INTRODUCTION

Neurofilaments function in axons by providing mechanical stability to these long processes and forming a scaffolding upon which other structural elements, including microtubules, microfilaments, mitochondria, and perhaps the cell membrane, are supported (1, 2). Like all intermediate filament proteins, neurofilament (NF)¹ proteins share a conserved sequence responsible for the formation of -helical coiled coil dimers that assemble further into tetramers and then into long filaments of 10-nm diameter. Neurofilaments are unique in being composed of three subunits (NF-H, NF-M, and NF-L) two of which (NF-H and NF-M) have long C-terminal extensions that influence, but are not absolutely required for filament assembly in vitro. On the other hand, at least one of the large subunits must be expressed in vivo to produce filaments (3), and the function of the long extensions appears to be formation of bridges to other NF (4, 5, 6), to mitochondria (7, 8), to microtubules (9), and perhaps to other cellular structures (10, 11).

Phosphorylation of numerous sites on repetitive sequences of NF side arms occurs during NF axonal transport and is locally modified in correlation with increased NF bundle density in Ranvier nodes (12). Hyperphosphorylated NF epitopes are associated with abnormal accumulation of NF in pathological situations (13, 14). How such phosphorylations affect interactions of neurofilaments with each other or with other molecules is not known. The hypothesis that transient interactions between NF and/or between NF and other subcellular elements such as microtubules (MT) participate in the slow transport of polymerized NF is suggested by the local accumulation of axonal NF after their separation from MT following iminodipropionitrile intoxication (15). This alteration of NF transport may result from the disruption by the drug of the MT-NF cross-bridges (16). The impairment of NF axonal transport in mice expressing a transgene bearing -galactosidase at the end of the NF-H carboxyl-terminal side arm also suggests the requirement for labile interactions between NF in the active mechanism of their export from the perinuclear cytoplasm. The presence of active -galactosidase in NF networks is likely to stabilize cross-links between NF polymers mediated by -galactosidase tetramerization (17).

One of the consequences of interactions between side arms of long filamentous polymers is that they alter the mechanical properties of networks formed by such filaments. Previous reports showed that purified NF form highly viscous gels (18) which are influenced by the phosphorylation level of NF subunits (19, 20). Furthermore, the extent of NF gelation is affected in vitro by iminodipropionitrile (21) and Al³⁺ ions (22), two compounds that induce NF accumulation in animal models of neuropathies (15, 23). Cross-bridges between purified NF in vitro exhibit the same morphological characteristics as those measured in situ (3, 6), suggesting that similar domains of the NF-H and NF-M subunits are involved in both situations.

We analyzed in this work the structure and viscoelastic, or rheologic, properties of NF networks using a variety of rheometric and optical methods. The structures of single NF in solution and within a gel were determined by electron microscopy and by video-enhanced fluorescence microscopy of labeled NF protein. To distinguish the specific effects of NF cross-bridges from viscoelastic features common to IF, similar experiments were also done with vimentin, an IF protein that lacks long C-terminal extensions. To assess the possibility that cross-bridge formation might be modulated in situ, effects of various potential modulators such as phosphorylation levels, inositol lipids, and actin filaments were also examined.

