Cotranslational assembly of some cytoskeletal proteins: implications and prospects

The cytoskeleton is a complex three-dimensional web in the interior of eucaryotic cells. The cytoskeleton orders many structures in the cell and performs many kinds of transport and motility for the cell. It is distinguished by a high degree of spatial differentiation and anisotropy. How does the cell construct something so heterogeneous and complex?

A widely used model of protein assembly postulates that proteins are synthesized on ribosomes and released into solution within the cell, before diffusing within the cell and assembling into cytoskeletal or other structures. This model is clearly appropriate for a variety of cytoskeletal proteins, including tubulin and actin (Mitchison, 1992; Theriot and Mitchison, 1992). Because these proteins assemble after translation, much has been learned about their assembly by using in vitro protein chemistry and in vivo methods such as FRAP (fluorescence recovery after photobleaching).

However, a growing body of evidence supports a different model of assembly for certain cytoskeletal proteins. Several observations, made with a variety of experimental protocols, suggest that assembly of some cytoskeletal structures is difficult to explain solely by post-translational assembly. Rather, these proteins undergo cotranslational assembly; they first associate with the cytoskeleton during translation, as nascent (not yet complete) peptides. The range of this evidence includes in situ autoradiography of cytoskeletal proteins, immunoprecipitation of nascent peptides, immunofluorescence and, most recently, fluorescent in situ hybridization.

The first evidence for localized assembly came from 3T3 cells, in which polyribosomes, stained with acridine orange, were examined both in intact cells and in cytoskeletons prepared by Triton extraction (Fulton et al., 1980). Polyribosomes were concentrated near the nucleus; this pattern was not affected by Triton extraction. By simultaneously staining with acridine orange and revealing newly synthesized protein with autoradiography of [³⁵S]methionine, protein synthesis and movement of these new proteins was observed in intact 3T3 cells and in skeletal frameworks prepared by Triton extraction. When intact 3T3 cells were examined this way, autoradiographic grains (representing proteins synthesized during the 10 minute pulse with [³⁵S]methionine) were fairly evenly distributed throughout the cytoplasm, both after a pulse and after a several hour chase period. When the cytoskeleton was examined after a pulse, autoradiographic grains were clustered over polyribosomes, rather than being evenly distributed. After release from ribosomes, therefore, it appeared that a significant fraction of cytoskeletal proteins remained close to their site of synthesis and that their assembly into the cytoskeleton was very rapid. Cytoskeletal proteins labeled during a 10 minute pulse were found throughout the 3T3 skeletal framework if protein synthesis occurred during a chase period, but cytoskeletal proteins remained localized near polyribosomes if emetine inhibited protein synthesis during the chase period. This sequence of local assembly followed by redistribution was particularly striking in hemangioma cells, which have nearly circular profiles and are especially wellspread (Fulton, 1984). These observations showed restricted exchange of cytoskeletal proteins between a soluble pool and the cytoskeleton under these conditions. These experiments suggested that many cytoskeletal proteins assembled onto the cytoskeleton near the time and place of synthesis.

To extend these observations, cytoskeletons were prepared from HeLa cells in an in vitro system that permitted translation to continue (Fulton and Wan, 1983). The cytoskeletons were diluted with varying amounts of a translation buffer and [³⁵S]methionine was added to examine the association between proteins translated in vitro and the cytoskeleton. If all cytoskeletal proteins enter a soluble pool after completion and before assembly into cytoskeletal structures, then the percentage of labeled protein on the cytoskeleton should have decreased as the cytoskeleton was diluted. However, the percentage of total radioactivity observed on the cytoskeleton was independent of the extent to which the cytoskeleton was diluted during translation. Association with the cytoskeleton could occur as completed polypeptides or while nascent polypeptides were still on polyribosomes. To distinguish between these possibilities, the association of cytoskeletal proteins early in translation was examined in the presence and absence of puromycin. Puromycin is a protein synthesis inhibitor that releases nascent polypeptides from polyribosomes (Munro, 1967; Grollman, 1968). Most nascent polypeptides are found associated with the cytoskeleton early in translation because of their association with polyribosomes (Lenk et al., 1977). Only nascent polypeptides that are associated with the cytoskeleton independent of the polyribosome remain on the cytoskeleton in the presence of puromycin. Nascent chains that are associated with the cytoskeleton solely via polyribosomes will be released into the soluble phase upon addition of puromycin. Fulton and Wan in this experiment showed persistent association of some, but not all, nascent chains with the cytoskeleton in the presence of puromycin (i.e. independent of polyribosomes). Two-dimensional gel electrophoresis confirmed that the proteins that bound to the cytoskeleton during in vitro translation were representative of those that assembled in vivo.

