Norman Y. Yao§*, Yi-Chia Lin*, Chase P. Broedersz?, Karen E. Kasza, Frederick C. MacKintosh?, and David A. Weitz*‡¶

§Harvard College 2008; *Department of Physics, Harvard University, Cambridge, MA 02138, USA;Department of Physics and Astronomy, Vrije Universiteit, 1081HV Amsterdam, The Netherlands;School of Engineering and Applied Sciences, Harvard University, Cambridge, MA 02138, USA.

Neurofilaments are found in abundance in the cytoskeleton of neurons, where they act as an intracellular framework protecting the neuron from external stresses. To elucidate the nature of the mechanical properties that provide this protection, we measure the linear and nonlinear viscoelastic properties of networks of neurofilaments. These networks are soft solids that exhibit dramatic strain stiffening above critical strains of 30–70%. Surprisingly, divalent ions, such as Mg2+, Ca2+, and Zn2+ act as effective cross-linkers for neurofilament networks, controlling their solid-like elastic response. This behavior is comparable to that of actin-binding proteins in reconstituted filamentous actin. We show that the elasticity of neurofilament networks is entropic in origin and is consistent with a model for cross-linked semiflexible networks, which we use to quantify the cross-linking by divalent ions.


The mechanical and functional properties of cells depend largely on their cytoskeleton, which is comprised of networks of biopolymers; these include microtubules, actin, and intermediate filaments. A complex interplay of the mechanics of these networks provides cytoskeletal structure with the relative importance of the individual networks depending strongly on the type of cell [1]. The complexity of the intermingled structure and the mechanical behavior of these networks in vivo has led to extensive in vitro studies of networks of individual biopolymers. Many of these studies have focused on reconstituted networks of filamentous actin (F-actin) which dominates the mechanics of the cytoskeleton of many cells [2-7]. However, intermediate filaments also form an important network in the cytoskeleton of many cells; moreover, in some cells they form the most important network. For example, in mature axons, neurofilaments, a type IV intermediate filament, are the most abundant cytoskeletal element overwhelming the amount of actin and outnumbering microtubules by more than an order of magnitude [8]. Neurofilaments (NF) are assembled from three polypeptide sub-units NF-Light (NF-L), NF-Medium (NF-M), and NF-Heavy (NF-H), with molecular masses of 68 kDa, 150 kDa and 200 kDa, respectively [8]. They have a diameter d ~ 10 nm, a persistence length believed to be of order lp ~ 0.2 µm and an in vitro contour length L ~ 5 µm. They share a conserved sequence with all other intermediate filaments, which is responsible for the formation of coiled dimers that eventually assemble into tetramers and finally into filaments. Unlike other intermediate filaments such as vimentin and desmin, neurofilaments have long carboxy terminal extensions that protrude from the filament backbone [9]. These highly charged “side-arms” lead to significant interactions among individual filaments as well as between filaments and ions [10]. Although the interaction of divalent ions and rigid polymers has been previously examined, little is known about the electrostatic cross-linking mechanism [11]. Networks of neurofilaments are weakly elastic; however, these networks are able to withstand large strains and exhibit pronounced stiffening with increasing strain [12, 13]. An understanding of the underlying origin of this elastic behavior remains elusive; in particular, even the nature of the cross-linkers, which must be present in such a network, is not known. Further, recent findings have shown that NF aggregation and increased network stiffness are common in patients with amyotrophic lateral sclerosis (ALS) and Parkinson’s. Thus, an understanding of the fundamental mechanical properties of these networks of neurofilaments is an essential first step in elucidating the role of neurofilaments in a multitude of diseases [14]. However, the elastic behavior of these networks has not as yet been systematically studied.