EXPERIMENTAL PROCEDURES

Bovine spinal cords were obtained from a local slaughterhouse. NF were purified according to Leterrier and Eyer (18) with some modifications; the crude NF pellet was purified by sedimentation at 200,000 × g for 3 h 30 min at 4 °C onto 5 ml of 1.5 M sucrose in buffer A (Mes 0.1 M, MgCl₂ 1 mM, EGTA 1 mM, pH 6.8) and 15 ml of 0.8 M sucrose in buffer A. This step recovers NF as a pellet, removing minor contaminants in the sucrose layer. Purified NF resuspended in buffer A containing 0.8 M sucrose were dialyzed against the same buffer at 4 °C (24 h). Dialyzed NF were gently homogenized with a Teflon-glass potter, and 1-ml aliquots were frozen in liquid nitrogen for storage at 135 °C. Samples were slowly thawed on ice before measurements of rheology at 24 °C of 3 mg of NF/ml in the presence of 5 mM MgCl₂ (see below: precautions for reproducible measurements) and a mixture of protease inhibitors at the final concentrations: N-p-tosyl-L-arginine methyl ester 0.1 mg/ml; aprotinin 0.05 unit/ml; pepstatin A 1 µM; leupeptin 1 µM; phenylmethylsulfonyl fluoride 1 mM; chloroquin 0.1 mM; soybean trypsin inhibitor 10 nM. Phosphatidylinositol 4,5-bisphosphate (PIP₂), protease inhibitors, and acid phosphatase (type II from potato; 0.93 mg/ml, 88 units/mg) were from Sigma.

Vimentin, purified according to Nelson and Traub (24), was a gift from Peter Traub and Manfred Schliwa. 3 mg/ml vimentin was dissolved in 10 mM Tris, pH 7.6, 6 mM dithiothreitol, and 6 M urea, and dialyzed overnight against the same buffer lacking urea. Vimentin polymerization was initiated by adding 150 mM KCl and protease inhibitors before incubation.

Actin was purified from rabbit skeletal muscle (25) and polymerized in 2 mM MgCl₂, 150 mM KCl, 0.2 mM CaCl₂, 0.2 mM dithiothreitol, 0.5 mM ATP, 2 mM Tris, pH 7.4. Gelsolin was obtained according to Kurokawa et al. (26).

Most viscoelastic measurements were made with a torsion pendulum (27) by measuring free oscillations after imposition of a momentary displacement. Frequency-dependent oscillatory measurements were made with a Rheometrics (Piscataway NJ) RFSII instrument (28). Dynamic measurements with the pendulum were made by applying a momentary impulse causing a low strain (<5%). The resonance frequency of oscillations of the sample after its initial perturbation allowed the determination of G. This measurement was made repeatedly on samples at rest during the time course of the gel formation. Alternatively, after gel formation was largely complete, G was measured from oscillations in samples that were deformed under a constant stress. In these cases the deformation (strain) induced by the constant stress was usually much larger than the additional strain during oscillations, and therefore such measurements reflect the viscoelastic resistance of the sample in a deformed state. A third condition was the application of a constant stress to gelled samples followed by the measurement of strain as a function of time.

The formation of NF gels in vitro is highly dependent on the conformation of the polymers (18). Several experimental conditions must be carefully controlled to form NF gels with reproducible viscoelastic properties. The homogenization forces must be standardized to solubilize the initial material but avoid breaking long NF polymers into fragments. Such fragments will not assemble from the first soluble extract into a sedimentable pellet after incubation with glycerol (see purification procedure) and will therefore be lost. On the other hand, the NF will remain with the membranous pellet if the homogenization strength is too low and the original NF network is not disrupted. An efficient procedure is to use an Omnimixer (for bovine spinal cords) at the lower speed (chop), five times for 15 s, at 4 °C, until the homogenate is of a milk shake appearance, and noticeably more viscous than the buffer (this latter indication is of major importance for the quality of the NF preparation, since viscosity is related to the average length of NF pieces).

The final NF preparation (after dialysis overnight at 4 °C against buffer A, 0.8 M sucrose) should be resuspended with a hand-driven Teflon-glass homogenizer before storage. Freezing and storage of NF samples should be done in liquid nitrogen. Some loss of the NF gelation rate occurs after storage (10-20%), although it remains constant for over a year. Samples kept at 80 °C do not exhibit the same stability after prolonged storage.

