or to form through a morphological re-arrangement of the pre-existing protoplasm. Alternatively, lamellae were also proposed to form the wall of elongated chambers. The theories of mitosis discussed at the time were excellently covered by E. B. Wilson (Wilson, 1902). The discovery of mitotic cell division in 1875 was followed by decades of research, in which the existence of the so-called spindle fibers and astral arrays was vigorously debated. Opponents of the theories suspected the fibers were nothing else but artifacts of the fixation and staining procedure (Wilson, 1902).

The development of polarized light microscopy was key to establishing the existence of spindle fibers, as spindles could be observed in living cells. Using this technique, the first observation of sea urchin spindles was performed by W. J. Schmidt in the 1930s (Schmidt, 1937, 1939) and provided evidence for the existence of such spindle fibers, even though distinct fibers could not be resolved. It still took several years before Hughes and Swann (Hughes and Swann, 1947) and Shinya Inoué (Inoué, 1951) demonstrated of the existence of spindle fibers. The application of polarized light microscopy, culminating in the development of the Pole-Scope, also opened an entire new field of research, in that mitotic spindles could be analyzed in living cells after specific perturbations and manipulations (Inoue and Oldenbourg, 1998). As early as in 1953, Chaetopterus eggs were treated with colchicine or exposed to cold temperatures; this resulted in the first description of what we know today to be the kinetochore fiber (k-fiber) (Inoué, 1953). In the end, it was the use of osmium tetroxide (Roth and Daniels, 1962) and glutaraldehyde (Sabatini et al., 1963) for specimen fixation for electron microscopy that finally proved the existence of spindle fibers, which were described and named ‘microtubules’ in 1963 (Ledbetter and Porter, 1963).

The composition of microtubules was only resolved in 1968 when Edwin Taylor and colleagues succeeded in the purification of tubulin by co-purifying it with colchicine, a drug known to destroy mitotic spindles (Weisenberg et al., 1968). The purification of tubulin and the development of assays for test-tube polymerization of microtubules eventually opened up the field of in vitro microtubule research, thereby significantly increasing our understanding of microtubule dynamics (Johnson and Borisy, 1975; Mitchison and Kirschner, 1984; Weisenberg, 1972). The application of different optical methods, such as fluorescence recovery after photobleaching (FRAP) (Axelrod et al., 1976) and the use of fluorescein-labeled tubulin, together with the injection of fluorescently labeled dyes (Salmon et al., 1984), allowed the characterization of both microtubule and spindle dynamics in vivo. Microinjection with fluorescent tubulin was extensively used for mammalian tissue cells because of the advantages of the fluorophores and the ease of the injection procedure (Waterman-Storer and Salmon, 1997).

The discovery of green fluorescent protein (GFP) in the 1960s (Shimomura et al., 1962) and the successful cloning of GFP in the early 1990s (Prasher et al., 1992) resulted in the first expression of GFP-labeled β-tubulin in the touch receptor neurons in Caenorhabditis elegans (Chalfie et al., 1994); this opened up the field of light microscopy to protein localization in living cells. Subsequently, GFP was expressed in distinct positions or organs in animals and at specific time points during the development of a number of model organisms, thus expanding the possibilities to observe and characterize mitotic spindles in vivo. Today a number of GFP derivatives, such as yellow fluorescent protein (YFP) and others are available to simultaneously label distinct spindle components in living cells (Rodriguez et al., 2017).

Over the past 20 years, advances in light microscopy, such as stimulated emission depletion (STED) microscopy (Klar et al., 2000) or saturated structural illumination microscopy (SSIM) (Gustafsson, 2005) have resulted in an improvement of resolution from ∼500 nm to 100 nm. Moreover, the development of super-resolution technology, such as photoactivated localization (PALM) (Betzig et al., 2006) and stochastic optical reconstruction microscopy (STORM) (Rust et al., 2006), has further pushed the resolution limit to ∼40 nm. So far, superresolution microscopy has been applied to analyze spindle components, such as centrosomes (Mennella et al., 2014, 2012) and kinetochores (Ribeiro et al., 2010).

In summary, over the past 50 years, remarkable improvements in reagents and technology for light microscopy enabled an explosion of knowledge about the structure and function of mitotic spindles. Although providing great tools to analyze spindle dynamics in living cells, light microscopy today cannot offer sufficient resolution to visualize each microtubule in a given spindle, to determine the length of individual microtubules and assign each microtubule to a distinct functional class according to their position within the spindle (Fig. 3A). Such a detailed analysis is only possible using 3D electron microscopy (Fig. 3B) and the major advances in this field for mitosis research are briefly summarized in the next section.