Here, we report the linear and nonlinear viscoelastic properties of networks of neurofilaments. We show that these networks form cross-linked gels; the cross-linking is governed by divalent ions such as Mg2+ at millimolar concentrations. To explain the origins of the network’s elasticity, we apply a semiflexible polymer model, which ascribes the network elasticity to the stretching of thermal fluctuations; this quantitatively accounts for the linear and nonlinear elasticity of neurofilament networks, and ultimately, even allows us to extract microstructural network parameters such as the persistence length and the average distance between cross-links directly from bulk rheology.

Materials and Methods


Neurofilaments are purified from bovine spinal cords using a standard procedure [9, 15, 16]. The fresh tissue is homogenized in the presence of buffer A (Mes 0.1 M, MgCl2 1 mM, EGTA 1 mM, pH 6.8) and then centrifuged at a K-factor of 298.8 (Beckman 70 Ti). The crude neurofilament pellet is purified overnight on a discontinuous sucrose g radient with 0.8 M sucrose (5.9 ml), 1.5 M sucrose (1.3 ml) and 2.0 M sucrose (1.0 ml). After overnight sedimentation, the concentration of the purified neurofilament is determined with a B radford Assay using bovine serum albumen (BSA) as a standard. The purified neurofilament is dialyzed against buffer A containing 0.8 M sucrose for 76 hours and then 120 μl aliquots are flash frozen in liquid nitrogen and stored at -80 °C.

Bulk Rheology

The mechanical response of the cross-linked neurofilament networks is measured with a stress-controlled rheometer (HR Nano, Bohlin Instruments) using a 20 mm diameter 2 degree stainless steel cone plate geometry and a gap size of 50 μm. Before rheological testing, the neurofilament samples are thawed on ice, after which they are quickly pipetted onto the stainless steel bottom plate of the rheometer in the presence of varying concentrations of Mg2+. We utilize a solvent trap to prevent our networks from drying. To measure the linear viscoelastic moduli, we apply an oscillatory stress of the form σ(t) = A sin(ωt), where A is the amplitude of the stress and ω is the frequency. The resulting strain is of the form γ(t) = B sin(ωt + φ) and yields the storage modulus and the loss modulus . To determine the frequency dependence of the linear moduli, G'(ω) and G”(ω) are sampled over a range of frequencies from 0.006–25 rad/s. In addition, we probe the stress dependence of the network response by measuring G'(ω) and G”(ω) at a single frequency varying the amplitude of the oscillatory stress. To probe nonlinear behavior, we utilize a differential measurement, an effective probe of the tangent elastic modulus, which for a viscoelastic solid such as neurofilaments provides consistent nonlinear measurements of elasticity in comparison to other nonlinear methods [17‑19]. A small oscillatory stress is superimposed on a steady pre-stress, σ, resulting in a total stress of the form σ(t) = σ + |δσ| sin(ωt). The resultant strain is γ(t) = γ + |δγ| sin(ωt + φ), yielding a differential elastic modulus and a differential viscous modulus [2].

Scaling Parameters

To compare the experiments with theory, we collapse the differential measurements onto a single master curve by scaling the stiffness K’ and stress σ by two free parameters for each data set. According to theory, the stiffness versus stress should have a single, universal form apart from these two scale factors. We determine the scale factors by cubic-spline fitting the data sets to piecewise polynomials; these polynomials are then scaled onto the predicted stiffening curve using a least squares regression.

Results and Discussion

To quantify the mechanical properties of neurofilaments, we probe the linear viscoelastic moduli of the network during gelation, which takes approximately one hour; we characterize this by continuously measuring the linear viscoelastic moduli at a single frequency, ω = 0.6 rad/s. Gelation of these networks is initiated by the addition of millimolar amounts of Mg2+ and during this process we find that the linear viscoelastic moduli increase rapidly before reaching a plateau value. We measure the frequency dependence of the linear viscoelastic moduli over a range of neurofilament and Mg2+ concentrations. To ensure that we are probing the linear response, we maintain a maximum applied stress amplitude below 0.01 Pa, corresponding to strains less than approximately 5%; we find that the linear moduli are frequency independent for all tested frequencies, 0.006–25 rad/s. Additionally, neurofilament networks behave as a viscoelastic solid for all ranges of Mg2+ concentrations tested and the linear storage modulus is always at least an order of magnitude greater than the linear loss modulus, as shown in Fig. 1. This is indicative of a cross-linked gel and allows us to define a plateau elastic modulus G0 [20].