Frozen samples must be thawed by leaving the tube on ice, without any perturbation of the sample before the complete disappearance of frozen pieces. Then, the NF suspension should be stirred on a vortex mixer at maximal speed for a controlled time: 3 times 20 s with standing in ice for 30 s. between stirring. This step disrupts any NF aggregate already formed during storage (NF interactions occur slowly in cold) and provides homogenous NF suspensions in which no gelation seed (small NF aggregate) remains. The same vigorous stirring should be applied to all NF samples mixed with other components, in order to allow the reproducibility of measurements. Avoiding any of these specific conditions will result in uncontrolled NF samples in which either no viscosity changes occur, or variable results are obtained from apparently identical samples.

The nature and pH of buffers can also affect NF gelation; Tris buffers should be systematically avoided. The optimum pH for gelation is 6.5, and no gel is obtained at pH above 7.5 or under 5. Furthermore, monovalent cations inhibit gelation, which is instead stimulated by divalent cations (18). Attempts to purify rapidly crude NF from contaminant proteins by high salt (such as 1 M KCl) result in variable NF preparations which, under current incubation conditions, will remain highly fluid for very long times before sudden transformation into a nearly solid gel.² The presence of either glycerol or sucrose, although not strictly necessary for interactions between NF, allows the stable conformation of the polymers during long incubation periods. In our hands, the best buffer for protecting gelation properties of purified NF was buffer A made in 100% D₂O, containing 0.5 M sucrose. Among sugars, sucrose was the most convenient, although a slightly different behavior of NF occurs when using a variety of other sugar species, glucose being the closest to sucrose in protecting NF gelation capacity under long time of incubation.² After the long dialysis at 24 °C needed for substantial NF dephosphorylation by phosphatases (19), the gelation behavior of NF is slightly modified in control samples (loss of some of the gelation capacity), suggesting that minor conformational changes of the polymers occur during dialysis at 24 °C. The presence of either glycerol or sucrose is an absolute requirement for protecting NF in these conditions.

NF gels in buffer A were fixed for 10 min with 1% glutaraldehyde in the same buffer. Fixed samples were extensively washed with distilled water to remove buffer. Platinum/carbon replicas of rapidly frozen, freeze-dried, and fractured samples were obtained as described (29).

Rhodamine N-hydroxysuccinimide ester was prepared from rhodamine B and N-hydroxysuccinimide by a standard dicyclohexylcarbodiimide coupling procedure. The three reactants were incubated at a ratio of 1:1.2:1.2 in methylene chloride for 1 h at O °C, and 20 h at 20 °C. Dicyclohexylurea was filtered off, and the residue was triturated with hexane. The crude product was recrystallized from isopropanol/ether. Thin layer chromatography was performed on Merck Alufolien and gave R_F = 0.565 (n-BuOH:AcOH:water, 4:1:1).

NF proteins were labeled by incubating 1.6 mg/ml NF (in buffer A containing 0.4 M sucrose) for 30 min at 24 °C in the presence of a 1000 times excess molar ratio of succinimidyl rhodamine B (estimated NF molecular mass 10³ kDa for a 2/3/7 molar ratio of NF-H, NF-M, and NF-L). Labeled NF were recovered by centrifugation at 332,000 × g for 1 h at 4 °C on a 0.8 M sucrose layer in buffer A. Pellets were resuspended in the same buffer, and aggregates were removed by centrifugation for 2 min at 15,000 × g. Observations of labeled NF were done on diluted samples (10 µg/ml) alone or mixed with 3 mg/ml unlabeled NF, using uncoated glass slides or slides precoated with 5 mg/ml casein in buffer A, sucrose 0.8 M, and washed with the same buffer.

RESULTS

Comparison of Dynamic and Static Viscoelastic Properties of NF and Vimentin Gels

When gelation of neurofilaments was initiated by addition of 6 mM MgCl₂, the elastic storage shear modulus G rose for 2 h following an apparent lag of a few minutes. This parameter is closely related to the extent of network formation in a polymerized system (28). Fig. 1A shows that the time course of increase in G for a NF gel is complex. A relatively slow increase in elasticity is followed by a more rapid rise that occurs approximately 90 min after addition of MgCl₂. Similar results were obtained from more than 10 similar experiments. The structural change responsible for the increased elasticity is not polymerization of NF de novo since the intensity of light scattered from similarly treated samples did not change over this time (data not shown).