Electron microscopy (EM) of in vitro assembled microtubules stimulated investigations on the structure of microtubules and the interaction of molecular motors with microtubule walls. Both negative staining and immobilization of microtubules in frozen-hydrated samples revealed information about protofilament numbers, functional end-morphologies of individual microtubules, and the mode of interaction of specific minus- and plus-end-directed motors (Downing and Nogales, 2010). It was the discovery of vitrification of water (Dubochet et al., 1982; Dubochet and McDowall, 1981) that laid the groundwork for the presentation of atomic maps of tubulin and culminated in the 2017 Nobel prize for chemistry for the development of cryo-EM for the high-resolution determination of macromolecular structure (Beck and Baumeister, 2016; Irobalieva et al., 2016). However, what about the structural analysis of microtubules within the cellular context?

Early EM studies used glutaraldehyde to routinely fix whole cells. After dehydration at room temperature and plastic embedding, thin (60–80 nm) sections were then imaged in a transmission electron microscope. Spindle components from a number of specimens were thus visualized in two dimensions. From the beginning, the aim was to combine the contextual information of light microscopy with the ultrastructural data as obtained from EM to gain information about specific mitotic stages and spindle components (Brinkley et al., 1967; Webster et al., 1978). Correlative light and electron microscopic (CLEM) approaches are still used in current cell biological research related to mitosis (McDonald, 2009; Müller-Reichert et al., 2007).

The next level of solving the ultrastructure of mitotic spindles was reached by the stacking of serial thin sections to quantify microtubule numbers as a function of their position along the spindle axis (Fuge, 1973; McIntosh and Landis, 1971). Later on, such serial-section analysis allowed the reconstruction of whole diatom spindles (McDonald et al., 1977, 1979), entire yeast spindles (Ding et al., 1993; Winey et al., 1995, 2005) and kinetochore fibers (McDonald et al., 1992), as well as of interpolar microtubules in PtK2 cells grown in culture (Mastronarde et al., 1993). The EM laboratory in Boulder (University of Colorado at Boulder, USA) pioneered the development of software tools (IMOD) to computationally stack the individual images in order to present 3D models of segmented microtubules and organelles (Kremer et al., 1996).

Glutaraldehyde is a standard fixative for EM, but its use is fraught with a number of problems. The biggest issues are the slow diffusion rate of chemical fixation and the subsequent dehydration procedure at room temperature for plastic embedding. A major step forward towards quantitative analysis of spindle structure was therefore the introduction of high-pressure freezing for biological samples (Moor, 1987). High-pressure freezing overcomes the thickness limitation of plunge freezing, where the ability to cryo-immobilize samples with a thickness of up to 0.2 mm is possible. The combination of high-pressure freezing with subsequent freeze substitution has significantly contributed to achieving superb ultrastructural preservation for EM. This approach is now a routine preparation method for ultrastructural investigations in current model systems, such as yeasts, worms, flies and mammalian samples (Müller-Reichert, 2010).

An in-depth analysis of the mitotic spindle architecture in 3D requires a high resolution in the z direction. Obviously, the z-resolution in serial-section reconstructions is rather low (as it is limited by the section thickness, typically 60–80 nm) and certainly a major limiting factor of this approach. This limitation is evident in early 3D reconstructions of serial sections of mitotic spindles in budding yeast (Winey et al., 1995), and it was one of the major motivations to turn to electron tomography to reconstruct microtubules at the yeast spindle poles (O'Toole et al., 1999). A major advantage of this method is that this technology allows a resolution of 5–6 nm in 3D. It was again the Boulder laboratory that pioneered the development of software tools for electron tomography, such as SerialEM for automatic data acquisition and the calculation of double-tilt tomograms (Mastronarde, 1997, 2005; O'Toole et al., 2017). In combination with high-pressure freezing and freeze substitution, electron tomography was initially applied to reconstruct small spindle components, such as centrosomes (O'Toole et al., 2003), centrioles in C. elegans (Pelletier et al., 2006) and mammalian kinetochores (Dong et al., 2007; McIntosh et al., 2008; VandenBeldt et al., 2006), as well as whole spindles of budding yeast (O'Toole et al., 1999). However, analyzing larger volumes by electron tomography has been a difficult and time-consuming task, and it took several years to develop tools to montage individual tomograms of a given section and to stack a number of consecutive tomograms to increase the reconstructed volume (Höög et al., 2007; Ladinsky et al., 1999; Noske et al., 2008). Last but not least, microtubules had to be segmented manually and the joining of data over several consecutive sections was difficult and also needed to be done manually. Therefore, 3D reconstruction of complete spindles containing hundreds to thousands of microtubules was a massive undertaking and fraught with numerous technical difficulties.