Figure 1
Figure 1. The frequency dependence of the linear viscoelastic moduli of cross-linked networks for a variety of neurofilament and Mg2+ concentrations. A) Variations of the moduli at constant Mg2+ concentration (5 mM) and changing filament concentration B) Variations of the moduli at constant neurofilament concentration (1.5 mg/ml) and changing Mg2+ concentration.

The elasticity of neurofilament networks is highly nonlinear; above critical strains γc of 30–70%, the networks show stiffening up to strains of 300% [21], as shown in Fig. 2. This marked strain-stiffening occurs for a wide variety of Mg2+ and neurofilament concentrations. In addition, by varying the neurofilament concentration cNF and the Mg2+ concentration cMg, we can finely tune the linear storage modulus G0 over a wide range of values, as seen in Fig. 3. The strong dependence of G0 on Mg2+ concentration is reminiscent of actin networks cross-linked with the incompliant cross-linkers such as scruin [2, 22, 23]; this suggests that in the case of neurofilaments, Mg2+ is effectively acting as a cross-linker leading to the formation of a viscoelastic network. Thus, the neurofilaments are cross-linked ionically on length scales comparable to their persistence length; hence, they should behave as semiflexible biopolymer networks. We therefore hypothesize that the network elasticity is due to the stretching out of thermal fluctuations. These thermally driven transverse fluctuations reduce neurofilament extension resulting in an entropic spring. To consider the entropic effects we can model the Mg2+-cross-linked network as a collection of thermally fluctuating semiflexible segments of length lc, where lc is the average distance between Mg2+ cross-links. A convincing test of the hypothesis of entropic elasticity is the nonlinear behavior of the network. When the thermal fluctuations are pulled out by increasing strain, the elastic modulus of the network exhibits a pronounced increase.

Figure 2
Figure 2. The strain-stiffening behavior of neurofilament networks at various Mg2+ and neurofilament concentrations. Close squares represent G', the elastic modulus and open squares represent G'', the viscous modulus. Dramatic nonlinearities are seen at critical strains ranging from 30–70%.
Figure 3
Figure 3. The linear elastic modulus can be finely tuned by varying the concentration of the cross-linker Mg2+ and the neurofilament concentration.

To probe this nonlinear elasticity of neurofilament networks, we measure the differential or tangent elastic modulus K’(σ) at a constant frequency ω = 0.6 rad/s for a variety of neurofilament and Mg2+ concentrations. If the network elasticity is indeed entropic in origin, this can provide a natural explanation for the nonlinear behavior in terms of the nonlinear elastic force-extension response of individual filaments that deform affinely. Here, the force required to extend a single filament diverges as the length approaches the full extension lc, since   [24-26]. Provided the network deformation is affine, its macroscopic shear stress is primarily due to the stretching and compression of the individual elements of the network. The expected divergence of the single-filament tension leads to a scaling of ; we therefore expect a scaling of network stiffness with stress of the form K’(σ) ~ σ3/2 in the highly nonlinear regime [2]. Indeed, ionically cross-linked neurofilament networks show remarkable consistency with this affine thermal model for a wide range of neurofilament and cross-link concentrations, as shown in Fig. 4. This consistency provides convincing evidence for the entropic nature of the network’s nonlinear elasticity [2, 25].

Figure 4
Figure 4. The dependence of K'(σ) on σ for a variety of neurofilament and Mg2+ concentrations. All data show an exponent of approximately 3/2 in agreement with the affine thermal model.