Fig. 1B shows a typical free oscillation of a NF gel measured 50 min after warming a sample in a torsion pendulum. The good fit of a damped sinusoidal function to the data allows the determination of the shear moduli, and the relatively low degree of damping implies that the ratio of the loss modulus (G") to the elastic storage modulus (G) is small.

Fig. 1C shows G and G" during NF gelation as measured in a Rheometrics RFS instrument by forced oscillations. The absolute value of G and its rate of increase are similar to those measured in the torsion pendulum, and the low value of G" relative to G, especially at longer times is consistent with the low damping shown in Fig. 1B and with the formation of a strong elastic network.

The frequency dependence of NF viscoelasticity is shown in Fig. 2. The values of G and G" are nearly constant over a range of deformation frequencies from 0.1 to 100 rad/s, and G  G" over the entire frequency range, as expected for a material with little viscous loss. The broad plateau in shear moduli demonstrates that very few free molecular motions occur in NF gels that can dissipate mechanical energy within 0.01-10 s. These data are consistent with either a network containing very long filaments that cannot diffuse to relieve stress within this time or a network of filaments linked together by cross-links with average lifetimes greater than 10 s.

Another aspect of the time-dependent viscoelasticity of NF gels is shown in Fig. 3 which depicts how the stress induced by a sudden strain to 5% relaxes in measurements made on the same sample at two different times after gel formation. When such a measurement is done 100 and 200 min after addition of MgCl₂, the initial static shear moduli G(t) are 70 and 200 Pa, respectively, and consistent with the dynamic shear moduli shown in Fig. 1C. Despite the difference in magnitudes of G at these two times, the rates of stress relaxation are nearly the same, and are very nearly fit by a single exponential function that appears to decay to zero at very long, experimentally unattainable times. These results suggest that a single type of molecular response dominates the stress relaxation, and that the magnitude but not the time constant of this response changes as the overall gel strength increases. All of these features suggest that the shear modulus increases because the number of transient but long-lived interfilament contacts, or cross-links, increases, rather than that the modulus increases because of changes in filament length, concentration, or overall geometry.

A common characteristic of biopolymer gels is the strong dependence of their viscoelastic properties on the amount by which they are deformed (30, 31). In order to compare the mechanical properties of vimentin and NF gels, increasing degrees of steady stress (resulting in a relatively constant strain) were applied to both samples after the same incubation time. At stresses below 40 Pa, causing strains near or below 100%, a large, strain-dependent increase in G (strain hardening) occurs for both vimentin and NF gels (Fig. 4A), consistent with previous measurements of vimentin (31). However, the vimentin gel ceases to harden and abruptly ruptures at stresses above 50 Pa, whereas the NF gel continues to increase resistance to stresses greater than 200 Pa. The initial increase of G with strain is predicted to be a function of semiflexible filament networks, as increasing strain begins to pull out the slack of filaments between points of cross-linking or entanglement (32), and the similar responses of NF and vimentin at low strains suggests that the basic intermediate filament structure of these proteins is similar. However, the much greater resistance of NF gels to high stresses suggests the dominant contribution of specific cross-links between NF that vimentin strands lack. The increased resistance of NF gels to increasing stresses is also shown in Fig. 4B, which depicts the resistance of a NF gel to the increasing strains applied by the Rheometrics device at a constant shear rate of 0.05/s. At strains below 4, the stress continuously increases, as expected for an elastic material, and the upward curvature of the plot confirms the strain hardening property of these networks, since the shear modulus is approximately the slope of this curve. However, once a critical strain above 5 is imposed, there is an abrupt fall in the resistance to additional strain, and the stress at constant shear rate declines to a stable value, characteristic of viscous flow, suggesting that either filament cross-links or the filaments themselves have broken.