To summarize, the first evidence for cotranslational assembly for some cytoskeletal proteins included: (1) the rapid and localized association of cytoskeletal, but not soluble, proteins with the cytoskeleton after a short pulse; (2) the association of cytoskeletal proteins during in vitro translation with the cytoskeleton, independent of the concentration of soluble proteins; and (3) the association of some, but not all, nascent polypeptides with the cytoskeleton during translation, independent of polyribosomes.

More specific methods were developed to examine the cotranslational assembly of specific cytoskeletal proteins. Myosin heavy chains form coiled coils in a rod region that then interacts with other myosin molecules to form thick filaments in muscle. Nascent myosin heavy chain (MHC) polypeptides were found on the cytoskeleton of cultured chick embryonic skeletal muscles, consistent with cotranslational assembly (Isaacs and Fulton, 1987). Cultures that had been extracted with Triton were exposed to [³H]puromycin to simultaneously label and release nascent polypeptides from ribosomes. A monoclonal antibody to MHC (with an epitope near the amino terminus) was used to immunoprecipitate MHC from the cytoskeletal and soluble fractions. In day 4 cultures 32% of the [³H]puromycin radioactivity in the MHC, representing only nascent polypeptides, was found on the cytoskeleton. A similar fraction of MHC nascent polypeptides was also detected on the cytoskeleton by using [³⁵S]methionine pulse-chase experiments, in which puromycin or RNase released nascent chains from polyribosomes, followed by immunoprecipitation of MHC and analysis on polyacrylamide gels.

Titin is a very long molecule, stretching from M line to Z line in muscles and associating with myosin in the A band; it undergoes more extensive cotranslational assembly in cultured muscle, as suggested by the exclusive association of full length titin with the cytoskeleton after a short pulse, and by the observation that 74% of titin nascent chains remain associated with the cytoskeleton after puromycin treatment (Isaacs et al., 1989). For comparison, less than 20% of total cellular nascent chains were found on the cytoskeleton. Therefore, nascent chains are not, as a class, insoluble simply because of their incomplete length.

Titin and myosin are closely associated within minutes of being synthesized (Isaacs et al., 1992). This association was detected by briefly labelling muscle cultures with [³⁵S]methionine, followed by a chase period. Cultures were extracted to make cytoskeletons and then crosslinked with ethylene glycol (bis-succinimidyl succinate) (EGS). Immunoprecipitation with monoclonal antibodies to titin and myosin heavy chain recovered specific proteins in cross-linked complexes. Even after a pulse period as short as six minutes, both titin and myosin heavy chain could be recovered with the antibody to the other protein, suggesting that during synthesis these proteins have a close enough association with each other to be crosslinked by EGS, a 15 Å long linker. The interaction of newly synthesized titin and myosin heavy chain, both of which undergo cotranslational assembly though to different extents, may contribute to the remodeling of nascent myofibrils into sarcomeres.

Vimentin, like all intermediate filament proteins, forms a coiled coil that associates with other intermediate filament proteins to form filaments; it is another protein that has been observed to undergo cotranslational assembly to a significant extent (Isaacs et al., 1989). By using pulse-chase experiments in the presence of puromycin or RNase, or by labeling with [³H]puromycin, nascent chains were consistently identified by immunoprecipitation from the cytoskeletons of chick embryonic muscle and fibroblast cells. In addition, a kinetic argument for cotranslational assembly in fibroblasts was made by observing that the fraction of full-length vimentin present on the cytoskeleton after any chase time is greater than would be expected if based solely on the rate of disappearance of full-length vimentin from the soluble pool. No qualitative differences in 2-D electrophoretic pattern or degree of phosphorylation were noted between the soluble and cytoskeletal fractions of vimentin. The nascent vimentin chains were removed from the cytoskeleton by 50 mM arginine, but not by 50 mM lysine or 1 M salt; thus their associations with the cytoskeleton are biochemically similar to those of full-length vimentin polypeptides.