The affine thermal model also suggests that the functional form of the data should be identical for all values of cMg and cNF. To test this, we scale all the data sets for K’(σ) onto a single master curve. This is accomplished by scaling the modulus by a factor G’ and the stress by a factor σ’. Consistent with the theoretical prediction, all the data from various neurofilament and Mg2+ concentrations can indeed be scaled onto a universal curve, as shown in Fig. 5. The scale factor for the modulus is the linear shear modulus G’G0, while the scale factor for the stress is a measure of the critical stress σc at which the network begins to stiffen. This provides additional evidence that the nonlinear elasticity of the Mg2+-cross-linked neurofilament networks is due to the entropy associated with single filament stretching.

Figure 5
Figure 5. Collapse of all data sets of the σ dependence of K' onto a single universal curve. The solid line represents the theoretical prediction of reference 2. The scaling parameters are G0, the linear elastic modulus and σc, the critical stress. These parameters are calculated using a least squares regression.

To explore the generality of this ionic cross-linking behavior, we use other divalent ions including Ca2+ and Zn2+. We find that the effects of both of these ions are nearly identical to those of Mg2+; they also cross-link neurofilament networks into weak elastic gels. This lack of dependence on the specific ionic cross-link lends evidence that the interaction between filaments and ions is electrostatic in nature. This electrostatic interaction would imply that the various ions are acting as salt-bridges, thereby cross-linking filaments into low energy conformations.

The ability to scale all data sets of K'(σ) onto a single universal curve also provides a means to convincingly confirm that the linear elasticity is entropic in origin. To accomplish this, we derive an expression that relates the two scaling parameters to each other. For small extensions δl of the entropic spring, the force required can be derived from the wormlike chain model giving . Assuming an affine deformation, whereby the macroscopic sample strain can be translated into local microscopic deformations, and accounting for an isotropic distribution of filaments, the full expression for the linear elastic modulus of the network is given by


where κ = kBTlp is the bending rigidity of neurofilaments, kBT is the thermal energy, and ρ is the filament-length density [2, 25, 27, 28]. The density ρ is also proportional to the mass density cNF, and is related to the mesh size ζ of the network by [29]. Furthermore, the model predicts a characteristic filament tension proportional to , and a characteristic stress


[2, 22, 25]. Thus, if the network’s linear elasticity is dominated by entropy, we expect the scaling cNF1/2G0 ~ σc3/2 , where the pre-factor should depend only on kBT and lp; although the pre-factor will differ for different types of filaments it should be the same for different networks composed of the same filament type and at the same temperature, such as ours. Thus, plotting cNF1/2G0 as a function of σc for different neurofilament networks at the same temperature should result in collapse of the data onto a single curve characterized a 3/2 power law; this even includes systems with different divalent ions or different ionic concentrations. For a variety of divalent ions, we find that cNF1/2G0 ~ σcz, where z = 1.54 ± 0.14 in excellent agreement with this model, as shown in Fig. 6. It is essential to note that the 3/2 exponent found here is not a direct consequence of the 3/2 exponent obtained in Fig. 5, which characterizes the highly nonlinear regime. Instead, the plot of cNF1/2G0 as a function of σc probes the underlying mechanism and extent of the linear elastic regime.

Figure 6
Figure 6. The dependence of cNF1/2G0 on σc. The solid line is the result of a regression fit to the data and depicts an exponent of 1.54. This is in agreement with the affine thermal model which predicts an exponent of 3/2. Closed squares are data obtained with Mg2+, open squares are data obtained with Ca2+, and crossed squares are data obtained with Zn2+. The inset shows the dependence of G0 on cNF and depicts an exponent of 2.5 obtained from regression. This is also consistent with the affine thermal model which predicts an exponent of 2.2.