Differences between NF and vimentin gels are also evident from their slow deformations under steady shear stress (Fig. 5). Immediately after imposition of a small stress both types of gels deform to the same extent, but the NF gel quickly reaches a steady level of deformation that remains constant for a period of minutes. In contrast, the vimentin network continues to deform (creep) without limit, as reported earlier (31). This difference again suggests that cross-links between network strands dominate the rheology of NF gels, whereas the rheology of vimentin is consistent with that of a network of long intertwined filaments that have no permanent connections.

Dynamic Light Scattering of Vimentin and NF

Dynamic light scattering provides information about the intramolecular motions of long overlapping polymers, and this technique has been used to measure the bending stiffness of actin filaments in solution (28, 33). Fig. 6 shows a qualitative comparison of intensity autocorrelation functions obtained from 0.2 mg/ml concentrations of vimentin, neurofilaments, actin filaments, and taxol-stabilized microtubules. This concentration is low enough so that NF gels containing large bundles do not form, and the scattering is presumably dominated by the thermal motions of individual polymers. Since dynamic light scattering probes molecular motions on a scale (~10-100 nm) that is much smaller than the interfilament distance, these measurements are relatively insensitive to filament length or to the presence of interfilament cross-links. This comparison suggests that, despite their larger diameters, both vimentin and neurofilaments are more flexible than F-actin, and much more flexible than microtubules. Neurofilaments appear to be slightly stiffer than vimentin filaments, but an exact comparison is complicated by that the fact that these filament types require different solution conditions for optimal polymer formation. Preliminary quantitative measurements of vimentin networks suggest that they are approximately 10 times more flexible than actin filaments.³

Visualization of Fluorescently Labeled NF

Single rhodamine-labeled neurofilaments adherent on a glass surface appear as thin, relatively straight filaments with lengths as large as 20 µm (Fig. 7). When these filaments are imaged in solution, before adhering to the surface of the microscope cover glass, only diffuse spherical images are seen, suggesting that the single filaments in solution are so flexible that they curl into coils whose structure is not resolved by light microscopy or that intrafilament cross-links constrain them to a compact shape. In contrast, single actin filaments and microtubules in solution are easily resolved by the same methods as stiff or semiflexible rods, consistent with their greater stiffness (34) (data not shown). The straight images of neurofilaments seen when they adhere on a glass surface, or when a trace of fluorescent filaments co-assemble with an excess of unlabeled neurofilaments to form bundles, appear to require an unfolding of the inherently flexible filament onto the surface, and perhaps stabilization by inter-NF cross-bridging.

Electron Microscopy of Gelled NF

Replicas of NF gels obtained from samples incubated for 90 min at 24 °C showed two types of filament arrangements. Very long thick bundles of dense parallel arrays of NF running throughout the sample are connected to less dense NF domains in which NF polymers are linked together in a spider-web fashion (Fig. 8). Such bundling was not observed in nongelled NF, in which unoriented and unaligned wavy NF predominate (35).

Regulation of NF Gels in Vitro

Previous experiments have shown that dephosphorylation on specific sites of NF subunits affects NF interactions in opposite manners, depending on the accessibility of the phosphorylated sites to phosphatases (19). The effect of adding acid phosphatase to begin NF dephosphorylation at the same time that NF gelation is initiated by magnesium was evaluated by measuring NF rheology (Fig. 9). Dephosphorylation of easily accessible phosphorylated sites of NF subunits eliminates the lag phase before a significant elastic modulus is measured and induces an approximately 2-fold stimulation of the gelation reaction (Fig. 9). This observation confirms previous findings (19) based on viscosity measurements.