Keeping these technical hurdles in mind, it is not surprising that we currently have only very limited ultrastructural 3D data on mitotic spindles. In fact, most of the data had already been obtained in the 1970s to 1990s in just a few model systems (Ding et al., 1993; Mastronarde et al., 1993; McDonald et al., 1992; Winey et al., 1995). However, recent advances in software tools for electron tomography (see below), have allowed us to obtain large-scale reconstructions of mitotic spindles in C. elegans and we will present the unexpected findings of this study in the next paragraphs.

Several advances in software development have led to automated approaches to segment microtubules in electron tomograms (Redemann et al., 2014; Weber et al., 2012), to stitch segmented microtubule across consecutive tomograms (Weber et al., 2014), and to quantify spindle structure by taking each individual reconstructed microtubule into account. This, in combination with developments in visualization technology, significantly pushed the 3D reconstruction of large spindles and enabled the first large-scale spindle reconstruction from the early embryo of the nematode C. elegans (Fig. 3B), a well established model system in mitosis research (Müller-Reichert et al., 2010).

In order to generate large-scale 3D reconstructions of mitotic spindles, we cryo-immobilized single-cell C. elegans embryos by high-pressure freezing, followed by freeze substitution and thin-layer embedding (Müller-Reichert et al., 2007). Such plastic-embedded embryos are then cut into 300-nm-thick serial sections. To cover the pole-to-pole volume of a metaphase spindle, for instance, we acquired 12 double-tilt tomograms per section, with ∼5 GB of tomographic data per section (Redemann et al., 2017). Applying the same procedure, we then acquired 24 of these sections to cover the z-axis, thus recording a significant volume of the mitotic metaphase spindle (Fig. 4). Following image acquisition for electron tomography, dual-axis tomograms were then computed for each montage panel.

The C. elegans spindles were modeled in two steps. First, microtubules were automatically segmented in each section, using a template-matching approach, followed by an automatic tracing algorithm (Redemann et al., 2014; Weber et al., 2012). Second, the segmented microtubules were then used to automatically stitch the serial sections to a single stack (Weber et al., 2014). However, we found that this automatic workflow for stitching was often problematic owing to the large size of the data, deformation of individual sections and missing visual validation steps. Therefore, we developed a novel stitching tool that allows users to achieve an alignment and matching of microtubules across a stack of serial sections by applying a combination of automatic steps, visual validation and interactive correction. This approach supports quality assessment by highlighting regions that have been checked for stitching errors and verified (Redemann et al., 2017).

In addition to the automatic segmentation and the stitching of microtubules across multiple sections, we also developed a large set of tools in collaboration with the Zuse Institute in Berlin (http://www.zib.de/projects/reconstruction-and-analysis-microtubule-spindle-using-electron-tomography) to analyze the obtained 3D data (Redemann et al., 2017). Our goal was to have a standardized set of quantitative tools to automatically analyze microtubule position and length, and the interaction of microtubules with each other, as well as with chromosomes. We are convinced that detailed 3D data, as obtained by the analysis of complete spindle reconstructions, is a huge leap forward in our efforts to gain a deeper understanding of spindle assembly and the mechanics of chromosome segregation.

We set out to reconstruct the complex spatial arrangements of microtubules in C. elegans mitotic spindles in order to understand how KMTs are assembled and interact with the holocentric kinetochore of C. elegans by combining electron tomography of serial sections with light microscopy and mathematical modeling (Redemann et al., 2017). Based on the large-scale reconstructions, we first classified individual microtubules according to their position within the spindle and the interaction of their minus-ends with target complexes. We thus identified AMTs, KMTs and SMTs. It is worth noting that such a classification required the reconstruction of complete microtubules (end-to-end) for at least KMTs and SMTs.