For a fixed ratio of cross-links R = cMg/cNF, we expect cross-linking to occur on the scale of the entanglement length, yielding lc cNF2/5 [25, 27, 30]. Thus, we expect the linear storage modulus to scale with neurofilament concentration as G0 cNF11/5 [25]. For R = 1000, we find an approximate scaling of G0 cNF25, consistent with the predicted power law, as shown in the inset of Fig. 6. Interestingly, the stronger concentration dependence of G0 may be a consequence of the dense cross-linking that we observe. Specifically, for densely cross-linked networks, corresponding to a minimum lc on the order of the typical spacing between filaments as we observe here, the model in Eq. (1) predicts G0 cNF25 [25]. The agreement with the affine thermal model in both the linear and nonlinear regimes confirms the existence of an ionically cross-linked neurofilament gel whose elasticity is due to the pulling out of thermal fluctuations.

The ability of the affine thermal model to explain the elasticity of the neurofilament network also suggests that we should be able to quantitatively extract network parameters from the bulk rheology. The model predicts that

Equation 3(3)


Equation 4(4)

where ρ 2.1×1013 m-2 for neurofilament networks at a concentration of 1 mg/mL. This yields a persistence length lp 0.2 µm which is in excellent agreement with previous measurements [31]. In addition, we find that lc 0.3 µm which is close to the theoretical mesh size ≈ 0.26 μm; surprisingly, this is far below the mesh size of 4 µm inferred from tracer particle motion [1]. Such particle tracking only provides an indirect measure: in weakly cross-linked networks, for instance, even particles that are larger than the average inter-filament spacing will tend to diffuse slowly.

To further elucidate the cross-linking behavior of Mg2+, we explore the dependence of lc on both cMg and cNF, based on Eq. (3-4). Based on the form of G0 and σc, we expect that . Assuming that Mg2+ is acting as the cross-linker and that lc is also the typical distance between binary collisions of filament chains we would expect that , where le is the entanglement length. Thus, for a given concentration of neurofilaments

Equation 5(5)

[25]. This yields

Equation 6(6)

where X is the exponent of the Mg2+ concentration. Naively, we would expect that X ≈ 1 which would imply that doubling the concentration of Mg2+ would halve the average distance between cross-links. Empirically we find a much weaker dependence on cMg, where . This weaker dependence suggests that mM concentrations of Mg2+ actually saturate our networks. This is consistent with a calculation of the percentage of Mg2+ ions, which actually act as cross-links. The number of cross-linking ions per cubic meter is ; the number density of ions, N in a standard 5 mM Mg2+ concentration is N ≈ 30 × 1023. Thus, there is an excess of Mg2+ ions available to act as cross-linkers; this may account for the weak cross-link dependence. A similarly weak dependence has been seen previously with actin networks in the presence of the molecular motor heavy meromysin where X was found to be 0.4 and thus, , where cA is the actin concentration and cHMM is the heavy meromysin concentration [32]. Utilizing our empirical power law for cMg, we are able to collapse the curves such that which is in excellent agreement with the predicted exponent 2/5, as shown in Fig. 7. The fact that the cross-linking distance lc scales directly with cMg further confirms the role of Mg2+ as the effective ionic cross-linker of the neurofilament networks. Thus, our findings demonstrate both the entropic origin of neurofilament network’s elasticity as well as the role of Mg2+ as an effective ionic cross-linker.

Figure 7
Figure 7. The dependence of G0/σc×cNF-1/5 on cNF. The solid line is the result of a regression fit and exhibits an exponent of 0.41. This is in agreement with the affine thermal model which predicts an exponent of 2/5. The inset shows the dependence of G0/σc×cNF-2/5 on cMg and depicts an exponent of 0.21 obtained by a regression fit; this empirical power law was used to collapse the data and to obtain the 0.41 exponent for the cNF dependence.