The possibility that NF gelation may be affected by other cellular elements was also investigated by rheologic assays. Motivated by evidence that phosphatidylinositol and its phosphorylated derivatives interact with several cytoskeletal proteins, including microtubule-associated proteins (36), we analyzed the influence of PIP₂ on NF gelation. In Fig. 10A are shown recordings of G as a function of time during NF gelation in the presence of micromolar concentrations of PIP₂. A mild increase in G was observed at 2 µM PIP₂, while 10 µM PIP₂ induced a 200% increase in G after 80 min (Fig. 10A). The hardening of NF gels at increasing stresses was also modified by both concentrations of PIP₂ (Fig. 10B). The strain-dependent increase in G in the NF control occurred similarly as in all other samples tested (Fig. 4) with the rupture of gels occurring in the same range of stress (150-220 Pa). However, NF gels formed in the presence of PIP₂ exhibited less strain hardening and broke at stress values larger than control NF (Fig. 10B). Since the ionic strength of the solution is relatively high (>100 mM), these observations are not likely to be simply due to electrostatic interactions with the acidic lipid micelles, but rather suggest structural modifications of NF gels induced by PIP₂ in the micromolar concentration range.

Since NF in vivo are found in regions containing other cytoskeletal elements, we examined the interaction of NF with actin filaments in vitro. NF gel formation was measured in the absence or presence of 2 mg/ml F-actin (average length of 20 µm) or F-actin that was preincubated with the actin filament severing protein gelsolin in a molar ratio 1/200 (average length 0.7 µm) (28). In Fig. 11A are shown the increase of G with incubation time in these mixtures. When the NF gel forms in the presence of long actin filaments, a relatively high value of G is measured at the earliest time point, due presumably to the viscoelasticity of the already formed actin network. The further rise in G parallels that of the NF gel formed in the absence of F-actin. However, the presence of gelsolin, which shortens the filaments to 0.7 µm inhibited the formation of a strong viscoelastic network (Fig. 10A). The strain hardening of the same samples after incubation for over 1 h (gel state in NF alone) showed a large influence of actin filaments on the resistance to increasing stress (Fig. 10B). The reduction of NF resistance to stress and the lower strain hardening of samples was greater in gelsolin-F-actin containing gels than in samples containing long F-actin filaments (Fig. 10B), possibly because the long actin filaments provide an additional, NF-independent, resistance to deformation at relatively small strains. When the actin filaments are too short to form an elastic network themselves, as is the case with gelsolin, the main affect of the short actin filaments is to weaken the NF gel to a state similar to that observed with vimentin in Fig. 4A. These results suggest that actin filaments may diminish the cross-links formed between neurofilaments.

DISCUSSION

The first aim of this work was to characterize quantitatively the viscoelasticity of gels formed by native NF suspensions (18). Previous measurements of viscosity changes in NF suspensions with a falling ball apparatus revealed that a gel was formed, but further measurements of the mechanical properties at steady state were not possible (18). On the other hand, the viscosity values obtained by this method reflect the initial cross-bridging density between filaments, and thus represent a convenient procedure for describing the conditions allowing NF gelation (18) as well as its regulation by the phosphorylation level of NF subunits (19, 20) or by neurotoxic drugs (21, 22). The present results allow a direct comparison of rheological parameters of NF gels with those of MT, actin filaments, and vimentin (34, 37).

One goal of the present study was to define the contribution of NF side arms to the physical characteristics of NF gels. The unusual molecular composition of NF polymers, with the periodic protrusion of large lateral extensions of the carboxyl-terminal domains of the two high molecular weight subunits NF-H and NF-M (5, 38, 39), suggests a specific contribution of these projection domains to the physical properties of NF and NF networks. In addition, ultrastructural studies suggest that NF interact physically with adjacent structures of the neuronal cytoplasm, such as other NF, MT, actin filaments, and membranous organelles (4, 40).

Viscoelastic measurements confirm the hypothesis that NF projections link individual filaments together, and that one consequence of this linkage is a large increase in the elastic strength of neurofilament networks. Intermediate filaments in general resist larger deformations compared to purified microtubules or actin filaments. However, NF gels, even when compared with vimentin have larger elastic moduli and are able to resist much larger stresses (Figs. 4 and 5). The difference appears to be due to the cross-bridges formed by the projections from NF-M and NF-H subunits.