Vimentin assembly in chick embryonic erythroid cells was also examined; the three methods used on fibroblasts were all applied to the erythroid cells. Low levels of cotranslational assembly (about 10%) were found with all three methods. These observations were consistent with earlier observations in erythroid cells (Blikstad and Lazarides, 1983), where it was shown that vimentin is assembled into intermediate filaments from a soluble precursor. Vimentin is therefore assembled into the cytoskeleton by two different pathways, with the relative contributions of these two pathways varying with cell type. Independent evidence for the existence of cotranslational assembly of vimentin using a dual isotope technique has been provided by Low et al. (1985), in cultured human fibroblasts.

A recent observation consistent with cotranslational assembly is the co-localization of vimentin mRNA and vimentin protein in the costameres of mature skeletal muscle differentiating in vitro (Cripe et al., 1993). Costameres overlie the myofibril above the Z line, with spacing the same as sarcomeres, i.e. ∼1.4 µm. This particular spacing of mRNA is too fine to generate stable patterns of soluble proteins, because proteins diffuse too quickly to maintain a grid with spacing of less than 2 microns. This mRNA pattern cannot, therefore, generate local gradients of soluble proteins.

A parsimonious explanation of this observation is that the mRNA is present at the site where vimentin protein is undergoing cotranslational assembly. Appropriate controls were performed in this experiment to exclude nonspecific binding of the antibody to digoxigenin, nonspecific binding of the probe to costameres or myofibril, and lack of specificity for DNA-RNA hybrids. It is interesting that vimentin mRNA in younger muscle cells displayed several different staining patterns, being bipolar in young myoblasts, perinuclear in elongated myoblasts, and diffuse in young myotubes. Perhaps mRNA localization will prove important for assembling and maintaining differentiated cytoskeletal structures in the developing embryo. The relationship between localized messenger RNA and cytoskeletal assembly, both cotranslational and post-translational, has been discussed elsewhere (Fulton, 1993).

It has been recently reported that behavior of microinjected X-rhodamine-labeled vimentin, followed by fluorescence recovery after photobleaching (FRAP), is consistent with a steady-state exchange of vimentin subunits along the length of intermediate filaments with a soluble pool (Vikstrom et al., 1992). Exchange of vimentin subunits between the microinjected soluble pool and intermediate filaments may at first glance seem inconsistent with cotranslational assembly of vimentin. Microinjection provides the cell with a sizeable pool of soluble vimentin subunits that is not normally present, however. A concentration effect, where the likelihood of assembly of microinjected vimentin into intermediate filaments is increased because of such a large soluble pool compared to nascent polypeptides available for assembly, cannot be excluded. There is no reason, however, to believe that assembly of vimentin (or other cytoskeletal proteins) must exclusively occur via either post-translational assembly from a soluble pool or cotranslationally from nascent chains on the polyribosome. The published data for vimentin points to about half of the vimentin protein assembling post-translationally in fibroblasts. Perhaps incorporation of soluble subunits into cytoskeletal structures such as intermediate filaments represents a repair mechanism of the cell, allowing continued cytoskeletal remodeling. In addition, initial assembly that occurs cotranslationally does not predict whether a protein may disassemble and then re-assemble.

Actin is a cytoskeletal protein whose mRNA has been shown to be localized in several cell types, including muscle (Lawrence and Singer, 1986). Actin mRNA appears to be concentrated at the periphery in young myoblasts and in fibroblasts, being particularly prominent in lamellipodia. High concentrations of actin mRNA were also observed in areas of cell contact. Myotubes and proliferating myoblasts had a much more uniform distribution of actin mRNA. The prominent localization of actin mRNA in peripheral regions of motile cells closely mimics the distribution of growing actin filaments in lamellipodia. Localization of actin message may be related, therefore, to cell motility. Actin protein appears to assemble from a soluble monomer pool. Thus, co-localization of message and protein to a particular region within a cell does not, per se, imply that cotranslational assembly is occurring. Vimentin mRNA in this study of early myoblasts and myotubes showed perinuclear localization, coincident with the primary location of vimentin protein.