We measure the linear and nonlinear viscoelastic properties of cross-linked neurofilament solutions over a wide range Mg2+ and neurofilament concentrations. Neurofilaments are interesting intermediate filament networks whose nonlinear elasticity has not been studied systematically. We show that the neurofilament networks form densely cross-linked gels, whose elasticity can be well understood within an affine entropic framework. We provide direct quantitative calculations of lp and lc from bulk rheology using this model. Furthermore, our data provides evidence that Mg2+ acts as the effective ionic cross-linker in the neurofilament networks. The weaker than expected dependence that we observe suggests that Mg2+ may be near saturation in our networks. Future experimental work with other multivalent ions is required to better understand the electrostatic interaction between filaments and cross-links; this would lead to a better microscopic understanding of the effects of electrostatic interactions in the cross-linking of neurofilament networks. Moreover, the effect of divalent ions on the cross-linking of networks of other intermediate filaments would also be very interesting to explore.


This work was supported in part by the NSF (DMR-0602684 and CTS-0505929), the Harvard MRSEC (DMR-0213805), and the Stichting voor Fundamenteel Onderzoek der Materie (FOM/NWO).


  1. S. Rammensee, P. A. Janmey, and A. R. Bausch, Eur. Biophys. J. Biophy. 36, 661 (2007).
  2. M. L. Gardel et al., Science 304, 1301 (2004).
  3. B. Hinner et al., Phys. Rev. Lett. 81, 2614 (1998).
  4. R. Tharmann, M. Claessens, and A. R. Bausch, Biophys. J. 90, 2622 (2006).
  5. C. Storm et al., Nature 435, 191 (2005).
  6. J. Y. Xu et al., Biophys. J. 74, 2731 (1998).
  7. J. Kas et al., Biophys. J. 70, 609 (1996).
  8. P. C. Wong et al., J. Cell Biol. 130, 1413 (1995).
  9. J. F. Leterrier et al., J. Biol. Chem. 271, 15687 (1996).
  10. S. Kumar, and J. H. Hoh, Biochem. Biophy. Res. Co. 324, 489 (2004).
  11. G. C. L. Wong, Curr. Opin. Colloid In. 11, 310 (2006).
  12. O. I. Wagner et al., Exp. Cell Res. 313, 2228 (2007).
  13. L. Kreplak et al., J. Mol. Biol. 354, 569 (2005).
  14. S. Kumar et al., Biophys. J. 82, 2360 (2002).
  15. A. Delacourte et al., Biochem. J. 191, 543 (1980).
  16. J. F. Leterrier, and J. Eyer, Biochem. J. 245, 93 (1987).
  17. N. Y. Yao, R. Larsen, and D. A. Weitz, J. Rheol. 52, 13 (2008).
  18. C. Baravian, G. Benbelkacem, and F. Caton, Rheol. Acta 46, 577 (2007).
  19. C. Baravian, and D. Quemada, Rheol. Acta 37, 223 (1998).
  20. M. Rubenstein, and R. Colby, Polymer Physics (Oxford University Press, Oxford, 2004).
  21. D. A. Weitz, and P. A. Janmey, P. Natl. Acad. Sci. USA 105, 1105 (2008).
  22. M. L. Gardel et al., Phys Rev Lett 93 (2004).
  23. A. R. Bausch, and K. Kroy, Nat. Phys. 2, 231 (2006).
  24. C. Bustamante et al., Science 265, 1599 (1994).
  25. F. C. Mackintosh, J. Kas, and P. A. Janmey, Phys. Rev. Lett. 75, 4425 (1995).
  26. M. Fixman, and J. Kovac, J. Chem. Phys. 58, 1564 (1973).
  27. A. N. Semenov, J. Chem. Soc. Faraday T. Ii 82, 317 (1986).
  28. F. Gittes, and F. C. MacKintosh, Phys. Rev. E 58, R1241 (1998).
  29. C. F. Schmidt et al., Macromolecules 22, 3638 (1989).
  30. T. Odijk, Macromolecules 16, 1340 (1983).
  31. Z. Dogic et al., Phys. Rev. Lett. 92 (2004).
  32. R. Tharmann, M. Claessens, and A. R. Bausch, Phys. Rev. Lett. 98 (2007).

Comments Closed