Comparative dynamic light scattering studies of NF, vimentin, MT, and actin filaments demonstrate that the two types of IF are more flexible polymers than either MT or F-actin (Fig. 6). This flexibility was also evident from fluorescence imaging of labeled NF in suspension which appeared as coiled structures that unraveled into long filaments when adhering to the glass surface or incorporating into NF bundles (Fig. 7). This bundle formation was confirmed by electron microscopy of gelled NF (Fig. 8). Previous studies showing that birefringent formations appeared progressively during NF gelation (18) also suggested that interactions between NF result in alignment of filaments in the gelled sample.

These findings support the hypothesis that interactions between NF, likely through charged domains, promote the formation of long bundles of aligned filaments in vitro, similar to the organization of interconnected NF networks in situ (4). The organization of long bundles of parallel NF in gelled samples (Figs. 7 and 8) in vitro is comparable with the organization of NF in parallel bundles found in Sf9 cells transfected with NF-L + NF-M molecules (3), specifically in cells where mutated NF-M subunits can form cross-bridges, in contrast to cells containing exclusively the transfected NF-L molecule alone or with NF-M mutants that are unable to cross-bridge (3). Although cross-bridges between NF are difficult to see in our samples, as a possible consequence of the density of NF bundles in vitro (Fig. 8), the similar longitudinal organization of NF bundles suggests common mechanisms of alignment caused by cross-bridging in both situations (Fig. 8) (3). Thus, our observations support the hypothesis raised by Nagakawa et al. (3) that cross-bridging between NF is a requirement for their bundling in parallel arrays.

The rigidity and stability of NF gels formed in vitro demonstrate the potential of cross-linking to alter the mechanical properties of NF networks, but these features are likely to be modulated by several factors in vivo. In the present study, the properties of NF gels were altered by either PIP₂ or acid phosphatase treatment. These results favor the hypothesis that PIP₂-dependent conformational changes of NF allow the formation of gels that are more flexible than those obtained with NF alone (Fig. 10B). At present the way in which PIP₂ micelles alter NF structure is unknown. The possibility of interactions between PIP₂ and NF was raised from the observation that NF are interconnected in situ with a number of membranous compartments (4), and from the increasing number of experimental data suggesting a direct interaction between IF and lipids (41, 42, 43). PIP₂ and other lipids might therefore control NF bundles in situ in a manner similar to the postulated role of NF phosphorylation previously investigated (19).

In a similar perspective of putative cross-talk between NF networks and the plasma membrane, the interactions between NF and actin filaments in vitro suggests a possible link between the most stable intraneuronal cytoskeletal network (NF) and the more dynamic actin filaments that line the plasma membrane (44). Based on the effects of actin filaments on both the gelation kinetics of NF and the resistance of NF gels to stress, it appears that actin filaments inhibit the NF bundling process. This inhibition could occur either by specific binding of actin to NF or by steric interactions between the two types of filaments that inhibit the lateral alignment of NF necessary for efficient cross-bridging. There are as yet no data demonstrating a specific binding site for actin on NF, but such a site has been reported for vimentin (45).

Connections between NF and actin filaments are of interest with regard to the organization of the axonal cytoskeleton. NF are interconnected with MT and membranous organelles in axons, but neither of these two polymers interact directly with the cell plasma membrane (4). In contrast, actin is present in axons both as a cortical submembranous network and as part of the cytoplasmic matrix (40, 44, 46). The present findings of direct effects of PIP₂ and actin filaments on NF networks in vitro, with opposite effects of these two components on the structure of NF gels, are strong indications for the existence of an intricate, metabolically sensitive network combining the three major neuronal polymers, MT, NF, and actin filaments, in which each component can alter the physical properties of the composite structure. The possibility that anchoring sites for actin filaments might exist on NF also supports the hypothesis of a continuous physical linkage among MT, NF bundles, and the plasma membrane, thus leading toward future investigations of a direct interdependence between plasma membrane surface topography and the organization of the cytoskeleton as an integrated organelle.