This classification of microtubules enabled us to further quantify the distinct properties of each microtubule class (Redemann et al., 2017). In fact, data of such high resolution in combination with light microscopy is a very powerful tool to investigate the properties of specific microtubules within the spindle. As an example, while our light microscopy data and mutant studies strongly suggested that microtubules are nucleated from the mitotic centrosome in C. elegans, our structural data showed that only a few KMTs are directly connected to the centrosomes (Fig. 5A) (Redemann et al., 2017). Our quantitative analysis suggested that minus-ends of KMTs selectively detach and depolymerize from the centrosome (Fig. 5B). Thus, our results show that the connection between chromosomes and centrosomes is mediated by an anchoring of KMTs into the entire spindle network and that there are only few direct connections through KMTs. In addition, these indirect connections are likely highly transient (Redemann et al., 2017). In general, this finding raised the question about the necessity of a direct chromosome-to-centrosome connection for the segregation process in anaphase. Another surprising outcome of our 3D study was that the ‘classic’ interdigitating microtubules were not detected in our reconstructions. Ultimately, we are interested in whether our findings are specific to the nematode system, or whether they point to general implications that are also applicable to mammalian spindles.

Certainly, there is no simple generic scheme that can be used to explain the diversity of the functional organization of the numerous mitotic spindles that have been described in the literature (Helmke et al., 2013). Differences in spindle structure are mainly related to the type of mitosis, the organization of the kinetochore and the preferred mechanism of chromosome segregation. First, mitotic systems differ in the degree and timing of nuclear envelope breakdown (NEBD). A closed mitosis with an intact envelope during the entire course of mitosis is found, for instance, in budding and fission yeast. In these systems, the spindle pole bodies, the functional equivalents of the animal centrosome, are embedded in the intact nuclear envelope (Ding et al., 1993; Winey et al., 1995). An open mitosis is found in the ‘typical’ mammalian cell with a complete NEBD occurring at mitotic prophase (McDonald et al., 1992). A semi-closed (or semi-open) mitosis is observed in C. elegans (Albertson, 1984) and Drosophila embryos (Harel et al., 1989). In the first embryonic division in C. elegans, the nuclear envelope is opened at the position of the two opposite centrosomes, a phenomenon called ‘polar fenestration’ (Rahman et al., 2015).

Second, embryonic or cellular systems differ in the organization of the kinetochore. The monocentric kinetochore of mammalian cells is the most common type and characterized by a distinct region on chromosomes (i.e. the centromere) to which KMTs are physically attached (Musacchio and Desai, 2017). In addition, KMTs in mammalian systems are organized into k-fibers (McDonald et al., 1992). In budding yeast, only a single microtubule is attached to a ‘point’ kinetochore (Winey et al., 1995). In contrast, chromosomes with a holocentric kinetochore show an attachment of KMTs along their entire length. This phenomenon of a dispersed kinetochore, however, is fairly common in nature and can be observed in nematodes, hemipteran insects and in a number of monocot plants (Melters et al., 2012). Interestingly, both types of kinetochores appear to have a similar protein composition, despite the difference in their functional organization (Dernburg, 2001).

Third, cellular systems show differences in their anaphase A (shortening of the pole-to-chromosome distance) and anaphase B (increase in the pole-to-pole distance) patterns. While some show either anaphase A or B, others show both phases, with either anaphase A or B occurring first (Oegema et al., 2001; von Dassow et al., 2009).

It is the diversity in microtubule dynamics and spindle organization that makes it difficult to draw general conclusions about spindle organization and raises a number of important questions. For instance, are there similarities in the length distribution of KMTs in C. elegans and k-fibers in mammalian cells? Is it possible to draw general conclusions about spindle structure and to work out underlying principles that are operating behind the different systems? What needs to be done next to answer these questions?

While we have succeeded in the reconstruction of a number of C. elegans mitotic spindles, there are still a number of technical issues that need to be solved. One demanding task is the further development of computational tools for image acquisition and processing, and quantitative analysis of the large-scale tomographic data. The bottleneck of serial electron tomography is still the tremendous amount of time needed to join numerous tomograms for montaging and stitching.

Automation of data processing is an important aspect for high-throughput analysis, and we have already pushed automation of several steps. In addition, the use of machine learning, or deep learning, is very attractive for future data analysis. As an example, this could be implemented to automatically group microtubule ends according to their morphology (either closed or open), giving us insights into their dynamic state, as has been achieved for microtubules assembled in vitro (Chretien et al., 1995; Mandelkow et al., 1991; Müller-Reichert et al., 1998).

