Creating a niche in the cytoskeleton: Actin reorganization by a protein kinase

One of the most apparent differences between normal cells and their malignantly transformed counterparts and between immature and fully differentiated cells is the shape of the cells (1). Cell morphology is controlled largely by the structure of the cytoskeleton, a system of three distinct types of filamentous polymers that assemble into networks and bundles of various kinds to link the cell interior physically with the plasma membrane and endow the cell with viscoelastic properties. The transformation of cytoplasm from a liquid to a solid was observed by early microscopists to be tightly associated with the cells' ability to move, and this sol-to-gel transition is now known to be caused by changes in the state of polymerization and organization of the protein actin, one of the three types of filaments comprising the cytoskeleton (2). When something goes wrong in the complex system of control proteins and messengers that signals for changes in the actin system, as can happen with genetic mutations or by the insertion of viral genes coding for maliciously altered control proteins, cells may migrate and divide inappropriately, because the signals for division or motility cannot be stopped. One such control factor is the protooncogene protein Abl. In addition to its implication in human leukemias, the tyrosine kinase Abl also is an important regulator of the actin system involved in neuronal cell function and early neuronal development (3, 4).

Abl is part of the signaling pathways that control the actin cytoskeleton and associates with large actin-containing cellular structures, because it has two actin binding sites at its C-terminal domain (6). Abl's ability to control cell function requires its actin binding function, presumably because such targeting to the cytoskeleton directs its kinase activity to appropriate targets (6). Most cells, including neurons, also express other proteins similar to Abl including most prominently a protein called the Abl-related gene (Arg; ref. 3). Although Abl and Arg share many features in common, Wang et al. (7) report in this issue of PNAS that Arg contains an additional actin binding site that allows it to reorganize cellular actin and to form actin filament bundles in vitro by a mechanism that does not require its kinase activity. The finding that a member of the Abl tyrosine kinase family can act directly as a cellular actin bundling factor suggests a new mechanism for this class of cell regulatory proteins.

The report by Wang et al. highlights several features of the actin cytoskeleton that are central to understanding its biological roles. One challenge to understanding the functions of actin-binding proteins is that there are so many of them, and they can be coexpressed at high levels in most cells, making it difficult to show unambiguously that a particular actin-containing structure has been directed to form by a specific protein. Although there are only a handful of proteins that bind actin monomers, there are over 100 that bind polymeric (F)-actin, and many of them such as like Arg cause F-actin to form bundles (8). The dozens of actin bundling factors fall into the three classes shown in Fig. 1. Some such as α-actinin appear to have only one actin binding site but can link two actin filaments together, because they form homodimers with the actin binding sites exposed on opposite faces of the dimer. A common actin binding motif, present in several bundling factors, is formed by two repeats of a calponin homology domain, also present in Arg. Other actin bundling factors are monomeric polypeptides, but they possess two distinct actin binding sites, each of which binds a separate actin filament. Arg falls into this class, although it also may self-assemble into dimers and higher oligomers. However, many proteins reported to bundle actin in vitro have no obvious sequence homology to other actin binding sites but have in common a large positive surface charge that interacts strongly with the highly anionic actin filament. The first two classes of actin bundlers rely on highly specific protein–protein binding interfaces such as those characteristic of binding between two protein monomers (9). The third type of interaction, while not excluding docking of two complementary protein interfaces, does not require it, and the binding energy depends on the entropy gained when small ions, mostly divalent Mg²⁺, are released as the higher valence polycation binds the side of the filament (10). This type of bundling is closely analogous to the polyelectrolyte effects driving condensation of DNA by multivalent counterions and is a topic of much current theoretical work. Of course the two kinds of bundling mechanisms are not mutually exclusive, and it is possibly relevant that the newly reported actin binding site in Arg resides within a cationic region of the protein.

Figure 1.

Three types of proteins that form bundles of actin filaments. In addition to specific protein–protein bonds formed by the Top and Middle examples, release of divalent and other small ions (small blue dots) by multivalent cations with surfaces compatible with the actin filament contour also contribute to the stabilization of actin bundles.

In principle the actin bundles formed by the first two classes of proteins should have the filaments arranged in a specific polarity. Actin filaments are asymmetric both in structure and dynamics due to the asymmetric structure of the monomers and the hydrolysis of ATP that follows the addition of monomers to the faster-growing end of the filament (11). Therefore, the polarity of the filaments in the bundles will, for example, depend on whether the homodimeric crosslinkers are arranged in parallel or antiparallel arrangements and on the spatial relationship of the two classes of actin binding sites in proteins such as Arg. Experimentally, a mixed polarity of filaments is found usually in actin bundles formed in vitro, and even in the cell actin-containing bundles that look similar in light micrographs display a range of polarities when examined by electron microscopy (12). Such a mixed polarity may reflect either the flexibility of bundling proteins or the contribution of electrostatic mechanisms that are only weakly affected by the polar nature of the filament. The polarity of the bundles made by Arg has not been determined yet, but this feature would affect the possible functions of such bundles for allowing contraction or other directed motions.