The present study demonstrates that NF gels formed in vitro are more resistant to stress than any other noncovalent gel of biological polymers previously studied (34). However, NF in vivo are thought to be weakly cross-linked (47). The differences in these results and the modulation of NF rheology in vitro by agents such as acid phosphatase, PIP₂, and F-actin support the hypothesis that NF cross-bridging is a NF-specific property that is strongly controlled in situ. These data also address issues concerning models for how NF interact with each other.

With regard to the hypothesis that no direct cross-bridging of pure NF occurs because phosphorylated side arms of adjacent NF should repel each other (48), the present data demonstrate that purified NF do interact strongly in vitro and that the specific mechanical properties of NF networks are the result of a cross-bridging mechanism that occurs slowly in the course of gelation (Fig. 3). Cross-bridges between pure NF in vitro were analyzed by electron microscopy (6). Furthermore, Gotow and Tanaka (49) and Gotow et al. (50) established a direct relationship between the phosphorylated state of NF-H and NF-M subunits and the extent of cross-bridging between NF in situ and in vitro. Modification of the phosphorylation level of the numerous sites of NF-H and NF-M side arms in vitro and in vivo modulates the extent of NF gelation in vitro (Fig. 9) (19, 20). Thus, the high phosphorylation level of NF side arms does not induce repulsion between NF either in vivo or in vitro, but, instead can promote interactions between NF in both situations and may be a major element of their regulation.

The direct contribution of NF side arms to NF bundles in situ has been recently analyzed (3). This study suggested that NF-M side arms participate in both NF polymer elongation and inter-NF cross-bridging. The use of mutated NF-M molecules further demonstrated that the core domain of the NF-M side arm containing the putative phosphorylation sites is required for the formation of cross-bridges (3). These authors proposed that NF bundle assembly occurs as a two-step process of copolymerization of NF subunits followed by establishment of cross-bridges. This hypothesis is consistent with our data. First, the evidence that single NF unraveled from nearly spherical structures into straight filaments during the gelation process (Fig. 7) suggests that bundling of NF occurs after their polymerization. Second, the present data and previous reports (18, 19, 20, 21, 22) describing the modulation of NF gelation in vitro by various agents (18), support strongly the hypothesis that the NF cross-bridging mechanism is regulated independently of NF subunit polymerization.

The self-organization of highly flexible NF into bundles mediated by specific interactions between charged domains of NF might be of biological relevance since NF are organized in parallel arrays in axons (4). The bundling of NF in situ occurs in the proximal domain of the axon, in parallel with a change in the phosphorylation level of NF subunits and the establishment of a dense cross-bridging between NF (51). Such a spatial pattern is in good agreement with the observation that interactions between highly flexible (phosphorylated) NF are required for their alignment into long bundles in vitro (Figs. 6 and 7).

The present work (Fig. 11) also demonstrates that the NF-subunit-specific cross-bridging mechanism, dependent on a fragile NF conformation (18), is strongly controlled by other cytoplasmic proteins in vitro. Several such potential modulators bind to NF in situ; MAPs, tubulin (18), and actin (Fig. 11) interfere with NF gelation in vitro, as do other soluble proteins in the tissue extract (7).⁴ These findings strengthen the hypothesis that NF organization in vivo involves the control of NF cross-bridging by several proteins, allowing the local regulation between interconnected and independent NF. Alterations of such a balance (by changes in the phosphorylation level of NF subunits or/and in the interactions between NF and other cytoskeletal elements) might induce local accumulations of NF in pathological situations, as suggested by the behavior in vitro of hyperphosphorylated NF from aging rats (20).