Whenever a highly novel finding is reported, it is appropriate to consider whether a simpler explanation for it can be found in some artifact of experimental design. It seems unlikely, however, that the whole body of observations that point to cotranslational assembly can be accounted for in this way. It is worthwhile to consider in detail two particular possible sources of artifact, the detergent Triton X-100 and the nascent chains themselves.

Triton X-100 is a non-ionic detergent that came into use for cytoskeletal studies during the late Seventies (Osborn and Weber, 1977; Brown et al., 1976; Lenk et al., 1977). In general, as is common for non-ionic detergents, Triton X-100 does not denature proteins (Helenius et al., 1979) and was chosen empirically because its presence in extraction buffers (of the appropriate composition) permitted the recovery of morphologically congruous cytoskeletal structures. The recent observation that some soluble proteins become bound to structures, especially nuclear structures, in its presence (Melan and Sluder, 1992) suggests that the use of Triton X-100 should be re-examined.

In the studies by Melan and Sluder, concentrations of Triton X-100 (0.1% or 0.2%) that were below or near the critical micelle concentration (CMC), when used with buffers of moderate ionic strength, lead to artifactual redistribution of some soluble proteins, most commonly to the nucleus but sometimes to fibrillar structures in the cytoplasm. Higher concentrations of Triton X-100 (1%, above the CMC) obliterated these artifactual patterns. The concentration of Triton X-100 used in the cotranslational assembly experiments (0.5%) was not tested by Sluder, but this concentration is well above the CMC. In separate studies, 0.5% Triton X-100 produced stable cytoskeletal preparations that did not shed protein even when the extraction buffers contained large amounts of competing nonspecific proteins (Gilbert and Fulton, 1985); with these buffers, 0.5% and 2% Triton X-100 gave the same results.

In addition, if Triton X-100 were causing artifactual binding to the cytoskeleton, it should do so independent of the time at which a protein is synthesized. However, in most of the studies discussed, either the spatial pattern or the amount of material found in the cytoskeleton changed as a function of time after synthesis. It is difficult to propose models for artifacts that vary over time.

Finally, whenever possible, comparisons were made between intact cells and cytoskeletons after extraction. Polyribosome patterns (Fulton et al., 1980), myosin and titin staining (Isaacs et al., 1992) and vimentin protein and messenger RNA patterns were all unchanged by this extraction procedure. Clearly, however, a blanket assumption of such preservation cannot be extended automatically to all buffers that contain Triton X-100; the appropriateness of a given buffer to a given procedure needs to be determined.

More than 85% of all nascent peptides labeled with puromycin are released by 0.5% Triton X-100 (Isaacs and Fulton, 1987); therefore insolubility is not a general property of nascent peptide chains. Puromycin becomes covalently attached to the nascent chain as it is released from the ribosome, so this radioactivity exclusively represents nascent chains. In addition, when assembly is measured kinetically by examining only the full length polypeptides, the rates of assembly agree well with measurements made on nascent chains (Isaacs et al., 1989). Competition with non-radioactive proteins did not release substantial amounts of nascent peptides (Isaacs and Fulton, 1987). Nascent peptides that initiated in vitro did not bind efficiently to the cytoskeleton (Fulton and Wan, 1983); it is difficult to see how artifactual binding could differentiate between in vivo and in vitro initiation. For vimentin, three biochemical criteria that characterize the association of full-length vimentin were satisfied by the nascent chains (Isaacs and Fulton, 1989). Titin and myosin nascent chains were cross-linked to molecules to which the full-length molecules become cross-linked (Isaacs et al., 1992). The cumulative weight of these diverse observations makes it unlikely that the nascent chains are displaying significant amounts of non-specific binding under these conditions.

One other laboratory has reported biochemical evidence for cotranslational assembly (Low et al., 1985). These studies measured full-length vimentin chains, so they offer independent support for our kinetic studies. A different form of evidence that is consistent with cotranslational assembly is the observation of homodimers and homopolymers of proteins that form heteropolymers if re-assembled in vitro. The most striking example of this is seen in the work of Gauthier (1990), who detected myofibrils that were largely composed of a single isoform of myosin, although the muscles were synthesizing two isoforms that co-assemble in vitro. Other examples, including one involving tropomyosin, have also been reported (Lin et al., 1985; Silberstein et al., 1986; Bandman, 1985).