A next step in understanding the cellular function of Arg-mediated actin restructuring is to relate the morphology of actin-rich networks to their mechanical properties, because remodeling the cortical actin network can both generate internal forces and resist those imposed from outside the cell. This is a difficult problem, in part because there is no complete theoretical model for the elasticity of polymer networks formed by the open meshworks of relatively stiff polymers as are found in the cytoskeleton. Unlike most synthetic gels that are formed by highly flexible coils where network elasticity can be predicted accurately from the polymer concentration and the density of crosslinks between the strands, the situation with F-actin is different. The fact that the polymers in the network are so stiff that on the length scale of the distance between filament contacts (the mesh size) they are approximately straight makes the macroscopic stiffness dependent on not only the number of crosslinks but also their geometry and flexibility. Therefore, it is not easy to predict whether areas of the actin-rich cell cortex at which Arg has induced actin to bundle become stiffer. The mechanical effect will depend on factors such as the kinds of crosslinks already formed by other proteins and the total polymer concentration.

Proteins such as Arg, which can link actin filaments together, do so in a variety of geometries as shown in Fig. 2. The first actin crosslinker reported, filamin A, is a large homodimer with actin binding regions at the ends of two 90-nm-long flexible arms (13). The crosslinks it forms tend to have angles distributed around 90°, and this finding has led to the idea that filamin A is especially potent at forming orthogonal networks that maximize elasticity at a minimum of polymer mass. A second class of crosslinkers is exemplified by the Arp2/3 complex, a relatively compact multisubunit complex that drives the formation of dendritic structures at which filaments are more prone to branch at angles centered around 70° than they are to form uniform three-dimensional networks (14). These types of crosslinkers have distinct mechanical effects in cells, as shown in a recent report that the flat lamellae required for motility of melanoma cells requires the crosslinking by filamin A, a function that cannot be taken over by Arp2/3 (15).

Figure 2.

Structures formed by three types of F-actin linking proteins. Actin filaments (green lines) are joined at different angles by proteins that form links with various degrees of flexibility. Note that although the shapes of filamin A and Arp2/3 are based on electron micrographs, their sizes are not drawn to scale. Then contour length of filamin A is near 180 nm, whereas the length of the Arp2/3 complex is ≈10 nm. The dimensions of Arg are not known, but its molecular weight is ≈1/3 that of the filamin A dimer. Domains that contact actin filaments are depicted as light green patches.

The effect of actin bundlers on the elasticity of actin is likely to be more subtle than the effects of linkers that form more open networks. In the absence of other crosslinks, the formation of actin bundles has only a small effect on elasticity, based for example on the smooth scaling of elastic modulus with concentration as actin undergoes a transition from isotropic to nematic liquid crystalline phases (16). However, in crosslinked networks the effects may be more pronounced. In one model of the elasticity of orthogonal actin networks linked by rigid 90° bonds, redistribution of some filaments into bundles leads to an ≈50% increase in stiffness (17). In other models where the networks are relatively sparse and heterogeneous such as those of actin formed by Dictyostelium gelation factor, formation of a small fraction of actin bundles may have a greater stiffening effect, because these larger bundles take up more of the stress imposed on the network (18).

The importance of the geometry of specific crosslinkers in determining the elasticity of actin networks is only beginning to be characterized. In flexible gels, the crosslink can be well modeled as a rigid point connecting two network strands, because the elastic response to deformation comes from pulling out the two ends of the coils between crosslinks. For a network of stiff polymers such as actin, imposition of shear or elongational deformations can be resisted only slightly by extending the two ends of the actin strand, because it is already nearly straight in the resting state (19, 20). Therefore, flexibility and movement of crosslinks, as well as the angles they form between filament strands, may be crucial to determining how specific crosslinking proteins alter the macroscopic stiffness of materials such as the cell cortex.

The finding by Wang et al. (7) that Arg may directly alter the actin cytoskeleton rather than doing so by exerting its protein kinase activity is bolstered by their argument that this relatively scarce protein is concentrated enough in particular structures to compete with more abundant actin-binding proteins. Although the total cytosolic concentration of cellular Arg (20 nM) would seem to be too low to have a significant impact on actin structures containing >100 μM actin, the localization of Arg to small structures such as dendritic spines increases its effective concentration to near micromolar levels that are sufficient to affect actin assembly in vitro. Likewise, the finding that Arg binds to actin in a highly cooperative manner implies that selected regions of the cytoskeleton will be particularly enriched in Arg. Of course these same arguments for localization apply to models in which the kinase activity of Arg is targeted to specific cellular sites, and it is likely that the cellular effect of Arg and other cytoskeletal-bound kinases will be a combination of both direct and kinase-mediated effects on the actin system.

The tight association between a critical regulatory protein such as Arg and the cytoskeleton emphasizes the importance of spatial localization for signal transduction processes. In many cases the spatial sequestration of signaling proteins may be as important as their total concentration or the fraction of proteins in their active state in processes where cells commit themselves to a fate that is dictated by the convergence of multiple messages. The cytoskeleton, in addition to serving mechanical roles, also is the main spatial organizer of the cytoplasm and is tied intimately to the structure of cellular membrane interfaces. The current report of Arg's ability to manipulate its own cytoskeletal environment suggests that there is much to learn about how the cytoplasmic space is partitioned to achieve levels of control that allow the cell to make sense of the numerous simultaneous messages that it constantly receives.