Furthermore, there are also alternative microscopy techniques for spindle reconstruction. Recent advances in serial block face scanning electron microscopy (SBF-SEM) now allow the reconstruction of large areas and volumes of cells and tissues (Nixon et al., 2017). Serial block-face imaging does not rely on the production and collection of serial sections to produce reconstruction of large volumes, and its biggest advantage is that deformations of the sample can be largely avoided. Unfortunately, the biggest advantage of the SBF-SEM, the sample being contained within a single block, is also its biggest disadvantage. In order to visualize cellular structures by EM, those structures need to be post-stained by heavy metals to increase the contrast of the sample. However, if the sample is contained in a big volume of resin, then the staining solution will not be able to penetrate the entire volume, resulting in a relatively low contrast. Therefore, at this point, the use of this technology is largely limited by a low contrast when using cryo-immobilized plastic sections for spindle reconstruction. Despite encouraging work, there is a long way to go before the same precision in the quantification of microtubule architecture can be achieved by this method as is currently obtained by tomographic reconstruction of semi-thick sections.

One future goal would be to apply cryo-electron tomography to image the mitotic spindle in a frozen-hydrated state, thus avoiding dehydration of the sample by freeze substitution. Such work has been carried out on the small spindles of Ostreococcus tauri (Gan et al., 2011), but applying these methods to larger more complex spindles would be a major technical challenge as sectioning of frozen samples is difficult due to the brittle nature of the ice. An alternative means to apply cryo-electron tomography could certainly be the production of cryo-lamellae as obtained by cryo-focused ion beam-scanning electron microscopy (cryo-FIB-SEM) (Mahamid et al., 2016; Villa et al., 2013). This technology uses a focused ion beam to produce a section (or cryo-lamella) of a frozen-hydrated cell, avoiding sectioning with an ultramicrotome. However, it is not possible to create serial sections with this technology as the regions above and below the lamella are milled away by the focused ion beam in the production process. Although this technique is technically highly demanding, cryo-FIB-SEM makes it possible to obtain high-resolution information of selected regions of spindles. Cryo-electron tomography of frozen-hydrated cells would push the level of resolution further and would be an excellent approach to resolve any motor proteins located in-between microtubules in the spindle, given that an averaging during steps of image processing would be possible.

In the future, the availability of large reconstructions with a single-microtubule resolution from different systems should make it possible to also revisit the mammalian spindle architecture and question any conclusions and hypothesis that were made on the basis of light microscopy observations. An important question to revisit, for example, would be the origin of microtubules in the spindle [i.e. centrosome-based nucleation (Wu and Akhmanova, 2017), nucleation on pre-existing microtubules (Goshima et al., 2008; Petry et al., 2013) and nucleation on or around chromosomes (Heald et al., 1996)]. Our understanding of the role of each nucleation mechanism and its contribution to spindle formation is certainly limited. By a combination of dynamic and ultrastructural data as obtained from both light and electron microscopy, it will be possible to further investigate the origin of microtubules within spindles. Another issue to analyze is the re-occuring question of the presence of microtubule bundles within spindles. In order to detect bundles within the spindles, however, one first has to define the properties of a bundle. How many microtubules make a bundle? How close or distant can those microtubules be? Do the microtubules have to be arranged in parallel and if so, over what distance does this parallelism have to persist? Although there are certainly ongoing discussions about this issue in the field (Muscat et al., 2015), a definite answer is currently lacking. In addition, we have also no clear understanding of how forces might be generated and transmitted within and outside of the spindle. What are the interactions of microtubules with each other as well as with the cell cortex or other components of the cells, such as organelles or membranes? It will also be very interesting to follow up on the structure of k-fibers in mammalian cells. Are all microtubules in those k-fibers directly connected to the centrosomes? Moreover, recently published light microscopy data presented evidence for a role of bridging fibers that interact with k-fibers in the segregation of mitotic chromosomes in mammalian cells (Vukusic et al., 2017). At this point, it is unclear how k-fibers and bridging fibers interact, and how this observation fits with a possible role of IMTs in mammalian spindles. We anticipate that serial-section electron tomography will offer the resolution necessary to analyze the fine structure of the microtubules in such bridging fibers. Now that we have the tools to answer these and other interesting questions in the field, we look forward to exciting new insights in the near future.

The authors would like to thank Dr Johannes Baumgart (MPI-PKS, Dresden, Germany) for extracting the density maps in Fig. 3 and Drs Sebastian Fürthauer, Ehssan Nazockdast and Michael Shelley (Flatiron Institute, Center for Computational Biology, New York, USA) for collaborating on mathematical modeling of spindle structure. The authors are grateful to Drs Eileen O'Toole and Richard McIntosh (Boulder, USA) for a critical reading of the manuscript.