For proteins to assemble during translation, they must begin folding as nascent peptides. Such folding appears to be possible, prime facie, as many proteins re-fold in less time than is required for their synthesis. Cotranslational folding of proteins was formally presented as a hypothesis by Purvis et al. (1987), with a particular emphasis on translational pause sites that could permit folding of domains during translation. Experimental support for cotranslational folding has recently been published for globin (Krasheninnikov et al., 1991), tryptophan synthetase β subunit (Fedorov et al., 1992), and the yeast TRP3 gene product, a bifunctional protein (Crombie et al., 1992). For some proteins, peptide fragments of them can fold into the secondary structures seen in the intact molecule; α helices do so more readily than do β pleated sheets (Dyson et al., 1992).

The hypothesis of protein folding during translational pauses suggests a new perspective on observations of discrete nascent peptides of myosin, titin and vimentin (Isaacs and Fulton, 1989). All three proteins displayed populations of nascent peptides that migrated as discrete bands, whether the label used was methionine or puromycin. For vimentin, one of these nascent peptides was the size that would result from translationally pausing at a 27 bp hairpin that lies within the coding sequence (Bloemendal et al., 1985). It might prove informative to replace this hairpin with isocoding sequences that would not form a hairpin.

Not all cytoskeletal proteins undergo cotranslational assembly; actin, tubulin and α-actinin are three that do not (Mitchison, 1992; Isaacs and Fulton, 1989). Do proteins that undergo cotranslational assembly have structural features in common? Vimentin and myosin heavy chain comprise parallel coiled coils; both undergo significant cotranslational assembly. The globular proteins tubulin and actin, and the fibrillar but antiparallel dimer α-actinin are not cotranslationally assembled, as measured by kinetic experiments used to detect assembly during translation for vimentin. Thus, our methods also detect post-translational assembly. The conformation of the protein probably plays a role in permitting cotranslational assembly to occur. If so, candidates for cotranslational assembly include fibrillar, self-associating proteins that assemble in parallel, not antiparallel (e.g. vimentin, MHC), or elongated proteins binding to fibrillar structures (e.g. titin). As tropomyosin has a coiled-coil fibrillar structure that undergoes parallel selfassembly, it offers an interesting example to test this model for the assembly mechanism of cotranslational assembly. Experiments are underway to test whether its assembly is cotranslational.

Nothing is known about the factors that determine the levels of cotranslational assembly for a particular protein. Vimentin exhibited different levels of cotranslational assembly in fibroblasts and erythrocytes. These cells differ both in the relative rates of vimentin synthesis and in vimentin content; in addition they are different cell types. The relative rate of synthesis might affect cotranslational assembly by affecting the probability of nascent chains encountering another nascent chain. The mass of assembled vimentin might affect cotranslational assembly by affecting the probability of nascent chains encountering vimentin polymerized in filaments. Cell type specific variation could result from chaperones or other accessory proteins. If one cell type had synthetic rates and a vimentin content that differed with developmental stage or growth conditions, it might be feasible to determine whether one or both of these variables affects cotranslational assembly significantly. In MDCK cells, vimentin synthesis is greatest in the subconfluent cells; vimentin mass is greatest in confluent cells. Cotranslational assembly should be highest in the subconfluent cells if relative rate is more important; it should be highest in the confluent cells if vimentin mass is the significant variable. These cotranslational assembly rates are presently being measured.

The advantage to a cell of cotranslational assembly may be that organizational information is provided during protein synthesis. This could facilitate the assembly of such complex cytoskeletal structures as the contractile apparatus in striated muscle. It may facilitate the maintenance of a differentiated configuration of the cytoskeleton. The price for this mechanism of assembly is severe restriction of time and location of protein assembly; these restrictions do not apply in post-translational assembly. The two processes may interact. Nucleation is the rate-limiting step for vimentin assembly and is more sensitive to protein concentration than is elongation (Stewart, 1993). If cotranslational assembly contributes a significant fraction of nuclei, it could influence the organization of the cytoskeleton out of proportion to its fractional contribution to assembly. Perhaps it is the balance of post-translational assembly, responsive to the environment of the moment, and cotranslational assembly, providing structural memory, that permits the cytoskeleton to display its wonderful variety and functional adaptations.